IWCE AbstractsBook Final2015

18th International Workshop on
Computational Electronics
West Lafayette, Indiana
September 2-4, 2015
September 2 - 4, 2015
West Lafayette, IN
18th International Workshop on

Computational Electronics
IWCE 2015
Book of Abstracts
www.nanoHUB.org/groups/IWCE2015
IWCE 2015 hosted by:
Partial sponsorship kindly provided by:

Foreword
Welcome to West Lafayette and Purdue University!
On behalf of the local organizing committee, we would like to welcome you to the 18th International
Workshop on Computational Electronics. Throughout the years, this workshop has served as the leading
international forum of discussion on the current trends and future directions for computational electronics.
This year’s IWCE covers a wide range of topics. Included in this list are the popular areas of Quantum
Transport, Semiclassical Transport, Electronic Structures, and Thermal Transport. Additionally, there is
demonstrated interest in other topics such as Devices and TCAD, Multiscale Simulations, Quantum Dots &
Photonics, and 2D & Graphene Devices.
IWCE 2015 is organized in a single session format and includes eight invited talks, 46 contributed oral
presentations, and 47 poster presentations. We will continue our tradition of very active poster sessions which
have been a very positive trademark of IWCE. All abstracts are collected in this book, printed from camera-
ready manuscripts submitted by the authors. We would like to express our deep appreciation to the authors for
their effort to provide high-quality contributions.
It is our hope that the discussions that take place between both young and experienced researchers are
both open and thought provoking. Finally, we hope that all participants enjoy their stay in West Lafayette.
Gerhard Klimeck
Chair of IWCE 2015

and the Organizing Committee
Committee Memebers
Local Organizing Committee
Gerhard Klimeck (Chair) Purdue University

James Fonseca Purdue University
Tillmann Kubis Purdue University
Michael Povolotskyi Purdue University
Rajib Rahman Purdue University
Technical Committee
M.P. Anantram University of Washington, USA

Marc Bescond University of Aix-Marseille / CNRS, France
Gianluca Fiori Pisa, Italy
Massimo Fischetti University of Texas, Dallas, USA
Neil Goldsman Delaware, USA
Ganesh Hegde Samsung, USA
Karol Kalna Swansea, UK
Robert Kelsall University of Leeds, UK
Neerav Kharche Brookhaven NL, USA
Irena Knezevic University of Wisconsin, Madison, USA
Hans Kosina Technical University of Vienna, Austria
Yiming Li National Chiao Tun University, Taiwan
Massimo Macucci University of Pisa, Italy
Neophytos Neophytou Warwick, UK
Mahesh R. Neupane USA
Branislav Nikolic University of Delaware, USA
Xavier Oriols Universitat Autonoma de Barcelona, Spain
Jerome Saint-Martin University of Paris-Sud, France
Marco Saraniti Arizona State University, USA
Mincheol Shin Korea Advanced Institute of Science and Technology
Muhammad Usman Melbourne, Australia
Zhiping Yu Tsinghua University, China
Cory Weber Intel, USA
Committee Members
Advisory Board Committee
Sheila Aboud Synopsis, USA

Aldo Di Carlo University of Rome “Tor Vergata”, Italy
Philippe Dollfus University of Paris-Sud/CNRS, France
Massimo Fischetti University of Texas, Dallas, USA
Stephen Goodnick Arizona State University, USA
Joseph Jerome Northwestern University, USA
Robert Kelsall University of Leeds, UK
Gerhard Klimeck Purdue University, USA
Irena Knezevic University of Wisconsin, Madison, USA
Paolo Lugli Technical University of Munich, Germany
Massimo Macucci University of Pisa, Italy
Nobuya Mori Osaka University, Japan
Wolfgang Porod University of Notre Dame, USA
Nobuyuki Sano Tsukuba University, Japan
Siegfried Selberherr Technical University of Vienna, Austria
Mincheol Shin Korea Advanced Institute of Science and Technology, Korea
Michael Stroscio University of Illinois-Chicago, USA
Dragica Vasileska Arizona State University, USA
IWCE 2015 Program
September 2
7:00 A.M. Breakfast/Registration South Ballroom, Purdue Memorial Union

8:00 A.M. Welcome and Opening Remarks: Gerhard Klimeck Purdue University

Session 1 - 2D & Graphene
Session Chair: Gianluca Fiori
8:15-8:45 A.M.
Invited Speaker: Mahesh Neupane (U.S. Army Research Laboratory, Aberdeen Proving
Ground, Maryland USA)
Electronic and Vibrational Properties of 2D Materials from Monolayer to Bulk 1
8:45-9:00 A.M.
F. Karimi and I. Knezevic (University of Wisconsin-Madison)
Self-Consistent Field and Master Equation Approach to Calculating the Dielectric Function
of Graphene 3
9:00-9:15 A.M.
J. Li, H. Miranda, Y.M. Niquet, L. Genovese, I. Duchemin, L. Wirtz and C. Delerue (Univ.
Grenoble; University of Luxembourg; UMR CNRS)
The Role of Dimensionality on Phonon-Limited Charge Transport: From CNTs to Graphene 5
9:15-9:30 A.M.
A. Kefayati, M. Pourfath, and H. Kosina (University of Tehran; Institute for Microelectronics,
TU Wien)
A Rigorous Study of Nanoscaled Transistors Based on Single-Layer MoS2 7
9:30-9:45 A.M .
V. Hung Nguyen, H. Viet Nguyen, J. Saint-Martin, and P. Dollfus (Institute of Fundamental
Electronics, Univ. of Paris-Sud; Center for Computational Physics, Institute of Physics, VAST)
Strain Effects on the Electronic Properties of Devices Made of Twisted Graphene Layers 9
9:45-10:00 A.M.
M. L. Van de Put, W. G. Vandenberghe, and M. V. Fischetti (Universiteit Antwerpen; University
of Texas)
Overlaps in Stacked Graphene Flakes Using Empirical Pseudopotentials 11
10:00-10:15 A.M.
Break
Session 2 - Devices & TCAD
Session Chair: Yiming Li
10:15-10:45 A.M.
Invited Speaker: Damien Querlioz (Institut d’Electronique Fondamentale, Univ. Paris-Sud,
CNRS, Orsay, France)
Computational Methods for the Design of Bioinspired Systems that Employ Nanodevices 13
10:45-11:00 A.M.
Á. Papp, G. Csaba, and W. Porod (University of Notre Dame; Pazmany Peter Catholic
University)
Short-Wavelength Spin-Wave Generation by a Microstrip Line 15
11:00-11:15 A.M.
D. Guo, D. Brinkman, T. Fang, R. Akis, I. Sankin, D. Vasileska, and C. Ringhofer (School of
ECE and School of Mathematics, Arizona State University; First Solar)
Diffusion-Reaction Modeling of Cu Migration in CdTe Solar Devices 17
11:15-11:30 A.M.
H. Dey, G. Csaba, G. H. Bernstein, and W. Porod (Center for Nano Science and Technology,
University of Notre Dame)
Oscillations in Exchange-Coupled Nanomagnets 19
11:30-11:45 A.M.
H. Nakamura, I. Rungger, A. Narayan, S. Sanvito, Y. Asai, and J. Tominaga (National Institute of
Advanced Industrial Science and Technology (AIST), Tsukuba Japan; School of Physics, MBER
and CRANN, Trinity College)
Computational Study on Interfacial Phase Change Memory by Topological Superlattices 21
11:45-12:00 P.M.
P. Long, E. Wilson, J. Z. Huang, M. Rodwell, G. Klimeck and M. Povolotskyi (Purdue
Univeristy; University of California Santa Barbara)
Design and Simulation of GaSb/InAs 2D Transmission Enhanced TFET 23
12:00-12:15 P.M.
T. Sadi, L. Wang, L. Gerrer, and A. Asenov (University of Glasgow; Gold Standard Simulations
Ltd)
Self-Consistent Physical Modeling of Silicon-Based Memristor Structures 25
12:15-12:30 P.M.
X. Oriols, Z. Zhen, and E. Colomés, D. Marian (Universitat Autònoma de Barcelona)
Dissipative Quantum Transport Using One-Particle Time-Dependent (Conditional) Wave Functions 27

12:30-1:30 P.M.
Lunch

Session 3 - Electronic Properties
Session Chair: Marc Bescond
1:30-2:00 P.M.
Invited Speaker: Qing Shi (McGill University, Montreal, Canada)
Analyzing Variability in Short-Channel Quantum Transport from Atomistic First Principles 29
2:00-2:15 P.M.
D. Valencia, E.M. Wilson, M. J. W. Rodwell, G. Klimeck, M. Povolotskyi (Network for
Computational Nanotechnology, Purdue University; University of California, Santa Barbara)
Effect of the High-k Dielectric/Semiconductor Interface on Electronic Properties in
Ultra-Thin Channels 31
2:15-2:30 P.M.
F. W. Chen, M. Manfra, G. Klimeck, and T. Kubis (Department of Physics and School of ECE,
Purdue University)
NEMO5: Why Must We Treat Topological Insulator Nanowires Atomically? 33
2:30-2:45 P.M.
J. Ghosh, D. Osintsev, V. Sverdlov, and S. Selberherr (Institute for Microelectronics, TU Wien)
Spin Lifetime Dependence on Valley Splitting in Thin Silicon Films 35
2:45-3:00 P.M.
J. Z. Huang, K. Wang, W. R. Frensley, and G. Klimeck (Network for Computational
Nanotechnology, Purdue University; University of Texas)
Finite Difference Schemes for k • p Models: A Comparative Study 37
3:00-3:15 P.M.
Break
3:15-3:30 P.M.
S. Markov, C. Y. Yam, B. Aradi, G. Penazzi, A. Pecchia, T. Frauenheim, and G. H. Chen (Department
of Chemistry, University of Hong Kong; University of Bremen; University of Rome)
Density Functional Tight Binding Modeling in the Context of Ultra-Thin SOI MOSFETs 39
3:30-3:45 P.M.
Y. P. Tan, M. Povolotskyi, T. Kubis, T. B. Boykin,and G. Klimeck (Network for Computational
Nanotechnology, Purdue University; University of Alabama)
Transferable Tight Binding Model for Strained Group IV and III-V Heterostructures 41
3:45-4:45 P.M.
Shotgun Poster Announcements
5:00 P.M.
Poster session with cocktail reception Buchanan Room/RossAde
7:00 P.M.
Advisory Board Meeting at McGraws Steak House
September 3
7:00 A.M. Breakfast/Registration South Ballroom, Purdue Memorial Union
Session 4 - Multiscale Simulations

Session Chair: Xavier Oriols
8:00-8:30 A.M.
Invited Speaker: Christian Jirauschek (Institute for Nanoelectronics, Technische Universität
München, Germany)
Modeling of Quantum Cascade Laser Sources with Giant Optical Nonlinearities 43
8:30-8:45 A.M.
T. Cusati, G. Fiori, A. Fortunelli, G. Iannaccone (Università di Pisa; CNR Institute for Chemical
and Physical Processes)
Multiscale Modeling of Graphene-Metal Contacts 45
8:45-9:00 A.M.
Y.-B. Shi, S. Mei, and I. Knezevic (University of Wisconsin)
From Single-Stage to Device-Level Simulation of Coupled Electron and Phonon Transport in
Quantum Cascade Lasers 47
9:00-9:15 A.M.
M. Claus, A. Fediai, S. Mothes, A. Pacheco, D. Ryndyk, S. Blawid, G. Cuniberti and M. Schröter
(Center for Advancing Electronics Dresden, Institute for Materials Science and Max Bergmann
Center of Biomaterials, and Center for Computational Materials Science, TU Dresden;
Laboratory for Devices and Integrated Circuits, Universidade de Brasilia)
Multi-Scale Modeling of Metal-CNT Interfaces 49
9:15-9:30 A.M.
P. Muralidharan, D. Vasileska, S. M. Goodnick and S. Bowden (Arizona State University)
A Multi-Scale Modeling Approach to Study Transport in Silicon Heterojunction Solar Cells 51
9:30-9:45 A.M.
W. R. Frensley (University of Texas)
Variational Formulation of Stable Discrete k • p Models 53
9:45-10:00 A.M.
Z. Jiang, M. Povolotskyi, N. Onofrio, D. Guzman, D. Lemus, S. Perez, J. Bermeo, A. Strachan,
G. Klimeck (Purdue University)
Multi-Scale Quantum Simulations of Conductive Bridging RAM 55
10:00-10:15 A.M.
Break
Session 5 - Quantum Transport
Session Chair: Irena Knezevic
10:15-10:45 A.M.
Invited Speaker: Supriyo Datta (Purdue University)
Non-Equilibrium Green’s Function (NEGF) Method: A Different Perspective 57
10:45-11:00 A.M.
A. Afzalian, J. Huang, H. Ilatikhameneh, J. Charles, D. Lemus, J. Bermeo Lopez, S. Perez
Rubiano,T. Kubis, M. Povolotskyi, G. Klimeck, M. Passlack, and Y.-C. Yeo (TSMC, Network for
Computational Nanotechnology, Purdue University; TSMC Hsinchu Taiwan)
Mode Space Tight Binding Model for Ultra-Fast Simulations of III-V Nanowire MOSFETs
and Heterojunction TFETs 59
11:00-11:15 A.M.
B. Novakovic and G. Klimeck (Network for Computational Nanotechnology, Purdue University)
Time-Resolved Computational Method for Atomistic Open System Simulations 61
11:15-11:30 A.M.
D. Verreck, M. Van de Put, A.S. Verhulst, B. Sorée, W. Magnus, A. Dabral, A. Thean and G.
Groeseneken (imec Leuven; KU Leuven; Universiteit Antwerpen)
15-Band Spectral Envelope Function Formalism Applied to Broken Gap Tunnel Field-Effect
Transistors 63
11:30-11:45 A.M.
J. Fang, W. Vandenberghe, and M. Fischetti (University of Texas)
Progress on Quantum Transport Simulation Using Empirical Pseudopotentials 65
11:45-12:00 P.M.
M. Bescond, N. Cavassilas, H. Carrillo-Nuñez, S. Berradaand M. Lannoo (Aix Marseille Universit;
Integrated Systems Laboratory ETH Zürich)
Phonon Interactions in Single-Dopant-Based Transistors: Temperature and Size Dependence 67
12:00-12:15 P.M.
L. Bourdet, J. Li and Y. M. Niquet (University Grenoble Alpes)
Contact Resistances in Trigate Devices in a Non-Equilibrium Green’s Functions Framework 69
12:15-1:30 P.M.
Lunch
Session 6 - Quantum Dots and Photonics
Session Chair: Robert Kelsall
1:30-2:00 P.M.
Invited Speaker: Katsuhisa Yoshida (The University of Tokyo, Tokyo, Japan)
Sequential Tunneling Effect on Quantum-Dot Intermediate-Band Solar Cells 71
2:00-2:15 P.M.
P. Paletti, R. Pawar, G. Ulisse, F. Brunetti, G. Iannaccone, G. Fiori (Università di Pisa; Università
di Roma)
Simulation of Organic Solar Cell with Graphene Transparent Electrode 73

2:15-2:30 P.M.
X. Meshik, M.S. Choi, M. Dutta, and M.A. Stroscio (University of Illinois)
Screening Effect on Electric Field Produced by Spontaneous Polarization in ZnO Quantum Dot
in Electrolyte 75
2:30-2:45 P.M.
M. Franckié, D.O. Winge, and A. Wacker (Division of Mathematical Physics, Lund University)
Validating Simple Approaches for Quantum Cascade Laser Modeling 77
2:45-3:00 P.M.
N. Cavassilas, C. Gelly, F. Michelini and M. Bescond (Aix Marseille Université; CNRS; Université
de Toulon)
Thermionic Escape in Quantum Well Solar Cell 79
3:00-3:30 P.M.
Break
3:30-3:45 P.M.
O. Jonasson and I. Knezevic (Department of Electrical and Computer Engineering, University of
Wisconsin-Madison)
Wigner Function Approach to Quantum Transport in QCLs 81
3:45-4:00 P.M.
Md. R. Nishat, S. Alqahtani, Y. Wu, V. Chimalgi, and S. Ahmed (Department of Electrical and
Computer Engineering, Southern Illinois University)
GaN/InGaN/GaN Disk-in-Wire Light Emitters: Polar vs. Nonpolar Orientations 83
5:00-6:00 P.M.
Cocktail Hour at Shively Club
6:00 P.M.
Conference Banquet at Shively Club
September 4
7:00 A.M Breakfast
Session 7 - Semiclassical Transport
Session Chair: Massimo Fischetti
8:00-8:30 A.M.
Invited Speaker: Mark Lundstrom (Purdue University)
Unified View of Electron and Phonon Transport 85
8:30-8:45 A.M.
K. H. Park, P. N. Martin and U. Ravaioli (Beckman Institute and Department of ECE, University
of Illinois)
Thermoelectric Transports in Geometry- and Doping-Controlled Nanostructures 87
8:45-9:00 A.M.
D. Nagy, M. A. Elmessary, M. Aldegunde, J. Lindberg, A. Loureiro, and K. Kalna (ESDC;
College of Engineering, Swansea University; Dept. of Mathematics & Engineering Physics,
Mansoura University; School of Engineering, University of Warwick; Universidade de Santiago
de Compostela)
Multi-Subband Interface Roughness Scattering Using 2D Finite Element Schrödinger Equation for
Monte Carlo Simulations of Multi-Gate Transistors 89
9:00-9:15 A.M.
M. A. Elmessary, D. Nagy, M. Aldegunde, J. Lindberg , W. Dettmer, D. Perić, A. Loureiro, and
K. Kalna (College of Engineering, Swansea University; Dept. of Mathematics & Engineering
Physics, Mansoura University; School of Engineering, University of Warwick)
Anisotropic Schrödinger Equation Quantum Corrections for 3D Monte Carlo Simulations
of Nanoscale Multigate Transistors 91
9:15-9:30 A.M.
P. Ellinghaus, M. Nedjalkov, and S. Selberherr (Institute for Microelectronics, TU Wien)
Improved Particle Annihilation for Wigner Monte Carlo Simulations on a High-Resolution Mesh 93
9:30-9:45 A.M.
R. Hathwar, M. Saraniti, and S. M. Goodnick (Arizona State University)
Modeling Quantum Acceleration of Bloch Waves in Nanowires 95
9:45-10:00 A.M.
Y. Wimmer, W. Goes, A.-M. El-Sayed, A.L. Shluger, and T. Grasser (Vienna University of
Technology; University College London)
On the Validity of the Harmonic Potential Energy Surface Approximation for Nonradiative
Multiphonon Charge Transitions in Oxide Defects 97
10:00-10:15 A.M.
Break
Session 8 - Thermal Transport
Session Chair: Massimo Macucci
10:15-10:45 A.M.
Invited Speaker: Davide Donadio (University of California Davis, Davis, CA USA)
Ultra-Thin Silicon Membranes and Silicon Nanowires as Nanophononic and Thermoelectric
Devices 99
10:45-11:00 A.M.
J. H. Oh, M.-G. Jang, and M. Shin (Korea Advanced Institute of Science and Technology
Daejeon; Hallym University)
Calculation of Phonon Transmission in Si/PtSi Heterostructures 101
11:00-11:15 A.M.
S. Mei,Z. Aksamija, and I. Knezevic (University of Wisconsin-Madison; University of
Massachusetts-Amherst)
Thermal Conductivity Tensor of InxGa1- xAs/InyAl1-yAs Superlattices and Application to Quantum
Cascade Lasers 103
11:15-11:30 A.M.
S. Qazi, A. Shaik, A. Laturia, R. Daugherty, X. Guo, E. Bury, B. Kaczer, K. Raleva and
D.Vasileska (School of Electrical Computer and Energy Engineering, School for Engineering
of Matter, Transport and Energy, Arizona State University; IMEC Leuven; Faculty of Electrical
Engineering and Information Technologies, Ss. Cyril and Methodius University)
Multi-Scale Modeling of Self-Heating Effects in Nano-Devices 105
11:30-11:45 A.M.
A. Kommini and Z. Aksamija (Electrical and Computer Engineering, University of Massachusetts
Amherst)
Enhanced Seebeck Coefficient Due to Quantum Confinement in Gated 2D Nanomembranes 107
11:45-1:00 P.M.
Lunch
1:00 P.M.
Adjourn
Posters
P1 H. Ilatikhameneh, F. W. Chen, J. Appenzeller, R. Rahman, and G. Klimeck (Purdue University)

Electrically Doped 2D Material Tunnel Transistors 109
P2 P. Long, M. Rodwell, G. Klimeck and M. Povolotskyi (Purdue Univeristy; University of California
Santa Barbara)
Performance Degradation of Superlattice Transistors by Scattering 111
P3 V. T. Tran, J. Saint-Martin, and P. Dollfus (IEF; Université Paris-sud; CNRS)

Hybrid States and Bandgap in Zigzag Graphene/BN Heterostructures 113
P4 P. Sarangapani, D. Mejia, J. Charles, H. Ilatikhameneh, T. Ameen, A. Roche, J. Fonseca, and

G. Klimeck (Purdue University; University of Louisiana)
Quantum Dot Lab: An Online Platform for Quantum Dot Simulations 115
P5 T. A. Ameen, H. Ilatikhameneh, D. Valencia, R. Rahman, and G. Klimeck (Purdue University)

Engineering the Optical Transitions of Self-Assembled Quantum Dots 117
P6 A. H. Davoody, A. J. Gabourie, and I. Knezevic (University of Wisconsin-Madison)

Theoretical Study of Excitonic Energy Transfer in Carbon-Nanotube-Based Photovoltaics 119
P7 A. Price and A. Martinez (Swansea University)

Effect of Confinement in III-V Nanowire Field Effect Transistors 121
P8 A. Martinez and R. Valin and J. R. Barker (Swansea University; University of Glasgow)

Impact of Discrete Dopants in Ultrascale FinFETs and the Effect of XC on Dopant Clustering 123
P9 A. Chananay, A. Sengupta, and S. Mahapatra (Indian Institute of Science; Indian Institute of
Engineering Science and Technology)
Analysis of Vacancy Defects in Hybrid Graphene-Boron Nitride Armchair Nanoribbon Based
n-MOSFET at Ballistic Limit 125
P10 C. Huang, S. Poduri, and M.A. Stroscio (University of Illinois)

Simulation of Current-Voltage Curves for CdS Cylinders Embedded in P3HT 127
P11 . N. Lai, C. Y. Chen, and Y. Li (Parallel and Scientific Computing Laboratory; Institute of
C
Communications Engineering, National Chiao Tung University)
Nanosized-Metal-Grain-Induced Characteristic Fluctuation in Gate-All-Around Si Nanowire
Metal-Oxide-Semiconductor Devices 129
P12 B. Liu, R. Akis, and D. K. Ferry (Arizona State University)

Conductance Fluctuations in Graphene 131
P13 E. Colomés, D. Marian, and X. Oriols (Universitat Autònoma de Barcelona)

Reformulation of Quantum Noise: When Indistinguishable Becomes Distinguishable? 133
P14 E. R. Hedin, M. B. Orvis, and Y. S. Joe (Ball State University)

Coupled Nano-Rings: Strain and Magnetic Field Effects 135
P15 G. P. Szakmany, A. O. Orlov, G. H. Bernstein, and W. Porod (University of Notre Dame)
Nanothermocouple Characterization Platform: Simulation and Experiment 137
P16 H. Sahasrabudhe, Y. Wang, M. Povolotskyi, R. Rahman, and G. Klimeck (Purdue University)
Low Temperature Electrostatic Simulations of Realistic Large-Scale Structures 139
P17 H.M. Rafferty, R.W. Kelsall and Z. Ikonic (University of Leeds)

Calculation of Band-to-Band Tunnelling Currents in Germanium 141
P18 J. A. Morales Escalante, I. M. Gamba (University of Texas)

Parallelization of Collision Integrals on DG Solvers for Full Band Boltzmann-Poisson Models
of Electron Transport 143
P19 J.W. Jerome (Northwestern University)

Mathematical Consistency of Time Dependent Density Functional Theory with Kohn-Sham
Potentials 145
P20 P. Kivisaari, T. Sadi, J. Oksanen, J. Li, P. Rinke, and J. Tulkki (Aalto University; University of
Glasgow)
Bipolar Monte Carlo Simulation of Hot Carriers in III-N LEDs 147
P21 K. Fukuda, Y. Morita, T. Mori, W. Mizubayashi, M. Masahara, T. Yasuda, S. Migita, H. Ota
(National Institute of Advanced Industrial Science and Technology (AIST))
Modeling of Parallel Electric Field Tunnel FETs 149
P22 K. Moors, B. Sorée, and W. Magnus (Institute for Theoretical Physics; Imec; Universiteit
Antwerpen)
Analytic Solution of Ando’s Surface Roughness Model with Finite Domain Distribution Functions 151
P23 K.H. Park and U. Ravaioli (University of Illinois)

Thermoelectric Transports in Geometry- and Doping-Controlled Nanostructures 153
P24 L. Wang, T. Sadi, M. Nedjalkov, A.R. Brown, C. Alexander, B. Cheng, C. Millar, A. Asenov
(University of Glasgow; Institute for Microelectronics, TU Wien; Gold Standard Simulations
Ltd)
An Advanced Electro-Thermal Simulation Methodology For Nanoscale Device 155
P25 M. Claus, D. Teich, S. Mothes, G. Seifert,M. Schröter (Center for Advancing Electronics
Dresden, TU Dresden; Physical Chemistry, TU Dresden;UC San Diego)
Multiscale-Modeling of CNTFETs with Non-Regular Defect Pattern 157

P26 P. Marconcini, M. Macucci, M. R. Connolly, and E. D. Herbschleb (Università di Pisa;
University of Cambridge; National Physical Laboratory)
Simulation of Transport Through a Cavity Defined in Graphene with Electrostatic Lithography 159
P27 M. S. Choi, S. Poduri, M. Dutta, M. A. Stroscio (University of Illinois)

Numerical Analysis of Electric Field Enhancement in ZnO Film with Plasmonic Au Quantum Dots 161
P28 M. S. Choi, M. Dutta, and M. A. Stroscio (University of Illinois)

Numerical Analysis of Scattering Cross Section of Non-Precious Metallic Plasmonic Nanodisks 163
P29 M. S. Choi (University of Illinois)
Non-Local Optical Response in Silver-Silicon-Silver Heterostructure System 165
P30 M. Thesberg, M. Pourfath, N. Neophytou and H. Kosina (Technical University of Vienna;
University of Warwick)
Thermoelectric Power Factor Optimization in Nanocomposites by Energy Filtering Using NEGF 167
P 31 P. Das, G. Yin, S. S. Sylvia, K. Alam, D. Wickramaratne, and R. K. Lake (University of California;
East West University)
The Impact of the Ring Shaped Valence Band in Few-Layer III-VI Materials on FET Operation 169

P32 S. Berrada, N. Cavassilas, L. Raymond, M. Lannoo and M. Bescond (IM2NP, UMR CNRS 7334,
Bât. IRPHE, Technopôle de Château-Gombert)
The Impact of Lead Geometry and Discrete Doping on NWFET Operation 171
P33 S. Mukherjee, P. Sarangapani, M. Povolotskyi and G. Klimeck (Purdue University)

Study of Transport in Multi-Layered High Electron Mobility Transistors 173
P34 M. T. Quint and S. Ghosh (University of California)

Nanoplasmonics for All-Optical Control of Devices 177
P35 S. Sundaresan, M. Rashid, T. Jayasekera, and S. Ahmed (Southern Illinois University)

VFF-Monte Carlo Framework for Phonon Transport in Nanostructures 179
P36 J. Bohorquez-Ballen, H. Sirikumara, S. Ahmed, T. Jayasekera (Southern Illinois University)

Phonon Engineering of Si/Ge Core-Shell Nanowires for Thermoelectric Applications 181
P37 K. A. Khair and S. S. Ahmed (Southern Illinois University)

Diffusive Transport in Monolayer MoS2: Role of Remote Coulomb Scattering 183
P38 S. Soleimanikahnoj and I. Knezevic (University of Wisconsin-Madison)

On the Role of Rashba Spin-Orbit Coupling in Graphene Nanoribbons 185
P39 S. Mukherjee, S. Farid, M. Dutta and M.A. Stroscio (University of Illinois)

Modeling Polycrystalline Effects on the Device Characteristics of CdTe Based Solar Cells 187
P40 M. C. Nguyen, V. H. Nguyen, H. V. Nguyen, J. Saint-Martin, and P. Dollfus (Univ. of Paris-Sud;
Institute of Physics, VAST)
Strong Negative Differential Resistance in Graphene Devices with Local Strain 189
P41 V. P. Georgiev, T. Ali, Y. Wang, L. Gerrer, S. M. Amoroso, E. Towie, A. Asenov (University of
Glasgow; Peking University; Gold Standard Simulations Ltd)
Influence of Quantum Confinement Effects Over Device Performance in Circular and Elliptical
Silicon Nanowire Transistors 191
P42 X. Guo, S. Qazi, R. Daugherty, D. Vasileska (Arizona State University)

Algebraic Multigrid Poisson Equation Solver 193
P43 Y. Lan, C. Tang, J. Shi, M. Dutta, M. Stroscio (University of Illinois)
Phononic Properties for Enhanced Signal-to-Noise Photodetector 195
P44 M. Y. Lee, Y. C. Tsai, Y. Liand S. Samukawa (National Chiao Tung University, Tohoku University)
Electronic Structure Dependence on the Density, Size and Shape of Ge/Si Quantum Dots Array 197
P45 L. Yin, M. Fang, L. Zeng, L. Zhang, G. Du, X. Liu (Peking University; National University of
Defence Technology, Beihang University)
Accelerated 3D Full Band Self-Consistent Ensemble Monte Carlo Device Simulation Utilizing
Intel MIC Co-Processors on TianHe II 199
P46 Y. Chu, D. F. Mejia, J. Fonseca, M.Povolotskyi and G. Klimeck (Purdue University)
Crystal Viewer: An Online Platform for Learning and Understanding Crystalline Structure 201
P47 A. Kommini, G.P. Szakman, A. O. Orlov, G. H. Bernstein, W. Porod, and Z. Aksamija
(University of Massachusetts Amherst; University of Notre Dame)
Size Dependence of the Seebeck Coefficient for Single-Material Thermocouples 203
P48 Z. Stanojević, M. Karner, F. Mitterbauer, C. Kernstock (Global TCAD Solutions GmbH.)

New Computational Perspectives on Scattering and Transport in III/V Channel Materials 205
Electronic and Vibrational Properties of 2D Materials
from Monolayer to Bulk
Mahesh Raj Neupane
U.S. Army Research Laboratory, Aberdeen Proving Ground, Maryland 21005, USA
e-mail: [email protected]
theoretical and experimental synergetic efforts in

INTRODUCTION AND MOTIVATION exploring the effect of in-plane and out-of-plane
The placement of two dimensional (2D) materials such geometries of 2D materials and their heterostructures on
as hexagonal boron nitride (h-BN) and transition metal their electronic and vibrational properties.
dichalcogenide (TMDC) at the forefront of materials
and device research was pioneered by the discovery of RESULTS AND DISCUSSION
graphene, an atomically thin 2D allotrope of carbon An atomistic representation of the
obtained through mechanical exfoliation [1]. These 2D experimentally observed single layer (1T) and bulk
materials possess a wide range of electronic behaviors (4H) phase of tin disulfide (SnS2) is shown in Fig. 1
from insulator to metallic, resulting from their in-plane (a). In this study, band structure measurements in
strong covalent bonds and their weaker out-of-plane conjunction with ab-initio calculations and
coupling [2]. The intrinsic bandgap of the photoluminescence (PL) spectroscopy show that SnS2
semiconducting TMDCs makes them materials of is an indirect bandgap semiconductor over the entire
choice for next-generation low-dimensional optical and thickness range from bulk to monolayer which is
electronic devices for defense and civilian applications. different from other TMDC materials such as
These 2D van der Waal (vdW) materials hold promise molybdenum disulfide (MoS2). The difference arises
for a range of electronic, thermoelectric and because of the different orbital compositions of the
optoelectronic devices such as field effect transistor valence band at Γ.
(FET), light emitting device (LED), energy harvesting Misorientation between the layers was also
devices and ultrafast optical devices [3]. shown to affect the thermophysical properties of
graphene/graphene. Since the commensurate bilayer
The in-plane geometry exhibited particularly in TMDC system size grows exponentially with decreasing
materials is governed by the location of the transition misorientation angle, large scale molecular dynamics
metal atom sandwiched between the two dichalcogenide simulations, which include the anharmonic effects, are
atoms which govern the phase of the materials from used to analyze the angle-dependent, experimentally
metallic (1T) to semiconducting (2H) [4]. The out-of- observed low-energy vibrational modes [7]. The
plane geometry between the vdW layers, however, can results are verified against the Raman signatures, as
be changed in various ways such as by misalignment shown in Fig. 2. Experimental observation, along with
(sliding), misorientation, and formation of the theoretical validation, suggests that the low-
heterostructures [5]. Misorientation between the layers energy vibrational modes are insensitive to the
originates from the weak interlayer coupling or vdW misorientation angle in the bilayer misoriented
interaction between the layers. Similarly, when different graphene system.
layers of materials are stacked on one another their
physical and quantum properties can, in effect, Modulation of the vdW gap and
combine, interfere or cancel one another, to varying bandstructure in few-layer MoS2 was achieved via
degrees depending on the materials, leading to the new partial intercalation of oxygen [8]. Changes in the
properties [5]. The overall properties of these film thickness and PL spectra due to partial
heterostructures rely on interlayer coupling between the intercalation are analyzed using ab-initio simulations.
layers. In addition, modifying the interlayer coupling Calculations employing hybrid functionals combined
through chemical and physical treatments can also lead with semi-empirical corrections to account for vdW
to new out-of-plane topology between the layers. interactions are used to quantitatively compare to
experimental results. It was found that, as the layer
Motivated by this and a growing interest in 2D vdW thickness increases from 2 layers to 4 layers, the
materials, in this talk I will give an overview of the indirect to direct transition vdW gap distance
ISBN 978-0-692-50554-0 1
increases by 45%. An increase of the vdW gap [8] Dhall, R. et al., Advanced Materials 27 (9), 1573-1578
distance by 30% with respect to the equilibrium (2015)
distance between each monolayer, is sufficient to [9] Dhall, R. et al., App. Phys. Lett. (under review) (2015)
observe the indirect to direct crossover in the [10] Ionescu, R. et al., Chem. Commun., 51, 11213-11216
(2015)
bandgap, as shown in Fig. 3 (a, b). This observation is
consistent with our recent experimental optical a b
measurements [9]. The intercalated MoS2 bilayers . .
exhibited circularly polarized PL consistent with the
spin-valley-layer selection rules predicted 4H
theoretically. This vdW-gap dependent indirect-direct
bandgap transition was also observed in the MoS2-
1T
WS2 heterostructures [10]. Finally, some of the
existing opportunities for combining 2D materials
with organic polymers will be identified and c
discussed. Combining these materials, a new device-
design, TMDC-based Excitonic Field Effect .
Transistor (TexFET) is conceptualized. The working
principle of this device which exhibits low-power and
ultra-fast switching mechanism will be presented.
Fig. 1. (a) 1T and 4H SnS2 atomistic structure, (b) ARPES mapped
CONCLUDING REMARKS bandstructure of bulk 4H SnS2, and (c) Ab-initio (HSE-level)
To conclude, I will review some of the limitations and bandstructure for monolayer (1T) and bulk (4H) SnS2. Reprinted with
permission from { Ref. 6 } Copyright 2014, Americal Chemical Society.
challenges associated with the application of 2D
a b
materials for real device applications. I will summarize
my talk by providing an overview of ongoing and . .
planned 2D materials based research activities at Army
Research Laboratory (ARL) which is mainly focused on
designing lighter, faster and robust optoelectronic
devices for future army applications.
ACKNOWLEDGMENT c d
I acknowledge helpful discussions with R. Lake (UCR), A. . .
Mulchandani (UCR), A. Balandin (UCR), O. Monti (UA), S. Cronin
(USC), J. W. Andzelm (ARL) and T. L. Chantawansri (ARL). MRN
is supported by an appointment to the Postgraduate Research
Participation Program at the U.S. Army Research Laboratory
administered by the Oak Ridge Institute for Science and Education
through an interagency agreement between the U.S. Department of Fig. 2. (a) Misoriented bilayer graphene atomistic structure, (b) SAED image
Energy and USARL. This work was supported by grants of computer
of misoriented bilayer graphene, (c) Force directions for misoriented bilayer
time from the DOD High Performance Computing Modernization
Program at the U.S. Air Force Research Laboratory and U.S. Army graphene with angle 21.78°, and (d) Calculated l low frequency modes (X)
Engineer Research and Development Center DoD Supercomputing overlaid on the Raman spectrum of the misoriented graphene samples.
Resource Centers. This work is supported in part by the NSF Grant
No. 1307671, FAME, one of six centers of STARnet, a SRC program a b
sponsored by MARCO and DARPA, and a U.S. Dept. of Education
GAANN Fellowship. This work used the resources at XSEDE (NSF . .
Grant no: OCI-1053575) and Purdue University.
REFERENCES c d
[1] Novoselov, K. S. et al., Phys. Scr. 2012, 01400 (2012) . .
[2] Gong, Y. et al., Nat Mater 13, 1135–1142 (2014)
[3] Lopez-Sanchez, Kis et al., Nature Nanotech. (2013);
Bertolazzi, Kis et al., ACS Nano (2013); Zhike Liu et al.,
Chem. Soc. Rev., Advance Article. (2015)
Fig. 3. (a) Multilayer MoS2 atomistic structure, (b) PL-spectra of pre-O2
[4] Kappera, R. et al., Nature Materials 13, 1128–1134
(2014) treatment (black) and post-O2 treatment (red) for multilayer MoS2, (c)
[5] Dumcenco, D. et al., ACS Nano, 9 (4) (2015); van der Illustration of vdW-gap modulated indirect to direct transition in
Zande, A. M. et al., Nano Lett., 17 (7) (2014); Wang X., bandstructure, and d) vdW gap dependent indirect to direct bandgap transition
Nature Materials 14, 264–265 (2015) for 2L,3L and 4L (bulk-like) MoS2.. Reprinted with permission from { Ref.8} Copyright 2015, John
[6] Huang Y. et al., ACS Nano 8 (10), 10743-10755(2014) Wiley and Sons.
[7] Tharamani P. et al., Nanoscale (under review) (2015)
2
Self-consistent field and master equation
approach to calculating the dielectric function of
graphene
F. Karimi and I. Knezevic
Department of Electrical and Computer Engineering, University of Wisconsin-Madison,
Madison, Wisconsin 53706-1691, USA
Graphene, a two-dimensional carbon material dielectric function of electronic materials with an

with a hexagonal lattice, has been of great research arbitrary band structure and with the inclusion
interest in modern electronics [1], [2]. In the vicinity of scattering. The dielectric function is calculated
of the K and K’ points in the Brillouin zone, the within the self-consistent field (SCF) and linear-
energy dispersion of electrons in graphene is linear response approximation: the external perturbation
and gapless, which leads to some unique optical and the induced potential due to electron-electron
and electrical properties. For instance, the screening interactions are replaced with an effective poten-
properties and collective excitations (plasmons) are tial. The induced charge density as a function of
quite different from those in conventional two- the effective potential is then calculated based on
dimensional semiconductor nanostructures. In order a Markovian quantum master equation within the
to study the optical and electrical properties of density matrix framework [6]. Namely, the elec-
graphene, its dielectric function should be known. tronic systems in semiconductors, as open quantum
A well-known approach to calculating the dielec- systems, interact with an environment that may
tric function of an electronic system is the Lind- include photons, phonons, ionized impurities, etc.
hard dielectric function method [3]. It captures the The evolution of the density matrix in such systems
electron-electron interaction self-consistently and may be described by Markovian quantum master
is able to capture the photon-assisted interband equations [6], which are memoryless (i.e. the time
electron-hole excitations. But, this method does evolution of the density matrix depends only on its
not account for scattering mechanisms, e.g. ionized current state) and ensure the positivity of the density
impurity or phonon scattering. In order to take matrix. By writing a Markovian master equation for
those effects into account, Mermin introduced the the density matrix in graphene and solving it numer-
Mermin-Lindhard dielectric function [4]. He modi- ically, we obtain the induced charge density as a
fied the Lindhard dielectric function via the relax- function of the external field. Finally, by combining
ation time approximation (RTA), in such a way that it with a relation between the induced charge density
it does not violate the conservation of the number and the induced potential from the electromagnetic
of electrons. This method has been used to calculate wave equations, the dielectric function of graphene
the dielectric function of graphene [5]. However, it is calculated.
should be noted that the Mermin-Lindhard dielectric In the calculations, we assume electron scattering
function was derived for materials with a parabolic with acoustic phonons, non-polar optical phonons,
energy band dispersion. Also, it works as long as and ionized impurities. The real and imaginary parts
the intraband dissipation mechanisms are dominant, of the graphene dielectric function versus the wave
so in the case of nanostructures with densely spaced vector and frequency are shown in Fig. 1 for the
energy subbands and/or in the presence of interband Fermi energy EF =354 meV and a sheet density of
dissipation mechanisms (e.g. optical phonons), this ionized impurities of NI = 5.2 × 1011 cm−2 , and
method reveals its shortcomings. in Fig. 2 for EF =170 meV and NI = 2.5 × 1011
Here, we introduce a theory for calculating the cm−2 .
ISBN 978-0-692-50554-0 3
2 2
60
30
1.5 1.5
40
20
ω/ωF
ω/ωF
1 1
20
10
0.5 0 0.5
0
−20
0.5 1 1.5 2 0.5 1 1.5 2
k/kF k/k
F
(a) Real part of ε(q, ω) (a) Real part of ε(q, ω)
2 60 2
50 30
1.5 1.5
40
20
ω/ωF
ω/ωF
1 30 1
20
0.5 10
0.5
10
0 0
0.5 1 1.5 2 0.5 1 1.5 2
k/kF k/k
F
(b) Imaginary part of ε(q, ω) (b) Imaginary part of ε(q, ω)
Fig. 1: The real (a) and imaginary (b) parts of Fig. 2: The real (a) and imaginary (b) parts of
the graphene dielectric function, ε(q, ω), as a func- the graphene dielectric function, ε(q, ω), as a func-
tion of the wave vector k (in units of the Fermi tion of the wave vector k (in units of the Fermi
wavevector, kF ) and angular frequency ω (in units wavevector, kF ) and angular frequency ω (in units
of the Fermi angular frequency, ωF ) for the Fermi of the Fermi angular frequency, ωF ) for the Fermi
level of 354 meV and the impurity sheet density of level of 170 meV and the impurity sheet density of
5.2 × 1011 cm−2 . 2.5 × 1011 cm−2 .
The plasmon dispersion follows the zeros of the [2] A. K. Geim and K. S. Novoselov, The rise of graphene,
real part of the dielectric function [panel (a) in Nat. Mater. 6, 183 (2007).
Figs. 1 and 2]. Increasing the carrier density and [3] E. H. Hwang and S. Das Sarma, Dielectric function, screen-
ing, and plasmons in two-dimensional graphene, Phys.Rev.
decreasing the impurity density extends the plasmon B 75, 205418 (2007).
dispersion to higher frequencies. [4] N.D. Mermin, Lindhard Dielectric Function in the
This work was supported by the U.S. Department Relaxation-Time Approximation, Phys. Rev. B 1, 2362
(1970).
of Energy, Office of Science, Materials Science Pro-
[5] M. Jablan, H. Buljan, and M. Soljačić, Plasmonics in
gram and Engineering Division, Physical Behavior graphene at infrared frequencies, Phys. Rev. B 80, 245435
of Materials Program, award DE-SC0008712. (2009).
[6] H. P. Breuer and F. Petruccione, The Theory of Open Quan-
R EFERENCES tum Systems, Oxford University Press, Oxford, (2002).
[1] K. S. Novoselov, A. K. Geim, S. V. Morozov, D. Jiang,
Y. Zhang, S. V. Dubonos, I. V. Grigorieva, A. A. Firsov,
Electric field effect in atomically thin carbon films, Science
306, 666 (2004).
4
The role of dimensionality on phonon-limited
charge transport: from CNTs to graphene
J. Li∗ , H. Miranda† , Y.M. Niquet∗ , L. Genovese∗ , I. Duchemin∗ , L. Wirtz† and C. Delerue‡

∗ Univ. Grenoble Alpes, INAC-SP2M, L Sim, Grenoble, France and CEA, INAC-SP2M, L Sim, Grenoble, France
† Physics and Materials Science Research Unit, University of Luxembourg, Luxembourg
‡ IEMN - Dept. ISEN, UMR CNRS 8520, Lille, France
Graphene and carbon nanotubes (CNTs) are two- possibly non monotonic) behaviors depending on
dimensional (2D) and one-dimensional (1D) al- the nature of the CNTs. The convergence is much
lotropes of pure sp2 carbon. Their intrinsic elec- faster at high carrier density (Fig. 5) and high
tronic and transport properties are, therefore, inti- temperature (Fig. 6). A simple explanation starts
mately related. In particular, the transport properties with the electronic band structure of CNTs – which
of CNTs are expected to approach those of graphene can be viewed as a sampling of the 2D band
when increasing tube diameter. However, this 1D to structure of graphene by a bundle of lines parallel
2D transition, which gives insights into the role of to the tube direction. In general, the mobility in
dimensionality, has never been carefully studied. CNTs approaches that of graphene when the number
In order to investigate this transition, an atomistic of bands in the transport energy window (a few
computational model for the phonon-limited carrier kT around the Fermi energy) is large enough to
mobility in carbon materials has been developed. smooth out the effect of 1D confinement on the
The carrier mobility is obtained from an exact band structure and on the electron-phonon interac-
solution of Boltzmann transport equation. A tight- tions. The absence of linearly dispersive bands in
binding model with the latest parameters extracted semiconducting nanotubes and the selection rules
from first principle calculations is used for the for intervalley scattering are mostly responsible for
electronic band structure. A force-constant model the differences between CNT chiralities. Details will
is used for the phonon band structure, and has be given at the conference.
been refined in order to reproduce the admixture This work was supported by the French National
of optical components into the acoustic phonon Research Agency (ANR) project Noodles (contract
modes at non-zero wave-vectors. This admixture ANR-13-NANO-0009-02). Part of the calculations
has significant impact on the carrier mobility. were run on the TGCC/Curie machine using allo-
This model has been validated against DFT cal- cations from GENCI and PRACE. H. M. and L.
culations and experimental data. At high carrier W. acknowledge support by the National Research
density, the agreement with the DFT calculations Fund (FNR), Luxembourg (projects OTPMD and
of Ref. [1] (Fig. 1) and the experimental data of NanoTMD).
Ref. [2] (Fig. 2) is very satisfactory. At low carrier
R EFERENCES
density, the agreement with experiment [3] (Fig. 3)
is also very good, when the additional scattering by [1] C.H. Park, N. Bonini, T. Sohier, G. Samsonidze, K. Kozin-
sky, M. Calandra, F. Mauri and N. Marzari, Nano Lett. 14,
surface optical phonons in the substrate is taken into 1113 (2014)
account [4]. [2] D. Efetov and P. Kim, Phys. Rev. Lett. 105, 256805 (2010).
The same model and parameters were used to [3] K. Zou, X. Hong, D. Keefer and J. Zhu, Physical Review
study the transition from 1D to 2D in CNTs with Letters 105, 126601 (2010).
[4] A. Konar, T. Fang, and D. Jena, Physical Review B 82,
diameter up to 16 nm. The mobility in CNTs con- 115452 (2010).
verges to the mobility in graphene when increasing
tube diameter (Fig. 4), but with very different (and
ISBN 978-0-692-50554-0 5
102 1.36×1013 cm−2
2.86×1013 cm−2
4.65×1013 cm−2
6.85×1013 cm−2
101 10.8×1013 cm−2
µph (cm2/V/s)
DFT (Ref. 1) 105
T = 300 K
n = 1 × 1012 cm−2
100
Armchair
Zigzag Metallic
104 Zigzag Semiconducting
Graphene
10 −1
101 102
T (K) 0 2 4 6 8 10 12 14 16
d (nm)
Fig. 1. Resistivity of graphene at high carrier (hole) density
compared with recent DFT calculations [1]. Fig. 4. Evolution of the mobility with the diameter of the
CNTs.
102 1.36×1013 cm−2

2.86×1013 cm−2
4.65×1013 cm−2
6.85×1013 cm−2 Armchair
Zigzag Metallic
101 10.8×1013 cm−2
Zigzag Semiconducting
ρph (Ω)
EXP (Ref. 2)
Graphene
µph (cm2/V/s) 105
T = 300 K
100 d ≈ 3.5 nm
10−1
101 102
T (K) 104
0.0 0.5 1.0 1.5 2.0

Fig. 2. Resistivity of graphene at high carrier (hole) density n (cm−2 ) ×1013
compared with experiment [2]. Note that our calculations only
include the intrinsic electron-ponon scattering, and therefore Fig. 5. Evolution of the mobility with the carrier (hole) density.
underestimate the experimental mobility.
550
EXP (Ref. 3)
500
Simulations 107 Armchair
450 Zigzag Metallic
12
1×10 cm −2
Zigzag Semiconducting
Graphene
400
µph (cm2/V/s)
ρ (Ω)
106
350 12
1.5×10 cm
−2
300
2×1012 cm−2
105
250 2.5×1012 cm−2
n = 1 × 1012 cm−2
3×1012 cm−2
200 d ≈ 3.5 nm
0 50 100 150 200 250 300
T (K) 104
0 100 200 300 400 500
T (K)
Fig. 3. Resistivity of graphene at low carrier (hole) density
compared with experiment [3]. Scattering by surface optical Fig. 6. Evolution of the mobility with temperature.
phonons in the substrate is also included here.
6
A Rigorous Study of Nanoscaled Transistors
Based on Single-Layer MoS2
A. Kefayati∗ , M. Pourfath∗† , and H. Kosina†
∗ School of Electrical and Computer Engineering, University of Tehran, Iran
† Institute for Microelectronics, TU Wien, Gußhausstraße 27–29/E360, 1040 Wien, Austria
Since the successful experimental isolation of Lch=20nm
graphene in 2004, ultra-thin two-dimensional struc-

tures are being widely studied as potential building
blocks for future electronic devices. Among various
two-dimensional materials, single-layer (SL) MoS2
has attracted much attention. For a SL of MoS2 a di-
rect bandgap of 1.8 to 1.9 eV has been reported [1],
which is suitable for various electronic applications.
Recently, FETs based on SL MoS2 with an Ion /Ioff
ratio as high as ∼ 108 and a sub-threshold swing of
∼ 70 mV/decade have been achieved [2]. However,
Fig. 1. Schematic view of the simulated device. An n-type top
the reported mobility is below that of ultra thin body gate SL MoS2 FET with ohmic contacts. The channel length
or strained Si and III-V materials. It is believed is 20 nm and the gate oxide is 10 nm thick HfO2 layer with
that extrinsic sources such as charged impurities κ = 22. The substrate is assumed to be SiO2 .
(CI) [3] and inevitable Schottky contacts [4] limit 6
the characteristics of devices based on SL MoS2 . (a) (b)
9
High-κ gate insulators, such as HfO2 , can reduce
Drain Current (mA/µm)
Drain Current (mA/µm)

−1
10
CI scattering effects and boost the mobility [3, 4

10
−4
6
5], however, they can degrade the mobility due to 10
−7
remote phonon (RP) scattering [6] . The source of 2

−1.2 −0.7 −0.2 0.3
3
this scattering is in the surrounding dielectrics via
long-range Coulomb interactions, provided that the
dielectrics support polar vibrational modes. 0
−1.2 −0.7 −0.2 0.3
0
0 0.4 0.8
Gate Voltage (V) Drain Voltage (V)
To study electronic transport in SL MoS2 we
solved the NEGF equations self-consistently with Fig. 2. (a) The transfer characteristics in the presence of
the Poisson equation based on the box integra- scattering at VDS =0.1 V. Transconductance is gm = 9.6 mS/µ
tion method. An effective mass of m∗ = 0.48m0 m. The inset shows transfer characteristics in logarithmic scale.
(b) The output characteristics for VGS = - 0.6 to +0.4 V with
has been assumed for both longitudinal and trans-
0.2 V step (the arrow indicates the direction of VGS increase).
verse directions [7]. We have considered intrinsic Current saturation is observed for VDS > 0.3 V.
electron-phonon interactions including the longitu-
dinal acoustic (LA), the transverse acoustic (TA),
the longitudinal optical (LO), and polar optical SiO2 substrate, see Fig. 1. In this study the dynamic
phonons (POP) with the parameters adopted from screening of remote phonon modes stemmed from
Ref. [7]. The mobility is calculated based on the HfO2 and static screening of charged impurities are
method explained in Refs. [8, 9]. For device sim- included. Figure 2 shows the transfer and output
ulation we assumed a channel length of 20 nm, a characteristics in the presence of intrinsic phonon
10 nm thick HfO2 gate insulator, and a 50 nm thick scattering. The results indicate a relatively high Ion
ISBN 978-0-692-50554-0 7
800
CI (HfO2)
CI (SiO2)
2 CI + RP (HfO2)
Velocity (107cm/s) 600
Mobility (cm2/Vs)
CI + RP + IP (HfO2)
µ =580 cm2/(Vs) Exp. Ref. [7] (HfO2)
600
400
Mobility (cm2/Vs)
1
400
200
200
0 50 100
Electric Field (KV/cm)
0 0
0 50 100 1012 1013
Electric Field (KV/cm)
Impurity Density (cm-2)
Fig. 3. The velocity as a function of the applied electric
field. The inset shows the mobility as a function of the applied Fig. 4. Charged impurity (CI) and remote phonon (RP) limited
electric field. The low field mobility is about 580 cm2 /Vs and mobility as a function of charged impurity density for 30nm
decreases to 130 cm2 /(Vs) under high electric fields. thick HfO2 and also SiO2 top gate insulator. The mobility due
to IPs+CI+RP with HfO2 gate insulator is in good agreement
with experimental results in Refs. [3, 10].
of 9 mA/µm and a high Ion /Ioff ratio of about

the smallest mean free path. The presented results
107 , which are close to the ballistic limit. Low
can be used for appropriate selection of the gate
field phonon-limited mobility is evaluated to be
insulator material for optimal device performance.
580 cm2 /(Vs) in good agreement with the result of
Ref. [7]. The effect of high fields on the mobility
This work has been supported by the Austrian
and carrier velocity are depicted in Fig. 3. The
Science Fund (FWF), contract 25368-N30.
significant drop in the mobility is due to increased
polar and non-polar optical phonon scattering at TABLE I
high electric fields. T HE MEAN FREE PATH (λ) AND LOW- FIELD MOBILITY (µ)
FOR EACH SCATTERING SOURCE . RP SCATTERING IS
Figure 4 shows the effect of CI scattering and
CALCULATED FOR FOR A 30 NM THICK H F O 2 .
RP scattering for an average carrier density of
∼ 1013 cm−2 . By using a high-κ insulator the Phonon Modes λ (nm) µ (cm2 /(Vs))
CI scattering is suppressed which enhances the LA+TA 35 923
mobility. On the other hand, high-κ insulators LO 157 4161
introduce RP scattering which in turn reduce the POP 87 2316
RP 5.7 151
mobility. Therefore, at low CI concentrations the
mobility for a device with SiO2 as the top gate
is higher than that with HfO2 . Figure 4 shows
R EFERENCES
the evaluated mobility in the presence of intrinsic
phonons (IP), CI, and RP (CI+RP) is in good [1] W. S. Yun et al., Phys. Rev. B 85, 033305 (2012).
[2] B. Radisavljevic et al., Nature Nanotech. 6, 147 (2011).
agreement with experimental results from Refs. [3, [3] B. Radisavljevic et al., Nature Mater. 12, 815 (2013).
10]. Table I compares the mean free path and the [4] J. Kang et al., Phys. Rev. X 4, 031005 (2014).
mobility for each scattering mechanism studied in [5] Z.-Y. Ong et al., Phys. Rev. B 88, 165316 (2013).
[6] M. V. Fischetti et al., J. Appl. Phys. 90, 4587 (2001).
this work. The results show that acoustic phonons
[7] K. Kaasbjerg et al., Phys. Rev. B 85, 115317 (2012).
play a more significant role in short channel devices [8] Y.-M. Niquet et al., J. Appl. Phys 115, 054512 (2014).
than other intrinsic phonon modes. However, RP [9] R. Rhyner et al., J. Appl. Phys. 114, 223708 (2013).
scattering due to a 30 nm high-κ HfO2 results in [10] D. Lembke et al., Nanoscale 7, 6255 (2015).
8
Strain effects on the electronic properties of
devices made of twisted graphene layers
V. Hung Nguyen1,2, H. Viet Nguyen2, J. Saint-Martin1, and P. Dollfus1
1
Institute of Fundamental Electronics, CNRS, Univ. of Paris-Sud, Orsay, France
2
Center for Computational Physics, Institute of Physics, VAST, Hanoi, Vietnam
Graphene is one of the most attractive materials dependence on the twist angle, the mentioned
for beyond-CMOS electronics because of its phenomena are also dependent on both the strain
specific electronic properties, which are a amplitude and its applied direction. By choosing
consequence of its two-dimensional honeycomb appropriately the amplitude and direction of strain,
lattice and relativistic-like charge carriers at low our study suggests the possibility of observing the
energy [1]. To enlarge its range of applications, the van Hove singularities at reasonably low energy in
modulation of electronic structure of graphene a large range of twist angle (i.e., larger than 10°)
nanomaterials has been the subject of intense [4].
research. Recently, the interest of the graphene Next, we investigate the effects of strain on the
community has also been oriented toward the transport properties of vertical devices made of
investigation of twisted graphene multilayer twisted graphene bilayers (see the schematic view
lattices, a specific type of Van der Waals structures on top of Fig. 4) [5]. Again, because of the
of graphene. These lattices appear as promising different orientations of the two graphene layers,
materials providing various possibility of the Dirac cones of left and right graphene sections
modulating their electronic properties by changing can be separated in the k-space and hence a finite
the twist angle [2,3]. In this work, we investigate transport gap can open in this device (Fig. 4). Our
the effects of uniaxial strain on the electronic study shows that besides its dependence on strain
properties of devices based on twisted graphene amplitude, this feature also depends on the twist
layers (see Fig. 1) [4,5]. First, we explore the angle and the strain direction. Important, we find
effects of strain on the low-energy bands of that a finite gap as large as a few hundred meV can
twisted graphene bilayer. Second, we demonstrate be achieved with a small strain of only a few
that the strain engineering is an efficient technique percent (see Fig. 4(d)). On this basis, the ON/OFF
to open finite transport gaps in vertical devices current ratio as high as a few ten thousands and
made of stack of twisted graphene layers. high thermoelectric power can be achieved in our
Our calculations are based on the atomistic proposed devices (see Fig. 5).
tight-binding model, similarly as in [6]. A unixial In summary, we demonstrate that strain
strain is applied in the in-plane direction (i.e., in engineering is a promising technique to modulate
the Oxy plane as schematized in Fig. 1). To the electronic properties of twisted graphene
compute the charge transport, the tight-binding bialyer systems and to enlarge their applications,
Hamiltonian is solved using the Green's function e.g., in transistors, strain and thermal sensors. This
formalism [5]. can be also a useful approach for Van der Waals
In Fig. 2, we display two sub-figures showing structures of other 2D planar materials.
the effects of strain on the low energy bands of
REFERENCES
twisted graphene bilayer. Interestingly, it is shown
[1] A. C. Ferrari et al., Nanoscale 7, 4598 (2015)
that since the two graphene layers have different [2] G. Li et al., Nat. Phys. 6, 109 (2010)
orientations, the strain can break degeneracy of the [3] W. Yan et al., Nat. Commun. 4, 2159 (2013)
bands around the Dirac points. As a consequence, [4] V. H. Nguyen and P. Dollfus (2015); arXiv:1412.7583
the number of Dirac cones can double (right panel [5] V. H. Nguyen et al., Nanotechnol. 26, 115201 (2015)
of Fig. 2) and the van Hove singularity points [2] [6] V. M. Pereira et al., Phys. Rev. B 80, 045401 (2009) and
Z. F. Wang et al., Nano Lett. 12, 3833 (2012)
are separated in energy (see Fig. 3). Besides their
ISBN 978-0-692-50554-0 9
Fig. 1. Twisted graphene bilayer lattice considered in this
work. The twist angle is φ TL and a strain is applied along the
direction θ.
Fig. 4. (E-ky)-maps of transmission probability showing the

strain effects on the transport properties of vertical devices
made of a stack of two twisted graphene layers.
Fig. 2. Strain effects on the low-energy bands around the

Dirac point.
Fig. 5. Strain engineering to improve ON/OFF current ratio

and Seebeck coefficient in vertical devices made of a stack of
Fig. 3. Strain effects on the bands connecting two Dirac cones two twisted graphene layers.
and the van Hove singularity points.
10
Overlaps in Stacked Graphene Flakes using
Empirical Pseudopotentials
M. L. Van de Put∗ , W. G. Vandenberghe† , and M. V. Fischetti†
∗ imec,Kapeldreef 75, 3001 Heverlee, Belgium
∗ Condensed Matter Theory, Universiteit Antwerpen, Antwerpen, Belgium
† Dept. of Material Sciences, University of Texas at Dallas, Texas, USA
I NTRODUCTION
Recently, resonant tunneling has been observed in J = −2πeh̄−1 (EiT − EjB )
stacked graphene flakes[1]. The rotational misalign-
ment introduces an offset between the Dirac-cones f T (EiB ) − f B (EjT ) |i, T| j, B|2 ,
ij
in both flakes, resulting in negative differential
resistance when shifting the Dirac-cones under bias. with f T,B (EkT,B ) the respective top and bottom
In this paper we investigate the finite-size effect Fermi levels determined by (electrostatic) doping,
in nano-scaled graphene flakes. Improving on the i, T| j, B the overlap between the top and bottom
bulk description, and because the structures are – eigenstates with their respective energies EiT and
atomistically speaking – large in size, we use the EjB . This approximation assumes that electrons are
empirical pseudopotential method[2]. always instantaneously refilled to their equilibrium
states. For the current at finite temperatures we
C OMPUTATIONAL METHOD broaden the overlap using a thermal Gaussian.
To calculate the electronic states, we implement R ESULTS & D ISCUSSION
a local plane-wave empirical-pseudopotential eigen- Following the setup in Ref. [1] we have opted for
valuesolver for large structures. We adopt the real- a stacked structure as depicted in Fig. 1 A finite-
space technique used by Kresse and Furthmüller length armchair-edged ribbon is placed on a ribbon
to solve the Schrödinger equations[3] in our prob- with a zigzag edge with an angle close to 90◦ . Note
lem. Using this method, we evaluate the kinetic that we have not considered an intermediate layer
energy in reciprocal space and the potential energy and separated the layers 2 Å by vacuum.
in real space. Using a diagonal operator in both The wavefunction of the first conduction states
spaces effectively reduces the complexity from a of the flakes is plotted in Fig. 2. As an example,
full rank matrix product with complexity O(n2 ) we show their overlap as a function of rotation in
to that of an FFT with complexity O(n log n). Fig. 3. The Fermi-level of the layers is taken at 0.1
Our eigensolver implements the Residual Mini- eV above the highest valence state for the armchair
mization Method by Direct Inversion of the Iter- ribbon and 0.1 eV below the highest valence state
ative Subspace (RMMDIIS) first developed by by for the zigzag ribbon, simulating doping. We show
Wood and Zunger[4] and later used by Kresse and the resulting IV curve in Fig. 4 for different rota-
Furthmüller[3]. We selected this method because it tions. We observe that as we rotate the structure
allows semi-independent solution of the eigenvec- away from the 90◦ axis more resonance occurs as
tors in parallel. This alleviates the lack of parallel is evidenced by the NDR in the IV curves.
computation on a k-grid, which is lacking in our
closed systems. ACKNOWLEDGMENTS
To calculate the current, we make an approxima- I would like to acknowledge my promotors
tion based on the Fermi-Golden rule. The current at Wim Magnus and Bart Sorée for supporting my visit
a bias equal to EkT − EkB reads, to the University of Texas at Dallas.
ISBN 978-0-692-50554-0 11
Fig. 3. The overlap of the first conduction states of the zigzag
and armchair ribbons (shown in Fig. 2) at different rotations
Fig. 1. The simulated structure. An armchair ribbon (horizon-

tal) and a perpendicular zigzag ribbon (vertical). The bounding
boxes show the supercell used.
Fig. 4. The IV curve for different rotations. Zero temperature

current is shown by discrete markers, for finite temperature
300 K the current is shown as a full line.
R EFERENCES
[1] A Mishchenko, JS Tu, Y Cao, RV Gorbachev, JR Wallbank,
MT Greenaway, VE Morozov, SV Morozov, MJ Zhu,
SL Wong, et al. Twist-controlled resonant tunnelling in
graphene/boron nitride/graphene heterostructures. Nature
nanotechnology, 9(10):808–813, 2014.
[2] Massimo V Fischetti and Sudarshan Narayanan. An
empirical pseudopotential approach to surface and line-
edge roughness scattering in nanostructures: Application to
si thin films and nanowires and to graphene nanoribbons.
Journal of Applied Physics, 110(8):083713, 2011.
[3] G. Kresse and J. Furthmüller. Efficient iterative schemes
for ab initio total-energy calculations using a plane-wave
basis set. Phys. Rev. B, 54:11169–11186, Oct 1996.
[4] D M Wood and A Zunger. A new method for diagonalising
large matrices. Journal of Physics A: Mathematical and
Fig. 2. The absolute squared wavefunction averaged along z General, 18(9):1343, 1985.
of the first conduction states of the armchair (left) and zigzag
(right) ribbons. Atom positions are shown for reference
12
Computational Methods for the Design of
Bioinspired Systems that Employ Nanodevices
Damien Querlioz, A. F. Vincent, A. Mizrahi, N. Locatelli, J. S. Friedman, D. Vodenicarevic
Institut d'Electronique Fondamentale, Univ. Paris-Sud, CNRS, 91405 Orsay, France
programming conditions can naturally implement

ABSTRACT
learning-capable “synapses” [1,3] and
Biological systems compute by exploiting the subthreshold magnetic oscillators can exhibit
rich physics of their natural “nanodevices”. In biosimilar synchronization features [4].
electronics, it is therefore attractive to design Computational electronics like Monte Carlo
bioinspired computing paradigms, which exploit methods can allow us to investigate these often
device physics more deeply than digital logic, in poorly understood regimes, as seen in the case of
order to try and approach the energy efficiency of MTJs in Fig. 2 [4]. The result can assist us in
biological systems. Here, we show how designing compact models appropriate for
computational tools can assist this research effort. designing bioinspired circuits and systems [4,6].
INTRODUCTION ARCHITECTURE LEVEL WORK
Nanodevices provide large opportunities for Bioinspired architectures often require a high
electronic systems, offering new features in an number of nanodevices to achieve complex tasks.
extremely energy efficient way. However, they Credible computational simulation is therefore
also possess less desirable features like variability. essential before a system can be fabricated. System
Biology can be an inspiration, in that it is able to level simulations that abstract some features of a
compute using very variable nanodevices in a design provide the possibility to test large
surprisingly efficient way, usually exploiting their architectures. However, unlike conventional
device physics deeply, and in highly parallel and architectures, it is essential to include a relatively
error-resilient architectures. It is therefore detailed description of device physics and not a
attractive to develop bioinspired computing highly abstracted version. This allows us to
paradigms, where device physics is at the core of investigate the impact of all nanodevices
computation. For example, bioinspired imperfections like variability, noise or drift effects,
programming of memory nanodevices can to which bioinspired systems are often found to be
naturally lead to complex inference engines, surprisingly tolerant [1], as seen in Fig. 3.
whose properties naturally emerge from device
physics, as seen in Fig. 1 [1]. In such approaches, CONCLUSION
computational electronics tools are useful to Computational electronics techniques can be
investigate device physics, and also to design full extremely useful tools to design bioinspired
architectures and investigate the impact of device architectures with nanodevices, as the latter tend to
properties on their performances. rely on device physics. It is also essential to
investigate the impact of nanodevices
DEVICE PHYSICS INVESTIGATION
imperfections, which system level Monte Carlo
When nanodevices are used for bioinspired simulations can address.
applications, they are often used in subthreshold or
near-threshold regimes where their behavior is ACKNOWLEDGMENT
richer and energy efficiency is higher. For This work is supported by Labex NanoSaclay (ANR-
example, transistors in weak inversion can 10-LABX-0035), ANR COGNISPIN (ANR-13-JS03-
naturally implement compact artificial neurons [2]. 0004-01), and FP7 ICT BAMBI (FP7-ICT-2013-C).
Magnetic tunnel junctions (MTJs, basic cells of
magnetic RAM) programmed in weak
ISBN 978-0-692-50554-0 13
REFERENCES
[1] D. Querlioz et al., “Bioinspired Programming of Memory
Devices for Implementing an Inference Engine”, Proc. IEEE,
vol. 103(8), pp. 1398-1416 (2015).
[2] G. Indiveri et al, “Neuromorphic silicon neuron circuits” ,
Front. Neurosci 5: 73 (2011).
[3] A.F. Vincent et al., “Spin-transfer torque magnetic
memory as a stochastic memristive synapse for neuromorphic
systems”, IEEE Trans. Biomed. Circuits Syst., vol. 9 (2),
p.166 (2015).
[4] A. Mizrahi et al, “Magnetic stochastic oscillators: Noise-
induced synchronization to under-threshold excitation and
comprehensive compact model, IEEE Trans Mag. (2015)
[5] A.F. Vincent et al., “Analytical macrospin modeling of
the stochastic switching time of spin-transfer torque devices”,
IEEE Trans. Electron Devices, vol. 62 (1), pp.164-170
[6] N. Locatelli et al, “Vortex-Based Spin Transfer Oscillator Fig. 2. Monte Carlo simulations of magnetization dynamics
Compact Model for IC Design”, Proc. of ISCAS (2015). allows studying the behavior of MTJs in weak programming
conditions. Plot: mean switching time as a function of
programming current. Inset: detailed statistics of switching
times for several programming currents.
Fig. 1. (a) Basic and simplified schematic of a bioinspired

classifier, where memory nanodevices implement artificial
synapses. The detailed subthreshold or near-threshold physics
of the nanodevices allows the system to learn tasks such as (b)
handwritten character recognition or (c) car detection task Fig. 3. Monte Carlo simulations of a full bioinspired system
within a video. performing car detection task using MTJs in weak
programming conditions. The simulations include a detailed
analytical model of device physics that reproduces Fig. 2, and
can include device variations. (a) Detection rate and (b)
proportion of false positives on the full task as a function of
device variability. Impressive robustness is seen (industrial
MTJs have a typical device variability of 5%).
14
Short-wavelength spin-wave generation by a
microstrip line
Á. Papp*+, G. Csaba*, and W. Porod*,
*Center for Nano Science and Technology, University of Notre Dame, Notre Dame, IN 46556, USA
+Pazmany Peter Catholic University, Budapest, Hungary
e-mail: {apapp,porod,gcsaba}@nd.edu
contribute to the wave generation, which results in

INTRODUCTION
anisotropic spin-wave generation (Fig. 2).
Spin-wave-based devices are currently studied We found that the wavelength of the generated
as possible candidates for beyond-CMOS spin waves is independent of the physical
computing, and both spin-wave-based logic gates dimensions of the MSL in the investigated range
[1] and non-Boolean, wave-based approaches [2] of the physical dimensions, it only depends on the
have been proposed. Spin waves can be generated applied biasing magnetic field Bbias and the
in a number of ways, including spin torque frequency f, as can be seen in Fig. 3 and Fig. 4.
oscillators (STOs) [3], magnetoelectric (ME) cells However, there is a strong correlation between
[4], and waveguides. Exchange-dominated short- the spin-wave amplitude and the physical
wavelength (λ<100 nm) spin waves have to be dimensions of the MSL. Figs. 5-7 show the
generated by nanoscale structures, where the amplitude of the generated spin waves in function
antenna sizes are comparable to the wavelength. of the MSL dimensions. We also found that
Here, we study nanoscale microstrip lines placing the MSL centered at the edge of the
(MSL) as potential sources of exchange-dominated magnetic film the spin-wave amplitudes are
spin waves. Since MSLs can generate wide considerably higher at small wavelengths
wavefronts, they can be energy efficient compared compared to placing it in the middle.
to point sources (such as STOs). They are also
more straightforward to design and fabricate. For CONCLUSION
proof-of-principle experiments, waveguides can be We are not aware of any prior work using
combined with Brillouin Light Scattering (BLS) waveguides for generating short-wavelength
techniques [3] with no electromagnetic coupling exchange waves. Our simulation studies show that
between the generation and the measurement this relatively straightforward device is a
sides. promising experimental test system and may
potentially serve in novel devices as well.
RESULTS
We used micromagnetic simulations to estimate ACKNOWLEDGMENT
the intensity and wavelength of spin waves We thank Giovanni Carlotti (University of
generated by a MSL depending on multiple Perugia) for stimulating discussions.
parameters. The geometry of the setup is depicted
REFERENCES
in Fig. 1. To saturate the magnetic film, a magnetic
[1] A. Khitun, M. Bao and K. L. Wang, Magnonic logic
field Bbias is applied in-plane parallel to the line. circuits, J. Phys. D: Appl. Phys. 43, 264005S (2010).
As a first approach, we calculated the magnetic [2] G. Csaba, A. Papp and W. Porod, Spin-wave based
field generated by the MSL applying Ampere’s realization of optical computing primitives, J. Appl. Phys.
law numerically, ignoring the effect of the ground 115, 17C741 (2014).
plane and the dielectric, and assuming [3] M. Madami et al, Direct observation of a propagating
spin wave induced by spin-transfer torque, Nature
homogenous current distribution in the wire. For Nanotechnology 6, 635–638 (2011).
the simulations, we assumed a 1 mA current. [4] S. Cherepov et al, Electric-field-induced spin wave
The magnetic field generated by the MSL has generation using multiferroic magnetoelectric cells, Appl.
two components: Bx is symmetric and Bz is Phys. Lett. 104, 082403 (2014).
antisymmetric. We found that both components
ISBN 978-0-692-50554-0 15
Fig. 1. Geometry of the MSL (cross section).
Fig. 5. Wavelength and amplitude of the generated spin

waves in function of microstrip width w (Bbias = 150 mT, f =
15 GHz, d = 100 nm, t = 50 nm, constant current)
Fig. 2. Anisotropic spin-wave generation by the MSL

centered at x = 0 µm (green), and spin waves generated by
applying only the x (blue) or z (red) field component of the
line.

waves in function of microstrip thickness t (Bbias = 150 mT, f
= 15 GHz, w = 100 nm, d = 100 nm, constant current density)

waves in function of Bbias (f = 15 GHz, w = 100 nm, d = 100
nm, t = 50 nm)

waves in function of dielectric thickness d (Bbias = 150 mT, f =
15 GHz, w = 100 nm, t = 50 nm)

waves in function of frequency f (Bbias = 150 mT, w = 100
nm, d = 100 nm, t = 50 nm)
16
Diffusion-Reaction Modeling of Cu Migration
in CdTe Solar Devices
D. Guo1, D. Brinkman2, T. Fang3, R. Akis1, I. Sankin3 D. Vasileska1 and C. Ringhofer2
1
School of ECEE, Arizona State University, Tempe, AZ, USA
2
School of Mathematics, Arizona State University, Tempe, AZ, USA
3
First Solar, Perrysburg, OH, USA
In this work, we report on development of is the amount of carried charge, GF is the free en-
one-dimensional (1D) finite-difference and two- ergy of formation, and the electrostatic potential is
dimensional (2D) finite-element diffusion-reaction provided by φ. Due to very high diffusivities of
simulators to investigate mechanisms behind Cu- free carriers, we considered them reaching steady-
related metastabilities observed in CdTe solar cells state concentrations at every time step when solv-
[1]. The evolution of CdTe solar cells performance ing (5) for much slower ionic species.
has been studied as a function of stress time in re- In our 2D FEM scheme, anisotropic diffusion
sponse to the evolution of associated acceptor and model for a single grain boundary of width δ lo-
donor states. To achieve such capability, the simu- cated at x=0 is utilized, and is given by
lators solve reaction-diffusion equations for the dC d ( D d [C ] dx) d ( D d [C ] dy )
defect states in time-space domain self-   (6)
dt dx dy
consistently with the free carrier transport. Re-
sults of 1-D and 2-D simulations have been com-  Dgb ,
 | x |  / 2
where D   (7)
pared to verify the accuracy of solutions.  Dg ,
 | x |  / 2
Evolution of concentration profile for an arbi-
For a finite difference mesh with size h=δ and a
trary defect C is simulated by solving reaction-
symmetric grain boundary, this model gives iden-
diffusion equation [2] tical discretization to the traditional Fisher model
dC dJ C [4].
   RC (1)
dt dx In summary, the diffusion-reaction model has
In (1), RC represents the net reaction rate of the de- been applied to Cu-related defects in CdTe PV de-
fect C, and for the case of Cu is calculated based vices. The simulation results from this study give
on the main reactions responsible for p-type dop- us a deeper understanding of the role of Cu on the
ing formation in CdTe [3]: performance of CdTe solar cells. The 1D diffu-
Cui  CdCd  CuCd  Cdi (2) sion-reaction code is installed on nanoHUB.org.
Cui  VCd  CuCd (3) ACKNOWLEDGMENT
In reaction (2), also known as the knock-off reac- This work was supported by the Department of
tion, forward and backward rates are calculated as Energy under award number DE-EE0006344.
 R
 f  K C
f s [Cu i ] REFERENCES
 (4)
 Rb  Kb [CuCd ][Cdi ]
 [1] K. D. Dobson, I. Visoly-Fisher, G. Hodes, and D. Cahen,
"Stability of CdTe/CdS thin-film solar cells," Solar
In (4), CS stands for the concentration of cation Energy Materials and Solar Cells, vol. 62 (2000).
sites in CdTe, while Kf and Kb represent tempera- [2] P. M. Fahey, P. B. Griffin, and J. D. Plummer, "Point de-
ture-dependent rate constants. The expression defects and dopant diffusion in silicon," Reviews of Modern
scribing the flux of species C in the gradient of its Physics, vol. 61 (1989).
electrochemical potential is given below. [3] S. Hu, "Nonequilibrium point defects and diffusion in sili-
con," Materials Science and Engineering: R: Reports, vol.
 C 
 
 13 (1994).
J C   DC    GF      C  (5) [4] J. C. Fisher, "Calculation of diffusion penetration curves
 kT
 
 for surface and grain boundary diffusion," Journal of Ap-
In (5), diffusivity of the defect C is given by DC, θ plied Physics, vol. 22 (1951).
ISBN 978-0-692-50554-0 17
INPUT CONDITIONS MODEL INPUTS
• Initial conditions (distributions) • Set of included models
1ST PRINCIPLE
• Stress conditions (light, bias, • Set of included reactions DFT calculations
Initial conditions (distributions)

temperature, time) • Device stack (opt.)
CORE SOLVER
• Defect / carrier kinetics Feedback
• Device simulations
SIMULATED OUTPUTS EXPERIMENTAL VERIFICATION

Evolutions of: Evolutions of:
• Device characteristics (I-V, C-V, QE-V, etc.) • Device characteristics (I-V, C-V, QE-V, etc.)
• Defect/complex/charge distributions • Charge segregation GBs, etc. (spec. methods)
• Atomic profiles • Atomic profiles (ToF-SIMS, GDOES, etc.)
Fig. 1. Unified diffusion-reaction/device solver – multi-scale computation with feedback (https://nanohub.org/tools/predicts1d/).
Fig. 4. Comparison of the concentration of Cui, Cdi and CuCd for the
1D finite-difference scheme and the 2D finite-element method
Fig. 2. Defect profiles and corresponding doping profiles simulated scheme finished by a standard 200oC 220s anneal from a finite Cu
after a 200oC 220s anneal and two-hour open circuit stress in a stand- source. Although the time step employed for the 2D case is 100 times
ard CdS/CdTe solar cell. Extra Cui moved into CdTe layer due to larger than the 1D time step, the difference between these profiles is
small built-in electric field under OC condition, which reduced the p- negligible.
type net doping during the stress.
Fig. 5. The 2D FEM scheme allowed the inclusion of grain-boundary

Fig. 3. Comparison between simulated and experimental IV curves of enhanced diffusion of Cu in the simulation. A fork-shaped grain-
the CdTe solar cell before and after the stress. Loss in p-doping boundary was assumed in this 2D simulation. Clear segregation of Cu
caused the reduction of VOC, hence the conversion efficiency. along the boundary was achieved.
18
Oscillations in Exchange-Coupled Nanomagnets
H. Dey, G. Csaba, G. H. Bernstein, and W. Porod
Center for Nano Science and Technology, University of Notre Dame, USA
domain magnets in order to understand the basic

INTRODUCTION
modes (see Fig. 2). Using a numerical mesh of 5 nm
Spins in ferromagnetic materials undergo x 5 nm x 3 nm, a full micromagnetic mode reveals
precession when subjected to an oscillating the role of the internal oscillation modes (Fig 3).
magnetic field. This so-called ferromagnetic
resonance (FMR) has been a widely-used RESULTS
characterization technique for magnetic materials. The single-domain calculations straight-
Recently, FMR techniques have been forwardly show the emergence of symmetric and
extensively applied to nanostructured magnets, antisymmetric modes, which can be tuned by the
where both shape and magnetostatic interactions exchange-coupling strength and the external field
play a role in shaping the FMR spectrum [1]. In this (Fig. 2). Considering the internal degree of freedom
paper, we study nanomagnets that are patterned for the nanomagnets, it is already known to result in
from exchange-coupled layers. In this case, the a number of additional resonances. Our simulations
interlayer exchange coupling will also have a role clearly show that the number of possible modes
in determining the resonance behavior of the multiplies due to interlayer coupling (Fig. 3). The
magnets. exchange coupling also alters the possible static-
Magnetic oscillations are currently being studied domain configurations, resulting in modes that are
for possible applications in computing devices [2], different from the known center and edge modes
and the structures studied here may find [3]. We are currently studying these modes both
applications there. experimentally and, via further simulation and
analytical methods.
PHYSICAL STRUCTURE AND EXPERIMENTAL
CHARACTERIZATION CONCLUSION
We used sputtering to fabricate the magnetic We are not aware of any studies on the dynamic
multilayers. Top and bottom ferromagnetic layers behavior of exchange-coupled and patterned films.
were 10 nm CoFe and 3 nm CoFe, with a 0.8 nm Besides being an interesting model system, such
thin Ru spacer layer sandwiched between them. The systems can be engineered to display a wide variety
resulting stack was measured in a vibrating sample of resonance phenomena, and can have applications
magnetometer (Fig. 1). From the M-H curve, we in oscillatory computing architectures.
obtained the exchange field (~ 100 mT), which was
ACKNOWLEDGMENT
used to extract the coupling strength ‘J’. We used
this ‘J’ in our simulations. The film was We are grateful to Hans-Werner Schumacher
subsequently patterned to 200 nm x 100 nm and Xiukun Hu (PTB, Braunschweig) for
nanomagnets for FMR measurements in the near experimental collaboration.
future. REFERENCES
NUMERICAL METHOD [1] Nembach, Hans. et al. "Mode-and size-dependent Landau-
Lifshitz damping in magnetic nanostructures: evidence for
Micromagnetic simulations were carried out nonlocal damping." Phys rev lett 110, no. 11 (2013):
using OOMMF in order to understand the 117201.
resonance modes of the coupled system. For [2] Papp, A., W. Porod, and G. Csaba, "Hybrid yttrium iron
different external fields (along easy axis), a small garnet-ferromagnet structures for spin-wave devices," J
Appl Phys 117, no. 17 (2015): 17E101.
oscillating field was applied perpendicular to the [3] Carlotti, G., S. Tacchi, G. Gubbiotti, M. Madami, H. Dey,
external field direction. Frequency of the oscillating G. Csaba, and W. Porod, "Spin wave eigenmodes in single
field was varied from 1-15 GHz. First, top and and coupled sub-150 nm rectangular permalloy dots," J
bottom CoFe layers were approximated as single- Appl Phys 117, no. 17 (2015): 17A316.
ISBN 978-0-692-50554-0 19
Figure 1: M-H curve for the exchange coupled stack (Green arrow: Magnetization of top
layer, Red arrow: Magnetization of bottom layer), stack geometry shown in inset.
Figure 2: Simulated FMR spectra for three different cases (10 mT external field applied
along easy axis); symmetric and antisymmetric modes for J=0 and J= 0.5 erg/cm2 are
zoomed in and shown (single-domain approximation).
Figure 3: Full micromagnetic model showing internal oscillation modes for the same
three cases shown above in Fig. 2; simulated magnetizations of different layers also
shown.
20
Computational Study on Interfacial Phase Change
Memory by Topological Superlattices
H. Nakamura*1,2, I. Rungger3, A. Narayan3, S. Sanvito3, Y. Asai 1,2, and J. Tominaga1,2
1
National Institute of Advanced Industrial Science and Technology (AIST), Umezono 1-1-1, Tsukuba, 305-8568, Japan
2
CREST-JST, 7 Gobancho, Chiyoda-ku, Tokyo 102-0076, Japan
3
School of Physics, MBER and CRANN, Trinity College, Dublin 2, Dublin, Ireland
ABSTRACT
We performed first principles simulation of CALCULATION OF IPCM DEVICE PROPERTIES
interfacial phase change memory device, which We adopted the model device structure
consists of GeTe-Sb2Te3 superlattice (GTS) and analyzed W/(QL)2(Ge2Te2)(QL)2/W, where QL
W(111) electrodes. On the contrary of standard represents the quintiple layer, i,e, Sb2Te2 unit. As
phase change memory, SET/REST switch is possible GeTe structures, we took three phases,
possible only by phase change of the interface inverted-Petrov, Ferro-GeTe, and Petrov, namely,
GeTe layer, i.e., crystalline-crystalline transition.[1] as shown in Figure 1. The Petrov phase is
To argue the high/low resistance states (HRS/LRS), considered as HRS while the last two phases are
we examined three model device structures, i.e., candidates of LRS. In Figure 2, our first-principles
inverted-Petrov, Petrov, and Ferro-GeTe phase, calculations of IV are presented for the above three
respectively, and carried out first-principles phases. As shown in Fig. 2, Petrov phase provides
nonequilibrium Green’s function theory better agreement of the resistive ratio of HRS/LRS
(NEGF)[2,3]. We present the mechanism of resistive with the experiments than that of Ferro-GeTe
switch and importance of spin-orbit interaction. In phase. Spin-orbit effect is important to ON/OFF
addition, we discuss application of 3D current ratio as shown in Figure 3. Since spin-orbit
homologous topological insulator to functional effect to IV characteristic is large only for HRS,
device. we calculated quasiparticle spectra, which is
defined as the pole of Green’s function on the
INTRODUCTION
QL/GeTe interface, as a function of energy and
The phase change memory (PCM) is one of momentum (k parallel to the interface). The result
most promising nonvolatile information storage is presented in Figure 4, and Dirac semi-metallic
technologies. Recently, the GTS structure is dispersion is observed.
proposed as PCM and is called interfacial PCM
(iPCM). SET and RESET states are realized only CONCLUSION
by the flip-flop transition of Ge atoms in crystal Our simulation confirms that distinct resistive
phase. Furthermore, the Sb2Te3 is topological switch is possible by only GeTe layer phase
insulator, and it is blocked by GeTe normal change. Spin-orbit effect changes resistance of
insulator; thus it is expected that the topological inverted Petrov phase largely, which relates to
states is found on each GeTe/Sb2Te3 interface in HRS; thus NEGF with spin-orbit coupling is
GTS[4] although GTS is embedded in the required to evaluate SET/RESET resistance
electrodes as the device cell. In the present study, qualitatively. GTS of inverted-Petrov phase in
we identify HRS/LRS states and topological device cell has characteristic quasiparticle
property in device cell by calculating Green’s dispersion at each interfacial 2D edge of
functions. Then we discuss relation of device topological material like Dirac semi-metal.
function and Dirac cone (and/or semi-metallic
REFERENCES
behavior) dispersion, which is feature of
[1] R. E. Simpson et al., Nat. Nanotech. 6, 501 (2011)
topological insulator. [2] H. Nakamura et al., Phys. Rev. B 78, 235420 (2008)
[3] A. R. Rocha et.al, Phys. Rev. B 73, 085414 (2006)
[4] B. Sa, et al., Phys. Rev. Lett. 109, 096802 (2012)
ISBN 978-0-692-50554-0 21
Figure 1. The models of GTS structures of the iPCM device.
The left pannel is schematic view of GTS. The right panel
consist of inverted-Petrov, Ferro-GeTe, and Petrov Phase, Figure 4. The contour plot of quasiparticle energy
respectively. The thick grey, thin grey, and purple circles dispersion, which relates to the 2D band dispersion at
represent Sb, Te, and Ge atom, respectively. The device cell Sb2Te3/GeTe interface of inverted-Petrov phase. The
consists of W(111) electrode layers. Fermi level of W electrode is set to zero. The right blue
dot line circle is eye guide to figure out Dirac cone-like
structure.
Figure 2. First principles IV characteristics of GTS iPCM

device for the three GeTe phases. HRS is inverted-Petrov
structure. Spin-orbit interactions are included in the
transport calculations.
Figure 3. The plot of ON/OFF current as a function of bias

voltage. On current is LRS of Petrov phase, and OFF current is
obtained by HRS (inverted-Petrov). The red dot line is the
result calculated by including spin-orbit interaction while the
black dot line is obtained by omitting spin-orbit interaction.
22
Design and simulation of GaSb/InAs 2D
Transmission enhanced TFET
Pengyu Long, Evan Wilson, Jun Z. Huang, Mark Rodwell*, Gerhard Klimeck and Michael
Povolotskyi
Purdue Univeristy, West Lafayette, Indiana, USA 47906
*
University of California Santa Barbara, Santa Barbara, California, USA 93106
doping. The transmission enhanced TFETs are PN

INTRODUCTION
heterojunction TFETs with three GaSb barriers in
Tunneling Field Effect Transistors (TFETs) the channel. The GaSb barriers are 3, 5.5, and 3
have been widely analyzed for potential use in monolayers (ml) thickness, and are separated by
low-power applications because of its steep InAs layers 9.5 ml thickness. The 3-barrier
subthreshold swing (SS). However, the on-currents structure is 16.5ml away from GaSb/InAs tunnel
(Ion) of tunneling FETs are much lower than that of junction..
to conventional MOSFETs because of low Ballistic transport properties are simulated by
tunneling probability [1]. Quantum transmitting boundary method (QTBM)
Avci [2] reported a modified TFET, adding a and band structure described by sp3d5s* tight
second barrier to introduce a resonant, bound state. binding basis [5]. The QTBM and Poisson
As this bound state is brought into resonance with equations are solved self-consistently, using
the source, the transmission peaks sharply, thus NEMO5 nanoelectronics modeling software [6].
deceasing (improving) the subthreshold swing.
Here we show supplementary barriers added to a RESULTS
TFET can improve the electron transmission over The width of wells and barriers was adjusted
a broad energy range, thus improving on-current. empirically with a series of simulations to peak the
Electron transmission enhancement is similar transmission over a broad energy range, thus
[3] to the design of optical anti-reflection coatings increasing the device Ion . With the addition of the
and of microwave impedance-matching networks. 3-barrier structure, the transmission is increased to
An electron wave incident on the PN tunnel barrier 71% at 210 meV incident energy range, and is
has some transmission probability less than unity, greater than 7% over an energy range from 207 to
and hence a nonzero wave reflection probability. If 275 meV and from 316meV to 340meV (Fig. 2).
a secondary reflector is added, the reflections from Fig. 3 shows the energy resolved current density..
the PN junction and of the secondary reflector will Transfer characteristics are shown in Fig. 4.
add and interfere; with appropriate adjustment of Using the ITRS general purpose (GP) logic off-
the barrier thickness and position, the reflections current specification of 10-3 A/m, and selecting
will add destructively, giving zero reflectivity, and a 0.3V supply voltage (hence Vgs=0.3V in the on-
100% transmission, at one particular energy. With state), the SS improved from 32mV/dec. to
multiple layers, the energy range of high 25mV/dec. and Ion is increased from 36 A/m to
transmission can be increased; there are, however, 102 A/m.
bounds [4] to energy range over which
transmission is high. CONCLUSION
Supplementary barriers added to a TFET can be
MODEL
designed such that they introduce reflections
The device is based on a p-GaSb/n-InAs double which interfere destructively with the reflection
gate TFET with 3.2nm thick body and 2.56nm from the main PN tunnel barrier, thereby reducing
thick gate oxide (  r =20), 20nm gate length, the net electron reflectivity. The tunnel barrier
-3
51019 cm source doping and 2  1019 cm
-3
drain transmission coefficient, and hence TFET on-
current, can thereby be increased. Multilayer
ISBN 978-0-692-50554-0 23
designs permit the transmission to be enhanced TFET, which has three supplementary reflectors inside the
over a broad energy range, thereby significantly channel.
increasing the TFET on-current.
ACKNOWLEDGMENT
The use of nanoHUB.org computational
resources operated by the Network for
Computational Nanotechnology funded by the US
National Science Foundation under Grant Nos.
EEC-0228390, EEC-1227110, EEC-0228390,
EEC-0634750, OCI-0438246, OCI-0832623 and
OCI-0721680 is gratefully acknowledged. This
material is based upon work supported by the
National Science Foundation under Grant Number Fig. 2. Left: Band diagram of a transmission enhanced TFET
(1125017). NEMO5 developments were critically and conventional TFET at ON-state. Right: Transmission
supported by an NSF Peta-Apps award OCI- probability at the zoomed-in energy range of BTBT at
transverse wave vector k=0
0749140 and by Intel Corp.
REFERENCES
[1] Seabaugh A C, Zhang Q. Proceedings of the IEEE, 2010,
98(12): 2095-2110.
[2] Avci U E, Young I A.. IEDM Tech. Dig, 2013: 96-99.
[3] A. N. Khondker, M. Rezwan Khan, A. F. M. Anwar,
Journal of applied physics 63.10 (1988): 5191-5193.
[4] R. M. Fano, J. Franklin Inst., vol. 249, pp. 57-83, Jan.
1960; and pp. 139-155, Feb. 1960.
[5] Luisier M, Schenk A, Fichtner W, et al. Physical Review
B, 2006, 74(20): 205323.
[6] J.E. Fonseca, T. Kubis, M. Povolotskyi, B. Novakovic, A.
Ajoy, G. Hegde, H. Ilatikhameneh, Z. Jiang, P. Sengupta,
Y. Tan, G. Klimeck Journal of Computational Electronics
12.4 (2013): 592-600. Fig. 3. Energy resolved current density at ON-state for
conventional GaSb/InAs heterojunction TFET (left) and
transmission enhanced TFET (right).
GaSb InAs InAs
oxide
GaSbmetal
barriers
3.3nm
20nm
GaSb barriers
Fig. 4. Transfer characteristics of conventional GaSb/InAs

GaSb InAs InAs heterojunction TFET compared with transmission enhanced
TFET aligned at GP off-current specification of 10-3 A/m.
oxide
metal
3.3nm
20nm
Fig. 1. Cross-section view of aconventional Double Gate

GaSb/InAs heterojunction TFET and transmission enhanced
24
Self-Consistent Physical Modeling of Silicon-
Based Memristor Structures
T. Sadia, L. Wanga, L. Gerrera, and A. Asenova,b
a
Device Modeling Group, School of Engineering, University of Glasgow, Glasgow G12 8LT, Scotland,
UK. b Gold Standard Simulations Ltd, Glasgow G3 7JT, Scotland, UK.
demonstrated experimentally by our collaborators

ABSTRACT
[1]. GARAND evaluates accurately the electric
We employ a newly-developed three- field and potential distributions by coupling the
dimensional (3D) physical simulator to study Si solution of Poisson's and density-gradient
resistive switching nonvolatile memory (RRAM) equations. Electrothermal coupling is achieved by
structures. We couple a stochastic simulation of solving the time-dependent HDE using the local
ion transport to the ‘atomistic’ simulator Joule heating. In the simulation process, we update
GARAND and a self-heating model to explore the the traps' electron occupancies, by evaluating the
switching processes in these devices. The model hopping rates between the traps and the tunneling
represents a significant improvement compared to rates between the traps and the electrodes, which
the phenomenological models based on the resistor are used to calculate the steady-state current [2].
breaker network. The simulator is calibrated with Fig. 2 illustrates the simulation procedure.
experimental data, and reconstructs nicely the
conductive filament (CF) formation and rupture in RESULTS, DISCUSSION AND CONCLUSION
the 3D space. A 3D volume under a poly-Si plate has been
INTRODUCTION simulated, by considering an oxide thickness of
10nm and a simulation contact area of 40nm long
Nonvolatile memories based on resistive and 20nm wide. Fig. 3 shows the I-V
switching (memristors) offer improved characteristics of the device, Fig. 4 shows typical
controllability of charge in semiconductor temperature maps, and Fig. 5 shows the vacancy
memories at small sizes [1]-[3]. The development distributions up to the CF formation. At biases
of memristor technology allows low cost-per-bit, below 10V, few vacancies are created, giving very
low power dissipation, and high endurance. In low currents. At around 10V, filament seeds
addition, memristors can be integrated in crossbar appear and grow as bias is increased to 16V. At
arrays stacked in multiple levels in the 3D space around 17V, an accelerated generation of oxygen
[4]. The development of Si memristors, in vacancies occurs, forming a CF and bridging the
particular, will result in a breakthrough in low-cost p-Si substrate and the poly-Si terminal. The CF
on-chip integration with Si microelectronics [1]. In formation gives rise to an abrupt jump in the
this work, we present a self-consistent current, corresponding to the transition from the
electrothermal simulation of intrinsic resistive high resistance state (HRS) to the low resistance
switching in Si RRAM devices. Previous state (LRS). Temperature rise due to self-heating
simulation work has relied on phenomenological (up to 250K) plays a major role, as it affects the
models based on the resistor breaker network [5]. oxygen ion/vacancy generation probability and
Furthermore, most work focuses on resistive diffusion. A more detailed analysis will be
switches based on metal oxides (e.g. HfO2). presented to show the effect of self-heating, oxide
MODEL thickness, and excess Si percentage on the device
We couple a kinetic Monte Carlo (KMC) operation, including the electroforming, set and
simulation of ion transport to GARAND [6] and a reset processes of the switching behavior.
time-dependent heat diffusion equation (HDE) ACKNOWLEDGMENT
solver to capture switching in the Si structure
We would like to thank the EPSRC (UK) for
illustrated in Fig. 1, whose operation has been
funding under the grant EP/K016776/1.
ISBN 978-0-692-50554-0 25
(a)
Fig. 1. The experimental memristor structure from Ref. [1].
(b)
Fig. 2. The simulation framework.
(c)
Fig. 5. The vacancy distribution within the oxide, for (a) 10V,
(b) 15V, and (c) 17V bias voltages.
REFERENCES
[1] A. Mehonic et al., “Resistive switching in silicon sub-
Fig. 3. The I−V characteristics up to the CF formation. The oxide films”, J. Appl. Phys., vol. 111, p. 074507, 2012.
currents are given for an equivalent width of 1mm. [2] S. Yu et al., “On the stochastic nature of resistive
switching in metal oxide RRAM: Physical modeling, Monte
Carlo simulation, and experimental characterization”. In
Electron Devices Meeting (IEDM), p. 17.3., 2011.
[3] D. B. Strukov et al., “The missing memristor found”,
Nature, vol. 453, pp. 80-83, 2008.
[4] The International Technology Roadmap for
Semiconductors [Online]. Available: http://www.itrs.net/
[5] S. C. Chae et al., “Random circuit breaker network model
for unipolar resistance switching”, Advanced Materials, vol.
20, pp. 1154-1159, 2008.
Fig. 4. Typical temperature [6] GARAND Statistical 3D TCAD Simulator [Online].
maps, at x=0 (through the Available: http://www.goldstandardsimulations.com/
middle of the structure), as
bias is increased (10V,
~16V then just below 17V).
26
Dissipative quantum transport using one-particle
time-dependent (conditional) wave functions
X.Oriols, Z. Zhen, E.Colomés
Departament d’Enginyeria Electrònica, Universitat Autònoma de Barcelona, Bellaterra 08193, Spain
I NTRODUCTION freedom are substituted by Bohmian trajectories ex-

A quantum device is a many-body open system cept ra . It has been demonstrated [6] that Ψa (ra , t)
driven far from (thermodynamic) equilibrium. For obeys the following wave equation:
ballistic systems, these difficult proble is tradition- ∂Ψa (ra , t) h̄2 2
ally solved from the one-particle Schrödinger equa- ih̄ = − ∇ + Ua (ra , R α (t), t)
∂t 2m a
b
tion which only captures non-dissipative (elastic)
α (t), t) + iJa (ra , R
+Ga (ra , R α (t), t) Ψa (ra , t). (2)
scattering of electrons with external potentials. In b b
principle, one would like to include dissipation
where the Bohmian trajectories are denoted by
into the previous one-particle wave functions of α (t)} with α an index for the pos-
aα (t), R
R(t) = {R
the electron. However, strictly speaking, such one- b
sible initial positions [1]. The term Ua (ra , rbα (t), t)
particle wave functions (for a quantum subsystem)
can be any type of many-particle potential defined in
do not exist in the Copenhagen formulation of
the position-representation. The explicit expression
quantum phenomena [1]. Thus, much less intu-
of the potentials Ga (ra , rbα (t), t) and Ja (ra , rbα (t), t)
itive mathematical entities are required: the density
can be found in reference [6]. It is important to spec-
matrix [2], its Wigner-Weyl transformation named
ify that (2), contrarily to the orthodox Schrödinger
Wigner distribution function [3] or the so-called
equation, provides a non-linear and non-unitary evo-
non-equilibrium Greens function (NEGF) formal-
lution of the one-particle wave function Ψa (ra , t)
ism [4]. Fortunately, the concept of one-particle
that allows the modeling of dissipation, decoherence
wave function (of a quantum subsystem) including
and (collapse) measurement with minimum compu-
dissipation can be rigorously formulated within the
tational difficulties [7].
Bohmian explanation of quantum phenomena [5].
O NE - PARTICLE CONDITIONAL WAVE FUNCTIONS E LECTRON -P HONON INTERACTION
In this conference, we show how dissipation in As an example, we consider the electron-phonon
quantum transport can be rigorously tackled from interaction in a typical AlGaAs resonant tunneling
the (conditional) wave function, with the com- structure. In particular we consider emission and
putational burden limited to solving one-particle absorption of optical phonons with energy h̄ωop =
wave functions with time-dependent potentials. Let 0.03613 eV . The scattering rates are directly ob-
us consider a many-particle wave function Φ(r, t) tained from the Fermi Golden Rule. Under the
solution of the unitary Schrödinger equation where assumption that the spatial dependence of the per-
r = {ra , rb } with ra the position of the a−electron turbing potential is smaller than its temporal depen-
and rb = {r1 , ...., ra−1 , ra+1 , ...., rN } the positions dence, it can be demonstrated that Ja (ra , rbα (t), t)
of the rest. Bohmian mechanics defines the so-called in (2) is negligible and Ga (ra , rbα (t), t) can be
conditional wave function [1], [6] as: easily approximated to satisfy the electron-phonon
α (t), t), energy conservation. The initial (quantum ensem-
Ψa (ra , t) = Φ(ra , R (1)
b ble) energy E of Ψa (ra , t) is modified by
which constitutes a slice of the whole multi- Ga (ra , rbα (t), t) until its final value E ± h̄ωop . See
dimensional wave function Φ(r, t). All degrees of figures 1 and 2 and the explanation therein.
ISBN 978-0-692-50554-0 27
C ONCLUSIONS
In this conference, we present a new formalism
Conditional wave−function modulus

to study dissipative transport with one-particle (con-
ditional) wave functions. This new feature has been 4
x 10
7
included, as the last step, into the general, versatile 3

time-dependent BITLLES simulator [8]. Previous 2
features included into the BITLLES are electron- 1
electron interaction beyond mean field [9], open 0
system boundaries [10], exchange (fermion) inter- 5
4 2.5
actions [11], [12], noise computation [7], [13] and −7 3 2
x 10 1.5
high-frequency displacement currents [14]. This last 2
1 x 10
−12
1
ingredient, the quantum dissipation, clearly justifies 0.5
Position (m) 0 0 Time(s)
why the BITLLES simulator is the natural (full)
quantum extension of the Monte Carlo algorithm
for the semi-classical Boltzmann equation, where Fig. 1. (Red) Time-evolution of a wave function and associated
the dynamics of the electron trajectories are guided trajectories solution of the one-particle Schrödinger equation
and (Blue) Time-evolution of a (conditional) wave function and
by the one-particle (conditional) wave function dis- associated trajectories solution of the one-particle wave in (2)
cussed in (2) rather than by the electric field. where the absorption of a phonon is considered. (Green) the
free space potential. Initial wave packet energy E = 0.05eV .
ACKNOWLEDGMENT The (blue) conditional wave function and its trajectories have
a larger velocity (kinetic energy) than the (red) Schrödinger
This work has been partially supported by the wave function after the interaction with the phonon. From a
“Ministerio de Ciencia e Innovación” through the computational point of view, only a proper shape of the term
Spanish Project TEC2012-31330 and by the Grant Ja (ra , rbα (t), t) in the time-dependent wave equation (2) is
agreement no: 604391 of the Flagship initiative required to discuss dissipation.
“Graphene-Based Revolutions in ICT and Beyond”.
R EFERENCES
[1] A. Benseny, G. Albareda, A.S. Sanz, J. Mompart, and X.
Conditional wave−function modulus
Oriols, European Physical Journal D, 68, 286 (2014). x 10

7
[2] G. Lindblad, Commun. Math. Phys. 48 (2): 119 (1976). 7
[3] W. R. Frensley, Rev. Mod. Phys. 62, 745 (1990). 6
[4] R Lake, G Klimeck, RC Bowen, D Jovanovic Journal of 5

Applied Physics 81 (12), 7845-7869 (1997). 4
[5] X.Oriols and J.Mompart, Applied Bohmian mechanics, 3
Editorial Pan Stanford (2012). 2
[6] X. Oriols, Phys. Rev. Let. 98(6), 066803, (2007).
1
[7] D. Marian, E. Colomés, Z. Zhan, X. Oriols, Journal of
0
Computational Electronics, in press (2015) 6 1
[8] http://europe.uab.es/bitlles 4
0.5 −12
[9] G. Albareda, H. López, X. Cartoixà, J. Suñé, X. Oriols, x 10
−7
2 x 10
Phys. Rev. B, 82, 085301, (2010). Position (m)

0 0
Time (s)
[10] G. Albareda, J. Suñé, and X. Oriols, Phys. Rev. B. 79,
075315, (2009).
[11] A. Alarcón, S. Yaro, X. Cartoixxà, and X. Oriols, Journal Fig. 2. The same data as in Fig. 1 for a double barrier
of Physics: Condensed Matter. 25, 325601 (2013) 1nm/4nm/1nm structure plotted in green. Barrier height 0.3 eV.
[12] D. Marian, E. Colomés and X. Oriols, J. Phys. Condens. Initial wave packet energy E = 0.1eV . The (blue) condi-
Matter (in press). arXiv:1408.1990v2 [quant-ph] tional wave function and its trajectories have a larger velocity
[13] F.L.Traversa et al., IEEE Transaction on Electron Devices, (kinetic energy) than the (red) Schrödinger wave function after
58, 2104 (2011). the interaction with the phonon. In addition, the interaction
[14] A.Alarcón and X.Oriols, Journal of Statistical Mechanics: with the phonon slightly increases the number of reflected
Theory and Experiment. P01051, (2009). trajectories. This is due to the fact that the energy of the electron
after interacting with the phonon deviates from the resonant
energy.
28
Analyzing Variability in Short-Channel Quantum Transport from Atomistic First Principles
Qing Shi1, Eric Zhu2, Leo Liu2, Yin Wang3 and Hong Guo1
1
Centre for the Physics of Materials and Department of Physics,
McGill University, Montreal, QC H3A 2T8 Canada
2
Nanoacademic Technologies Inc., 7005 Blvd. Taschereau, Brossard, QC J4Z 3A7
3
Department of Physics, University of Hong Kong, Hong Kong
Due to random impurity fluctuations, the device-to-device variability is a serious challenge to

emerging nanoelectronics. In this talk I shall present a theoretical formalism and its numerical
realization to predict quantum-transport variability from atomistic first principles. Our approach
is named the non-equilibrium coherent-potential approximation (NECPA) which can be applied
to predict both the average and the variance of the transmission coefficients such that
fluctuations due to random impurities can be predicted without lengthy brute force computations
of ensemble of disordered configurations. As an example, we quantitatively analyzed the off-
state tunnel conductance variability in Si nanosized field-effect transistor channels with channel
lengths ranging from 6.5 to 15.2 nm doped with different concentrations of boron impurity atoms.
The variability is predicted as a function of the doping concentration, channel length, and the
doping positions. The device physics is understood from the microscopic details of the potential
profile in the tunnel barrier. Time permitting, other systems will also be presented as examples.
ISBN 978-0-692-50554-0 29
30
Effect of the High-k Dielectric/Semiconductor
Interface on Electronic Properties in Ultra-Thin
Channels
D. Valencia, E.M. Wilson, M. J. W. Rodwell*, G. Klimeck, M. Povolotskyi
Network for Computational Nanotechnology, Birk Nanotechnology Center, Purdue University, West Lafayette,
Indiana 47907-1285, USA
*
Department of Electrical and Computer Engineering, UCSB, Santa Barbara, California
e-mail: [email protected], [email protected]
interface using EHT. Density of States (DOS)
ABSTRACT SUBMISSION
(Fig. X) for the interface cases are increased
As logic devices continue to downscale, an relative to the hydrogen-terminated UTB case due
increasing fraction of the channel atoms are in to the coupling of the semiconductor with the
close contact with oxide atoms of the gate. These oxide. It is noted that though the DOS is increased
surface atoms experience a chemical environment for the interface case (Fig. 6), the conduction
that is distinct from the bulk-like environment bands become nearly flat away from the Γ-point.
found in thicker channels. Using the non-
orthogonal tight-binding method Extended Huckel SUMMARY
Theory (EHT), III-IV/High-k dielectric interfaces EHT calculations were performed on
are constructed and electronic structure in the two semiconductor/oxide interfaces and band structure
transverse directions in the plane of the interface is and DOS results were obtained. These results are
calculated. then compared with corresponding results for
hydrogen-terminated UTB’s.
INTRODUCTION
In the ultra-small channels predicted by the ACKNOWLEDGMENT
continued scaling of transistors, an ever increasing This material is based upon work supported by
fraction will be in contact with the gate oxide the National Science Foundation under Grant
material as seen in Fig 1 based on ITRS Number (1125017). We would also like to
projections for single gate devices1. However, acknowledge the Rosen Center for Advanced
most existing atomistic models of semiconductor Computing at Purdue University for the use of
nanowires (NW) and ultra-thin bodies (UTB) use their computing resources and technical support.
hydrogen termination2-4 to approximate the oxide
REFERENCES
semiconductor interface. Semiconductor/oxide
[1]Wilson, Linda. "International Technology Roadmap for
interfaces are constructed using AtomistixToolKit Semiconductors (ITRS)."Semiconductor Industry Association
(ATK)5, minimizing strain between the two (2013).
surfaces as in Fig. 2. The structure is then relaxed [2] M. Luisier, A. Schenk, W. Fichtner, and G. Klimeck,
using ATK’s LDA DFT implementation. Atomistic simulation of nanowires in the sp3d5s* tight-
Electronic structure is then computed within our binding formalism: From boundary conditions to strain
calculations, Phys. Rev. B, 74, 205323, (2006).
own nanoelectronic modeling tool NEMO5. [3] S. R. Mehrotra, S. Kim, T. Kubis, M. Povolotskyi, M. S.
Where available, existing EHT parameters are for Lundstrom, and G. Klimeck, Engineering nanowire n-
semiconductors6. For oxides with no existing MOSFETs at Lg< 8 nm, IEEE Trans. Electron
parameterization, EHT parameters are fitted to Devices60,2171 (2013).
[4] C. Y. Yam et al., multi-scale modeling of junctionless
band structure and wave function DFT targets (Fig field-effect transistors Appl. Phys. Lett., 103, 062109(2013).
3) following the process described by Y. Tan et [5]AtomistixToolKit version 2014.2, QuantumWise A/S
al7. (www.quantumwise.com).
[6] J. Cerdá and F. Soria, Phys. Rev. B 61, 7965, Accurate
DISCUSSION and transferable extended Hückel-type tight-
Electronic structure (Figs. 3 and 4) is calculated bindingparameters (2000).
in the two transverse directions in the plane of the
ISBN 978-0-692-50554-0 31
[7] Y. Tan,et al., Empirical tight binding parameters for GaAs
and MgO with explicit basis through DFT mapping, J. Comp.
Elect 12, 56(2013).
Fig. 2 .InAs/TiO2 (rutile crystal phase) interface. This

structure was structurally relaxed with LDA DFT
Fig. 1. Band structure comparison of InAs/TiO2 with H-
terminated UTB
Fig. 4. Band structure comparison of InAs/TiO2 with H-

terminated UTB. Inset: Interface coordinate system and
Fig. 3 .Example fitting of EHT parameters to DFT Brillouin zone
Targets for rutile phase of bulk TiO2
Fig. 5. InAs/HfO2 interface Fig. 6. Density of states comparison of InAs/TiO2 with H-

terminated UTB.
32
NEMO5: Why must we treat topological
insulator nanowires atomically?
Fan W. Chen1,2, Michael Manfra1, Gerhard Klimeck2, and Tillmann Kubis2
1
Purdue, West Lafayette, Department of Physics
525 Northwestern Avenue, West Lafayette, IN 47906
2
Purdue, West Lafayette, Department of ECE
298 Nimitz Dr, West Lafayette, IN 47906
INTRODUCTION MODEL
Surface electrons in topological insulator (TI) In this work, atomistic sp3d5s* (20 band) tight
based devices such as Bi2Te3 nanowires face no binding bandstructure calculations of Bi2Te3
backscattering in the absence of magnetic nanowires are calculated by the multipurpose
impurities. Their surface conductance is expected to NanoElectronics Modelling Tool (NEMO5) [6].
be only limited by the surface Fermi velocity [1]. Magnetic fields are included in terms of the Peierl’s
However, experimental values of the Bi2Te3 surface phase factor in symmetric gauge [8]. All presented
Fermi velocity varies more than 10% [2, 3]. atomistic calculations are numerically very intense
So far, theoretical studies of TI wires assume and required typically about one million CPUs on
rotational symmetry along the wire axis. In the case the Blue Waters supercomputer. The surface
of Bi2Te3 this assumption is only true for wires bandstructures of these sophisticated results are
grown along [001] direction [4]. In contrast, very well fit with a new analytical model. This
fabricated Bi2Te3 nanowires are grown in [110] model is an extension of the one in Ref. [4] to
direction and often have rectangular cross sections include different Fermi velocities on different wire
[5]. The facets of [110] Bi2Te3 nanowires show facets.
different chemical compositions: Two facets are
RESULTS
composed of Te atoms only and the other two
contain both Te and Bi atoms. Such details of the The atomistic tight binding model with the
surface chemistry require atomistic representations. Peierl’s phase factor of NEMO5 reproduces the
In this work, atomistic sp3d5s* (20 band) tight band gap dependence on magnetic fields as
binding bandstructure calculations of Bi2Te3 discussed in Refs. [4] The calculated oscillations of
nanowires are presented. In agreement with Ref. [4] the magnetoconductance in these nanowires for
the band gap of the Bi2Te3 nanowires close in this different gate voltages agree qualitatively with
work when the magnetic flux through the wire cross experimental data of Ref. [5] (see Fig.1). One
section agrees with half-integer flux quanta. dimensional helical states are observed in the wires
Deviations from literature are found in the details of agreeing with Ref. [7] (see Fig.2). Figure 3
the surface state energies and surface Fermi illustrates differences in the bandstructures of
velocities: Fermi velocities of chemically different rectangular Bi2Te3 nanowires with different ratios
surfaces differ. This creates an effective potential of pure Te and mixed atom type surfaces. The
on the wire surface which can confine TI surface surface states of these atomistic calculations are
states on specific surface facets. Guided by the subject to confinement depending on the facet’s
atomistic results, the analytical Fermi velocity dimension and chemistry.
model of Ref. [4] is augmented to cover the impact The surface bandstructures of the atomistic
of the detailed wire surface chemistry. calculations serve as fitting targets for the surface
Fermi velocities of the new analytical model. The
analytical model can efficiently predict effective
surface potentials which confine surface states on
specific facets.
ISBN 978-0-692-50554-0 33
CONCLUSION
In summary, the atomistic representation of
Bi2Te3 wires unveils chemically different wire
surfaces and surface state dispersion relations. The
presented analytical model is able to predict the
atomistic results for the surface state dispersion and
confinement.
ACKNOWLEDGMENT
We acknowledge discussion with Luis Jauregui,
Yaohua Tan, Sicong Chen and James Charles as
well as support from Blue Waters sustained-
petascale computing project (awards OCI-0725070
and ACI-1238993) and the state of Illinois. This
Fig. 2. 1D helical surface states of a 48x12nm2 Bi2Te3 nanowire
work was also supported by the Semiconductor
with a magnetic field of 3.5949T solved with NEMO5 for the
Research Corporation’s (SRC) Nanoelectronics
two (E,k) tuple as indicated in the bandstructure in (a). The
Research Initiative and National Institute of color represents different spin orientations in (b). Electrons of
Standards & Technology through the Midwest the same energy and opposite spin orientation propagate into
Institute for Nanoelectronics Discovery (MIND), opposite direction as illustrated in (c).
SRC Task 2141, and Intel Corporation.
REFERENCES
[1] Hwang, Choongyu, et al. Scientific reports 2 (2012).
[2] Qu, Dong-Xia, et al. Science 329.5993 (2010): 821-824.
[3] Li, Yao‐Yi, et al. Advanced materials 22.36 (2010):
4002-4007.
[4] Bardarson, Jens H., P. W. Brouwer, and J. E.
Moore. Physical review letters 105.15 (2010): 156803.
[5] Jauregui, L. A., Pettes, M. T., Rokhinson, L. P., Shi, L.
& Chen, Y. P. arXiv:1402.2659 (2014).
[6] S. Steiger, M. Povolotskyi, H.-H. Park, T. Kubis, and G.
Klimeck, IEEE Transactions on Nanotechnology 10,
1464 (2011)
[7] Hong, Seung Sae, et al. Nano letters 14.5 (2014): 2815-
2821.
[8] Graf, M., and P. Vogl. Physical Review B 51.8 (1995):
4940. Fig. 3. (a) Atomic structures of 60x150 nm2 Bi2Te3 nanowires
grown along the 110 direction. The facets with only Te1 atoms
are larger in type I than in typeII wires. (b) Bandstructures of
the nanowires of (a).
Fig. 1. (a) Experimental Magnetoconductance as function of Fig. 4. Surface state quantization energies as a function of the
magnetic flux Φ in units of magnetic flux quanta Φ0 = h/e for inverse wire circumference C for Bi2Te3 nanowires grown in
magnetic fields applied parallel to the axis of the nanowire in [110] direction. Different facets of [110] wires differ in their
Ref.[5] at two different gate voltages. NEMO5 simulation chemistry and give different quantization energies. Surface
results (b) agree qualitatively with (a). Fermi velocities can be extracted from linear fits to these data.
34
Spin Lifetime Dependence on Valley Splitting
in Thin Silicon Films
J. Ghosh, D. Osintsev, V. Sverdlov, and S. Selberherr
Institute for Microelectronics, TU Wien, Gußhausstraße 27–29/E360, 1040 Wien, Austria
Electron spin attracts much attention as an alter- spin lifetime enhancement is less pronounced when
native to the electron charge degree of freedom for ΛΓ �= 0, although even in such a case εxy can
low-power re-programmable logic and non-volatile boost τ S by orders of magnitude. Figure 2 shows
memory applications. In contrast to charge, spin is the increase in τ S with εxy for different t. We notice
not a conserved quantity, and having sufficiently that at high εxy , the values of τ S become the same
long spin lifetime is critical for applications. We with respect to ΛΓ .
investigate spin relaxation in (001) SOI structures In order to understand this behavior, we con-
depending on the splitting between the equiva- sider the spin-flip caused by the intra- and inter-
lent [001] valleys. In order to evaluate the wave valley transitions. Figure 3 shows the energy levels
functions and spin flip matrix elements we use of the two lowest unprimed subbands primarily
the effective k·p spin-Hamiltonian, written at the responsible [5] for the spin relaxation for several
vicinity of the X -point for the two relevant valleys ΛΓ . Figure 4 depicts the inter- and intra-subband
along the Z -axis in the Brillouin zone [1]. The scattering components of τ S at two distinct temper-
valley-orbit interaction leads to the energy splitting atures (T ). It is observed that the major contribution
between the equivalent unprimed subbands in the to spin relaxation comes from the inter-subband
confined electron system [2] or the valley split- processes, which is greatly reduced at high strain, in
ting. The valley splitting in unstrained films is [2]: accordance with Figure 3. At high valley splitting,
2
△EC = 2π ΛΓ 2π
(k0 t)3 · | sin(k0 t)|, where k0 = 0.85 a ,
the spin relaxation is determined by intra-valley
a is the silicon lattice constant, and t is the film scattering which is independent on △EC and ΛΓ .
thickness. ΛΓ is a parameter defining the strength In Figure 5 we plot the variation of τ S with the
of the valley-orbit interaction with its value ranging total valley splitting (ref: Figure 3). It is observed
between 2eV [3] and 5.5eV [4], depending on the that τ S strongly depends on ΛΓ at the lower ranges
approach used to evaluate the band structure. Both of the valley splitting. This behavior is correlated
methods reproduce the features of the conduction with the suppression of the spin hot spots (Figure 6)
and valence band equally well, but require addi- with increasing △EC [5]. Most importantly in all
tional experimental verification at higher energies cases, the spin lifetime is boosted by several orders
where the discrepancy appears. The valley-orbit of magnitude.
interaction is significantly enhanced in (001) sili- This work is supported by the European Research
con films uniaxially stressed along [110] direction Council through the grant #247056 MOSILSPIN.
(εxy ) [5]. We consider the surface roughness (SR) R EFERENCES
and the longitudinal (LA) and transversal (T A)
[1] Y. Song and H. Dery, Physical Review B vol. 86, 085201,
acoustic phonons as the prominent spin relaxation 2012.
mechanisms for our UTB sample [5]. [2] T. Ando et al., Review of Modern Physics vol. 54, 437,
We analyze the variation of the spin lifetime 1982.
(τ S ) with εxy versus the electron concentration [3] D. Rideau et al., Solid-State Electronics vol. 53, 452,
2009.
(NS ) in Figure 1. The spin relaxation becomes less [4] T.B. Boykin et al., Applied Physics Letters vol. 84, 115,
severe for higher carrier concentrations for all three 2004.
considered mechanisms. It is to notice that the [5] D. Osintsev et al, IWCE Proc., doi: 10.1109/IWCE.2014.
ISBN 978-0-692-50554-0 35
2
10
3 total T=150K
1 10
10 inter
intra
Spin lifetime [ns]

Spin lifetime [ns]
0 2
10 10
T=300K
-1
10
1
13 -2
NS=10 cm , ΛΓ=5.5eV 10
-2
10 13 -2
NS=10 cm , ΛΓ=0
0
12 -2 10
-3 NS=10 cm , ΛΓ=5.5eV
10
12 -2
NS=10 cm , ΛΓ=0
-4 -1
10 0 10 0 0.003 0.006 0.009 0.012 0.015
0.003 0.006 0.009 0.012 0.015
εxy εxy
Fig. 1. Variation of the spin lifetime with εxy with the electron Fig. 4. Spin relaxation time with the respective components.
concentration as parameter. t = 2.7 nm, T = 300 K. NS = 1012 cm−2 , t = 2.7 nm, and ΛΓ = 5.5 eV.
2
10
1
1 10
10
Spin lifetime [ns]
Spin lifetime [ns]

0 0
10 10
-1
10
-1 ΛΓ=0
10
t=1.36nm, ΛΓ=5.5eV ΛΓ=2eV
-2
10
t=1.36nm, ΛΓ=0 ΛΓ=5.5eV
-2
-3 t=2.7nm, ΛΓ=5.5eV 10
10
t=2.7nm, ΛΓ=0
-4
10 10 0
-3
0 0.003 0.006 0.009 0.012 0.015 0.01 0.02 0.03 0.04 0.05 0.06
εxy Valley splitting [eV]
Fig. 2. Variation of the spin lifetime with εxy with the sample Fig. 5. Variation of the spin lifetime with the valley splitting
thickness as parameter. NS = 1012 cm−2 , T = 300 K. results shown in Figure 3.
ΛΓ=5.5eV 0
10
0.075 nd ΛΓ=0
ΛΓ=2eV 2 subband
relaxation matrix elements
Normalized intersubband
ΛΓ=2eV
-1
10 ΛΓ=5.5eV
Energy [eV]
0.05 st
1 subband
-2
10 -1
kx=0.5nm-1,
ky=0.8nm
0.025 -3
10
valley splitting
-4
10
0 0 0.003 0.006 0.009 0.012 0.015 0.018
0 0.003 0.006 0.009 0.012 0.015
εxy εxy
Fig. 3. The energies of two lowest unprimed subbands with Fig. 6. Variation of the SR spin relaxation inter-subband
εxy at two distinct ΛΓ values. NS = 1012 cm−2 , t = 2.7 nm, matrix elements with εxy and with ΛΓ around the spin hot
T = 300 K. spot.
36
Finite Difference Schemes for k · p Models:
A Comparative Study
Jun Z. Huang, KuangChung Wang, William R. Frensley∗ , and Gerhard Klimeck
Network for Computational Nanotechnology, Purdue University, West Lafayette, IN 47907, USA
∗ University of Texas at Dallas, Richardson, TX 75080, USA
Multi-band k · p models discretized with finite For BF operator ordering, the one-sided difference
difference method (FDM) have been widely used to also gives results free of spurious bands (not shown
study electronic properties of semiconductor nanos- here). The convergence behaviors of the one-sided
tructures. However, different schemes of FDM exist differences for both SYM and BF ordering are
in the literature, some of them are numerically plotted in Fig. 3. It can be observed that the error
unstable leading to spurious states [1][2], while decreases with the mesh size and thus they are
others are stable but require special treatment of the numerically stable.
boundary conditions and/or the material interfaces The implementation is simple and straightforward
[3][4][5][6]. Therefore, a comparison of their nu- for the centered difference case, but a bit com-
merical behaviors (and implementation tricks) will plicated for the one-side difference case. For the
be very helpful for selecting a suitable scheme and latter case, in order to avoid artificial spin splitting,
obtaining reliable results. To this end, we have interchange of forward and backward schemes is
implemented into NEMO5 simulation software [7] required when one goes from one spin group to the
the following options, (a) centered difference for other spin group (Fig. 4). In addition, the imposing
symmetrized (SYM) Hamiltonian [1], (b) centered of Dirichlet and Neumann boundary conditions is
difference for Burt-Foreman (BF) Hamiltonian [8], tricky. As illustrated in Fig. 5, to enforce Dirichlet
(c) one-sided differences for SYM Hamiltonian [3], (Neumann) boundaries at the two ends, one needs
and (d) one-sided differences for BF Hamiltonian to eliminate one row and one column corresponding
[6]. For all cases, eight-band and six-band models to the spin-down (up) conduction band of the first
for both zincblende and wurtzite type materials are node, and those corresponding to the spin-up (down)
available. conduction band of the last node.
In Fig. 1, we compare the dispersion relations For BF ordering, further cares are needed in order
of bulk InAs (operator ordering is irrelevant here) to preserve the geometric symmetry and eliminate
discretized with centered and one-sided differences. slope anomalies of the wave functions. In partic-
Note that Foreman’s Ac=0 strategy [8] has been ap- ular, averaged band edges should be used at the
plied. It is clear that centered difference causes the heterojunction interfaces and the mesh points where
bowing down of the conduction band, implying that they should be placed matter [6]. As a result, the
when the structure is confined there will be high-k electron wave function is smooth at the material
oscillating states polluting the low energy window. interfaces, but shifted forward (backward) by half a
The one-sided difference instead approximates the mesh for the first (second) spin group, shown in Fig.
original continuum band structure quite well. 6. As a comparison, the wave functions of the SYM
This has been confirmed by plotting the band Hamiltonian using one-sided difference [3] are also
structure of a GaAs-InAs quantum well with SYM plotted, it is observed that they are not shifted but
operator ordering (Fig. 2). The centered difference there are slope discontinuities around the interfaces.
produces spurious bands appearing almost every- In summary, centered difference produces spu-
where around the band gap, whereas the one-sided rious solution; one-sided difference gives stable
difference is free of any of those unphysical bands. results but it needs proper cares regarding its im-
ISBN 978-0-692-50554-0 37
plementation and interpretation. [2] X. Cartoixà, D. Z.-Y. Ting, and T. C. McGill, J. Appl. Phys.
This research was supported in part by Intel Cor- 93, 3974 (2003).
[3] O. Stier and D. Bimberg, Phys. Rev. B 55, 7726 (1997).
poration, and in part by the Taiwan Semiconductor [4] H.-B. Wu, S. J. Xu, and J. Wang, Phys. Rev. B 74, 205329
Manufacturing Corporation, through memberships (2006).
in the Network for Computational Nanotechnology. [5] Y. Jiang, X. Ma, Y. Xu, and G. Song, J. Appl. Phys. 116,
173702 (2014).
R EFERENCES [6] W. R. Frensley and R. Mir, (2015), arXiv 1412.7201.
[7] S. Steiger, M. Povolotskyi, H.-H. Park, T. Kubis, and G.
[1] S. L. Chuang and C. S. Chang, Semicond. Sci. Technol. Klimeck, IEEE Trans. Nanotechnol. 10(6), 1464 (2011).
12, 252 (1997). [8] B. A. Foreman, Phys. Rev. B 56, R12748 (1997).
4 One−Sided
Forward
Continuum
Energy E [eV]
2
Backward
Centered
0 Γ Backward
L X
−2 Forward
−20 −10 0 10
Momentum K [1/nm]
Fig. 4. Interchange of forward and backward schemes for
Fig. 1. Eight-band E-k relations of bulk InAs. For discrete different spin groups of the eight-band model.
E-k relations, ∆X = ∆Y = ∆Z = 0.25nm.
5 5
0.5 10 10
Energy E [eV]
15 15
0 20 20
L Γ X 25
5 10 15 20 25
25
5 10 15 20 25
−0.5
Fig. 5. An example of enforcing Dirichlet (left) and Neumann

−2 −1 0 1 2
Momentum K [1/nm] (right) boundary conditions for a matrix with 3 nodes (the
matrix size is thus 24). The discretization scheme for each node
Fig. 2. E-k relations of a GaAs-InAs quantum well. The is shown in Fig. 4.
crosses (circles) are results of centered (one-sided) difference.
0.06
−2 CB CB
10 0.04
GaAs InAs GaAs GaAs InAs GaAs
0.02
Deviation [eV]
10
−4 0
0.08 up spin
0.06 VB down spin VB
Ec BF
−6
Ev BF 0.04
10
Ec SYM 0.02 GaAs InAs GaAs GaAs InAs GaAs
Ev SYM 0
0 5 10 15 0 5 10 15
−8
10 1 2 3 SYM BF
10 10 10
Number of Grids
Fig. 6. Wave functions of the lowest (highest) electron (hole)
Fig. 3. Deviation of the 1st conduction and valence band states, using one-sided differences, for both SYM and BF
energies with respect to number of grids. Hamiltonian. The spin up and down states are degenerate.
38
Density functional tight binding modeling in the
context of ultra-thin SOI MOSFETs
S. Markov*, C. Y. Yam*, B. Aradi+, G. Penazzi+, A. Pecchia^, T. Frauenheim+ and G. H. Chen*
*
Department of Chemistry, The University of Hong Kong, Hong Kong
+
BCCMS, University of Bremen, Bremen, Germany, Ûniversity of Rome “Tor Vergata”, Rome, Italy
We investigate the applicability of density and the penetration of states in the insulator from
functional tight binding (DFTB) theory [1][2], both the metal and the semiconductor. We attribute
coupled to non-equilibrium Green functions the somewhat inaccurate band-offsets between
(NEGF), for atomistic simulations of ultra-scaled SiO2 and Si to the weakly expressed CB minimum
electron devices, using the DFTB+ code [3][4]. In of SiO2 at Γ. This should have relatively small
the context of ultra-thin silicon-on-insulator (SOI) influence on the confinement effects in the
transistors we adopt atomic models that include channel, because of the abrupt change in gap after
not only the Si channel, but also the interfacial 0.7 nm from the interface. Indeed, Fig. 3 shows
SiO2, and look at the change of electronic, very good agreement between experimental [7],
dielectric and transport properties as Si film and calculated dependence of Si band-gap and
thickness is reduced from 10 nm to less than 1 nm. CB/VB-edge shifts on Si film thickness. This is in
We build on our previous reports [5][6], and draw contrast to the simulations with H-passivated Si
a systematic comparison against a corresponding channel, which overestimates the effects ~2 times.
model that employs H-passivation of the channel, Similar comparison is drawn in Fig. 4 for the
and against known experiment. dependence of Si permittivity on Si thickness,
DFTB derives as an approximation of density calculated as described in [8]; experimental data is
functional theory (DFT). It is computationally from [10]. The microscopic permittivity profile
more efficient since part of its Hamiltonian is pre- across 0.8 nm SOI structure is shown in the inset
computed in tight-binding terms over extended of Fig. 4, and the interface transition agrees with
neighbors, employing non-orthogonal basis. This more accurate studies [9].
makes DFTB readily applicable to larger structures Beyond the electronic structure and dielectric
even with disorder, e.g. as found in SiO2. An properties, we compare IDVG characteristics of
additional part in the DFTB Hamiltonian junctionless-SOI transistor with 3 nm gate length
approximates effects of second-order charge and 0.8 nm Si body thickness, calculated in
fluctuations, which allows DFTB to capture more DFTB+NEGF (ballistic limit) and the measured
accurately not only permanent charge transfer due from a similar device [11] in Fig 5. Model
to bond asymmetry, but also induced polarization compares well with experiment in the sub-
due to applied electric field, and hence to model threshold regime, after small work-function
dielectric response. adjustment. There is significant difference
DFTB relies however on 1 to 3 parameters however, between the models with H- and SiO2-
(typically) per chemical element. We recently passivated Si channel, as seen in Fig. 6.
reported good parameterization for Si and O that
REFERENCES
yields accurate band-structure for bulk Si, as
[1] [T. Frauenheim et al, J. Phys. Cond. Matt 14 3015 (2002).
shown in Fig 1. Although the band-structure of
[2] M. Elstner et al, Phys. Rev. B 58 7260 (1998).
bulk oxide is not as accurately reproduced, the [3] A. Pecchia et al, New J. Phys. 10 065022 (2008).
error is still around 20%, and a large-gap insulator [4] B. Aradi et al, J. Phys. Chem. A 111 5678 (2007)
results, as seen in Fig 1. Fig. 2 shows a fragment (www.dftb-plus.info).
of an atomic model of an Al–amorphous-SiO2–Si [5] S. Markov et al, Proc. SISPAD 65 (2014).
[6] S. Markov et al, IEEE Trans. Elec. Dev. 62 696 (2015).
structure, and the corresponding density of states [7] Z. Lu, D. Grozea, Appl. Phys. Lett. 80 (2002) 255.
as a color-map. From the on-atom projected DOS [8] J. Nakamura et al, J. Appl. Phys. 99 (2006) 054309.
we extract the profile of the conduction and [9] F.Giustino, A.Pasquarello, Phys.Rev. B 71 (2005) 144104.
valence band-edges across the MOS junction, [10] H. Yoo, M. Fauchet, Phys. Rev. B 77 (2008) 115335.
superposed on the color-map. Notable is the [11] S. Migita et al Proc. IEDM 191 (2012).
gradual initial opening of the gap on the Si side,
ISBN 978-0-692-50554-0 39
Fig.1. Band-structure of bulk Si (left) and β-cristobalite SiO2
(right) show very good agreement for Si, and a large gap for
SiO2 at the same time, with the same DFTB parameters for Si. Fig. 4. Permittivity of Si versus Si film thickness for SiO2- or
H-passivated channel. Inset: microscopic permittivity profile
for SiO2 passivated, 0.8 nm Si film and the atomic model.
Fig. 2. Density of states, and conduction and valence band

profiles across Al–SiO2–Si structure, calculated in DFTB, Fig. 5. Transfer characteristics of a junctionless SOI MOSFET
show penetration of states in SiO2 from both Al and Si. with 3 nm gate length and 0.8 nm Si channel compare well in
the subthreshold regime against experiment with the given
critical dimensions
Fig. 3. Band-gap widening and edge-shift of the conduction

and valence bands of ultra-thin Si vs. Si thickness, for SiO2- Fig. 6. Transfer characteristics of ETSOI transistors modeled
and H-passivation. Band-edge shifts lead to appreciable with H- and SiO2-passivation show large difference due to the
change in gate work-function difference, with thinning Si. differences in the electronic and dielectric properties.
40
Transferable tight binding model for strained
group IV and III-V heterostructures
Yaohua P. Tan, Michael Povolotskyi, Tillmann Kubis, Timothy B. Boykin∗ and Gerhard Klimeck
Network for Computational Nanotechnology, Purdue University, West Lafayette, Indiana, USA,47906
∗ University of Alabama in Huntsville, Huntsville, Alabama, USA, 35899
Modern semiconductor devices have reached effect is included using strain induced onsite and
critical device dimensions in the range of sev- interatomic couplings which depend on the change
eral nanometers. Devices such as superlattice- of bond lengths of the first nearest neighbours and
FETs[1],Ultra Thin Body(UTB)-FETs and FinFETs bond angles between the first nearest neighbours.
consist of strained materials with different lattice Band structures of Selected semiconductors (bulk
constant. Quantitative analysis of those devices Si, Ge, Si0.5 Ge0.5 , AlAs, GaAs and InAs) are shown
requires the reliable prediction of the bandgaps, in Fig. 1. The ETB band structures match the corre-
effective masses in strained heterostructures. The sponding hybrid functional calculations results well.
Empirical Tight Binding (ETB) methods are appro- Compared with corresponding HSE06 results, band
priate for atomistic device modeling due to their nu- edge at high symmetry points are within 0.05eV
meric efficiency[2]. However, the accuracy of ETB and important effective masses have less than 10%
calculations dependents on the transferability of the error. Fig. 2 and 3 show InAs and Si conduction and
ETB parameters. In this work, transferable ETB valence band edge splitting under strains produced
parameters of strained IV and III-V group semicon- by stress along 001 and 111 directions. The splitting
ductors are generated from ab-initio calculations[3], of conduction and valence band edges at high
[4]. The ETB parameters show good transferability symmetry points such as Γ, L and X are correctly
when applied to strained bulk materials as well as captured by the strain model. Fig. 4 and Fig. 5
ultra-thin superlattices. show the band structure of GaAs/AlAs and Si/Ge
ETB parameters are obtained through ab-initio superlattices respectively. The TB band structure
mapping process[4]. During the parameterization agree with the HSE06 bands, demonstrating good
process, ETB parameters and basis functions are transferability of ETB parameters for group IV and
adjusted to match the corresponding ab-initio band III-V semiconductors.
structures and wave functions. In this work, ab- The use of nanoHUB.org computational re-
initio bands of the strained bulk materials and sources operated by the Network for Computational
superlattices are calculated using VASP. Hybrid Nanotechnology funded by the US National Sci-
functional HSE06 is used to produce correct band ence Foundation under Grant Nos. EEC-0228390,
gaps. Group IV and III-V materials are parameter- EEC-1227110, EEC-0228390, EEC-0634750, OCI-
ized using the sp3d5s* ETB model. The parameter- 0438246, OCI-0832623 and OCI-0721680 is grate-
ized group IV and III-V materials include Si, Ge, fully acknowledged.
Si0.5 Ge0.5 and compounds XY with X = Al,Ga and
R EFERENCES
In, Y = P, As and Sb. To have transferable ETB
parameters, following constraints are imposed. a) [1] P. Long, et al, IEEE Electron Device Letters 35,
1212(2014)
Onsites of each atom depend only on the atom type [2] G. Klimeck, et al, IEEE Trans. Electron Devices 54, 2079
instead of materials. b) Both strained and unstrained (2007).
ETB band structures are fitted to ab-initio results. c) [3] Y. Tan,et al, Journal. Comp. Elec 12, 56 (2013).
Variation of interatomic coupling parameters among [4] Y. Tan,et al, arXiv.1503.04781 [cond-mat.mtrl-sci] .
different materials is less than 0.3eV. The strain
ISBN 978-0-692-50554-0 41
5 Strained Si
0.4
4 (a) 001 stress, VB and CB (G) (b) 111 stress, VB and CB (G)
3
0.2
2
HSE06
Band edges splitting (eV)

1 ETB 0
Energy (eV)
Si Ge HSE06 Si0.5Ge 0.5 HSE06

0 ETB
ETB
−1 −0.2
−2
−0.4
−3 0.4
(c) 001 stress, CB (X) (d) 111 stress, CB (X)
−4
−5
0.2
5
3 0
HSE06 HSE06
2
ETB ETB
HSE06 HSE06
1 HSE06
−0.2
Energy (eV)
AlAs ETB GaAs ETB InAs

ETB −4 −2 0 2 4−4 −2 0 2 4
0 001 stress (GPa) 111 stress (GPa)
−1
−2 Fig. 3. Band edge at high symmetry points(Γ,X) splitting of

−3 strained Si under 001 and 111 stress. The 0 energy of (a) and
−4 (b) corresponds to the unstrained top valence bands; while 0 of
−5 the (c),(d) correspond to lowest unstrained conduction band at
L G X L G X L G X
X points.
Fig. 1. Bulk band structure of Si, Ge, SiGe, AlAs, GaAs 5
and InAs. ETB bands shows good agreement with the hybrid
functional bands(HSE06). Energy (eV)
Strained InAs
(a) (b)
(a) 001 stress, VB and CB (G) (b) 111 stress, VB and CB (G) 0
0.5
0
HSE06 HSE06
ETB ETB
−5
−0.5 X G K X G K
0.5 (c) 001 stress, CB (L) (d) 111 stress, CB (L) Fig. 4. Band structure of GaAs/InAs superlattices(HSE06 vs
Band edges splitting (eV)
ETB). Band structure of GaAs/InAs superlattices with 4 atomic

layers (a) and 8 atomic layers (b) in the primitive unit cell.
0 The ETB parameters are transferable when applied to ultra thin
HSE06 HSE06
superlattices.
ETB ETB
−0.5
(e) 001 stress, CB (X) (f) 111 stress, CB (X)
1
4
0.5
2
Energy (eV)
0 (a) (b)
0
−0.5
−4 −2 0 2 4 −4 −2 0 2 4 −2
001 stress (GPa) 111 stress (GPa)
HSE06 HSE06
−4 ETB ETB
Fig. 2. Band edge at high symmetry points(Γ,L,X) splitting of X G K X G K
strained InAs under 001 and 111 stress. The 0 energy of (a) and
(b) corresponds to the unstrained top valence bands; while the 0 Fig. 5. Band structure of Ge/Si superlattices(HSE06 vs ETB).
energy of the (c),(d) and (e),(f) correspond to lowest unstrained Band structure of Ge/Si superlattices with 4 atomic layers
conduction band at L and X points respectively. Compared (a) and 8 atomic layers (b) in the primitive unit cell. The
with the HSE06 result, the ETB model capture the band edge TB parameters are transferable when applied to ultra thin
splitting at high symmetry point under different strains. superlattices.
42
Modeling of Quantum Cascade Laser Sources
with Giant Optical Nonlinearities
C. Jirauschek
Institute for Nanoelectronics, Technische Universität München, Arcisstr. 21, D-80333 München, Germany
EMC [5]. By coupling the carrier transport to

INTRODUCTION
optical simulations as illustrated in Fig. 1, a self-
The quantum cascade laser (QCL) is an consistent modeling scheme for THz DFG
extremely versatile light source, employing structures is obtained. This approach has been
artificially engineered optical transitions in the applied to an experimental device [2]. Figure 2
conduction band of a semiconductor nanostructure. shows the simulated MIR and THz powers as a
By careful design of the active region, the optical function of lattice temperature, along with
gain characteristics can be custom-tailored for the available experimental data at 300 K [2]. Good
envisaged application over a wide range of the agreement between theory and experiment is
mid-infrared (MIR) and terahertz (THz) spectral observed. Furthermore, the results indicate that the
region. Recently, the possibility to additionally temperature degradation of the THz power can be
integrate giant artificial nonlinearities into the gain mainly ascribed to the decrease in MIR powers.
medium has been exploited for various innovative On the other hand, the optical nonlinearity is
applications, including THz difference frequency temperature insensitive, and also depends only
generation (DFG) at room temperature [1,2] and moderately on the applied bias (see Fig. 3).
QCL-based frequency combs [3,4]. Besides frequency conversion, giant optical
The further development of such advanced nonlinearities are used to induce coupling between
QCL sources requires careful modeling, providing the longitudinal modes, which is exploited for the
an improved understanding of the carrier-light generation of picosecond pulses [6] and frequency
interaction and allowing for systematic design combs [3,4]. To describe the underlying coherent
optimization. These devices combine aspects from nonlinear optical phenomena, Maxwell-Bloch
various areas such as nanoelectronics, nonlinear equations are commonly employed [6,7], which
photonics, and plasmonics, which must be however contain unknown parameters such as
adequately considered in state of the art simulation scattering rates. These are often estimated from
tools. This can be achieved by an integrated multi- experimental data or used for fitting. In our
domain modeling approach, taking into account approach, these quantities are extracted from EMC
the electron transport in the active region and simulations, resulting in a self-consistent modeling
optical field propagation in the resonator on an approach. In Fig. 4, the simulated spectral intensity
equal footing. In the following, we discuss the is shown for a THz QCL frequency comb source
application of such simulation approaches to QCL- [4]. The spectrum features equidistant comb lines
based DFG structures and frequency combs. over an extended spectral range and is split into
SIMULATION two lobes, as also observed in experiment [4].
Conventional THz QCLs only work at In conclusion, multi-domain simulation
cryogenic temperatures. A promising alternative is approaches combining EMC with advanced
THz DFG, which uses an active region design models for the nonlinear optical effects provide a
supporting two MIR lasing modes at frequencies f1 powerful tool for the modeling of innovative QCL
and f2 and featuring a second order optical sources using giant optical nonlinearities.
nonlinearity. Frequency mixing gives rise to an ACKNOWLEDGMENT
additional field component at f1-f2, enabling QCL- This work was funded by the Heisenberg
based THz generation at room temperature. The program of the German Research Foundation
nonlinear susceptibility can be computed using (DFG, JI115/4-1).
ISBN 978-0-692-50554-0 43
REFERENCES
[1] M. Belkin et al., Room temperature terahertz quantum
cascade laser source based on intracavity difference-
frequency generation, Appl. Phys. Lett. 92, 201101
(2008).
[2] Q. Y. Lu et al., Room temperature single-mode terahertz
sources based on intracavity difference-frequency
generation in quantum cascade lasers, Appl. Phys. Lett.
99, 131106 (2011).
[3] A. Hugi et al., Mid-infrared frequency comb based on a
quantum cascade laser, Nature 492, 229 (2012).
[4] D. Burghoff et al., Terahertz laser frequency combs,
Nature Photon. 8, 462 (2014).
[5] C. Jirauschek et al., Monte Carlo study of terahertz
difference frequency generation in quantum cascade
lasers, Opt. Express 21, 6180 (2013).
[6] V.-M. Gkortsas et al., Dynamics of actively mode-locked
quantum cascade lasers, Opt. Express 18, 13616 (2010). Fig. 2. Simulated MIR and THz powers vs. temperature
[7] C. Jirauschek and T. Kubis, Modeling techniques for (solid lines), and available experimental data (asterisks).
quantum cascade lasers, Appl. Phys. Rev. 1, 011307
(2014).
Schrödinger-Poisson
Wave fcts./ Subband

energies populations
Carrier transport (EMC)
Fig. 3. Nonlinear susceptibility |(2)| as a function of

temperature and applied bias.
Electromagnetic simulation
Mode profiles
Frequency mixing
f1
f1
f1-f2
f2
f2
Fig. 1. Schematic illustration of a multi-domain simulation Fig. 4. Simulated optical power spectrum of the QCL
approach for QCL frequency conversion structures. frequency comb source in [4].
44
Multiscale modeling of graphene-metal contacts
T. Cusati, G. Fiori, A. Fortunelli*, G. Iannaccone
Dipartimento di Ingegneria dell’Informazione Università di Pisa, Via Caruso 16, 56122 Pisa, Italy
*
CNR Institute for Chemical and Physical Processes, Via G. Moruzzi, 1 - 56124 Pisa, Italy
achievable Rc would provide physical insights on

ABSTRACT
the metal-graphene interface.
The quality of contacts between metals and two-
dimensional materials is a critical aspect for the APPROACH
performance of transistors based on two-
dimensional materials. In this talk we focus on an We have performed ab-initio calculations with
approach to multiscale modeling of graphene- Quantum Espresso of the structures shown in
metal contacts, considering both chemisorbed and Fig. 1, using a plane wave basis set and a gradient-
physisorbed materials. We show that it is possible corrected exchange-correlation functional
to use density functional simulations of contacts to (Perdew-Burke-Ernzerhof (PBE)). We have
extract a set of physical properties that enable considered four different metals (Ni, Cu, Pt, Pd)
accurate device-level simulations at a higher level divided in two categories, based on the binding
of abstraction. We also show that – properly energy and the metal-graphene distance:
evaluated - the intrinsic conductance of graphene- chemisorbed metals (Ni and Pd), with stronger
metal contacts is higher than that predicted in bonds and physisorbed metals (Cu and Pt), that
previous studies. lead to weaker bonding (Fig. 2).
We have then computed the transmission

INTRODUCTION coefficient via the solution of the scattering
One of the most important challenges related to the problem considering incoming and outgoing Bloch
use of two-dimensional materials in electronic and states as implemented in the PWCOND module of
optoelectronic devices is to understand and control Quantum Espresso (Fig. 3). Finally, we have
the contact between metal and 2D materials. For extracted for each contact the interface
example, a low contact resistance between transmission coefficient and the relative position
graphene and metal electrodes is required for of the Dirac point at the interface with respect to
obtaining high performance in graphene devices, the electrochemical potential. This information can
as it directly impacts the fMAX for analog be used to accurately describe the contact at a
applications and the propagation delays of digital higher level of abstraction (Fig. 4). At the device
gates. The transport properties of the graphene- level, using the Landauer-Buttiker linear response
metal interface have experimentally been formalism, we have computed the contact
investigated through transfer length and four-probe conductance G at room temperature.
methods, in order to measure the contact resistance
(Rc) between graphene and different metals (i.e. CONCLUSIONS
Cr, Ti, Cu, Au, Ni, Pd and Pt). Unfortunately, We have developed a multiscale procedure for
graphene contact fabrication technology is not taking into account the essential physics of
mature and fully reproducible, and therefore a graphene-metal contacts in device simulation. We
broad range of experimental values of Rc is show that according to our model the intrinsic
obtained even for the same metal [1-5]. Indeed, the conductance of contacts is higher than that
contact resistance is highly dependent on several extracted with previous measurements and
factors including metal work function, number of simulations (compare table 1), because the latter
graphene layers, deposition temperature, process. typically include also the effect of surrounding
A systematic study of the electrical characteristics partially depleted graphene, which can be
of metal-graphene contacts is highly desirable. optimized with doping or contacts. Further
From this perspective, a theoretical study of the investigation on Pd-graphene is needed.
ISBN 978-0-692-50554-0 45
ACKNOWLEDGMENT
Authors would like to acknowledge support
from the EC 7FP through the Projects GRADE
(Contract 317839), GONEXTs (Contract 309201),
Graphene Flagship (Contract 614391), and from
Quantavis s.r.l.
REFERENCES
[1] Z. Chen, J. Appenzeller, VLSI Symp. Tech. Dig. 13, 128
(2009).
[2] E.J.Lee et al., Nat. Nanotechnology, 3, 486 (2008)
[3] W. S. Leong et al., Nano Lett. 8, 994 (2014)
[4] J. Maassen et al., Appl. Phys. Lett. 97, 142105, (2010).
[5] K. Stokbro, et al, Phys. Rev. B, 85, 165442 (2012) Fig. 3. Transmission for metal-graphene systems obtained
[6] X. Ji et al., Phys. Chem. Chem. Phys. 15, 17883, (2013). with PWCOND for different metals (Cu,Ni,Pd,Pt) compared
[7] E. Watanabe et al., Diamonds & Related Materials. with the transmission of the ideal graphene monolayer.
24, 171, (2012).
Fig. 4. Power conversion efficiency of the investigated OSC

with graphene transparent electrode for different materials as
active layer.
Fig. 1. Simulated structure. Three regions are highlighted: the
left and the right leads and the so-called scattering region Materials Intrinsic Data from the literature -
(central part). Left and right leads are ideally connected to Resistance (Ω•µm)
Resistance
semi-infinite bulk metal leads (left side) and graphene
(Ω•µm)
monolayer (right side). Since we want to directly compute
transport from metal to graphene, we need to ensure that the Ni 80 300 [3] 600 [6]
right lead contains only the metal and the left one only
graphene. Cu 350 44 [4] – 627 [7]
Pd 4500 403 [6], 600 [7]
Pt 10 764 [6]
Table 1: Comparison between the intrinsic resistance of

metal-graphene contacts and results obtained from the
literature. Resistances from the literature are usually higher
because they typically include the effect of the partially
depleted graphene region surrounding the contact, which is
not strictly part of the “intrinsic” graphene-metal contact, and
Fig. 2. Side and bottom views of the scattering region can therefore be optimized with electrostatics and/or doping.
considered for the DFT and transmission calculations, for the
Ni-, Cu-, Pt- and Pd-graphene contacts.
46
From single-stage to device-level simulation of
coupled electron and phonon transport in quantum
cascade lasers
Y.-B. Shi, S. Mei, and I. Knezevic
Department of Electrical and Computer Engineering, University of Wisconsin - Madison
1415 Engineering Dr, Madison, WI 53706-1691, USA
Email: [email protected], [email protected]
A critical challenge for quantum cascade lasers Simulation of coupled electron and phonon transport
(QCLs) is achieving reliable room–temperature in QCLs is a multiscale problem. The electron transport
continuous–wave (RT-cw) operation at high powers is confined in the active core consisting of many repeated
required by many applications. However, the high stages, typically with a total thickness of several microns.
operating lattice temperature, accompanied by the high In contrast, the only constraints for thermal transport
electronic temperature, has a detrimental impact on the are the thermal boundary conditions, as specified by the
threshold current density, wall plug efficiency (WPE), waveguide and mounting configurations, in a much larger
and device reliability [1]. Therefore, fully understanding spatial scale, typically several hundreds of microns. The
the coupled transport of electrons and phonons and large-scale thermal transport causes the lattice temper-
connecting the insights to device performance metrics ature in the active core to become stage dependent.
is critical for further developments of QCLs. Consequently, the periodic electron transport feature in
In QCLs, the high electric field and current density the active core is no longer justified, since the electron-
required for lasing pump considerable energy into the phonon scattering rate varies from stage to stage. The
electronic system. The accelerated electrons can scatter scale difference and the coupling between electron and
with each other, with phonons, with layer interfaces, phonon transport makes the simulation challenging, and
imperfections, or impurities. Among these possible scat- a single-stage simulation, which most existing studies
tering mechanisms, the electronic system relaxes its en- adopt, is not sufficient to accurately describe these pro-
ergy mainly through the emission of longitudinal optical cesses.
(LO) phonons [3]. Owing to the low group velocities, Here, we present a multiscale simulation technique for
LO phonons are not efficient at carrying the heat away, coupled electron and phonon transport on the example
and act as a temporary energy storage system. The main of a 9 µm GaAs-based midinfrared (mid-IR) QCL [2].
lattice cooling mechanism is the anharmonic decay of The core of the simulator is a very detailed “table”
LO phonons into acoustic (LA) phonons [4], and the of electronic current densities J and the corresponding
high group velocities of LA phonons make them the heat generation rates Q obtained for a large range of
primary heat carriers to transfer the energy, through the assumed effective lattice temperatures TL and given
waveguide and substrate, to the heat sink. The electron– electric fields F ; the table is calculated using a single-
LO phonon interaction is fast process, with a time scale stage coupled ensemble Monte Carlo (EMC) simulator
on the order of 0.1 ps, while the anharmonic decay of LO for electron and phonon transport within a single stage,
phonons into LA phonons is much slower and typically developed previously [6]. The table essentially provides
takes several ps [5]. As a result, the population of LO a generic and compact representation of the J–F and heat
phonons is built up over time in the active core and their generation characteristics of a single stage under a wide
distribution is driven far from equilibrium [3], [5]. The range of possible operation conditions.
abundance of nonequilibrium LO phonons will also feed Next, based on the charge-current continuity equation
back to the electronic transport by affecting LO phonon and solving the global heat diffusion equation, we obtain
scattering rates. Therefore, both the electron and phonon the temperature and field profiles. Namely, we start from
subsystems are essential to laser characteristics. an assumed current density J and assumed tempera-
ISBN 978-0-692-50554-0 47
ture profile, with each stage becoming a temperature-
dependent source for the heat diffusion equation. We
iterate the temperature profile until the required thermal
boundary conditions are satisfied, updating the field and
heat generation rate profiles as we go. Once we have the 160
final temperature profile, we also have the appropriate

150
field profile for the given current and temperature, and
temperature (K)
we can calculate the total voltage drop V over the whole 140
structure. As a result, we will get a realistic temperature
profile and I −V curves, directly relatable to experiment. 130
The simulated device is a GaAs-based mid-infrared

120
QCL designed for emission at 9 µm [2]. Figure 1 shows T0 = 77 K
the lattice temperature over the entire active core of 110
0 10 20 30 40
the QCL for the heat sink temperature (T0 = 77 K). stage number
A maximum temperature of 160 K indicates strong self
Fig. 1: Lattice temperature distribution in the active core
heating. The stage-dependent electric field is plotted in
of the QCL (heat sink is at 77 K).
Fig. 2, which shows the electric field difference among
50.5
all the stages can be as large as 2.7 kV/cm at J = 10
kA/cm2 . The integration of the stage-dependent electric 50
electric field (kV/cm)

field gives the voltage across the active core, and Fig. 3 49.5
shows the calculated current density–voltage character-
49
istics along with experiment results. The discrepancy
between simulation and experiment likely results from 48.5
the simplified cross-plane thermal conductivity model of
48
the active core used in the simulation; we are presently J = 10 kA/cm2, Vtot = 7.9 V
working on an improved microscopic calculation [7]. 47.5
0 10 20 30 40
This work was supported by the U.S. Department of stage number
Energy, Basic Energy Sciences, Materials Science and Fig. 2: Electric field distribution in the active core of the
Engineering Division, Physical Behavior of Materials QCL.
Program, award DE-SC0008712. 15
R EFERENCES
[1] M. S. Vitiello, G. Scalari, B. Williams, and P. D. Natale, 10
voltage (V)
“Quantum cascade lasers: 20 years of challenges,” Opt. Express

23, 5167 (2015).
[2] H. Page, C. Becker, A. Robertson, G. Glastre, V. Ortiz, and
C. Sirtori, “300 K operation of a GaAs-based quantum-cascade 5
laser at λ ≈ 9 µm,” Appl. Phys. Lett. 78, 3529 (2001). sim
[3] G. Paulavicius, V. Mitin, and M. A. Stroscio, “Hot-optical- exp
phonon effects on electron relaxation in an AlGaAs/GaAs quan-
tum cascade laser structure,” J. Appl. Phys. 84, 3459 (1998). 0
0 5 10 15 20 25 30
[4] S. Usher and G. P. Srivastava, “Theoretical study of the anhar- current density (kA/cm2)
monic decay of nonequilibrium LO phonons in semiconductor
structures,” Phys. Rev. B 50, 14179 (1994). Fig. 3: A comparison between the current density–
[5] R. C. Iotti, F. Rossi, M. S. Vitiello, G. Scamarcio, L. Mahler, voltage characteristics obtained in the simulation and
and A. Tredicucci, “Impact of nonequilibrium phonons on the
electron dynamics in terahertz quantum cascade lasers,” Appl. measurement [2]. Correction of the slope requires an
Phys. Lett. 97, 033110 (2010). accurate microscopic model for the thermal conductivity
[6] Y. B. Shi and I. Knezevic, “Nonequilibrium phonon effects in tensor, which we are currently developing [7].
midinfrared quantum cascade lasers,” J. Appl. Phys. 116, 123105
(2014).
[7] Y. B. Shi, S. Mei, and I. Knezevic, “Multiscale simulation
of coupled electron and phonon transport in quantum cascade
lasers,” in preparation (2015).
48
Multi-scale modeling of metal-CNT interfaces
M. Claus∗§ , A. Fediai‡∗†† , S. Mothes∗ ,§ A. Pacheco§ , D. Ryndyk‡∗†† ,
S. Blawid∗∗ , G. Cuniberti‡∗†† and M. Schröter§
∗ Center for Advancing Electronics Dresden, TU Dresden, Germany, [email protected]
§ Chair for Electron Devices and Integrated Circuits, TU Dresden, Germany
‡ Institute for Materials Science and Max Bergmann Center of Biomaterials, TU Dresden, Germany
∗∗ Laboratory for Devices and Integrated Circuits, Universidade de Brasilia, Brazil
†† Dresden Center for Computational Materials Science, TU Dresden, Germany
I NTRODUCTION DOS for the investigated (17,0)-tube with Pd and Al

It is well known that the metal-CNT interfaces in contacts, respectively. Compared to Al, Pd does not
CNTFETs have been a key factor limiting the device change the DOS of the CNT underneath the metal
performance. Efforts have been made to improve the contact significantly. Fig. 3 shows atomistically cal-
understanding of physics at these interfaces and the culated transfer characteristics for the two different
contact length scaling behavior seen in experiments contact materials.
[1]. However, the related interface phenomena are The atomistic simulation results are employed
not fully understood [1], which makes it difficult to to adjust the contact model used within the trans-
improve the device performance. port studies at the device level [5] which offers
Typically, the impact of the metal-CNT interfaces greater flexibility regarding device architecture and
on the device characteristics is lumped into a re- materials and a significantly reduced computational
sistance which is commonly labeled as the contact burden. The greater flexibility is possible due to a
resistance Rc (including the contribution of both, simplified hetero-junction contact model [5] which
the source and the drain contact). has been verified by atomistic simulations. The key
While for digital applications the contact length contact parameters are the interface potential step
is an critical parameter in terms of overall device Φ (see also [6]), the Schottky barrier height Φsb ,
size, for analog HF applications as described in the DOS within the metal coated tube portion and
[2] the contact length is in µm scale which allows direct tunneling of metal states to the channel [7].
an investigation of the metal-CNT interfaces in Fig. 4, 5 and 6 show the impact of the contact
the long-contact-limit. As suggested in [3], contact model parameters on the contact resistance and on
engineering and thus a detailed understanding are the transconductance. ( For a ballistic channel and
mandatory for optimizing the device linearity of fully transparent contacts, the total device resistance
CNTFETs. approaches its ballistic limit of Rq = h/4q 2 ≈
6.4 kΩ.) These plots exemplify the challenges for
R ESULTS AND D ISCUSSION contact engineering.
A multi-scale modeling and simulation frame- ACKNOWLEDGMENT
work for CNTFETs is employed to study the impact
The authors acknowledge the Cfaed and the DFG
of different contact materials on the device behavior.
projects CL384/2, SCHR95/6 for financial support.
By means of an atomistic simulation platform [4]
which combines large-scale ab initio calculations R EFERENCES
with a Green function approach a good agreement [1] A. D. Franklin, et al. ACS Nano, 8(7), 7333–7339, 2014.
between atomistic simulation studies and experi- [2] M. Schroter, et al. IEEE JEDS, 1(1), pp. 9–20, 2013.
mental data for Pd [1] and Al contacts has been ob- [3] S. Mothes, et al. IEEE Nanotechn., 14(2), 372–378, 2015.
[4] A. Fediai, et al. Phys. Rev. B, 91, pp. 165404, 2015.
served demonstrating a strong correlation between [5] M. Claus, et al., J. Comp. Elect., 13(2), 689–700, 2014.
metal-CNT coupling strength, contact length and [6] G. Giovannetti, et al. PRL, 101, 026803, 2008.
the contact resistance. Fig. 1 and 2 show the local [7] V. Perebeinos, et al. PRL, 111, 236802, 2013.
ISBN 978-0-692-50554-0 49
0.01 0.3 0.9
15
Rc − Rq (kΩ)
10
0
0.0 0.1 0.2 0.3 0.4 0.5
△Φ (eV)
Fig. 4. Simulated contact resistance for different Φ and

Fig. 1. Atomistically calculated local DOS for a 10 nm long different DOS (i. e. effective masses) within the metal coated
CNTFET with Pd contacts. tube portion.
15
Rc − Rq (kΩ)
10
0.01 0.05 0.1 0.3 0.9

0
0.0 0.05 0.1 0.15 0.2
Φsb (eV)
Fig. 5. Simulated contact resistance for different Schottky

Fig. 2. Atomistically calculated local DOS for a 10 nm long barrier heights and different DOS within the metal coated tube
CNTFET with Al contacts. portion.
40
0.9 0.05 0.01
30
gm (µS)
20
10
0
-1.0 -0.8 -0.6 -0.4 -0.2 0.0
Vgs (V)
Fig. 6. Transconductance for different contact DOS (i. e. the

contact effective mass). The error bars indicate the impact of
Fig. 3. Comparison of atomistically calculated transfer char-
the Schottky barrier heights which are changed in between 0 eV
acteristics for Pd and Al contacted 10 nm long CNTFET.
and 0.2 eV.
50
A Multi-Scale Modeling Approach to Study
Transport in Silicon Heterojunction Solar Cells
Pradyumna Muralidharan, Dragica Vasileska, Stephen M. Goodnick and Stuart Bowden
Arizona State University, Tempe, Arizona, USA.
hole and split off bands to calculate hole

INTRODUCTION
properties. We also consider the warping and non -
Single junction solar cells based on Silicon parabolicty of the bands which can play a
continue to be relevant and commercially significant role for hot carriers. At the a-Si/c-Si
successful in the market due to their high interface the EMC solver is coupled to the KMC
efficiencies and relatively low cost processing. solver. While the EMC monitors the progress of
Heterojunction solar cells based on crystalline (c- carriers in the high field region in the c-Si near the
Si) and amorphous (a-Si) silicon (HIT Cells) have heterointerface [3], the KMC simulates defect
paved the way for devices with high VOC's (>700 assisted transport in the intrinsic a-Si. The EMC
mV) and high efficiencies (>20%) [1]. Last year solver calculates the distribution function at the
Panasonic announced a world record efficiency of heterointerface and delivers it to the KMC which
25.6% for its trademark a-Si/c-Si HIT cell [2]. simulates 'hopping' of discrete carriers via point
The novel structure of the device precludes the like defects till they exit the barrier via various
usage of traditional methods (such as drift extraction mechanisms [4].
diffusion) to accurately understand the nature of
transport. Theoretical models used by commercial RESULTS AND CONCLUSIONS
simulators assume a Maxwellian distribution of Fig. 1. shows the velocity vs. field character-
carriers and lack the sophistication to study defect ristics for holes in warped non parabolic bands.
transport. In this work we utilize a combination of The EMC solver shows that the carrier distribution
Ensemble Monte Carlo (EMC) simulations to at the heterointerface is highly non-Maxwellian
study the high field behavior of the photogenerated (Fig. 2) due to high fields while most commercial
minority carriers at the interface and Kinetic simulators using the Drift-Diffusion method
Monte Carlo (KMC) simulations to understand assume a Maxwellian distribution of carriers at all
defect assisted transport in the amorphous silicon points within the device. The KMC solver utilizes
layer. this distribution to conduct defect assisted
transport and calculates the energy distribution
THEORETICAL MODEL
function (Fig. 3) and average time taken for
In this work we simulate a HIT cell structure 'hopping' transport. Fig.4 shows how the average
with a p+ front emitter (a-Si) of 10 nm, an intrinsic transit time reduces with increase in the average
buffer layer (a-Si) of 10 nm and a n type c-Si energy of the carrier distribution at the interface.
absorber layer. The potentials and fields in the The transit time is a great indicator of how much
device were calculated using the commercial current the device can support and the current
software SILVACO at a maximum power point of suppression that is occurring at the heterointerface.
0.6 V. The goal of our work is to study the The KMC solver results give us many valuable
behavior of the photogenerated minority carriers in insights into device transport, such as the effect of
the high field region near the heterointerface and phonon emission on the carrier distribution and the
its transport through the intrinsic a-Si buffer layer identification of the dominant extraction
via defect assisted transport. mechanism (which was Poole-Frenkel emission)
The transport in the high field region in the c-Si for the device under consideration. We can also
region near the heterointerface is studied by evaluate device characteristic for barriers with
implementing the EMC method. As the different defect densities which allows us to
photogenerated carriers under consideration are explore different material systems which can
holes, the EMC considers the heavy hole, light potentially be incorporated into a HIT cell.
ISBN 978-0-692-50554-0 51
REFERENCES
[1] M. Taguchi et al., Prog. Photovolt: Res. Appl. 2005.
[2] K. Masuko et al, IEEE Journal of Photovoltaics, vol.4,
no.6, pp.1433,1435, Nov. 2014.
[3] P. Muralidharan, K. Ghosh, D. Vasileska and S. M.
Goodnick, 40th IEEE Photovoltaics specialists conference,
pp 2519-2523, 2014.
[4] G.C. Jegert, "Modeling of Leakage Currents in High-κ
Dielectrics", PhD Thesis.
Fig. 4. The energy distribution function of the carriers after

collection as calculated by the KMC.
Fig. 1. Schematic diagram of a HIT cell with simulation

domains. Overall electric field distribution is calculated using
drift-diffusion solver coupled to a global Poisson solver
(Silvaco). EMC is used to account for hot holes, which is, in
turn, coupled with KMC to study in details transport through
defect states in the a-Si.
6 Drift Velocity vs. Electric Field

10
Experimental
Simulation
Drift Velocity (m sec-1)
Fig. 5. Average transit time vs. average energy of carriers at

10
5 the heterointerface.
4
10 5 6
10 10
Electric Field (V/m)
Fig. 2. Velocity vs. Electric Field for holes in c-Si.
Fig. 6. Average transit time vs. characteristic decay energy (it

is an indicator of the defect density present in the barrier).
Fig. 3. Energy distribution function for a strongly inverted

device at the a-Si/c-Si interface as calculated by the EMC.
52
Variational Formulation of Stable Discrete k · p
Models
William R. Frensley,
University of Texas at Dallas, Richardson, TX 75080, USA
The longstanding problem of spurious states in k · p each derivative yields a row of the discrete Schrödinger
models of semiconductor nanostructures has been shown equation.
to be an artifact of the use of the centered-difference The γ terms produce variations on the three-point
approximation to the gradient, and it has been shown difference formula for the Laplacian that are well known
that stable models may be constructed on the basis of in the study of single-band effective-mass models [5].
lower-order one-sided differences [1]–[4]: The P terms require special attention. Straightforward
evaluations of the action produce centered-difference
†
K̂L f = K̂R f = (−i/Δ) (fn − fn−1 ) . (1) discretizations, which result in spurious states. We must
n n
therefore impose a form which produces a one-sided
Key results are illustrated in Figs. 1–3. These demon-
difference, such as:
strate that the conventional approach of deriving the ∗ ∗
formulation in the continuum and then invoking approx- iPαβ ψα;j (ψβ;j − ψβ;j+1 ) − ψβ;j ψα;j ∗
− ψα;j+1 .
imation theory to obtain the discrete model is seriously (4)
awed. The aws should be obvious: the bandstructures The variational treatment of the V terms produces an
produced by the k·p approximation can never be periodic unexpectedly rich structure, and leads to the elimination
functions, and their results are therefore nonphysical at of a major cause of slope anomalies [2] in the resulting
short distances. Approximation theory then emphasizes wavefunctions. Let Bα;j be the energy of band α for the
delity to these nonphysical results by focusing on short- material associated with meshpoint j. Then we normally
distance behavior. Every discrete formulation will yield assume Vαα;jj = Bα;j . A calculation using this assump-
a periodic bandstructure; thus discrete models may be tion is shown in Fig. 5, and shows a slope anomaly.
approximations, but they can be far more credible de- But, if we take the model of Fig. 4(a) that associates
scriptions of the physical systems than continuum k · p a material with a mesh interval j + 1/2, the crudest
approximations. approximation to the functional (ψ piecewise constant)
gives Vαα;jj = (Bα;j−1/2 + Bα;j+1/2 )/2. Fig. 6 uses
VARIATIONAL F ORMULATION this form. Ma et al. [3] found this remedy, but could
To avoid approximation theory we invoke a more only attribute it to a grading of the heterojunction. We
fundamental formulation and separately derive the con- see that it is a natural consequence of modeling an abrupt
tinuum and discrete models from it. A k · p model of the heterojunction within the detailed description demanded
form: by the variational formulation. Evaluating the functional
Ĥαβ = γαβ kz2 + Pαβ kz + Vαβ , (2) under the assumption of piecewise-linear ψα∗ and ψβ
yields a weakly nonlocal V matrix, with small super-
can be derived by minimizing the action: and sub-diagonal terms. The implications of this form
are still being explored.
S = dz [γαβ ∂z ψα∗ ∂z ψβ + Pαβ ψα∗ ∂z ψβ In summary, the variational approach provides a sys-
tematic procedure for deriving discrete envelope-function
+Vαβ ψα∗ ψβ − Eψα∗ ψβ ] . (3) models. Abstract discussions of operator ordering and
symmetrization are replaced by concrete considerations
To derive the discrete Schrödinger equation, we must of how the wavefunctions and material parameters vary
make explicit assumptions about how the wavefunction between meshpoints.
and material parameters vary within each mesh interval,
as illustrated in Fig. 4. The best choice is to assume that R EFERENCES
the ψβ (z) vary linearly within a mesh interval. The action [1] O. Stier and D. Bimberg, Phys. Rev. B 55, 7726 (1997).
(3) then becomes a sum over mesh intervals of bilinear [2] Y. Jiang, X. Ma, Y. Xu, and G. Song, J. Appl. Phys. 116, (2014).
∗ [3] X. Ma et al., J. Appl. Phys. 116, (2014).
expressions in the discrete wavefunctions ψα;j and ψβ;j . [4] W. R. Frensley and R. Mir, (2015), arXiv 1412.7201.
For this discrete formulation, the functional derivative [5] W. R. Frensley, Heterostructures and Quantum Devices (Academic
reduces to the set of all derivatives ∂S/∂ψα;j ∗
, and Press, 1994), pp. 273–303.
ISBN 978-0-692-50554-0 53
2.0
(a) (b)
1.5
ψ ψ
K(k)Δ
1.0
KC B B
|KL|
0.5 KL
k
6th order
0.0 j−1 j j+1 j−1 j j+1
0.0 0.2 0.4 0.6 0.8 1.0
k(π/Δ)
Fig. 4. Assumed details of a discrete model within a mesh interval. B
Fig. 1. Dispersion relations for the different K̂ operators. represents a band-edge energy. Heterojunctions are taken to be either
coincident with a mesh point (a), or centered within an interval (b).
2 (a) 0.30
0.20
E (eV)
0 0.10
0.00
−2
−0.10
−4 −0.20
(b) 0.30
−6
0.20
0.10
−10 −8 −6 −4 −2 0 2 4 6 8 10
0.00
k (nm−1)
−0.10
1st Order Continuum 2nd Order
−0.20
0 2 4 6 8 10 12 14
Fig. 2. Band structure resulting from the rst-order and centered- Position (nm)
diference models, compared to the continuum solution. The Hamil-
tonian is the 3-band model employed by Cartoixà applied to |CB |LH |SO
In0.53 Ga0.47 As with a mesh spacing of 0.29 nm which equals a/2.
Fig. 5. Lowest-energy conduction band state of an InAs-AlSb structure
(a) 0.30 showing slope anomaly due to simplistic treatment of V terms. (a) left-
hand difference, (b) right-hand difference.
0.20
0.10
0.00
−0.10
(b) 0.30 (a) 0.30
0.20
0.20
0.10
0.10 0.00
0.00 −0.10
−0.10 −0.20
(c) 0.4 (b) 0.30
0.20
0.2
0.10
0.0 0.00
−0.2 −0.10
−0.4 −0.20
0 2 4 6 8 10 12 14 0 2 4 6 8 10 12 14
Position (nm) Position (nm)
|CB |LH |SO |CB |LH |SO
Fig. 3. The state which should be the highest-energy hole state of Fig. 6. Calculation of Fig. 5 but using material-averaged V terms.
a model superlattice. The discrete formulations are: (a) left-hand, (b)
right-hand and (c) centered difference. Case (c) is clearly spurious.
54
Multi-Scale Quantum Simulations of Conductive
Bridging RAM
Zhengping Jiang, Michael Povolotskyi, Nicolas Onofrio, David Guzman, Daniel Lemus, Santiago Perez,
Jose Bermeo, Alejandro Strachan, Gerhard Klimeck
School of Electrical and Computer Engineering, Purdue University, West Lafayette, IN, 47906, USA
have a Gaussian distribution centered at atom.

INTRODUCTION
Empirical Tight Binding Hamiltonian is
With shrinking sizes of transistors, scaling of constructed for an atomic configuration. The
memory devices has achieved great success these simulation flow is shown in Fig.1.
years. The conductive bridging RAM (CBRAM) The device is partitioned into slabs based on
as one of the emerging memory technologies has coupling with the Breadth-first search method[6].
advantages over the flash memory on high Such slab structure is suitable for transmission
scalability, low power consumption, high coefficient calculation with the recursive Green’s
resistance ratio and long retention[1]. function (RGF) technique. Because the charge
The basic cell of CBRAM is composed of a self-consistent calculation is not required at this
channel electrolyte between two metal electrodes. stage, only forward RGF is performed. The self
Under biases, atoms from the anode will turn into energy is calculated with the Sancho-Rubio
ions by oxidation reaction and diffuse through the method[7].
solid electrolyte to the cathode. At cathode ions The calculation is performed at Conte
gain electrons and turn into atoms. The atoms start supercomputer. Each compute node is equipped
to form metallic conductive filaments from with two 8-core Intel Xeon-E5 processors and two
cathode. Devices will switch from high resistance 60-core Xeon Phi coprocessors. The calculation is
to low resistance when the filaments keep growing paralleled at each node with 2 MPI ranks and 8
and finally reach anode. OMP threads. The computationally heavy matrix
The conductance changes with respect to operations have been offloaded to Xeon Phi
morphologies of conducting filaments are coprocessors to gain maximum performance which
important for evaluating performance and gives speed up 3.2×. It takes roughly 800 seconds
reliability. Tunneling through electrolyte is to finish 1 energy point on 1 MPI rank.
potentially harmful for achieving high resistance
ratio. Quantum simulations are important in RESULTS
predicting conductance changes in the switching In this work, silicon oxide is considered as
processes. electrolyte sandwiched between electrodes made
from Cu. Cu is parameterized with sp3d5 orbitals
METHODS
based on the environmental dependent empirical
The electrochemical process modeling requires tight binding (ETB) model[8] with coupling up to
application of different physical models in 2nd nearest neighbor. The model has been
different scales. The filament formation is optimized to match ab-initio bandstructure
simulated by molecular dynamics (MD) with the calculations of several ultra-thin nanowire unit
ReaxFF reactive force field [2-4]. Quantum cells.
transport is simulated with non-equilibrium SiO2 parameterization is done with two
Green’s function [5]. crystalline forms (α-quartz and β-cristobalite) and
The atom positions and the Mulliken charges used for the amorphous system as shown in Fig. 2.
calculated from charge equilibrium calculation Interactions between Cu and O atoms are strong
under external electrochemical potentials are taken and favored in the interface relaxation. The Cu-O
from the MD simulations. The charges are used to interaction is parameterized from Cu2O.
calculate the electrostatic potential with Poisson The simulated structures contain 6651 atoms
equation. The atomic charge density is assumed to with a cross section of 4nm×4nm. The separation
ISBN 978-0-692-50554-0 55
of two electrodes in the initial structure is 1.5nm as
shown in Fig.3a. The filaments generated at
different stages (t=0, 250, 500, 1000 ps in Fig. 3 a-
d) from MD are taken as inputs for the ETB
calculations for ballistic conductance. The time
step for MD simulations is 0.5fs.
With the generated structures, current through
the devices are calculated at Vd=-1V. Figure 4
shows the calculated current at different times at
Vd=-1V and the transmission coefficient at t=0ps
and 250ps. It is shown as soon as the Cu atoms Figure 2 Bandstructure of (a) α-quartz and (b) β-
start to form clusters inside SiO2, the current cristobalite calculated with LDA and tight-binding.
increases. Current ratio of I(t0=0ps)/I(t1=250ps) =
195 is obtained. When more filaments are
constructed, the resistance is lowered and the
current will keep increasing.
REFERENCES
[1] M. Kund, et al., Electron Devices Meeting, 2005. IEDM
Technical Digest. IEEE International, pp. 754-757, 2005.
[2] N. Onofrio, et al., Nat Mater, vol. 14, pp. 440-446, 2015.
[3] A. C. T. van Duin, et al., The Journal of Physical
Chemistry A, vol. 105, pp. 9396-9409, 2001.
[4] S. Plimpton, Journal of Computational Physics, vol. 117,
pp. 1-19, 1995.
[5] S. Steiger, et al., IEEE Trans on nano, vol. 10, 2011.
[6] M. Wimmer and K. Richter, Journal of Computational
Physics, vol. 228, pp. 8548-8565, 2009. Figure 3 Diffusion of Cu atoms in SiO2 at t=0, 250, 500,
[7] M. P. L. Sancho, et al., Journal of Physics F: Metal 1000ps. Si and O atoms are not plotted for better visibility. (a)
Physics, vol. 15, p. 851, 1985. Distance between electrodes is 1.5nm in initial structure. (b)
[8] G. Hegde, et al., Journal of Applied Physics, 115, 123703 Clusters are formed in SiO2. Two electrodes are not connected
2014. by filaments. (c) Two electrodes are connected by Cu
filaments. The connectivity is plotted based on a coupling
radius of 0.39nm. (d) More filaments are formed.
Figure 1 Simulation flow. Structures and charge profiles

are generated by MD simulations. The electrostatic potential Figure 4 (a) Current for structures at Fig.3 at V d=-1V. (b)
is calculated based on atomic charges. Current is calculated by Total transmission at t=0ps and t=250ps.
NEGF and conductance is extracted.
ACKNOWLEDGEMENT
This work was supported by the Semiconductor Technology Advanced Research Network (STARnet) Center of
Function Accelerated Nanomaterial Engineering (FAME) (Task 2382.004), one of six centers supported by the
STARnet phase of the Focus Center Research Program (FCRP). Funding for STARnet is sponsored by DARPA and
MARCO. The use of nanoHUB.org computational resources operated by the Network for Computational
Nanotechnology funded by the US National Science Foundation under Grant Nos. EEC-0228390, EEC-1227110,
EEC-0228390, EEC-0634750, OCI-0438246, OCI-0832623 and OCI-0721680 is gratefully acknowledged.
56
Non-Equilibrium Green's Function (NEGF)
Method: A Different Perspective
Supriyo Datta
Electrical Engineering Building, 465 Northwestern Avenue,
West Lafayette, IN 47907, USA
The second viewpoint seems more in keeping

INTRODUCTION
with the Boltzmann approach identified with the
The NEGF method was established in the 1960’s equation:
through the classic work of Keldysh [1] and others ∂f   
using the methods of many-body perturbation + ν .∇f + F.∇ p f = Sop f
∂t (3)
theory (MBPT) and this apporach is widely used.
By contrast I have introduced a different approach and NOT with the evaluation of the scattering
starting with the one-electron Schrodinger operator Sop which is analogous to the sigma’s in
equation [2] and in this talk I will try to answer the (1) and (2). Indeed Boltzmann himself was
quaestions I often get regarding the relation unaware of the Fermi’s golden rule commonly
between them. used nowadays to evaluate the Sop appearing in the
equation bearing his name.
NEGF EQUATIONS Similarly, totally new approaches for
Our central equations evaluating the sigma’s have been and will be
GR = [EI − H − Σ] −1 developed as we apply NEGF to newer problems
(1) like Coulomb blockade.
n R in A
G = G Σ G (2) CONTACT-ING SCHRODINGER
are essentially the same as Eqs.(75)-(77) of [1] In other words, we feel that the scope and
which is one of the seminal founding papers on the utility of Eqs.(1) and (2) transcends the MBPT-
NEGF method. based approach originally used to derive it. It
We obtain Eqs.(1) and (2) directly from a one- teaches us how to combine quantum dynamics
electron Schrodinger Equation using elementary with “contacts”, much as Boltzmann taught us how
arguments and use them to discuss many problems to combine classical dynamics with “contacts”,
of great interest like quantized conductance, using the word “contacts” in a broad figurative
(integer) quantum Hall effect, Anderson sense to denote all kinds of irreversible processes.
localization, resonant tunneling and spin transport. We feel that this should be a part of the training
All these problems involve quantum transport of all science and engineering students so that they
but do not require a systematic treatment of many- can apply it effectively to a wide variety of basic
body effects. On the other hand it goes beyond and applied problems that require “connecting
purely coherent transport allowing us to include contacts to the Schrodinger equation“.
phase-breaking interactions (both momentum-
relaxing and momentum-conserving) within a self- REFERENCES
consistent Born approximation (see Figure 2 for an [1] Keldysh L.V. (1965) Diagram Technique for
illustrative example). non-equilibrium processes. Sov.Phys. JETP 20,
1018. We have changed the notation somewhat,
SHOULD WE CALL THIS NEGF?
writing Σ for Σ R , G n for − iG < and Σ in for − iΣ <
* The answer is NO, if we associate NEGF with [2] S. Datta, Electron Transport in Mesoscopic
the MBPT used to obtain the sigma’s appearing in Systems, Chapter 8, Cambridge (1995); Quantum
Eqs.(1) and (2). Transport, Atom to Transistor, Cambridge (2005);
* The answer is YES, if we associate NEGF with Lessons from Nanoelectronics, World Scientific
Eqs.(1) and (2) irrespective of how the sigma’s are (2012)
obtained.
ISBN 978-0-692-50554-0 57
Fig.1. The NEGF-based quantum transport model
allows us to model current flow given the
Hamiltonian matrix [H] describing the channel,
the self-energy matrices [Σ1, 2 ] describing the
connection of the channel to the contacts, and [Σ 0 ]
describing interactions within the channel.
The NEGF equations can be obtained from a one-

electron Schrodinger equation modified to include
the effect of contacts through the terms of Σ and
s. An NEGF model that includes phase relaxation
smooths out the oscillations while a model with
momentum relaxation adds a slope to the potential
as expected for a distributed resistance. All models
assume elastic scattering.
Fig.2. A simple example involving the

transmission through a scatterer. The occupation
factor f (proportional to the electrochemical
potential) calculated from a coherent NEGF model
shows oscillations around the semiclassical result.
58
Mode space tight binding model for ultra-fast
simulations of III-V nanowire MOSFETs and
heterojunction TFETs
A. Afzalian, J. Huang*, H. Ilatikhameneh*, J. Charles*, D. Lemus*, J. Bermeo Lopez*, S. Perez
Rubiano*,T. Kubis*, M. Povolotskyi*, G. Klimeck*, M. Passlack, and Y.-C. Yeo.†
TSMC, Kapeldreef 75, 3001 Leuven, Belgium and †TSMC Hsinchu, 300-75 Taiwan
*
Network for Computational Nanotechnology, Purdue University, West-Lafayette, IN, USA
As transistors are scaled into the nanoscale high (typically above 100) compared to the Si case
regime, III-V channel materials and gate-all-around reported in [3] which has only a few UM.
(GAA) nanowire (NW) MOSFETs and TFETs can Nevertheless, we demonstrate here the possibility
be used to achieve performance boost and low to clean III-V MS basis from the UM and achieve
power operation [1,2]. Quantum effects play a a significant speed up ratio (Fig. 3 and 4).
crucial role in affecting device performance. The first step that consists in creating an initial
Atomistic tight binding (TB) simulations within the MS basis by sampling the NW slab eigenmodes at
NEGF framework has emerged as a powerful and selected E-k points is the crucial step upon which
versatile technique for full band transport studies relies the successful derivation of an optimized
that goes beyond the limitation of effective mass basis. A successful initial sampling is illustrated in
(EME) approximation [3,4]. Atomistic transport Fig.1 for a [100] InAs NW in a sp3s* basis
simulations in realistic nanostructures having a few including Spin Orbit coupling (SO) and the
tens of thousands of atoms typically translate into number of UM after 5 and 75 optimization steps is
Hamiltonian matrices that have a rank of ~106 and shown in Fig. 2 and 3, respectively. As shown in
require substantial computational resources. Fig. 1, the key point is to start with sampling a
Using an incomplete basis or low rank small number of eigenmodes within a reduced
approximation, the mode space (MS) approach has energy window and optimize this basis first. It is
been widely used to significantly speed up EME usually not possible to optimize a larger basis
models while keeping accuracy in a reduced directly. If needed a larger energy range can then
energy window [5-7]. A direct transcription of the be achieved by performing a second initial
MS technique to TB models usually fails. In a full sampling (e.g. adding new basis vectors at higher
band model, the device Hamiltonian is no longer a energy) to the first optimized basis (i.e. the
bounded operator. Thus, the basis transformed optimized basis of Fig. 3) and perform a second
model is plagued with unphysical modes (UM) optimization.
(Fig. 2), even if the chosen basis reproduces well Using the procedure illustrated in Fig. 1 to 3,
the physical modes of interest (Fig. 1) [3]. MS basis for [100] InAs and GaSb NWs with a
A solution to this problem was proposed and its square cross-section of size d = 5.45 nm were
applicability to Si NWs demonstrated. The derived from an initial sp3s* SO TB basis and slab
technique enlarges the MS basis by adding new Hamiltonian sizes in the MS were 5% and 6%,
basis vectors specially “tailored” to remove one by respectively, from their original size. The MS
one the unphysical branches in the MS model [3]. basis sets were used to simulate an InAs MOSFET
In this work, we have implemented such a MS and an InAs-GaSb heterojunction TFET. I-V
algorithm in Nemo5 [4] and explored its suitability curves obtained from the MS models match well
to various III-V homo- and heterojunction NW those obtained in the full model (Fig. 4). Using the
devices. We found that in III-V materials, the MS models, speed up factors larger than 150 were
number of unphysical modes to eliminate is very achieved, when compared to the initial TB models,
ISBN 978-0-692-50554-0 59
for both the MOSFET and TFET cases, using a [3] G. Mil’Nikov et al., Equivalent transport models in
cluster having 96 cores. atomistic quantum wire, Phys. Rev. B 85, 035317 (2012).
[4] S. Steiger et al., NEMO5: A Parallel Multiscale Nanoelectronics
In conclusion, the achievement of significant Modeling Tool, IEEE TNANO 10, 1464 (2011). J. Wang, E.
speed up factors demonstrates the potential of MS Polizzi and M. Lundstrom, Journal of Applied Physics, Vol. 96
tight binding models and offers unprecedented (4), pp. 2192-2203 (2004).
[5] J. Wang et al., A Three-Dimensional Quantum Simulation of
possibilities for the full band simulation of Silicon Nanowire Transistors with the Effective-Mass
nanostructures. Approximation, JAP 96 (4), 2192 (2004).
Acknowledgment. The authors thank Dr. Y. C. [6] A. Afzalian et al., A new F(ast)-CMS NEGF Algorithm for
Sun for management support. efficient 3D simulations of Switching Characteristics
enhancement in constricted Tunnel Barrier Silicon Nanowire
REFERENCES MuGFETs, JCE 8, 87 (2009).
[7] A. Afzalian, Computationally Efficient self-consistent Born
[1] J. del Alamo, Nanometre-scale electronics with III–V
approximation treatments of phonon scattering for Coupled-
compound semiconductors, Nature 479, 317 (2011).
Mode Space Non-Equilibrium Green’s Functions, JAP 110,
[2] A. Dey et al., High Current Density InAsSb/GaSb Tunnel
094517 (2011).
Field Effect Transistors, EDL 34, 211 (2013).
Fig. 2. Band structure obtained using MS basis (×) for the

Fig. 1. Band structure (E-k dispersion in the first half of the same structure as in Fig. 1 after 5 optimization steps, i.e.
Brillouin Zone) of a [100] InAs NW slab (d = 5.45 nm) adding 5 new basis vectors to remove 5 unphysical modes.
computed from the original (Real Space) TB model in a sp3s* The original band structure (▼) obtained using Real Space
basis including SO coupling (▼). The E-k points used to (RS) basis is also shown. Many UM are still present in the
construct the initial MS basis (from the related eigenvectors) MS model, which is not suitable for transport studies.
are indicated by green circles.
Fig. 3. Band structure of the MS model (×) of Fig. 1 after 75

optimization steps. The original band structure (▼) obtained Fig. 4. ID(VG) characteristics of a [100] d = 5.45nm 15 nm
using RS basis is also shown for comparison. The lowest part long InAs NW MOSFET and InAs-GaSb TFET computed
of the conduction band and higher part of the valence band from the original (RS) TB model in a sp3s* basis including SO
are free of the UM and correctly reproduce the band structure coupling and from the reduced InAs MS basis of Fig. 3. For
obtained using RS basis. The MS model is suitable for the TFET case a GaSb MS basis was also derived. Using the
transport in a reduced energy window. The MS basis size is MS models, good matching of the I-V characteristics together
only a few % of the original NW slab TB model. with large speed up factors were achieved when compared to
the original model.
60
Time-resolved computational method for
atomistic open system simulations
B. Novakovic and G. Klimeck
Network for Computational Nanotechnology, Purdue University, West Lafayette, USA
I NTRODUCTION scattered-mode space DOFs in hE (t) contain multi-

Time-resolved quantum transport simulations are ple energies due to the time-dependent Hamiltonian.
useful for accurate predictions of turn on energy- Closely spaced energies may lead to an unstable
delay characteristics of logic devices [1]. Further- matrix preventing straightforward inclusion of ever
more, time-resolved quantum simulations provide more energies in the scattered mode space. The
means to study novel device phenomena, e.g. elec- wide-band approximation (WBA) can be applied if
tron pumps [5]. They can also be utilized for the excitation rate is not too high. It amounts to
dissipation calculations due to phonon scattering keeping only the injected energy in the scattered-
with a multiscale time-stepping algorithm where mode space, equivalent to duΣ(t − u)ΨE (u) →
phonons are real space lattice vibrations that enter Σ(E)ΨE .
the Hamiltonian as inhomogeneous strain.
R ESULTS
We present a method based on the mixed energy-
time space scattering matrix approach. It has clear Time resolved simulations are performed for a
connection to Green’s functions [2] and the fol- Si nanowire with 2.1x2.1nm cross section and
lowing advantages. First, as an integral part of ∼ 20nm length in atomistic sp3 d5 s∗ basis. The sys-
our comprehensive simulation tool [4], it can be tem has ∼ 49 · 103 orbital DOFs and ∼ 270 modes
used to simulate realistic materials and geometries per energy. The time-dependent bias is a 4nm wide
in atomistic tight-binding basis. Second, for not rectangular barrier in the center of device, whose
too high excitation rates it directly relies on the 10mV amplitude has a variable mV/ps rising edge.
efficient transfer matrix method (TMM) for lead The time resolved data is calculated in the WBA.
self-energies [3]. Fig. 1 shows the 3D atomic wf magnitude and plots
its evolution along a chain of atoms. Initial homo-
T HEORY AND C OMPUTATIONAL ASPECTS geneous left injected state evolves into the wf that
Partitioning the wave function (wf) into injected shows reflection due to the potential barrier with
and scattered lead Bloch modes and exposing the 10mV amplitude and 1mV/ps rising edge. There
scattering matrix, we arrive at i∂ψE (t) /∂t = are no reflection artifacts at the open boundaries.
[hE (t) − E (t)] ψE (t) + IE (t) , where ψE , hE , and Crank-Nicholson time-stepping scheme is used with
IE are the wf, Hamiltonian, and lead injection term dt = 1fs. Fig. 2, shows the corresponding time-
in the mixed scattered-mode/orbital space, and E(t) resolved reflection R and transmission T proba-
is the injection energy. Equivalently, the scattered- bilities, while Fig. 3 shows the probability current
mode space degrees of freedom (DOF) can be conservation (R + T) for 1mV/ps, 2mV/ps, and
eliminated by explicitly calculating the self-energy 10mV/ps rising edges. Probability conservation is
Σ(t) in orbital space. less rigorous for higher excitation rates in the WBA,
When the above equation is used directly with the due to a larger energy variation of the injected state.
lead TMM the phase coherence between the device The upper limit for the validity of the WBA is
and leads may be lost, due to the arbitrary phase of ∼ 1mV/ps, which corresponds to a few GHz in
the eigensolver. This can be resolved by consistent realistic transistors. Interestingly, all three excitation
phase normalization in the leads. Furthermore, the rates show identical long time wf, Fig. 4.
ISBN 978-0-692-50554-0 61
Fig. 2. T and R corresponding to data in Fig. 1.
Fig. 1. Top: s orbital wave function magnitude along a chain

of atoms at 1.8x0.4nm for left injection at energy 1.515eV (2
propagating modes). The barrier rising edge is 1mV/ps and the
total rise time is 10ps. There are no visible artificial reflections
at device/lead interfaces. Middle and bottom: diagonal cross
section of the nanowire and the selected chain of atoms
corresponding to the top figure at 15ps.
ACKNOWLEDGMENT Fig. 3. Probability current conservation (T + R) for 1mV/ps,

2mV/ps, and 10mV/ps rising edges. Upper validity limit for
Funding provided by the U.S. Army Re- the wide-band approximation around 1mV/ps, corresponding
search Office contract No. W911NF-12-1-0607 to a few GHz in realistic transistors.
and W911NF-08-1-0527. Computations done on
nanoHUB.org funded by the NSF grant No. EEC-
1227110.
R EFERENCES
[1] D. E. Nikonov and I. A. Young, IEEE Journal on Ex-
ploratory Sold-State Computational Devices and Circuits,
DOI: 10.1109/JXCDC.2015.2418033 (2015).
[2] B. Gaury, J. Weston, M. Santin, M. Houzet, C. Groth, and
X. Waintal, Phys. Rep. 534, 1-37 (2014).
[3] M. Luisier, A. Schenk, and W. Fichtner, Phys. Rev. B 74,
205323 (2006).
[4] J. E. Fonseca, T. Kubis, M. Povolotskyi, B. Novakovic,
A. Ajoy, G. Hegde, H. Ilatikhameneh, Z. Jiang, P. Sen-
gupta, Y. Tan, and G. Klimeck, J. Comput. Electron. 12,
592 (2013). Fig. 4. Unlike Fig. 3, long time open boundary wave function
[5] G. Stefanucci, S. Kurth, A. Rubio, and E. K. U. Gross, magnitudes match well for three different excitation rates.
Phys. Rev. B 77, 075339 (2008).
62
15-band spectral envelope function formalism
applied to broken gap tunnel field-effect transistors
D. Verreck∗ , M. Van de Put† , A.S. Verhulst, B. Sorée †∗ , W. Magnus† ,
A. Dabral∗ , A. Thean and G. Groeseneken∗
imec, Kapeldreef 75, 3001 Leuven, Belgium
∗ Department of Electrical Engineering, KU Leuven, Belgium
† Department of Physics, Universiteit Antwerpen, Belgium
A carefully chosen heterostructure can signifi- in the spectral expansion prevents highly oscillating
cantly boost the performance of tunnel field-effect out-of-zone spurious solutions from polluting the
transistors (TFET) [1], [2]. Modelling of these first Brillouin zone (see Fig. 2). Fig. 3 shows the
hetero-TFETs requires a quantum mechanical (QM) full simulation procedure.
approach with an accurate band structure to allow In order to use the same EF basis set through-
for a correct description of band-to-band-tunneling out the heterostructure, the matrix elements pnm
(BTBT). Reflections at the heterojunction, as well and Hnm (see Fig. 1) are transformed [3]. This
as field-induced and size-induced confinement, also transformation requires a common eigenvalue de-
need to be considered, as they can heavily influence composition, and therefore commutativity, of the
TFET performance. We have therefore developed a momentum matrices. For the exact case of infinite
fully QM 2D solver, combining for the first time a bands, this is satisfied. Momentum matrix elements
full zone 15-band envelope function (EF) formalism in literature for finite band models, however, have
with a spectral approach, including a heterostructure lost this feature in the fitting process. We have
basis set transformation. therefore developed a constrained optimization al-
Our formalism is based on the EF formalism with gorithm, to search for k·p-parameters that reproduce
quantum transmitting boundary conditions [3]. In the most relevant band structure features, while
going to a 15-band description, we replace the finite retaining commutativity of the momentum matrices,
difference (FD) scheme in the confined direction hence allowing for a basis set transform (see Fig. 4).
with a spectral approach, in which we expand the Figs. 5 and 6 show simulation results for a
odd (Fn ) and even (Fm ) EFs, and the external broken gap diode and p-n-i-n TFET. The wider
potential Ve as follows: body device has a deteriorated SS as a result of
+∞
poorer electrostatic gate control. The EF formalism
Fn (x, z) = F̃n (x, kzµ ) sin(kzµ z) (1) allows for a transparent analysis of the transmission
µ=−∞ spectra of individual subband modes (see Fig. 5).
+∞
The computational efficiency of the spectral method
Fm (x, z) = F̃m (x, kzµ ) cos(kzµ z) (2) permits the simulation of wide structures: one 30 nm
µ=−∞ body TFET biaspoint only takes about 15 min using
+∞
10 cores on a state-of-the-art server.
Ve (x, z) = Ṽe (x, kzµ ) cos(kzµ z) (3) In conclusion, we showed a full-zone EF for-
µ=−∞ malism using a spectral method, along with a con-
with x (z ) the transport (confinement) direction, and strained optimization to obtain k·p-parameters that
translational invariance in y . The parity of the EFs is retain commutativity of the momentum matrices.
determined by the band coupling. Inserting Eqs. (1)- We demonstrated the formalism for hetero-diodes
(3) in the EF system finally results in the system in and TFETs, highlighting the transparency of the
Fig. 1. If Nx x-grid points, N bands and Nkz spec- analysis and the computational efficiency compared
tral components are considered, the system has di- to existing QM BTBT solvers.
mensions Nx × Nkz × N after discretization, instead
D.Verreck acknowledges support by IWT-Vlaanderen. This
of Nx × Nz × N in a FD approach. Since spectral work was supported by imec’s Industrial Affiliation Program.
methods provide exponential accuracy, compared to [1] H. Lu et al., IEEE J. Elec. Dev. Soc. 2, 44-49 (2014).
only polynomial accuracy for FD, Nkz can be much [2] U. Avci et al., IEEE J. Elec. Dev. Soc. 99, 1 (2015).
lower than Nz , especially when the EFs vary slowly [3] D. Verreck et al., J. Appl. Phys. 115, 053706 (2014).
in the confined z -direction. Additionally, the cut-off [4] S. Ben Radhia et al., Semicond. Sci. Technol. 22, 4 (2007).
ISBN 978-0-692-50554-0 63
Initialize P-matrices
Add Calculate band structure Constrained
constraint
minimization
Evaluate scoring function
Check visually
Yes No
Irregularities? Finish
Fig. 4. Parameter search algorithm. The scoring function is

a weighted sum of differences of calculated and experimental
values from literature of effective masses and bandgaps at the
Γ, X-and L-valleys. The constraints ensure commutativity of
© IMEC 2014 / CONFIDENTIAL DEVIN VERRECK
the momentum matrices Px , Py and Pz .
Fig. 1. System of EF equations after the spectral transform.

The plus(minus)-sign in the potential term corresponds to an
even(odd) EF. me is the free electron mass. pnm and Hnm
represent the k·p interband momentum and Hamiltonian matrix
elements respectively, which couple band n to band m. Note
that the potential term becomes the sum of a convolution and
an autocorrelation of the spectral components of the potential.
5 5 54 5 5 5
4 4 43 4 4 4
3 3 32 3 3 3
2 2 21 2 2 2
E (eV)
E (eV)
E (eV)
E (eV)
E (eV)
E (eV)
E (eV)
1 1 10 1 1 1 Fig. 5. Individual subband transmission probabilities at ky = 0

for a broken gap diode, illustrating the transparant data analysis.
0 0 0-1 0 0 0 Parameters have been obtained with the algorithm in Fig. 4.
Doping is 5x1019 cm−3 in the p- and n-regions. The arrow
-1 -1 -1-2 -1 -1 -1 indicates increasing subband number.
-2 -2 -2-3 -2 -2 -2-0.5
-1 -1 -0.5 -1 -1-0.5 -0.5
-1-0.5 00 0.5
0-0.5 0.5
-1-0.5
0.50 0.5 -0.5 1 00 0.5
-11 1 0.5
0-0.5 0.5 0.511
1 0.5 1
kx (1/m)kx (1/m) kx (1/m) 10 kx
xx10 x 10 kx
1010 (1/m)10
kx(1/m)
x 10 (1/m)
10
x 10
x 10
10
10
x 10
10
Fig. 2. 15-band kx (1/m) band structure calculation for a 10 nm

confined
InAs slab using a finite difference scheme (left) and a spectral
approach with 3 spectral components (right) in the confined
z-direction. The spectral approach prevents spurious solutions
from folding back onto the zone center and hence results in a
correct confined band structure description. Colors indicate the
dominant band contribution to each state.
Each energy and each ky
Solve contact eigenvalue problem

Each mode
Construct boundary conditions

Fig. 6. I-V curves for broken gap p-n-i-n TFETs with
Solve for envelope functions Tbody 30 nm and 20 nm. One I-V curve is generated in
about 2.5 h on a 10 core server, allowing for an efficient
Calculate current architecture optimization. Source, pocket and drain doping are
1x1019 cm−3 , 1x1019 cm−3 and 5x1017 cm−3 respectively.
Fig. 3. Flowchart of the developed formalism. Pocket thickness is 1 nm. The gate stack has an equivalent
oxide thickness of 0.6 nm and a metal workfunction of 4.5 eV.
64
Progress on quantum transport simulation using
empirical pseudopotentials
Jingtian Fang∗ , William Vandenberghe, and Massimo Fischetti
The University of Texas at Dallas, 800 W. Campbell Road, Richardson, Texas 75080 USA
∗ Email: [email protected]
I. I NTRODUCTION the contacts (self-energies). Using finite difference (FD) and

After performing one-dimensional simulation of electron discretizing a device into Nz slices along the transport di-
transport in narrow quantum wires without gate control in rection with a step-size ∆, the Hamiltonian H takes a block
Ref [1], [2] using the open boundary-conditions full-band tridiagonal form
plane-wave transport formalism derived in Ref [3], we now  
· · · · · · ·
extend the work to simulate three-dimensionally field-effect · T− Di−1 T+ 0 0 ·
transistors (FETs) with a gate bias applied and obtain their  

H = · 0 T −
D i
T +
0 · (3)
transport characteristics. We optimize multiple procedures for ,
· 0 0 T− Di+1 T+ ·
solving the quantum transport equation (QTE), such as using · · · · · · ·
a selected eigenvalue solver, the fast Fourier transform (FFT),
block assignment of matrices, a sparse matrix solver, and where the operator Di is a dense matrix of rank NG that in-
parallel computing techniques. With an expanded computing cludes the lattice and external potential (gate and source/drain
capability, we are able to simulate the transistors in the sub- bias) in the i-th slice, and the Hermitian operators T+ and T−
1 nm technology node as suggested by the ITRS [4], which are diagonal matrices also of rank NG . Therefore, the rank
features ∼5 nm physical gate length, ∼2 nm body thickness, of H is NG × Nz . The self-energy matrix ΣL is added to
∼0.4 nm effective oxide thickness (EOT), ∼0.6 V power D1 and ΣR is added to DNz . Solving Eq. (2) and Poisson’s
supply voltage, and a multi-gate structure. Here we simulate equation self-consistently for different drain-source bias and
an armchair graphene nanoribbon (aGNR) FET using a gate- gate bias, the current-voltage characteristics of the device can
all-around architecture and obtain its transport properties. We be obtained. Since we have not yet implemented in our com-
will discuss the numerics concerning the matrix size of the puter program the capability to account for bound and quasi-
transport equation, memory consumption, and simulation time. bound states, we are unable to perform full self-consistent
calculations. Therefore, for now we solve Eq. (2) with an
II. T HEORY
external potential obtained from a self-consistent calculation
As shown in Ref. [2], the Schrödinger equation describing within a simple semiclassical drift-diffusion approximation.
the electronic band structure in one-dimensional (1D) nanos- This is performed by solving the two-dimensional Schrödinger
tructures in the (x-z) plane, surrounded by a large vacuum equation in all slices, calculating the charge density using
padding in the (x-y) plane to decouple the structures, using local quasi-Fermi level, and then solving Poisson’s equation
the empirical pseudopotential method is to update the potential. An approximate device current is
h̄2 calculated from the obtained envelope wavefunctions φ.
|k + G| δG G + VG G ukGz = Ekz ukGz ,
2 lat
(1)
2m
G III. S IMULATIONS AND RESULTS
where k = (0, 0, kz ) is the wavevector along the axial z- An atomistic model of an aGNR with six dimer lines along
direction. This constitutes an eigenvalue problem of rank given the width direction x is shown in Fig. 1a and a device model
by the number of reciprocal-space vectors NG within the of 6-aGNRFET is shown in Fig. 1b. The channel has p+
energy-cutoff sphere of radius Ecut . The QTE describing the doping and the source and drain contact have n+ doping. The
1D electron transport along the z direction is a linear system discretization is shown in Fig. 1c and 1d. The band structure
of the form of the 6-aGNR is calculated by solving Eq. (1) and shown in
inj
(H − EI + ΣL + ΣR )φ = [RHS]L + [RHS]R ,
inj
(2) Fig. 2. The self-consistent external potential and charge distri-
bution within the aforementioned semiclassical drift-diffusion
where H is the Hamiltonian for the closed system, E is simplification are shown in Fig. 3 and their distribution after
the injection energy of electrons, I is the identity matrix, non-equilibrium quantum transport simulation are shown in
ΣL and ΣR are the self-energies of the contacts, φ is the Fig. 4. The I-V characteristics are shown in Fig. 5. Some
inj inj
envelope wavefunction, and [RHS]L and [RHS]R are the information about the numerical details of the calculations is
terms representing the amplitudes of the waves injected from shown in Table. I.
ISBN 978-0-692-50554-0 65
1.0 0.30
x [nm]
0.5 0.15
H termination a3 0.0
armchair edge 0 2 4 6 8 10 12 0.00
1
2
3
z [nm]
4
x 5
(a) Potential distribution

6
z
×1026
(a) 6-aGNR model (a3 =3dC=C = 0.43nm) (b) 6-aGNRFET model
1.0 4.5
x [nm]
0.5 3.0
0.0 1.5
0 2 4 6 8 10 12 0.0
z [nm]
(b) Charge distribution
Fig. 3: Approximate potential and charge distribution, aver-

aged along y, in the (x-z) plane after self-consistent calculation
(c) Discretization along z (d) Discretization in (x-y)
with the semiclassical assumption for a 6-aGNRFET with a
Fig. 1: Device model of 6-aGNRFET with an overall length drain-source bias VDS =0.4 V and a gate bias VGS =0.2 V.
L=30a3 =12.78 nm, a physical channel length Lg =5.12 nm,
a body width W=0.62 nm, and an EOT=0.62 nm. Electron
1.0 0.4
x [nm]
transport is along the z direction. The device is discretized 0.5 0.2
0.0
along z into Nz =121 slices with ∆=a3 /4=0.11 nm and 0 2 4 6 8 10 12
0.0
discretized in the cross-sectional (x-y) plane using a finite z [nm]
element mesh. Notice that this particular triangular mesh is
(a) Potential distribution
for FFT calculations.
×1027
1.0 1.2
7 x [nm] 0.5 0.8
0.4
6 0.0
Energy (eV)
5 0 2 4 6 8 10 12 0.0
4 z [nm]
3 4
2 6
y [Å]
1 2 4 (b) Charge distribution

0 2
0 0.2 0.4 0.6 0.8 1 0
0 2 4 6 8 10 0 Fig. 4: Potential and charge distribution, averaged along y, in
k (π/(3dC=C))
x [Å] the (x-z) plane from Eq. (2) for a 6-aGNRFET. Quasi-bound
(a) 6-aGNR band structure (b) |ψ(r|| )|2 distribution in (x-y) states can be observed in the region close to the drain contact.
Note that variation of the electron density corresponds to the
Fig. 2: The electronic dispersion of the first six low-energy location of individual atoms.
conduction bands (CBs) is shown in (a). Within 100 meV
from the bottom of the CB, only one conducting channel
is observed. The electron probability distribution |ψ(r|| )|2 is 4
101
shown in (b). This is obtained by averaging the square of 3 -0.2 10

0
IDS (µA)
IDS (µA)
the cross-sectional wavefunction over one unit cell along z. A 0.0

0.2 10 -1
2
selected eigenvalue solver from IBM ESSL package is used to 0.4
0.6
10-2
0.1
0.2
1
calculate the bands in the interesting energy range, and FFT 0.3
0.4
0 10-3
is used to calculate |ψ(r|| )|2 from Eq. (1). 0 0.1 0.2 0.3 0.4 -0.4 -0.2 0 0.2 0.4 0.6
VDS (V) VGS (V)
(a) ∆ = az /4 (b) ∆ = az /4
R EFERENCES
Fig. 5: Current-voltage characteristics of a 6-aGNRFET.
[1] J. Fang, W. Vandenberghe, and M. Fischetti, “Full-band ballistic quantum
transport in nanostructures using empirical pseudopotential,” in Compu-
tational Electronics (IWCE), 2014 International Workshop on. IEEE,
TABLE I: Rank of the matrices, simulation time, and
2014, pp. 1–3. memory consumption when simulating a 6-aGRNFET
[2] B. Fu and M. Fischetti, “Open boundary-condition ballistic quantum
transport using empirical pseudopotentials,” in Computational Electronics NG t (s, Eq. (1)) t (h) LDM t (h) Memory (GB)
(IWCE), 2013 International Workshop on. IEEE, 2013, pp. 28–29. 701 10 2.5a 84821 0.2b 5.0c
[3] B. Fu, “Quantum transport: from effective mass approximation to full a
Simulation time required to obtain the starting potential self-
band,” Ph.D. dissertation, The University of Texas at Dallas, Richardson, consistently for each applied bias.
TX, 2013. b
Simulation time required to solve Eq. (2) for a single injection
[4] International Technology Roadmap for Semiconductor, “Process integra-
tion, devices, and structures,” [Online], 2013.
energy E. MPI is used to solve Eq. (2) in parallel for all the
injection energies.
c
Peak memory required to solve Eq. (2). A sparse matrix
solver UMFPACK is used to save memory.
66
1
Phonon interactions in single-dopant-based

transistors: temperature and size dependence
M. Bescond∗ , N. Cavassilas∗ , H. Carrillo-Nuñez† , S. Berrada∗ and M. Lannoo∗
∗ Aix Marseille Université, CNRS, IM2NP UMR 7334, 13397, Marseille, France
† Integrated Systems Laboratory ETH Zürich, Gloriastrasse 35, 8092 Zürich, Switzerland
Theoretical studies have shown that single-dopant- scattering decreasing in wider nanowires [4], Fig.3 first
based transistors could generate remarkable room- shows ID − VG characteristics for a 3×3 nm2 cross-
temperature effects like current hysteresis [1] or negative section at T =300 K with the same LCH =8 nm. The
differential resistance [2]. Unfortunately these effects are current hysteresis still occurs in the ballistic regime but
potentially suppressed by electron-phonon interactions, it is removed in the presence of phonon interactions.
and in particular by acoustic phonon scattering [3]. In this Figure 4 explains this behavior by showing the local
work we investigate the dependence of electron-phonon density of states (LDOS) and the current spectra in the
scattering in single dopant-based nanowire transistor with ballistic and scattering regimes at VG =0.7 V. For the
respect to temperature and dimensions (Fig.1). We use a ballistic regime this bias corresponds to an impurity
3D real-space non-equilibrium Green’s function (NEGF) level located below the source energy sub-band (Fig.4-
approach where electron-phonon scattering is treated a). The current spectra presents then a maximum at
within the self-consistent Born approximation (SCBA) the subband-edge of the source (Fig.4-b). Adding the
through self-energies [3]. phonon scattering strongly couples the impurity state to
Within the effective mass approximation, the self- the contact (Fig.4-c) and generates a current spectrum
energy for acoustic phonons is commonly expressed as with important electron relaxations (Fig.4-d). In order
follows [3]: to reduce the coupling to the contacts we also plot in
Fig.3 ID −VG characteristics with a longer LCH =12 nm.
Ξ2 k B T < Interestingly current hysteresis is almost removed in both
Σ<
ac (E) = G (E), (1)
2ρu2s ballistic and scattering regimes. The main reason of the
where Ξ is the scalar deformation potential, ρ is the removal of the hysteresis is shown Fig.5 which shows
density and us the sound velocity of silicon. However LDOS and current spectra in the ballistic regime. When
Eq.(1) is an approximate expression only valid at high the level is above the source energy sub-band (Fig.5-
temperatures. Starting from the general formulation of a-b) the current occurs in the energy range around the
the self-energy, we first show that expression (1) remains impurity level. On the other hand, current spectra occurs
valid on the impurity site at the condition of replacing in higher energies than the source sub-band energy when
T by an effective temperature T ∗ given by: the impurity level is below (Fig.5-c-d) since the potential
barrier in 12 nm long channel increases and makes
∗ h̄ωc h̄ω1
T = T × exp − + , (2) negligible the tunneling component. Finally, Fig.6 shows
2kB T kB T
a strong dependence of the impurity binding energy with
where ωc is a critical frequency. T ∗ remains then close the cross-section. For a 4×4 nm2 nanowire the binding
to T if kB T ≥ h̄ω2 c and tends to a limiting value energy of the impurity is significantly reduced so that the
h̄ωc
2kB 60 K when T → 0. Based on this result, impurity level does not come anymore below the source
Fig.2 shows the effect of phonon scattering on current energy subband, making the presence of hysteresis even
characteristics for different temperatures in a 2×2 nm2 more complicated to obtain.
cross-section nanowire transistor with single impurity
located on the center of a 8 nm long channel. We see R EFERENCES
that the current hysteresis appears at T =150 K due to
[1] G. Mil’nikov et al. , Phys. Rev. Lett. 102, 036801 (2009).
the reduction of phonon scattering. As an illustration of
[2] M. Bescond et al. J. Appl. Phys. 107, 093703 (2010).
the influence of phonon scattering, current characteristics
[3] H. Carrillo-Nuñez et al. J. Appl. Phys. 116, 164505 (2014).
in the ballistic regime is also plotted for T =200 K, and
[4] M. Luisier et al., Phys. Rev. B 80, 155430 (2009).
shows a much larger hysteresis. The impact of phonon
ISBN 978-0-692-50554-0 67
2
-0.22 2
-0.24
1.5
Energy(eV)
-0.26
1
-0.28
0.5
-0.3
(a) (b)
-0.32 0
-0.22 2
-0.24
1.5
Energy(eV)
-0.26
Fig. 1. Schematic representation of the considered nanowire MOS- 1
FET with a dopant located at the center of the channel. In all the study -0.28
the source-drain bias VDS is 0.1 V, the source/drain donor doping ND -0.3
(c) (d)
0.5
equals 1020 cm−3 whereas TSiO2 = 1 nm. -0.32 0

4 8 12 16 20 4 8 12 16 20
x(nm) x(nm)
Fig. 4. Local density of states (first column) and current spectra

(second column) for LCH =8 nm in the ballistic regime (a-b) and with
phonons (c-d). In each sub-figure, the white line is the first subband
energy profile along the transport direction. Nanowire cross-section
is 3×3 nm2 . VG = 0.700 V.
1.4
-0.15
1.2
-0.2 1
Energy(eV)
0.8
-0.25
0.6
-0.3 0.4
(a) (b) 0.2

-0.35
0
0.6
-0.15
0.5
-0.2 0.4
Energy(eV)
Fig. 2. ID − VG characteristics in presence of phonon scattering -0.25

0.3
at T=100 K (squares), T=150 K (triangles) and T=200 K (circles). -0.3

0.2
As a comparison, current characteristics in the ballistic regime for (c) (d) 0.1
T=200 K is also represented. Nanowire cross-section is 2×2 nm2

-0.35
0
4 8 12 16 20 24 28 4 8 12 16 20 24 28
and LCH =8 nm. x(nm) x(nm)
Fig. 5. Ballistic local density of states (first column) and current

spectra (second column) for LCH =12 nm at VG = 0.450 V (a-b)
and VG = 0.475 V (c-d), for which the impurity level goes below
the source energy subband. Nanowire cross-section is 3×3 nm2 .
0.2 30
0.15
25
0.1
20
0.05
Energy(eV)
15
10
-0.05
5
-0.1
(a) (b)
0
4 8 12 16 20 4 8 12 16 20
x(nm) x(nm)
Fig. 3. ID − VG characteristics at T=300 K in a 3×3 nm2 cross-

section nanowire in presence of phonon scattering (circles) and in the Fig. 6. Ballistic LDOS in (a) a 2×2 nm2 cross-section nanowire for
ballistic limit (squares) for LCH =8 nm (solid lines) and LCH =12 nm VG =0.4 V and (b) a 4×4 nm2 cross-section nanowire for VG =0.3 V.
(dotted lines). LCH =8 nm.
68
Contact resistances in Trigate devices in a
Non-Equilibrium Green’s Functions framework
Léo Bourdet, Jing Li and Yann-Michel Niquet
University Grenoble Alpes, INAC/SP2M, L Sim, Grenoble, France
CEA, INAC/SP2M, L Sim, Grenoble, France
As the gate length L of field-effect transistors therefore be extrapolated to L = 0 (Fig. 3). The
is reaching the sub-20 nm range, the contact resis- contact resistance Rc = R(L → 0) in a 10x10
tances are increasingly limiting the electrical per- nm Trigate is plotted as a function of the carrier
formances of the devices. The “apparent” contact density in the channel in Fig. 4 (red line with
resistance Rc can be defined as the extrapolation to dots), for the “Reference” doping profile of Fig.
zero gate length of the total resistance of the device, 2. It is compared to the ballistic resistance of the
R(L) = Vds /Ids , where Ids is the drain current and channel (no scattering), and to the contact resistance
Vds the source-drain voltage. In the low field (low extracted in a device without surface roughness nor
Vds ) regime, this contact resistance is dominated by impurity scattering in the source/drain (continuous
i) the quality of the metal-semiconductor contact, background dopant distributions). The contact resis-
ii) the transport through the lowly doped regions tance is much larger than the ballistic resistance, and
of the devices such as the spacers, and iii) the is clearly limited by scattering by dopant impurities
“ballistic” resistance of the channel. Although the and surface roughness. It represents a significant
latter is intrinsic to the channel, and has nothing to part of the total resistance of a L = 30 nm long
do with the design of the source/drain, it is usually device (green line with diamonds). A careful anal-
mixed into the apparent contact resistance as it is ysis of the data (to be discussed at the conference)
independent on the gate length. shows that the lowly doped spacers are the most
In this work, we compute components ii) and iii) resistive elements.
of the contact resistance in Fully-Depleted Silicon- The contact resistances for the different doping
on-Insulator (FDSOI) Trigate devices in a Non- profiles of Fig. 2 are plotted in Figs. 5, and for
Equilibrium Green’s Functions (NEGF) framework different nanowire cross sections in Fig. 6. Rc
[1]. The channel is a rectangular nanowire with decreases when doping increases under the spacers,
width W and height H , and the gate stack is made at the expense of a loss of electrostatic control
of 0.8 nm SiO2 and 2.2 nm HfO2 . It is connected (larger DIBL). Also, Rc increases when the cross
to bulk source and drain contacts by 6 nm long sectional area decreases. Other results about the
spacers (see Fig. 1). Point-like dopants are added design of the spacers and the dependence of the
to the source and drain according to the different contact resistance on Vds will be discussed at the
target distributions plotted in Fig. 2, in order to conference.
capture impurity scattering in these regions. Surface This work was supported by the French National
roughness, Remote Coulomb Scattering (RCS) in Research Agency (ANR) project Noodles. The cal-
the channel, and electron-phonon interactions are culations were run on the TGCC/Curie machine
also included in the simulations and have been thanks to allocations from PRACE and GENCI.
calibrated on the experimental mobility. The current
R EFERENCES
is computed with a NEGF code in the effective mass
approximation. [1] Y.-M. Niquet, V.-H. Nguyen, F. Triozon, I. Duchemin,
O. Nier, and D. Rideau, Journal of Applied Physics 115,
At low bias, the resistance R(L) of the devices 054512 (2014).
is linear with L in the 20-100 nm range and can
ISBN 978-0-692-50554-0 69
Fig. 1. A longitudinal cross section of a W = 10×H = 10 nm
Trigate device, with overgrown source and drain contacts. The
dots in the contacts are single dopant impurities. The spacers
are 6 nm long.
Fig. 4. The contact resistance Rc extracted from Fig. 3

as a function of the carrier density n in the channel. Rc is
normalized to the effective width W + 2H of the device.
It is compared to the ballistic resistance of the channel, to
the contact resistance extracted without surface roughness nor
impurity scattering, and to the total resistance of a L = 30 nm
long device.
Fig. 2. Target doping profiles in the source of Fig. 1 (doping

profiles are symmetric in the drain). They are used to generate
random dopant distributions.
Fig. 5. The contact resistance Rc as a function of n for the

different doping profiles of Fig. 2.
Fig. 3. The resistance R(L) as a function of channel length L

for a particular realization of the “Reference” doping profile of
Fig. 2. The carrier density in the channel is n = 1013 cm−2 .
The contact resistance is the extrapolated Rc = R(L → 0),
while the slope gives an estimate of the carrier mobility in the Fig. 6. The contact resistance Rc as a function of n for
channel. different nanowire cross sections. The target doping profile is
the “Reference” profile of Fig. 2.
70
Sequential Tunneling Effect on Quantum-Dot
Intermediate-Band Solar Cells
K. Yoshida, Y. Okada, and N. Sano*
Research Center for Advanced Science and Technology, The University of Tokyo, Tokyo, Japan
*
Institute of Applied Physics, Univeristy of Tsukuba, Ibaraki, Japan
INTRODUCTION � � � �� ∗ ∗
− ∫� �� 𝑑𝑑𝑑𝑑 = ∫� ��(𝐺𝐺�� − 𝐺𝐺�� )𝑑𝑑𝑑𝑑 , (1)
� ��
Multi-stacked quantum dot (QD) superlattice � �
structure has been applied to realize intermediate where 𝑞𝑞 is the elementary charge, and 𝐺𝐺��∗ ∗
and 𝐺𝐺��
band solar cells (IBSCs) [1]. The device structure are the net-generation rates of IB-VB and CB-IB,
of IBSCs looks similar to the single-junction solar respectively, which are defined as a difference
cell but IBSCs can utilize lower energy photons between the optical generation rate and the
than the host material bandgap for optical carrier recombination rate. 𝑥𝑥�� and 𝑥𝑥�� are the lower and
generations via IB as presented in Fig. 1. This the upper boundaries of the integral for the i-th IB
additional carrier generation path can increase the dot. The device structure employed in the preset
output current density and, thus, IBSCs can exceed study is shown in Fig. 2.
the Shockley-Queisser limit of the single-junction
solar cell. The state-of-the-art epitaxial growth RESULTS
technology can make spacer layers between QDs The current continuity is satisfied by expressing
layers thin and as a result, the coupling between the total current density as a summation of current
multi-stacked QDs layers has been reported to components of CB electrons, holes and IB
modify the selection rule in optical absorption [2]. electrons as presented in Fig. 3. The strength of IB
In addition, it is expected that such coupling can dot coupling is expressed as TC. The IB current
introduce electron tunneling between the confined shows a strong position dependence induced by
states of QDs. The tunneling would be effectively profiles of the electrostatic potential profile and
make IB electron lifetime longer than the IB cell the quasi-Fermi level of IB as shown in Fig. 4. The
without the tunneling since the electrons can leave IB current density is smaller than the other
from QDs to the neighboring QDs before they components since IB dots are not directly
recombine with holes. In this study, we, therefore, connected to both contacts in the emitter and the
include the tunneling effect in the IBSC drift- base regions. To evaluate the tunneling effect on
diffusion method [3] and study its effect on QD- the cell performance, we define JIB as a
IBSC performance. difference between the current components of IB
JIB with and without the coupling where JIB is a
NUMERICAL METHOD
product of the elementary charge and the
In the present simulation, the Poisson equation integrated value of net generation rates through IB
and continuity equations for electrons in the in the cell. That difference indicates that the
conduction band (CB) and IB, and holes in the tunneling can improve the net carrier generation
valence band (VB) are solved self-consistently in rates via IB (see Fig. 5). This is because
the steady state. The confined state of QD is photogenerated IB electrons in the area close to the
modeled as IB dot structure as shown in Fig. 2 and top-emitter can sequentially tunnel to the
the tunneling is modeled by the sequential neighboring IB dots before they recombine with
tunneling which is known as the weak coupling holes. And finally these electrons can be excited
limit of quantum well superlattice [4]. The into CB by absorbing another incident photon. As
continuity equation of IB electrons is solved in the a result, the tunneling can improve QD-IBSC
integrated form using gauss’s theorem on each performance as shown in Fig. 6.
surface of IB dot and given by,
ISBN 978-0-692-50554-0 71
CONCLUSION
The sequential tunnelling effect is incorporated
in the drift-diffusion simulation for QD-IBSC as
IB tunnelling current. The tunnelling can improve
the cell performance making IB electron life time
long.
REFERENCES
[1] Y. Okada et al., Intermediate band solar cells: Recent
progress and future directions, Appl. Phys. Rev. 2,
021302 (2015).
[2] T. Kita et al., Polarization-insensitive optical gain
characteristics of highly stacked InAs/GaAs quantum dots,
J. Appl. Phys. 115, 233512 (2014).
[3] K. Yoshida et al., Device simulation of intermediate band Fig. 4. Calculated band diagram of QD-IBSC with the IB
solar cells: Effects of doping and concentration, J. Appl. coupling (TC = 0.5 meV) in short-circuit condition under 1 sun
Phys. 112, 084510 (2012). illumination. Ei and i are the band edge and the quasi-Fermi
[4] A. Wacker, Semiconductor superlattices: a model system level where i = C, V and I for CB, VB and IB, respectively.
for nonlinear transport, Phys. Rep. 357, 1 (2002).
0.12 Short−circuit condition
JIB (mA/cm )
0.08
2
Fig. 1. Schematics of optical absorption process in the single- 0.04

junction solar cell and IBSC. The arrow lengths represent
incident photon wavelengths.
0
0.001 0.01 0.1
TC (meV)
Fig. 5. Difference between IB current densities with and
without IB coupling TC in short-circuit condition. JIB = JIB(i)
- JIB(TC = 0.0 meV) where JIB is a product of the elementary
charge and the integrated value of net generation rate through
IB in the cell.
Fig. 2. Schematic of QD-IBSC structure employed in the
present study. NA and ND denote accepter and donor
concentrations. The cell temperature is 300 K and the incident 40
sun light is 6300 K blackbody radiation. TC = 0.0 meV TC = 0.5 meV
Current density J (mA/cm )
2
Jn + Jp + JI
40
Current density (mA/cm )
2
Jn + Jp Jn 1.5 TC = 0.5 meV

JIB (mA/cm )
2
Short−circuit condition
20 TC = 0.0 meV
1
20 TC = 0.5 meV
0.5
Jp
JI
0 0 0.5 1
0 1 2 3 4 V (V)
Position (m) 0 0.4 0.8
Fig. 3. Current profile of QD-IBSC with the IB coupling (TC
Voltage V (V)
= 0.5 meV) in short-circuit condition under 1 sun illumination.
Fig. 6. Current-voltage characteristics of QD-IBSCs with and
Jn, Jp and JI denote current densities of CB electrons, holes
without the sequential tunneling. Inset: JIB versus voltage.
and IB electrons, respectively.
72
Simulation of organic solar cell
with graphene transparent electrode
P. Paletti, R. Pawar, G. Ulisse*, F. Brunetti*, G. Iannaccone, G. Fiori
Dipartimento di Ingegneria dell’Informazione Università di Pisa, Via Caruso 16, 56122 Pisa, Italy
*
Dipartimento di Ingegneria Elettronica, Università di Roma Tor Vergata,
Via del Politecnico 1, 00133 Roma, Italy
COMSOL [3] package. The I−V characteristics

ABSTRACT
have been calculated with NanoTCAD ViDES [4],
We present a simulation study of the solving self-consistently the electrostatic, the
performance of organic solar cell (OSC) exploiting continuity and the exciton diffusion equations [5].
graphene as transparent electrode. The approach is
based on a multi-scale/multi-physics simulation RESULTS
framework, which is able to provide relevant In Fig. 1, the simulated OSC with graphene
information regarding the design guidelines of electrode is shown. In Fig. 2, we show the WF
efficient OSC. computed with DFT for different MoO3
thicknesses, and two different geometries of the
INTRODUCTION
oxide: oxigen- and Mo-terminated MoO3. On the
Organic solar cells (OSCs) hold promise as next- same picture, we show experimental results,
generation devices for energy conversion, due to showing good agreement with simulation results.
their low cost, the large area and their flexibility. We have then computed the I−V characteristics of
Exploiting graphene as transparent electrode in OSCs with P3HT, PTB7 and Perovskite active
OSCs could represent the killer application of layer. In Fig. 3, we show the Fill Factor (FF) and
graphene technology [1] because of i) its the energy efficiency for the Perovskite case. It is
transparency in the whole visible spectrum, ii) its apparent that OSCs are extremely sensitive to the
flexibility and iii) its large conductivity. series resistance per unit cell area. WF tuning
improves both the fill factor and the efficiency.
In this work, through a multi-scale approach, we CONCLUDING REMARKS
will provide design guidelines, assessing the
technological issues, which mostly affect device In Fig. 4 we show the efficiency of cells with
different HTL materials and the efficiency
performance, as for example, the quality of
graphene and the electrode work-function (WF). improvement provided by each cell engineering
Our multi-scale investigation exploits ab-initio option, as obtained by our multi scale simulation
approach. As can be seen, low contact resistance,
calculations to evaluate work-function tuning of
graphene contact by means of doping through the work function tuning, and grating can jointly
deposition of different MoO3 layers, provide a boost in power conversion efficiency.
We remark that graphene quality is a serious issue,
electromagnetic simulations for light management
when pattering the graphene electrode, and device since a very low defect density is required for an
modeling for computing the electrical acceptable sheet resistance.
characteristics of the cell and its main figures of ACKNOWLEDGMENT
merit. Authors would like to thank the support from
the EC. 7FP through the Project GONEXTs
MODEL (Contract 309201), Quantavis s.r.l. and support
The WF of Graphene on MoO3 has been from the Graphene Flagship (Contract 614391).
computed with ab-initio simulations exploiting
Quantum ESPRESSO [2], while light-management
of patterned electrode has been addressed with
ISBN 978-0-692-50554-0 73
REFERENCES ΦG=4.6eV
ΦG=4.9eV
0.80
[1] Y. Shi et al., ACS Nano. 4, 2689 (2010).
Fill Factor, VOC [V]

1.00 a) b)
[2] P. Giannozzi et al., J. Phys.: Condens. Matter, 21, 395502
Fill Factor
(2009). 0.80 0.60
[3] Info available at http://www.comsol.com 0.60
FF
[4] Available at http://vides.nanotcad.com RS=0 VOC 0.40
0.40
[5] B. Bay et al., IEEE Journal of Photovoltaics, 3, 310
4.6 4.8 5.0 5.2 5.4 0 5 10 15 20
(2013). Work Function [eV] 2
Series Resistance [Ω/cm ]
15.0 15.0
c) d)
Efficiency [%]
Efficiency [%]
12.0 12.0
9.0 9.0
6.0 RS=0 6.0
4.6 4.8 5.0 5.2 5.4 0 5 10 15 20

2
Work Function [eV] Series Resistance [Ω/cm ]
Fig. 3. Simulated results for a) FF, VOC and c) PCE
considering different WF for the graphene electrode for the
Perovskite-based OSC. Both FF [b)] and PCE [d)] degrade
with increasing series resistance.
15
Fig. 1. Sketch of the simulated OSC. When light illuminates Pristine 13.2%
the OSC, excitons are generated in the active layer, and then WF Tuning
Grating
separate at the interface between the active layer and the Hole
9.98% 6.08
Transporting layer (HTL). Holes are then collected at the 10
Efficiency [%]
1.13
transparent electrode, travelling through the hole-transporting
layer (HTL), while electrons are collected at the metal 3.43
contact. 7.14
5 4.66% 5.42
0.74
0.78
3.14
0
P3HT PTB7 Perovskite
Fig. 4. Power conversion efficiency of the investigated OSC

with graphene transparent electrode for different materials as
active layer.
Fig. 2. WF as a function of the MoO3 layer thickness for both

oxygen-terminated MoO3 top surfaces and Mo-terminated
surfaces. The blue solid line represents experiments from
[S.W. Tong et al., Adv. Func. Mat. Vol. 21, p. 4430, 2011].
74
Screening Effect on Electric Field Produced by
Spontaneous Polarization in ZnO Quantum Dot
in Electrolyte
X. Meshik, M.S. Choi, M. Dutta, and M.A. Stroscio
Dept. of Electrical and Computer Engineering, University of Illinois at Chicago, Chicago, IL 60607, USA
Email: [email protected]
quantum dots decay exponentially due to the

INTRODUCTION
screening:
Voltage-gated ion channels are instrumental in electrolyte spon
living cells, as they play key roles in processes EQD =EQD e−(r−R)/λD,(r>R) (3)
such as neuronal communication and muscle
movement. At any given time, a cell has a specific where r is the distance from the center of the
membrane potential due to the separation of quantum dot, and λD is the Debye length. λD in
intracellular and extracellular ions such as K+, Na+, electrolyte can be computed by
Ca2+ and Cl-. When the membrane potential
reaches a certain threshold, voltage-gated ion εkBT
channels open, resulting in the flow of ions across λD= (4)
2NAe2I
the membrane. Theoretical and experimental
studies have suggested that semiconductor
nanoparticles with built-in spontaneous where kB is Boltzmann constant, T is temperature,
polarizations that are in close proximity to these NA is Avogadro’s number, e is elementary charge,
ions channels can generate sufficient electric fields and I is ionic strength of the electrolyte, which is
to affect the channels’ dynamics [1-3]. This work 1 n 2
calculates the strength of the electric field given by I= 2  cizi , where ci is the molar
produced by ZnO quantum dots in physiological i=1
electrolyte. concentration of the ith ion in the electrolyte and zi
is the charge of the ith ion.
THEORY
Considering a wurtzite quantum dot such as DISCUSSION
ZnO with its spontaneous polarization grown We found that, in the electrolyte consisting of 4
along the c-axis [4]. For such a quantum dot, it is mM K+, 126 mM Na+, 3 mM Ca2+, 1 mM Mg2+
possible to approximate the ZnO quantum dot as a and 138 mM Cl- the Debye length is 16.02 nm.
uniformly polarized sphere as shown in Figure 1. The screening effect caused by the electrolyte is
Then the surface charge σ is given by shown in Figure 2. Considering that the minimum
potential difference needed to switch a voltage-
σ=P⋅̂n=Pcos(Θ) (1) gated ion channel is of the magnitude of 6 mV,
which requires the electric field across the 7 nm
By solving Gauss’s law, the electric field bilipid membrane of about 0.86×106 V/m, a ZnO
produced from a ZnO quantum dot is quantum dot should be within 8.27 nm of the cell.
This knowledge is potentially relevant for
spon P R3
EQD = 3ε [2cos(Θ)̂r+sin(Θ)̂Θ],(r>R) (2) therapeutic applications to treat medical conditions
r3 related to voltage-gated ion channels.
where ε is the dielectric function of the
surrounding media. However in water-based
electrolyte, the electric field produced from ZnO
ISBN 978-0-692-50554-0 75
ACKNOWLEDGMENT
This material is based upon work supported by,
or in part by, the National Science Foundation. a b
REFERENCES
[1] Stroscio, M. A., Dutta, M., Nariwani, K., Shi, P.,
Ramadurai, D. and Rufo, S. Integrating and Tagging
Biological Structures with Nanoscale Semiconducting
Quantum-Dot Structures, in Michael A. Stroscio and
Mitra Dutta, editors, Biological Nanostructures and
Applications of Nanostructures in Biology: Electrical,
Mechanical & Optical Properties, (Kluwer Academic
Publishers, New York, 2004). Fig. 3. Electric field due to spontaneous polarization of a 5-
[2] Pappas, T. C., Wickramanyake, W. M. S., Jan, E., nm ZnO quantum dot in water (a) and in electrolyte (b).
Motamedi, M., Brodwick, M. and Kotov, N. A. Nanoscale
Engineering of a Cellular Interface with Semiconductor
Nanoparticle Films for Photoelectric Stimulation of
Neurons. Nano Lett 7, 513-519 (2007).
[3] Lugo, K., Miao, X., Rieke, F. and Lin, L. Y. Remote
Switching of Cellular Activity and Cell Signaling Using
Light in Conjunction with Quantum Dots. Biomed. Opt.
Express 3, 447-454 (2012).
[4] Yamanaka, T., Dutta, M. and Stroscio, M. A. Spontaneous
Polarization Effects in Nanoscale Wurtzite Structures, J.
of Computational Electronics, 6, 313-316 (2007).
Fig. 1. Diagram of cell membrane with intracellular and

extracellular concentrations of major ions.
Fig. 2. Electric field produced from the spontaneous

polarization of a ZnO quantum dot (diameter=5 nm) versus
the distance from the quantum dot. The minimum distance
required to switch an ion channel (8.27 nm) is indicated.
76
Validating Simple Approaches for Quantum
Cascade Laser Modeling
M. Franckié∗ , D.O. Winge, and A. Wacker
Division of Mathematical Physics, Lund University, Bx 118, 22100, Lund, Sweden
∗ Previously M. Lindskog
The quantum cascade laser (QCL) is an results to RE and DM simulations. The THz sample
intersubband-based light source operating in a wide simulations are shown Fig. 1. For this structure with
range of wavelengths from the THz to mid-IR. In scattering injection, both the RE and NEGF models
order to improve their performance and broaden agree excellently with the experimental results. Sim-
their usage in general, much effort is put into similar results were also found for a similar structure
ulating these devices. Recently, genetic algorithms [1].
have been employed for optimization[1], [2], [3]. A detailed comparison between NEGF and two
As the number of parameters affecting the transport DM models was performed in Ref. [10], where the
is large, an efficient model is necessary. At the same main results are summarized in Fig. 2. Here it was
time, the model has to treat the complex many-body found that in order for the DM model to produce
problem sufficiently detailed to avoid un-physical reliable results, so-called second order currents have
results. to be considered, which are naturally included in the
Three classes of simulation models are mainly NEGF model; for these tunneling injection designs,
used for the simulation of QCLs. A rate equation a higher sophistication is needed. Although the gain
(RE) model [4] uses scattering rates between sub- is quite different in the NEGF and DM models, they
bands calculated via Fermi’s golden rule, in order to do predict very similar output powers and currents
give quick estimates of the level populations. Here, under irradiation so that they are indiscernible with
coherences between near-degenerate states have to the experimental data available.
be included as a tunneling rate. The transport prob- Our findings show that simple approaches can be
lem is then solved via the Schrödinger equation. used if the relevant features are properly accounted
Density matrix (DM) approaches [5], [6], [7] in for. In this context a fully self-consistent model
addition include coherences in the off-diagonal el- like NEGF is a good tool for validating different
ements of the density matrix. Both RE and DM approximations.
models assume thermalized sub-bands where the
R EFERENCES
momentum-dependence is averaged out. In contrast,
the non-equilibrium Green’s function (NEGF) the- [1] S. G. Razavipour et al. J. Appl. Phys., 113(20):203107,
2013.
ory [8] includes inter-subband scattering and treats [2] A. Bismuto et al. Appl. Phys. Lett., 101(2):021103, 2012.
the many-body problem consistently. The Green’s [3] A. Gajić et al. J. Appl. Phys., 115(5):053712, 2014.
functions of the system are calculated via a set [4] E. Dupont et al. J. Appl. Phys., 111(7):73111, 2012.
of self-consistent equations involving the scattering [5] S. Kumar and Q. Hu. Phys. Rev. B, 80(24):245316, 2009.
[6] E. Dupont, S. Fathololoumi, and H. C. Liu. Phys. Rev. B,
via self-energies. The theory is formulated as a 81(20):205311, May 2010.
perturbation expansion, where the expansion series [7] R. Terazzi and J. Faist. New J. Phys., 12(3):033045, 2010.
is truncated at a suitable point. Thus, the NEGF is [8] A. Wacker, M. Lindskog, and D.O. Winge IEEE
the most general of these models, and provides the J. Sel. Top. Quantum Electron., 19(5):1200611, 2013.
[9] A. Bismuto et al. Appl. Phys. Lett., 96(14):141105, 2010.
most information. [10] M. Lindskog et al. Appl. Phys. Lett., 105(10):103106,
We have performed NEGF simulations of a THz 2014.
[4] and a mid-IR QCL [9], and compared our
ISBN 978-0-692-50554-0 77
Fig. 1. Data taken from Ref. [4]. Left: Current density for the RE and NEGF models, as well as the experimental sample. Right:
laser spectrum (a), NEGF gain simulations (b), and RE gain simulations (c). The colored circles in (c) indicate the points which
correspond to the biases used in (a) and (b).
Fig. 2. Data taken from Ref. [10]. Left: Current density in the NEGF, two DM models, as well as experimental data, where
crosses show current densities under irradiation. The inset shows the simulated and experimental radiated output power. Right:
Gain and gain peak positions in the DM and NEGF models.
78
Thermionic escape in quantum well solar cell
Nicolas Cavassilas, Clémentine Gelly, Fabienne Michelini and Marc Bescond
Aix Marseille Université, CNRS, Université de Toulon, IM2NP UMR 7334, 13397, Marseille, France
InGaN alloys are considered as promising corresponding tQW=3.4 and 1.6 nm (devices A and
candidates for high-efficiency photovoltaic B). In both devices electrons and hole are strongly
devices, since their direct bandgap spans almost localized in the quantum well since band offset
the whole solar spectrum from 0.64 eV (InN) to between InGaN and GaN is high. A strong impact
3.46 eV (GaN). However, growing of high-quality of the electron-phonon scattering on carrier
InGaN layers with a large In content is difficult. extraction is thus expected. As shown in the insets
Nevertheless, the synthesis of nanowires can be of Fig. 3, this is confirmed by the secondary peaks
virtually substrate-free, which reduces the located at ±ħωop from the e1 and/or e2, where
formation of dislocations [1]. ħωop=92 meV is the polar optical phonon energy.
We propose a theoretical investigation of The major difference between both devices is the
InGaN/GaN core-shell nanowires for solar presence of the second state e2 in the quantum well
applications. The lattice mismatch between the of device A. In device B, due to stronger
InGaN and the GaN requires the use of multiple confinement, the e2 shift in the energy continuum.
quantum wells [2] and limits both the In content Fig. 4 shows the photo-current spectra
and the thickness of the InGaN layers. Fig. 1 corresponding to the LDOS presented in Fig. 3. It
shows the calculated e1-h1 transition between the results from phonon assisted thermionic escape as
topmost hole energy level (h1) and the deepest confirmed by the peaked shape of the spectra.
electron level (e1) in an InGaN quantum well Indeed, energy between peaks corresponds to ħωop.
embedded in GaN barriers. This e1-h1 transition is The corresponding total currents are 83 and 3.8
represented as a function of the In content and the (Arb. Unit) for devices A and B respectively. This
quantum well thickness tQW. We also show a unexpected ratio of about 20 has been established
prohibited pink area representing the limit due to for several photon energies. Even if photon
the lattice mismatch. The minimum e1-h1 transition absorption depends on the thickness of the
obtained without any dislocation is 3 eV. These quantum well, the main reason of this high ratio is
high-quality wells may be obtained if the GaN the thermionic escape. The rate of phonon
barriers are thick enough, i.e. ≥5 nm. In the absorption depends on the density of states of the
following, we then propose to compare quantum final electronic energy. In device B, due to the
wells exhibiting a e1-h1 transition equals to 3 eV strong confinement, density of states between e1
with various In contents. These calculations are and the continuum exhibits a minimum ~10-2 (Arb.
made with a photon energy of 3.2 eV, GaN Unit) as shown in the inset of Fig. 3. This
barrier’s thickness of 5 nm and 15 InGaN wells. minimum limits the phonon absorption of
No bias is applied. electrons and then their escape into the continuum.
The considered device and the corresponding In device A, since the confinement is weaker, the
calculated band-diagram are both represented in minimum of density of states between e1 and e2 is
Fig. 2. This result is obtained with a 1D non higher (~10-1 Arb. Unit). Moreover, the state e2,
equilibrium Green function formalism [3]. located between e1 and the continuum, also
Electron-photon interaction is assumed in the contributes to the phonon absorption and finally to
central well of the device only. This permits to the thermionic escape.
focus on the quantum efficiency of this particular [1] Y. Dong, B. Tian, T. J. Kempa, and C. M. Lieber, Nano
Lett. 9, 2183 (2009).
well connected to the other parts of the device. [2] J. R. Lang, N. G. Young, R. M. Farrell, Y. R. Wu, and J.
Fig. 3 shows the electronic Local Density of S. Speck, Appl. Phys. Lett. 101, 181105 (2012).
States (LDOS) obtained in the central quantum [3] N. Cavassilas, F. Michelini, and M. Bescond, J.
well for In contents of 14 and 19 % with Renewable Sustainable Energy 6, 011203 (2014).
ISBN 978-0-692-50554-0 79
Fig. 2. Schematic representation of the considered core-shell
nanowire and the corresponding 1D band-diagram obtained
Fig. 1. Iso-energy curves for the e1-h1 transition versus the In
with V=0 V, NA=7 1016 cm-3 and ND=3.5 1018 cm-3. The low
content and the quantum well thickness tQW. The pink area
electric field in the active region is due to the low p-doping.
represents the conditions for which dislocations are expected.
Fig. 3. The local density of states versus the energy and the
position in central quantum well of devices with a) In content
of 14 % and tQW=3.4 nm (device A), and b) In content of 19 % Fig. 4. Current spectra versus energy and position in the
and tQW=1.6 nm (device B). The insets represent the central quantum well of a) device A (In content of 14 %,
corresponding integrated density of states in the conduction tQW=3.4 nm), and b) device B (In content of 19 %, tQW=1.6
band versus the energy. nm).
80
Wigner Function Approach to Quantum
Transport in QCLs
O. Jonasson and I. Knezevic
1415 Engineering Dr., Madison, Wisconsin 53706-1691, USA, e-mail: [email protected]
Quantum cascade lasers (QCLs) are semiconduc- from Ref. [10], where lasing at a frequency of
tor heterostructures where lasing is achieved using 1.8 THz was obtained for temperatures up to 168 K.
transitions between localized quantum states. QCLs Figure 1 shows the calculated conduction band pro-
have received a considerable amount of attention file, which is obtained from the WBTE simulation
ever since their first experimental realization in coupled with Poisson’s equation. Also shown are the
1994 [1] and a wide range of modeling techniques, relevant subbands and their occupation at the design
at varying degrees of approximation, have been electric field of 15 kV/cm. From the figure we can
employed to simulate electron transport in these see that our results predict a relative occupation
structures [2], [3], [4], [5], [6]. A popular tech- n5 /n4 1.64, where n5 (n4 ) is the occupation
nique is to assume transport is semiclassical, due of the upper (lower) lasing level. The results are
to scattering between quasi-bound states, with the consistent with the device achieving lasing at the
rates calculated using Fermi’s golden rule [3], [4]. design electric field of 15 kV/cm.
Such an approximation is valid as long as transport Figure 2 shows the calculated current density vs
is mostly incoherent. However, in THz QCLs, co- electric field for lattice temperature in the range
herent transport and dephasing play important roles 50-300 K. At low temperatures, the relationship
and have to be included [7]. Here, we demonstrate between current density and electric field is highly
that the Wigner function formalism can accurately non-linear with a plateau around 7 kV/cm and an
describe the effects of coherence and dephasing NDR region around 13 kV/cm. However, at room
in THz QCLs, while avoiding the computational temperature, the fine features are nearly washed out
requirements of approaches using nonequilibrium due to enhanced electron-phonon scattering, and the
Green’s functions (NEGF) [5]. current vs field relationship is almost linear.
We solve the Wigner-Boltzmann transport equa- Figure 3 shows the steady-state Wigner function
tion (WBTE) using a particle-based numerical for different values of the applied electric field for a
method [8], [9] where periodic boundary conditions lattice temperature of 50 K. At zero field, electrons
are easily implemented. This approach is valid as mostly occupy the ground state in the widest well
long as the electron coherence length is shorter than and coupling between adjacent stages is low. For
the simulation domain. One of the advantages of intermediate field strength (7.5 kV/cm), electrons
this method is that the computational burden scales also occupy the narrower wells and transport can
linearly with the system length. This is because the be described as coupling between ground states in
Wigner function f (z, k) is defined in phase space, individual wells. At a higher field (15.0 kV/cm)
so doubling the system length only doubles the significant delocalization occurs and states that are
number of needed mesh points (and particle num- delocalized over multipe wells are occupied. The
ber) in the z direction. This is in contrast to density region around k = 0 in which the Wigner function
matrix approaches where doubling the system size has negative values are signs of high occupation
results in a 4-fold increase due to the density matrix of states that are delocalized between two or more
ρ(z1 , z2 ) dependence on two positions. wells, such as the upper and lower lasing level
To demonstrate the validity of the Wigner func- shown in Fig. 1.
tion approach, we simulated a THz frequency QCL We have demonstrated that the Wigner function
ISBN 978-0-692-50554-0 81
formalism can be used to efficiently model electron
transport in THz frequency QCLs. Effects of coher-
ence and dephasing due to scattering are important
in such devices and are included in our method. We
were able to calculate microscopic quantities such
as occupation of lasing states, which is crucial in
QCL device design.
This work was supported by the U.S. Department
of Energy, Office of Science, Materials Science Pro-
gram and Engineering Division, Physical Behavior
of Materials Program, award DE-SC0008712.
R EFERENCES
[1] J. Faist et al. Science, 264, 553 (1994).
[2] D. Indjin et al. Appl. Phys. Lett., 82, 9, 1347 (2003).
[3] H. Callebaut et al. Appl. Phys. Lett., 83, 2, 207 (2003).
[4] Y. B. Shi and I. Knezevic. J. Appl. Phys., 116, 12, 123105
(2014).
[5] S.-C. Lee and A. Wacker. Phys. Rev. B, 66, 245314
(2002). Fig. 2. Current density vs electric field for different lattice
[6] R. Iotti and F. Rossi. Phys. Rev. Lett., 87, 146603 (2001). temperatures for the THz QCL structure from Kumar et al.
[7] H. Callebaut and Q. Hu. J. Appl. Phys., 98, 10, 104505 [10].
(2005).
[8] D. Querlioz et al. J. Comput. Electron., 5, 443 (2006).
[9] M. Nedjalkov et al. Phys. Rev. B, 70, 115319 (2004).
[10] S. Kumar et al. Nature Phys., 7, 166 (2011).
Fig. 1. Conduction band (thick black) along with the relevant

subbands and their relative occupation (occupations are nor-
malized such that the highest occupation is one) for the THz
QCL structure from Kumar et al. [10]. The radiative transition Fig. 3. The steady-state Wigner function for zero electric field
is between subbands 5 and 4. Vertical dashed lines denote the (top), 7.5 kV/cm (middle), and 15.0 kV/cm (bottom) for the
extent of a single stage, which has a length of 54.5 nm. THz QCL structure from Kumar et al. [10]. Potential barriers
are shown in white. The simulation domain is three stages, with
a total length of 163.5 nm.
82
GaN/InGaN/GaN Disk-in-Wire Light
Emitters: Polar vs. Nonpolar Orientations
Md. R. Nishat, S. Alqahtani, Ye Wu, V. Chimalgi, and S. Ahmed
Department of Electrical and Computer Engineering, Southern Illinois University at Carbondale
1230 Lincoln Drive, Carbondale, IL 62901, USA. (E-mail: [email protected]; [email protected])
INTRODUCTION contain ~10M atoms. The undoped In0.25Ga0.75N

Nanoscale InGaN heterostructures are promising quantum disk is positioned at the center of the GaN
candidates for efficient, full-spectrum optical nanowire and has a thickness of ~3 nm. The
emitters and for use in conventional (e.g. lasers, calculated built-in potentials are shown in Fig. 1.
solid-state lighting) as well as novel (e.g. near-field While, in both systems, the pyroelectric contribution
photolithography, single phonon based quantum is negligible (not shown here), the piezoelectric
cryptography, diagnostic medicine and biological component is still significantly large and
imaging) applications. However, when grown along atomistically symmetric in nature. However, the
the [0001] polar axis, these structures exhibit large magnitude of the net potential for the m-plane
polarization fields, which significantly affect their structure is reduced by ~36.7 mV as compared to the
electronic and optical properties and lead to poor c-plane counterpart. Fig. 2 shows the wavefunctions
internal quantum efficiency (IQE). Recently, for for three cases: w/out internal fields, with strained
larger (mainly quantum well based) structures, it has lattice, and including net polarization fields. It is
been demonstrated that IQE could improve, if grown clear that, compared to c-plane, the m-plane
in the non-polar crystallographic orientations (such structure exhibits increased HOMO-LUMO overlap
as m-plane or a-plane) [1]. This work aims to as well as a larger bandgap (Fig. 3). This, in effect,
computationally investigate the validity of some of results in a higher average interband transition rate
these claims, and in particular, for recently reported (0.18 in the m-plane and 0.082 in the c-plane) and a
reduced dimensionality disk-in-wire structures [2]. greater degree of isotropy in the optical emission
characteristic (Fig. 4). Finally, when coupled to the
SIMULATION MODEL Synopsys’ TCAD toolkit, all of these facts result in
The overall simulation strategy is based on our in- higher spontaneous emission and improved (along
house QuADS 3-D computational platform. Here, with a delayed droop) IQE characteristic (Fig. 5). To
first, to compute the electronic structure and the conclude, a GUI enabled educational version of the
interband optical transition rates, we have employed simulator (dubbed nanoSSL and shown in Fig. 6),
a 10-band sp3s* tight-binding framework as available which currently computes the electronic and optical
in the NEMO 3-D software toolkit [3][4]. However, transition properties, has been developed and made
extensions have been made, where the freely available on NSF’s nanoHUB.org.
crystallographic orientation for the non-polar m-
plane structure is described via appropriate rotations ACKNOWLEDGMENT
in the lattice vectors. Strain and the polarization Supported by the National Science Foundation
fields have been calculated via a coupled atomistic Grant No. ECCS-1102192. Access to the XSEDE
valence-force field (VFF) and 3-D Poisson solver. and the nanoHUB computing platforms is also
Finally, the microscopically determined optical acknowledged.
transition parameters are incorporated into a TCAD REFERENCES
toolkit to obtain the terminal characteristics and
[1] P. Waltereit et al., Nature, vol. 406, pp. 865–868, 2000.
compare how atomicity and the net polarization
[2] S. Deshpande et al., Nature Comm., vol. 4, number. 1675,
field affect the IQE of these two systems. DOI: 10.1038/ncomms2691, 2013.
RESULTS AND DISCUSSION [3] G. Klimeck, et al., IEEE Trans. on Elect. Dev., vol. 54, 9,
pp. 2079–99, 2007.
The c-plane and the m-plane GaN/InGaN/GaN [4] S. Ahmed, et al., IEEE Trans. on Elect. Dev., vol. 57, 1,
disk-in-wire structures both have height and base pp. 164–173, 2010.
length of 100 nm and 14 nm, respectively, and
ISBN 978-0-692-50554-0 83
Fig. 1. Potential distributions in the growth direction.
Fig. 2. Wavefunctions due to the internal fields.
Fig. 3. Bandgap as a function of the internal fields.
Fig. 4. Optical transition rates.

droop kicks in at 80 A/cm2 and 110
A/cm2 for the c-plane and m-plane
devices, respectively.
Fig. 5. IQE as a function of injected current.
Fig. 6. nanoSSL simulator: https://nanohub.org/resources/nanossl.
84
Unified View of Electron and Phonon Transport
Jesse Maassen and Mark Lundstrom
Birck Nanotechnology Center
Purdue University, West Lafayette, IN 47907 USA
ABSTRACT TWO SIMPLE EQUATIONS

A simple, unified view of electron and phonon Two simple equations will be used. The first is
transport is presented. Similarities and differences the Landauer approach,
are identified, and new insights that come from
2q
addressing phonon transport from an electron T ( E )M ( E ) ( f1 − f 2 ) dE
h ∫
I= (1)
transport perspective will be discussed.
INTRODUCTION and the second is the McKelvey-Shockley version
Computational electronics has focused on of the Boltzmann equation,
simulating electron transport, and over the years as dI + (x) I + (x) I − (x) dI − (x)
device dimensions have shrunk, a suite of =− + = . (2)
dx λ λ dx
powerful tools has been developed. Devices
dissipate power and generate heat, but the The talk will discuss how these two approaches
treatment of heat transport has largely been the can be applied to both electrons and phonons and
work of other communities, but the evolution of how they can be connected to rigorous
technology with device structures such as simulations.
extremely Thin Silicon On Insulator (ETSOI) RESULTS
MOSFETs and FinFETs has increased the
importance of self-heating in nanodevices. The Figures 1-6 are some of the results that will be
average mean-free-path of phonons in bulk silicon discussed.
is over 100 nm, meaning that ballistic phonon CONCLUSION
transport may be even more important than As discussed in this talk, the approaches that
ballistic electron transport. Just as we have learned have been developed for electron transport can be
how to think about electrochemical potentials at readily extended to phonon transport, and they
the nanoscale, we must now learn how to think provide useful tools as well as new insights.
about temperature at the nanoscale. The problems
we face in modern electronic devices require us to ACKNOWLEDGMENT
treat electrons and phonons on an equal footing. This work was supported by the National
Those of us who come to this field from Science Foundation through the NCN-NEEDS
electronics naturally ask: “How can we adapt the program, contract 1227020-EEC and by the
tools, techniques, and concepts developed for Semiconductor Research Corporation.
electronics to phonons?” “What is different for
REFERENCES
phonon transport and what is similar to electron
[1] C. Jeong, R. Kim, M. Luisier, S. Datta, and M.
transport.” This talk will use two simple
Lundstrom, “On Landauer vs. Boltzmann and Full Band
approaches to address these questions. Another vs. Effective Mass Evaluation of Thermoelectric
important question is: “How do we connect Transport Coefficients,” J. Appl. Phys., 107, 023707,
rigorous simulations to the phenomenological (2010).
equations that device designers and [2] J. Maassen and M. Lundstrom, “Steady-State Heat
Transport: Ballistic-to-Diffusive with Fourier's Law,” J.
experimentalists need?” This talk will also address Applied Phys., 117, 035104, (2015).
this important question. [3] J. Maassen and M. Lundstrom, “A Simple Boltzmann
Transport Equation for Ballistic to Diffusive Transient
Heat Transport,” J. Appl. Phys., 117, 135102, (2015).
ISBN 978-0-692-50554-0 85
Fig. 1. The electron dispersion for bulk Si as computed by Fig. 4. The phonon window function, which indicates which
DFT simulation. ( )
channels are occupied and M ph !ω for Si. Note that all
phonon channels are occupied at 300 K.
Fig. 2. The number of channels vs. energy for bulk Si. The Fig. 5. Illustration showing that in bulk Si, phonons with a
initial increase from the band edges is linear for parabolic MFP greater than one micrometer contribute 50% to the
energy bands. thermal conductivity.
Fig. 3. The electron window function (which indicates which Fig. 6. Temperature vs. position in a Si films showing
( )
channels are occupied) and M el E for Si. Note that only temperature jumps at the contacts due to ballistic transport.
Symbols, BTE solutions, lines: Fourier’s Law when properly
electron channels very near the bottom of the band are
implemented (from [2]).
occupied.
86
Thermoelectric Transports in Geometry- and
Doping-Controlled Nanostructures
K. H. Park, P. N. Martin and U. Ravaioli
Beckman Institute and Department of Electrical and Computer Engineering, University of Illinois,
Urbana, Illinois 61801, USA
which is free from carrier concentration gets

ABSTRACT SUBMISSION
interfered by the structural variation.
This work investigates electron and phonon
transports in semiconductor nanowires and METHODS
nanoribbons with geometrical fluctuation and non- Change in the electronic transport due to the
uniform carrier doping concentration, expecting an geometry variation and non-uniform doping is
efficient phonon blockade with a minimal impact examined by computing an energy dependent
on electronic transport, so that enhancement of electrical quantum conductance of the structures
thermoelectric efficiency is expected. This under consideration employing a Recursive Green
comprehensive study examines trade-offs to find an Function approach [4]. The 3D conduction channel
optimal design for thermoelectric applications. is modeled by a 1D tight binding chain, and the
quantum nature is accounted by solving the
INTRODUCTION
Schrödinger equation in a series of 2D cross
Nanostructured semiconductors have attracted sections in the transverse direction. The quantum
considerable attention due to their potential transmissions of the sinusoidal nanowires are
applications in thermoelectric devices [1]. Utilizing compared with those of straight uniform nanowires
their large phonon mean-free-path (MFP) compared in Fig. 3. The nanowire undulations directly affect
to electron MFP, nano-structures have achieved the resistivity particularly at higher carrier energy
improved thermoelectric efficiency dictated by the but have little effect on low energy transport,
dimensionless thermoelectric figure of merit (ZT = leading this structure be potentially good for use of
S2σT/κ) [2]. In such nanowires, the phonon thermoelectric applications. Quantum
confinement starts developing ahead of the electron transmissions of the nanoribbons can be also
confinement and reduces thermal transport with investigated using the same method with the highly
hardly affecting electronic transport [3]. In this doped region brought into the calculation by
study, we investigate thermoelectric properties of solving the Schrödinger equation of the cross
silicon nanowires and nanoribbons with their sections. The Seebeck coefficient and electrical
geometry- and doping-modulation, expecting an conductivity of the structures will be computed
enhancement of the figure of merit ZT. through Landauer formula as a function of quantum
MODEL transmission.
We will consider sinusoidally undulated The lattice thermal conductivity of those non-
nanowires and non-uniformly doped nanoribbons uniform structures can be computed with molecular
with rough edges, as schematically depicted in Fig. dynamics at a constant temperature. We adopt the
1 and 2, respectively. It is expected to possibly reverse non-equilibrium MD method, which
realize an efficient phonon blockade through the imposes a heat flux first and measures the resulting
structural barriers, without degrading appreciably temperature gradient, performed with the Stillinger-
the electronic current. The high doping Weber interatomic potential. The MD simulation
concentration along the center region of the should reveal a decrease in thermal conductivity as
nanoribbon is anticipated to confine electrons structural barriers are imposed to the system. In Fig.
adequately, so that the rough edges have a minimal 4, it is shown that the lattice thermal conductivity of
impact on electron transport while phonon transport the sinusoidal nanowire is lower than that of the
straight nanowire.
ISBN 978-0-692-50554-0 87
CONCLUSION
We will show comprehensive results illustrating
possible trade-offs to optimize the thermoelectric
figure of merits of uniform and non-uniform nano-
structures. As we consider geometrical fluctuation
and non-uniform doping as the sole perturbation to
our system, the ratio of the figure of merits should
provide an insight of an optimum design for
thermoelectric applications.
REFERENCES
[1] A. I. Hochbaum, R. Chen, R. D. Delgado, W. Liang, E. C. Fig. 3. Comparison of quantum transmissions of a sinusoidal
Garnett, M. Najarian, A. Majumdar, and P. Yang, nanowire and a straight uniform nanowire (w = 4.34 nm, L =
Enhanced thermoelectric performance of rough silicon 108.6 nm)
nanowires, Nature 451, 7175 (2008).
[2] K. H. Park, M. Mohamed, Z. Aksamija, and U. Ravaioli,
Phonon scattering due to van der Waals forces in the
lattice thermal conductivity of Bi2Te3 thin films, J. Appl.
Phys. 117, 1 (2015).
[3] P. Martin, Z. Aksamija, E. Pop, and U. Ravaioli, Impact of
Phonon-Surface Roughness Scattering on Thermal
Conductivity of Thin Si Nanowires, Phys. Rev. Lett. 102,
12 (2009).
[4] P. N. Martin and U. Ravaioli, Green function treatment of
electronic transport in narrow rough semiconductor
conduction channels, J. Phys. Conf. Ser. 193, 012009
(2009).
Fig. 4. Comparison of lattice thermal conductivity of a

sinusoidal nanowire and a straight nanowire
Fig. 1. A schematic of a semiconductor nanowire with

sinusoidal undulations
Fig. 2. A schematic of a non-uniformly doped semiconductor

nanoribbon with edge-variation
88
Multi-Subband Interface Roughness Scattering
using 2D Finite Element Schödinger Equation for
Monte Carlo Simulations of Multi-Gate Transistors
D. Nagy∗ , M. A. Elmessary∗‡ , M. Aldegunde§, J. Lindberg† , A. Loureiro¶ , and K. Kalna∗
∗ ESDC, † C2EC, College of Engineering, Swansea University, Swansea SA2 8PP, Wales, United Kingdom
‡ Dept. of Mathematics & Engineering Physics, Faculty of Engineering, Mansoura University, Mansoura, Egypt
§ WCPM, School of Engineering, University of Warwick, Coventry CV4 7AL, England, United Kingdom
¶ CITIUS, Universidade de Santiago de Compostela, 15782 Santiago de Compostela, Galicia, Spain
Phone: +44 1792 602816, e-mail: [email protected]
Non-planar multi-gate transistors [1] are the pre- calculate form factors as:
ferred solutions for sub-16 nm digital CMOS technology ∗
fn,n (s) = ψn,k ψn ,k ΔV (1)
thanks to superior electrostatic integrity and delivering
a large ON-current. In this scaled process, interface where n, n� are the initial and final energy levels,
roughness scattering (IRS) is considered to be a crucial respectively, ΔV is the potential barrier between Si
factor for carrier transport [2], [3]. channel and high-κ dielectric. The IRS rate from the
initial state n to the
final state n is:
�
In this work, we have developed a new multi-subband 1

=
1
|Fn,n (q⊥ )|2 C(q)dq⊥ g1D,n (E) (2)
IRS model which uses wavefunctions and energy levels τn,n (E) 2h̄
from our 3D finite element (FE) Monte Carlo (MC)
R
where the Fn,n (q⊥ ) is a fast Fourier transform (FFT)
toolbox [4] with 2D FE Schrödinger equation quantum
of the form factors, C(q) is the exponential power
corrections [5]. The wavefunctions and energy levels
spectrum and g1D,n is the 1D density of states (DOS).
are used to calculate a form factor entering the multi-
The rate as function of initial energy is plotted in Fig. 2.
subband scattering rate. The new multi-subband IRS
Figs. 3 and 4 show the I D -VG characteristics normal-
model within 3D FE MC is then applied to 10.7 nm
ized by the area and Fig.5 illustrates average electron
gate length SOI FinFETs with rectangular-like and
velocity along the channel of the TRI FinFET. The drive
triangular-like (Fig. 1) cross-sections describing their
current, the drain induced barrier lowering (DIBL) and
3D geometry by the FE method and compared with a
sub-threshold slope (SS) are collected in Table I. The
3D Ando model.
average IRS rate monitored during the simulations as a
function of initial electron energy at VD = 0.05 V and
A new multi-subband IRS model is incorporated into
0.7 V (VG -VT = 0.7 V) is in Figs. 6 and 7. Fig. 8
a 3D Finite Element Monte Carlo simulator using 2D
shows electron density in the middle of the gate.
Schrödinger equation based quantum corrections. The
In summary, we have found that the multi-subband
new 3D code is then used to forecast the performance
EPN IRS is stronger than the 3D Ando IRS at a large
of 10.7 nm gate length SOI Si FinFETs with two
electron kinetic energy (from above 0.18 eV in the
cross-sections: rectangular-like (REC) and triangular-
simulations of 10.7 nm gate length Si SOI FinFETs).
like (TRI). We found that multi-subband IRS is much
The on-current is overestimated by the 3D Ando model
stronger at large electron kinetic energy resulting in
by 36%, 44% for low and high drain biases, respectively.
reduction in the drive current to 600 mA/μm2 for the
R EFERENCES
REC shaped channel and to 498 mA/μm 2 for the TRI
[1] X. Huang et al., IEDM Tech. Dig., pp. 67-70, 1999.
channel when compared to the results using the 3D [2] J. Seonghoon, M. V. Fischetti and T. Ting-Wei J. Appl. Phys.
Ando model. The multi-subband EPN model is based 102, 083715-1–14, 2007.
on a 2D formulation of Prange & Nee scattering model [3] J. Wang et al. Appl. Phys. Lett. 87 043101-1–3, 2005.
[4] M. Aldegunde et al., IEEE TED 60, 1561–1567, 2013.
for arbitrary paths [6]. Wavefunctions at each eigenstate [5] J. Lindberg et al., IEEE TED 61, 423-429, 2014.
are extracted along the interface for all the slices (21 [6] Z. Stanojević and H. Kosina in Proc. SISPAD 2013, pp. 352–
in our case) and unfolded onto an equidistant grid to 355.
ISBN 978-0-692-50554-0 89
p = 35.8 nm 16
1×10
3
250
1×10 TRI Ando
A = 57 nm2
Drain Current [mA/µm ]

Ando
2
TRI EPN
dielectric
EPN 2
Scattering Rate [s ]
1×10
-1
High-k
1×10
15 200
1
E 1×10
AT 14 150
17.67 nm
1×10
G 0
1×10
SOURCE
1×10
13 100
-1
1×10
1×10
12
-2 50
1×10
BO m VD = 0.05 V
X
.7n 11 -3
0
5.8
nm 10 1×10 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 1×10 0 0.2 0.4 0.6 0.8
Energy [eV] Gate Voltage [V]
Fig. 1: Schematic of the simulated triangular- Fig. 2: IRS rate from 3D (Ando) and multi- Fig. 3: ID -VG characteristics at VD =
like shape Si SOI FinFET with a Gaussian subband Prange & Nee (EPN) models. The 0.05 V using a normalised-to-area current
doping profile where σ is 3.45 nm in the EPN is for the 20 lowest electron subbands of for the 10.7 nm gate length TRI FinFET
transport direction and has an effective oxide the device at VD = 0.05 V and VG = 0.0 V comparing the 3D interface roughness model
thickness (EOT) of 0.62 nm following the up to 0.77 eV but calculated up to 3 eV. (Ando) with the multi-subband model (EPN).
ITRS 2014.
5
0
1×10
3
1000 ×1
2.0
Average Velocity [m/s]
TRI Ando
Drain Current [mA/µm ]
2
TRI EPN
2 5
800
1×10 10
1.5
× Cross-section TRI
Lg [nm] Ando EPN
Source
1
1×10 600
Drain
5
×1
0 SSLOW [mV/dec] 66
1.0
1×10
0
400 SSHIGH [mV/dec] 66
10
4
DIBLDD [mV/V] 34
× TRI Ando
1×10
-1
200 5.0 TRI EPN DIBLMC [mV/V] 64 65
VD = 0.7 V
0 ILOW [mA/µm2 ] 161 103
1×10 0
-2
0 0.-20 -10 0 10 20
MC
0.2 0.4 0.6 0.8 IMC
HIGH [mA/µm2 ] 885 498
Gate Voltage [V] Distance along transport [nm]
Fig. 4: ID -VG characteristics at VD = 0.7 V Fig. 5: Average velocity along the channel TABLE I
using a normalised-to-area current for the of the TRI FinFET at VG -VT = 0.7 V and S UB - THRESHOLD SLOPE (SS) AT V D =
TRI FinFET with a gate length of 10.7 nm VD = 0.7 V. 0.05 V AND 0.7 V FROM THE DD, DIBL
comparing the 3D interface roughness model FROM DD AND MC, AND DRIVE CURRENTS
(Ando) with the multi-subband model (EPN). (IVMC
D
) FOR THE TRI F IN FET.
log10[n(cm-3)]
19.7
15 19.6
19.5
11 11
0 0 19.4
×1 ×1
4.0 Ando VG = 0 V 2.0 Ando VG = 0 V 19.3
-1
-1
11 11 EPN VG = 0 V 10 19.2
0 EPN VG = 0 0
×1 ×1 19.1
3.0 1.5 Ando VG = 0.9 V
Z [nm]
Ando VG = 0.9 V
11
EPN VG = 0.9 V 11 EPN VG = 0.9 V
0 10
×1 ×
2.0 1.0 5
11 10
0 0
×1 ×1
1.0 5.0
0.00 0.00 0
200 400 600 800 1000 200 400 600 800 1000 -2 0 2
Energy [meV] Energy [meV] Y [nm]
Fig. 6: Average scattering rate for TRI device Fig. 7: Average scattering rate for TRI device Fig. 8: Electron density in the middle of the
at VD = 0.05 V with 3D (Ando) and multi- at VD = 0.7 V with 3D (Ando) and multigate in the TRI 100 FinFET at VG -VT =
subband (EPN) IRS models. subband (EPN) IRS models. 0.7 V, VD = 0.7 V using the new multi-
subband IRS model.
90
Anisotropic Schrödinger Equation Quantum
Corrections for 3D Monte Carlo Simulations of
Nanoscale Multigate Transistors
M. A. Elmessary∗† , D. Nagy∗ , M. Aldegunde‡, J. Lindberg ∗ , W. Dettmer∗ ,
D. Perić∗ , A. Loureiro∗ , and K. Kalna∗
∗ College
of Engineering, Swansea University, Swansea SA2 8PP, Wales, United Kingdom
† Dept of Mathematics & Engineering Physics, Faculty of Engineering, Mansoura University, Mansoura, Egypt
‡ WCPM, School of Engineering, University of Warwick, Coventry CV4 7AL, England, United Kingdom
Phone: +44 1792 602816, Fax: +44 1792 295676, Email: [email protected]
Multi-gate FETs like FinFETs and GAA nanowire ter (elliptical circumference/π )DNW =12.8 nm) and EOT
(NW) are leading solutions for sub-10 nm technology = 1.5 nm. We reverse engineer the doping profile at the
because of their superior electrostatic integrity [1]. sub-threshold region at VD = 0.05 V and VD =1.0 V
However quantum confinement and the complex 3D as shown in Figs. 1 and 2. We then simulated ID -
device geometry need to be realistically modelled for VG characteristics of the 22 nm gate length GAA Si
accurate physical simulations of carrier transport. NW at both low and high drain biases using 3D SEQC
Here, we report on anisotropic finite element (FE) MC toolbox without any free parameter with excellent
Schrödinger equation based quantum corrections (SE- agreement as can be seen in Fig. 3. We then scaled
QCs) for our in-house 3D FE Monte Carlo (MC) the Si GAA NW according to ITRS specifications to
device toolbox [2], [3]. The MC has already anisotropic a gate length of 10 nm and EOT of 0.8 nm. Fig. 4
bandstructure using k-vector transformations [4] but the shows the ID -VG characteristics for the scaled 10 nm
Schrödinger equation QCs [5] were approximated by GAA NW at VD =0.05 V and VD =0.7 V predicted by
isotropic electron effective mass tensor (EMT) [2]. We the 3D FE SEQC MC. Table II collects all devices
extend the calibration-free Schrödinger equation QCs characteristics for Si GAA NW with gate lengths of 22
into three ∆ valleys using longitudinal and transverse and 10 nm predicting that scaling to 10 nm gate will
electron effective masses. Table I collects the EMT of ensure superior electrostatic integrity and on-current
Si ∆ valley for �100� and �110� channel orientations. performance of NW FETs.
The 3D device FE mesh contains predefined 2D planes To show how anisotropic QC can affect the ID -
perpendicular to the transport direction, that are used VG characteristics, we simulated two 8.1 nm gate
to obtain the 2D potential for 2D Schrödinger
equation length FinFETs, rectangular-like (REC) and triangular-
[6]: − h̄2 ∇⊥ · (m∗ )−1 · ∇⊥ ψ(y, z) +U (y, z)ψ(y, z) = like (TRI), with the �100� and �110� channel orien-
2
Eψ(y, z), where E is the energy, (m∗ )−1 is the inverse tations, a channel perimeter of 26.5 nm, and area of
EMT with components define as (m∗ij )−1 = ωij , i, j = 49.5 nm2 (REC)/29.6 nm2 (TRI), and EOT=0.55 nm.
y, z ; ψ(y, z) is the wavefunction penetrating into sur- We have found (Figs. 6–9) that QC anisotropy effects
rounding oxide, and U (y, z) = −[qV (y, z) + χ(y, z)] play the strongest role in the �100� channel TRI device
is the potential energy with χ(y, z) being the electron increasing the drain current by about 15% and slightly
affinity [2]. The Schrödinger Eq. is solved separately for decreasing the current by 2% in the �100� channel REC
each of the three ∆ valleys. The resulting wavefunctions device. The QC anisotropy has negligible effect in any
(Fig. 5) are then used to calculate 2D quantum density device in the �110� orientation (see Table III).
for each of the valleys. The 2D density is interpolated R EFERENCES
onto the 3D simulation domain to obtain a separate [1] S. Bangsaruntip et al., IEDM Tech. Dig., pp. 526-529 (2013).
[2] J. Lindberg et al., IEEE TED, pp. 423-429 (2014).
quantum correction potential for each valley. [3] M. Aldegunde et al., IEEE TED, pp. 1561-1567 (2013).
We tested our new MC toolbox against experimental [4] D. Esseni et al., “Nanoscale MOS Transistors: Semi- Classical
data for a 22 nm gate length GAA Si NW [1] with �110� Transport and Applications,” Cambridge Univ. Press, 2011.
[5] B. Winstead et al., IEEE TED, 50, pp. 440-446, 2003.
channel orientation. The NW has elliptical cross-section [6] D. Nagy et al., IEEE Trans. Nanotechnol, pp. 93-100 (2015).
with diameters of 11.3/14.22 nm, with effective diame-
ISBN 978-0-692-50554-0 91
-3
1×10
-3 2.0×10
-3 IBM VD=0.05 V
1×10 IBM VD=1 V
Drain Current [A/µm]

-4
1×10
-4 MC VD=0.05 V, X/2, extended s/D, doping=5e19
1×10 -3
MC VD=1 V, X/2, extended s/D, doping=5e19
1.5×10
ID [A/µm]
19
Doping Concentration [cm ]
5×10 -5 -5
1×10 1×10
-3
IBM VD=0.05 V
19 Source volume
-6 IBM VD=1 V
4×10 -3
1×10
Drain volume
-6
1×10 MC VD=0.05 V 1.0×10
Gate
19 -7 MC VD=1 V
3×10 1×10 -7
-8
1×10
-4
2×10
19
X X 1×10 DD VD=0.05 V, X/2, extended s/D, doping=5e19 5.0×10
22 nm DD VD=1 V, X/2, extended s/D, doping=5e19 -8
-9 1×10
1×10
19 1×10 DD VD=0.05 V, X, small s/D, doping=5e19
DD VD=1 V, X, small s/D, doping=5e19
-9
-10
1×10 -0.1 1×10 0 0.0
0 0 0.1 0.2 0.3 0.4 0.2 0.4 0.6 0.8 1
-30 -20 -10 0
Distance along x-axis [nm]
10 20 30
Gate Voltage [V] Gate Voltage [V]
Fig. 1: Cross-section of Gausian-like Fig. 2: Devising doping profile for the 22 nm Fig. 3: ID -VG characteristics for the 22 nm
doping profile along the transport x- GAA nanowire at VD =0.05 V and VD =1.0 V GAA nanowire from the 3D FE MC with
direction for the GAA nanowire. with DD simulations by changing the S/D anisotropic SEQCs with no free parameter
region and the doping spread X, and a final (full lines) compared against experimental data
MC simulation. (dashed lines) [1].
500
2
1×10
Drain Current [µA/µm]

1×10
-2
2.0×10
-3 TRI, VD=0.05 V 400
MC VD=0.05 V
1
1×10
-3 MC VD=0.7 V 1×10
Drain Current [A/µm]
-4 1.5×10
-3 300
1×10 0
1×10
-5
1×10
1.0×10
-3
-1 200
1×10
-6 1×10
-7 isotropic <100>
1×10 -4 -2 100
5.0×10 1×10 isotropic <110>
-8 anisotropic <100>
1×10 anisotropic <110>
-3
1×10 0 0
0.2 0.4 0.6 0.8
-9
1×10 0 0.0
0.2 0.4 0.6 0.8 1
Gate Voltage [V] Gate Voltage [V]
Fig. 4: ID -VG characteristics for the Fig. 5: The first wavefunctions for three ∆ Fig. 6: ID -VG characteristics at VD =
scaled 10 nm gate length GAA nanowire valleys, in the middle of the 100 channel 0.05 V for the TRI shape FinFET with
predicted by the quantum corrected 3D for the REC (top) and TRI (bottom) cross- the 100 and 110 channel orientations
FE MC. sections at VG =0.8 V and VD = 0.6 V. showing the effect of anisotropic SEQCs.
3
1×10 3000
3 REC, VD=0.05 V 600
1×10 3
2000 1×10

2
2
1×10 500 2500
1×10 2
TRI, VD=0.6 V 1 1×10 REC, VD=0.6 V
1×10
1
1500 1×10 400 2000
1
0
1×10
0
1×10 1×10 300 1500
0
-1 1000 1×10
1×10 -1
1×10 200 -1 1000
-2 1×10
1×10 isotropic <100> 500 isotropic <100> isotropic <100>
-2
-3 isotropic <110> 1×10 isotropic <110> 100 1×10
-2 isotropic <110> 500
1×10 anisotropic <100> anisotropic <100> anisotropic <100>
anisotropic <110> anisotropic <110> anisotropic <110>
-3
1×10 0
-4
0 1×10 0 0 1×10 0
-3
0
0.2 0.4 0.6 0.8 1 0.2 0.4 0.6 0.8 1 0.2 0.4 0.6 0.8 1
Gate Voltage [V] Gate Voltage [V] Gate Voltage [V]
Fig. 7: ID -VG characteristics at VD = Fig. 8: ID -VG characteristics at VD = Fig. 9: ID -VG characteristics at VD =
0.6 V for the TRI shape FinFET with the 0.05 V for the REC shape FinFET with 0.6 V for the REC shape FinFET with
100 and 110 channel comparing the the 100 and 110 channel illustrating the 100 and 110 channel comparing
effect of anisotropic SEQCs. isotropic/anisotropic SEQCs. isotropic/anisotropic SEQCs.
Orien- Rectangular Triangular
FinFET
tation Valley ωyy ωzz m∗Tr iso aniso iso aniso
100 ∆1 1 1
ml VT [V] 0.25 0.25 0.25 0.25
mt
1
mt
1 Gate length [nm] 22 10 SSLOW [mV/dec] 72 67
100 ∆2 ml mt
mt
VT [V] 0.3 0.3 SSHIGH [mV/dec] 74 68
SSLOW [mV/dec] 74 66
1 1
100 ∆3 mt DIBLMC [mV/V] 82 76 70 48
�100�
mt ml
110 ∆1 mt +ml 1 mt +ml SSHIGH [mV/dec] 77 67 DIBLMC [mV/V] 84
�110�
77 77 51
DIBL [mV/V] 48 35
2mt ml mt 2
IMC [µA/µm] 2128 2075 1537 1667
mt +ml 1 mt +ml
110 ∆2
DIBLMC [mV/V] 64 39
�100�
2mt ml mt 2
IMC [µA/µm] 1150 1196 IMC [µA/µm] 1766 1728 1258 1251
1 1
110 ∆3 mt ml
mt �110�
TABLE I TABLE II TABLE III

T HE EMT OF ∆ VALLEY FOR 100 AND VT AND SUB - THRESHOLD SLOPE (SS) VT AND SS AT VD = 0.05 V (LOW)
110 CHANNEL ORIENTATIONS WHERE ωij AT VD = 0.05 V (LOW) AND 1.0/0.7 V AND 0.6 V (HIGH) FROM THE DD,
=1/m∗ij , ωyz =0 AND DEGENERACY =2. (HIGH) FROM THE DD, DIBL FROM DIBL FROM THE MC, AND DRIVE
WAFER ORIENTATION IS [100]. THE DD AND FROM THE MC, AND CURRENTS (I MC ) FOR REC AND TRI
DRIVE CURRENTS (I MC ) AT V G =1.0 V F IN FET S COMPARING ISOTROPIC ( ISO )
COMPARING 22 AND 10 NM GAA FET S . AND ANISOTROPIC ( ANISO ) SEQC S .
92
Improved Particle Annihilation for Wigner Monte
Carlo Simulations on a High-Resolution Mesh
P. Ellinghaus, M. Nedjalkov, and S. Selberherr
E-mail: {ellinghaus|nedjalkov|selberherr}@iue.tuwien.ac.at
The effects of surface roughness on quantum transport addresses these large memory requirements through a
in modern transistor architectures, e.g. UTB-SOI, have spatial domain decomposition [3]. However, the com-
become significant [1]. However, to properly model the putational demands of the SPWMC simulator allow it
surface roughness of a Si/SiO2 interface requires a fine to be run on a typical desktop computer, therefore, its
spatial resolution. We investigate the implications of memory demands should also follow suit. Furthermore,
the latter on the Signed-Particle Wigner Monte Carlo a finer resolution also reduces the effectiveness of the
(SPWMC) method, which presents a convenient for- annihilation step: there are more cells and each cell
malism to investigate transient and stationary processes is filled by less particles. Thereby, the likelihood of
in multi-dimensional semiconductor structures, ranging particles with opposite signs to meet and annihilate each
from quantum-coherent to scattering-dominated trans- other is reduced.
port. Fig. 1 shows how the effective channel width, and The highlighted problems can be remedied by re-
thereby the current (Fig. 2), is increased when using an ducing the spatial resolution of the grid on which the
insufficient spatial resolution. The current is calculated particles are recorded for annihilation; the k -values are
by the Ramo-Shockley theorem and a steady-state is discretized and remain fixed. The concept is illustrated
achieved by the periodic injection of individual elec- in Fig. 3. This reduces the memory requirement and
trons, represented by minimum uncertainty wave pack- increases the effectiveness of the annihilation procedure.
ets. Only the surface roughness is considered; phonons To counteract the loss in precision, a statistical fitting of
are switched off. the spatial distribution of the particles in the enlarged
SPWMC simulations rely on the generation of numer- spatial cells is performed before annihilation ensues; the
ical particles with + and − signs [2], generated at a rate obtained distribution is then used to regenerate the par-
in the order of 1015 s−1 . Particle annihilation is used ticles that remain after the annihilation. Fig. 4 compares
to counteract the exponential increase in the number of the regeneration of particles, annihilated on a coarsened
particles: the phase space is divided into many cells; grid, using a uniform distribution and a Gaussian distri-
within each cell particles with opposite signs annihilate bution. The former follows the true solution much better
each other since their contributions to the calculation of than the Gaussian distribution. This is consistent with
any physical quantity cancel out. The particles remaining the observations made in [4]. Both approaches reduce
after the annihilation are regenerated to represent the pre- the memory consumption by a factor of 16, however,
annihilation ensemble. the Gaussian distribution requires some extra memory
The memory required to represent the phase space and computation to calculate the additional moment. The
grid, on which particles are recorded for annihilation, same holds true for more sophisticated fitting techniques,
quickly becomes exorbitant in multi-dimensional simu- using more moments of the pre-annihilation distribution,
lations with a fine spatial resolution, since the number of which eradicates the achieved memory reduction.
cells increases with the power of the dimensionality of It is concluded that the debilitating increase in the
the phase space (the resolution also affects the number memory requirements to perform the annihilation step in
of k -values which must be retained to ensure a unitary SPWMC simulations with a fine spatial resolution can be
Fourier transform). A 30 nm×30 nm domain and a three- effectively remedied by reducing the spatial resolution of
dimensional k -space with a coherence length of 30 nm the phase space grid used for recording and performing
results in an array size exceeding 141 GB at a resolu- a statistical fit by calculating the mean position of the
tion of 0.2 nm. A distributed-memory (MPI) approach particles before the annihilation.
ISBN 978-0-692-50554-0 93
R EFERENCES
[1] J.M. Sellier, M. Nedjalkov, I. Dimov, S. Selberherr, Decoherence
and Time Reversibility: The Role of Randomness at Interfaces,
Journal of Applied Physics, 114, 174902 (2013).
[2] M. Nedjalkov, P. Schwaha, S. Selberherr, J.M. Sellier and
D. Vasileska, Wigner Quasi-Particle Attributes - An Asymptotic
Perspective, Applied Physics Letters, 102, 163113 (2013).
[3] P. Ellinghaus, J. Weinbub, M. Nedjalkov, S. Selberherr and
I. Dimov, Distributed-Memory Parallelization of the Wigner
Monte Carlo Method Using Spatial Domain Decomposition,
Journal of Computational Electronics, 14, 151 (2015).
[4] P. Ellinghaus, M. Nedjalkov and S. Selberherr, Optimized Particle
Regeneration Scheme for the Wigner Monte Carlo Method,
Lecture Notes in Computer Science, 8962, 27 (2015).
[5] S.M. Goodnick, D.K. Ferry, C.W. Wilmsen, Z. Liliental, D. Fathy
and O.L. Krivanek, Surface Roughness at the Si(100)-SiO2
Interface, Phys. Rev. B, 32, 8171 (1985).
Figure 3. A distribution of + and − particles in four spatial cells
7 7 (dashed lines). The gray cell, encapsulating all four cells, represents a
6 6 cell of the coarsened phase-space grid used to perform annihilation.
5 5 It is enlarged by a factor 2 in both directions, reducing the array
size by a factor of four. The effectiveness of annihilation is improved
y [nm]
y [nm]
4 4
3 3 from 3 to only 1 surviving particle.
2 2
1 1
0.0003
0 0 True solution
Uniform
0 2 4 6 8 10 0 2 4 6 8 10 Gaussian
x [nm] x [nm]
Figure 1. The geometry represents a 3 nm wide silicon channel 0.0002

Probability density
between two oxide layers (in black). The roughness of the Si/SiO2
interface is characterized by an exponential auto-correlation function
with a mean displacement, obtained from experiments in [5], used to
perturb the smooth interfaces of the channel. The interface roughness, 0.0001
modeled with identical statistical parameters, results in different

effective channel widths when using a 0.2 nm (left) or a 1 nm
resolution (right).
0
0 10 20 30 40 50 60 70 80 90
x-Position [nm]
0.0002
1e-05 True solution
Uniform
1.0 nm mesh Gaussian
0.2 nm mesh
Current [a.u.]
Probability density
5e-06 0.0001
0
0
0 10 20 30 40 50 0 10 20 30 40 50 60 70 80 90
Time [fs] x-Position [nm]
Figure 2. Comparison of the steady-state currents calculated for the Figure 4. A slice of the two-dimensional probability density of a
geometries in Fig. 1 with a spatial resolution of 0.2 nm and 1.0 nm, wave packet, evolving freely in a domain with a spatial resolution
show an increase in the latter due to a greater effective channel width. of 0.25 nm, after 40 (top) and 80 (bottom) forced annihilation
The current increases as the first wave packet is injected into the steps. The annihilation is performed on a coarsened grid with a
channel from the left and reaches a steady state once the channel 1 nm resolution and the particles are regenerated using uniform and
is saturated with particles. The current for the 0.2 nm case appears Gaussian distributions. The ’true solution’ – the evolution when the
noisier over the time shown, because it takes longer to saturate the annihilation step is omitted – is followed the best when particles are
channel with particles on the finer mesh. regenerated by a uniform distribution.
94
Modeling Quantum Acceleration of Bloch
Waves in Nanowires
R. Hathwar, M. Saraniti, and S. M. Goodnick
Arizona State University, Tempe, Arizona, USA.
INTRODUCTION
The X-matrices are defined as
The drift model used in a ‘free-flight drift’
routine of a Monte Carlo simulation of
semiconductor devices is a single band model that −i ∫ un*,k ( t ) ( x).∇ k un ',k ( t ) ( x).d 3 x (3)
X n ,n ' (k (t )) =
Ω
does not work well at band crossings or when the
energy separation between the bands are small. where Ω is the volume of the primitive cell. The
Applying strong electric fields to semiconductors derivative in eqn. 3 is calculated using Rayleigh-
can induce inter band transitions [1]. This effect Schrodinger perturbation theory as shown in eqn.
becomes prominent in materials such as nanowires 4. For degenerate bands the degenerate
where the energy spacing between the bands is perturbation theory is used.
small and the bands intercross more frequently un ',k ∂H / ∂k un ,k
across the Brillouin Zone (BZ). Various ∇ k un ( k ) =∑
n ',n'≠ n e n ' (k ) − e n (k )
un ' ( k ) (4)
approaches have been proposed to incorporate
carrier transitions between different bands such as Equation 1 is solved using the 4th order Runge-
the overlap test [2] and the velocity continuity Kutta method for coupled partial differential
method [3]. Both these methods have a drawback equations [6]. Special care is taken to ensure the
that the transition does not depend on the electric phases of the X-matrices are continuous across
field and is sensitive to the mesh size [4]. Krieger different k-points as noted in [7]. The probability
and Iafrate (KI) have developed a set of equations distributions obtained from eqn. 1 are calculated
which when solved give the transition probability for every k point in the BZ for a given free-flight
of a carrier as a function of time under an electric time step for all possible initial conditions before
field [5]. The KI method has been applied here to the simulation starts. The band structure and
the case of nanowires and the method and results scattering rates of the nanowires are calculated as
discussed in the following sections. shown in [8]. A uniform field Monte Carlo
simulation is then run to calculate the energy field
characteristics. At the beginning of every free-
MODEL flight step the carrier’s momentum is updated
The KI method solves the time dependent according to eqn. 2. Based on the carrier’s initial
Schrodinger equation for Bloch waves under an momentum value and band index the
electric field to give the following set of coupled corresponding probability distributions are
partial differential equations. obtained from the pre-calculated tables and a
∂Cn (t ) random number is used to decide the final band
i = e n (k (t )).Cn (t ) + eE (t ) index. The carrier is then placed in that band.
∂t (1)
× ∑ X n ,n ' (k (t )).Cn ' (t )
n'
RESULTS
where E(t) is the electric field, (-e) is the electronic The nanowires considered in this study are
charge and the wave vector k(t) is determined by 3nmx3nm rectangular Si and InAs nanowires
the semi-classical relation, along the [100] direction. The band structure of a
∂k (t ) 3nmx3nm InAs nanowire is shown in Fig.1. At
 = −eE (t ) (2) high energies an electron has the chance of getting
∂t
ISBN 978-0-692-50554-0 95
trapped within a band unless it can tunnel to higher
bands using the electric field. In Fig.2 we see the
difference the quantum acceleration model
produces on the average energy of the electrons in
a uniform field simulation in a 3nmx3nm InAs
nanowire. Without it the electron energy begins to
saturate as the carriers are unable to gain sufficient
energy from the electric field. In Fig. 3 we see a
specific example of a band crossing in the Si
nanowire highlighted by the circle. In Fig. 4 the
probability of finding an electron in each band is
shown after the electron is drifted across the band
crossing under an electric field of -100kV/cm. The
electron is initially placed in the 9th conduction
Fig. 2. Average energy of the electrons in a uniform field
band. simulation with the traditional (semi-classical) drift model
REFERENCES (blue) and with the quantum acceleration model (green) as a
function of electric field. This is for a 3nmx3nm InAs
[1] H-E. Nilsson, A. Martinez, E. Ghillino, U. Sannemo, E. nanowire.
Bellotti and M. Goano, Numerical modelling of hole
interband tunnelling in wirtzite Ga Nans SiC, Journal of
Applied Physics, 90,6 (2001).
[2] E. Bellotti, H.-E. Nilsson, K. F. Brennan, P. P. Ruden, and
R. J. Trew, J. Appl. Phys. 87, 3864 (2000).
[3] K. F. Brennan, E. Bellotti, M. Farahmand, H.-E. Nilsson,
P. P. Ruden, and Y. Zhang, IEEE Trans. Electron Devices
ED-47, 1882 (2000).
[4] Francessco Bertazzi, Michele Moresco and Enrico
Bellotti, Theory of high field carrier transport and impact
ionizationin wurtzite GaN. Part 1: A full band Monte
Carlo model. Journal of Applied Physics,106, 063718
(2009).
[5] Krieger J B and Iafrate G J 1986 Phys. Rev. B 33 5494–
500.
[6] W. H. Press, B. P. Flannery, S. A. Teukolsky, and W. T.
Vetterling, Numerical Recipes. The Art of Scientific
Computing _Cambridge University Press, Cambridge,
(1986). Fig. 3. Band structure crossing for a 3nmx3nm Si nanowire
[7] U. Lindefeldt, H-E. Nilsson and M Hjelm, Semicond. Sci. along the 100 direction. The black dots represent the discrete
Technol. 19 (2004). eigenvalues obtained from the eigensolver.
[8] R.Hathwar,M.Saraniti,S.M.Goodnick,IEEE Nano, Pg. 645,
2014.
Fig. 4. Probability distribution of the electron as it drifts

Fig. 1. Band structure of a 3nmx3nm InAs nanowire along
across the band crossing under an electric field of -100kV/cm.
the 100 direction.
96
On the Validity of the Harmonic Potential Energy Surface Approximation
for Nonradiative Multiphonon Charge Transitions in Oxide Defects
Y. Wimmer∗ W. Goes∗, A.-M. El-Sayed∗,†, A.L. Shluger† , and T. Grasser∗
∗
Vienna University of Technology, † University College London, UK , Email: [email protected]
Abstract: Hole trapping in oxide defects in the gate insulator to two points for each parabola, with q0 being the position of
of MOSFET transistors has been linked to a wide range of the minimum [2, 19] (see Fig. 2). However, as demonstrated in
phenomena like random telegraph noise, 1/f noise, bias temper- Fig. 2, EPar tends to underestimate the crossing point. Since τc
ature instability (BTI), stress-induced leakage current and hot and τe are extremely sensitive to the height of this barrier [8], a
carrier degradation [1–5]. Charge capture (τc ) and emission (τe ) good match in this region is essential for a good approximation.
times of a defect are theoretically described by nonradiative- Very promising results wereobtained using a V-approximation:
multiphonon (NMP)-theory, where the potential energy surface EVap (q) = γ( 1 + β(q − q0 )2 − 1) (2)
(PES) along the reaction coordinate is usually approximated being parabolic near the minimum at q0 but linear for larger
by a parabola [2, 6, 7]. In the classical high-temperature limit, values of |q − q0 |. It should be pointed out that although we
transitions between the different charge states occur at the refer to this function as V-approximation this does not indicate
intersection of those parabolas, which also defines the classical that EVap has a jump discontinuity at its minimum, but rather
reaction barrier. This harmonic approximation has been in use that its shape resembles more a V than a parabola. The constant
ever since the introduction of NMP-theory [8–10]. However,the γ defines the opening angle, for simplicity reasons its value was
quality of this approximation is rarely investigated [7]. In this set to 1.0 eV in this work.
work, we compare different approximations of the PES in order Calculating τc and τe for the OV (see Fig. 3 left), EPar actually
to calculate τc and τe and compare them to the results obtained by is a very good approximation, in this case even slightly better
Density Functional Theory (DFT) calculations. For this we use than EVap . However, for all the hydrogen-related defects, this is
the DFT-PES as well as two approximations of the PES of several not the case and EPar fails to reproduce the τc and τe values.
defects to calculate τc and τe in a pMOS device. Our study covers Although EPar only tends to slightly underestimate the reaction
different possible defect candidates, which have been previously barrier (as shown in Fig. 2 left), τc and τe are lower by several
identified as sources of degradation in silicon dioxide [11–13]: orders of magnitude (Fig. 3 right). Using EVap the agreement
The oxygen vacancy (OV, see Fig. 1b) in α-quartz, the hydrogen improves considerably.
bridge (HB, see Fig. 1c) in both α-quartz and in amorphous This observation also holds for a statistical comparison for
silicon dioxide (a-SiO2 ) and the hydroxyl E’ center (H-E’, see several defects in a-SiO2 , as can be seen in Fig. 4. Correlation
Fig. 1d) in a-SiO2 . Since the amorphous structures differ from improves significantly when using EVap instead of EPar . This is
each other, also the PES varies. Therefore we have investigated valid for all five voltages presented in these plots.
11 HB defects and 12 H-E’ defects from [14] to provide statistical The slope of the τc (VG ) curve determines the dependence of τc
data. We show that the parabolic fit to the PES underestimates on the gate bias [2]. We determined the τc ′ (VG ) curves in the
τc and τe by several orders of magnitude. Therefore we propose range VG = 1.0 V to 2.5 V by a least square fit to each data
a different approximation that yields to more accurate τc and set. The slope for the different approximations is again plotted
τe and captures several desired features, also in cases where the in correlation-plots (Fig. 5). Like in Fig. 4 EPar gives a poor
parabolic approximation fails. match to the behavior calculated using EDFT , whereas EVap fits
Simulation Framework: For our DFT calculations we used well. This is observed for all four temperatures investigated.
the CP2K framework [15] with the non-local PBE0 TC LRC Finally, a good approximation should also be able to reproduce
hybrid functional [16]. Large a-SiO2 structures containing 243 the temperature dependence of τc (the activation energy). In
atoms (crystalline) or 216 atoms (amorphous) were created using Fig. 6 a least square linear fit for τc (T) at VG = 1.0 V is provided
ReaxFF [17]. Using DFT, the defects were relaxed in their neutral for the τc (T) calculated using EDFT and both approximations
and positive state. This results in slightly different atomic con- discussed. Also here EVap reproduces the results in α-quartz and
figurations for each charge state. The normalized reaction coor- for the H-E’ better than EPar . Interestingly here the fit for the HB
dinate (NRC) is defined as the normalized atomic displacement- in a-SiO2 seams slightly worse than the parabolic approximation.
vector between them. For constructing the PES we used this Conclusions: We have calculated τc and τe for several defects
displacement vector to interpolate configurations between the and several approximations of the PES along the reaction coor-
two states (range 0.0 to 1.0), or extrapolate configurations in dinate. By comparing them to results obtained using DFT, we
the range -1.0 to 0.0 (or 1.0 to 2.0 respectively). The PES was show that the widely used parabolic approximation is poor for
then calculated by computing the energies for 30 configurations most of the investigated defects in this work, since it leads to
along the NRC in the range -1.0 to 2.0, and spline interpolation an underestimation of τc and τe by several orders of magnitude.
in between (see Fig. 2). Furthermore it also fails to represent the slope τc ′ (VG ) or the
These PESs calculated by DFT (EDFT ) were taken as a reference activation energy of the DFT-results. We propose a different
to evaluate the quality of possible approximations. When a gate approximation which captures the curves in the area of interest
bias is applied, the PESs shift relative to each other along the (near the crossing point) much more accurately. This leads to a
y axis, changing the reaction barrier and therebyτc and τe . In much better approximation of τc and τe and their characteristics
the following we will focus on these quantities, investigating for most of the investigated defects.
whether the aforementioned approximations accurately reproduce References:
τc and τe and their dependences. For this purpose the reaction
time constants vs. gate-voltage curves (for VG = −2.5 V to [1] P. Blöchl, Phys. Rev. B 62, 6158 (2000). [2] T. Grasser, Microel. Rel. 52, 39
2.5 V) were calculated for each defect in a pMOS device with (2012). [3] T. Grasser et al., (2015), (IRPS) 2015, accepted. [4] F. Schanovsky
oxide thickness of 2.5 nm, for different temperatures (T= 50 °C, et al., Proc. SISPAD (2013), pp. 1–4. [5] P. Lenahan et al., IEEE Trans.Nucl.Sci.
100 °C, 150 °C and 200 °C). For these calcualtions we use a semi- 48, 2101 (2001). [6] A. Palma et al., Phys. Rev. B 56, 9565 (1997). [7] J. Ji et al.,
classical approach based on [9] and [18]. An example of these Proc. IEDM (2014), pp. 21–4. [8] K. Huang et al., Proc .R.Soc.A 204, 406 (1950).
curves for different defects and approximations is depicted in [9] C. Henry et al., Pys. Rev. B 15, 989 (1977). [10] F. Schanovsky et al., JCE
Fig. 3. 11, 218 (2012). [11] F. Feigl et al., SSC 14, 225 (1974). [12] D. Griscom, Phys.
Results and Discussion: An analytic approximation should not Rev. B 22, 4192 (1980). [13] A.-M. El-Sayed et al., PRL 114, 115503 (2015).
only be able to fit the τ (VG ) curves (Fig. 3 and Fig. 4) but also [14] T. Grasser et al., Proc.(IEDM) (2014), pp. 530–533. [15] B. G. Lippert
the dependence on the gate bias (Fig. 5) and the temperature- et al., Mol. Phys. 92, 477 (1997). [16] M. Guidon et al., J.Ch.Theor.Comp. 5,
dependence (Fig. 6). The functions for the parabolic approxima- 3010 (2009). [17] A. van Duin et al., J.Ph.Chem.A 105, 9396 (2001). [18] F.
tion (EPar ) are usually calculated by fitting the parameter α in Schanovsky, Ph.D. thesis, 2013. [19] W. Gös et al., Bias Temp. Inst. for Dev.
EPar (q) = α(q − q0 )2 (1) and Circ., edited by T. Grasser (Springer, 2013), pp. 409–446.
ISBN 978-0-692-50554-0 97
O
defect free OV
O O O
O
Si O O
O Si Si
O Si
O
O
a) O b) O
H
HB H-E’
H
O O O
O
Si Si O Si O
Si
O O O
O O
c) d) O
O
Fig. 1: Schematics of the different defects (b,c,d)

used for the calculations compared to the defect-free
structure (a). In the oxygen vacancy (b) an O-atom Fig. 2: Cut through the potential energy surface along the norm. reaction coordinate for the neutral to the
is missing, while for the hydrogen bridge (c) it is positive state of the hydrogen bridge (see Fig. 1 c) in α-quartz and the two different approximation variants
replaced by a H-atom. In the case of the hydroxyl E’ discussed in this paper. In the parabolic approximation (left), the neutral PES (grey) is constructed using (1)
center (d) a H-atom is bound to the O-atom having fitted to the two grey points. These points are the value of the neutral PES at the neutral configuration and at
broken up one of the Si-O bonds. The latter defect the positive configuration when charged neutrally. The positive PES is constructed in a similar way.
needs bond distances larger than 1.65A [14] which The V-approximation (2)(right) represents the EDFT near the intersection point much better. Taking the neutral
are not present in α-quartz and therefore does only PES (light blue) as an example, one can see that here the points at 0.0 and 0.5 on the norm. reaction coordinate
exist in amorphous structures. are used for the approximation. When a voltage is applied, the PESs shift relative to each other along the y
axis, changing the intersection point which defines the reaction barrier in the classical high-temperature limit.
Fig. 3: Depending on the applied bias conditions, the PESs in Fig. 2 move up or down changing the energy-barrier that has to be overcome for charge-capture
or emission. The above plots show the reaction time-constants τc and τe at T=100°C. It can be seen that they are very sensitive to a change of the barrier height.
Although the parabolic approximation is good for the oxygen vacancy (left) (even slightly better than the V-approximation), it fails by several orders of magnitude
for all the hydrogen-related defects (right). For the defects in a-SiO2 , the behavior of τc (VG ) (τe (VG )) is of course slightly different for each investigated structure.
The two respective graphs are examplary for one of the investigated structures. One can see a considerable improvement when the V-approximation (2) is used. To
quantify the improvement, a l2 norm for log(τc (DFT))-log(τc (appr.)) (similar for τe ) is shown as score function.
Fig. 4: Double logarithmic correlation-plots for the absolute values of τc at

T=100°C for the hydrogen bridge in a-SiO2 (top) and hydroxyl E’ center (bottom). Fig. 5: Correlation-plots for the slope of log10 (τc (VG )). Like in Fig. 4, the
Correlation of τc calculated using EDFT and τc calculated using EPar (left) hydrogen bridge in a-SiO2 is depicted on top and the hydroxyl E’ center at the
or EVap (right) are shown. When τc for both approximations is the same, the bottom. Correlation of τc ′ calculated using EDFT and τc ′ calculated using EPar
respective point would lie along the diagonal, therefore the off-diagonality can (left) or EVap (right) are shown. It is clearly visible that τc ′ (VG ) is overestimated
be used as a criterion determining the quality of the fit. It is clearly visible that in the parabolic approximation. Also, to quantify the improvement, a l2 norm for
τc tends to be underestimated in the parabolic approximation. To quantify the the distance to the diagonal is shown as a score function.
improvement a l2 norm for the distance to the diagonal is shown as a score
function.
Fig. 6: Arrhenius plots showing the temperature dependence of τc (T) at VG = 1.0 V in the four investigated defects. In the amorphous structures, the values
are distributed over a wide range. The lines are a least square fit to the data points, also providing the slope to calculate the activation energy Ea . To quantify the
improvement, a l2 norm for the differences to the results using EDFT (black) is depicted as score function. For the case of the HB in α-quartz and for the H-E’,
EVap (light blue) is a better approximation than EPar (grey). Interestingly, for the HB in a-SiO2 the score for EVap is slightly worse.
98
Ultra-thin silicon membranes and silicon
nanowires as nanophononic and thermoelectric
devices
Daniele Selli, Shiyun Xiong, Claudia Mangold, Sanghamitra Neogi, and Davide Donadio∗
Max Planck Institute for Polymer Research, Ackermannweg 10, 55128 Mainz, Germany
∗ University of California Davis, One Shields Ave. Davis, CA, 95616
Engineering silicon at the nanoscale paves the

way to new applications of this cheap, abundant, and
technologically and environmentally friendly ma-
terial. Transistors in nanoelectronics have reached
the 10 nm size limit, implying very high density
but also critical issues regarding heat dissipation.
The fabrication of crystalline silicon nanowires and
ultra-thin silicon membranes with characteristic size
Fig. 1. Atomistic models of crystalline, rough and surface ox-
of the order of 10 nm or less gives rise to thermal idized silicon membranes, for which we have compute phonon
and electronic properties that enable the use of dispersion relations, thermal conductivity and phonon mean
silicon as a low-temperature thermoelectric material free paths.
for sensing and energy harvesting. At the same
time, nanowires, thin films and membranes can
be potentially employed as the basic platform to A combination of large scale molecular dynam-
develop nano-resonators, phononic crystals or meta- ics, lattice dynamics and electronic structure cal-
materials, with a variety of applications. It follows culations provide parameter–free recipes to opti-
that an efficient exploitation of nanoscale silicon mize materials for thermoelectric, nanoelectronic
stems from a better understanding of its electronic, and phononic applications. Indeed from these calcu-
phononic and thermal properties, and especially on lations we envisage strategies to finely tune silicon
the impact of confinement and surfaces. membranes to further improve their thermoelectric
In this work we unravel the effect of dimen- performances.
sionality reduction and the prominent role of sur- ACKNOWLEDGMENT
face structure and surface resonances in tuning the
This work has been supported by the European
phonon dispersion relations and the thermal conduc-
Commission FP7-ENERGY-FET project MERG-
tivity of silicon nanostructures, including ultrathin
ING, contract number 309150.
membranes [1], [2] and silicon nanowires [3], [4].
In particular we show how surface roughness and R EFERENCES
surface oxidation affect phonon transport in silicon [1] S. Neogi and D. Donadio, Eur. Phys. J. B 88, 73 (2015).
nanostructures, reducing dramatically their thermal [2] S. Neogi, J. Sebastian Reparaz, L. F. C. Pereira, B.
conductivity 1. We also compute the electronic Graczykowski, M. R. Wagner, M. Sledzinska, A. Shchep-
properties of silicon membranes as a function of etov, M. Prunnila, J. Ahopelto, C. M. Sotomayor Torres,
and D. Donadio, Acs Nano 9, 3820 (2015).
thickness and the concentration of dopants, identify- [3] D. Donadio and G. Galli, Nano Lett 10, 847 (2010).
ing the conditions that maximize the thermoelectric [4] I. Duchemin and D. Donadio, Appl Phys Lett 100, 223107
figure of merit, both for periodic extended materials (2012).
and for open systems in device configuration.
ISBN 978-0-692-50554-0 99
100
Calculation of phonon transmission in Si/PtSi
heterostructures
Jung Hyun Oh, Moon-Gyu Jang∗ , and Mincheol Shin
Department of Electrical Engineering, Korea Advanced Institute of Science and Technology,
Daejeon 305-701, Korea
∗ Department of Materials Science and Engineering, Hallym University, Chuncheon 200-702, Korea
e-mail:[email protected]
I NTRODUCTION atoms in the Si layer occurs roughly within one unit

In thermoelectric devices, the thermal conduc- cell (aSi = 0.546 nm). We also find that most atoms
tance of materials is an important factor to enhance in the Si layer near the interface deviate from the
the figure of merit because it is responsible for 4-bond geometry as shown in the two lower panels
maintaining the temperature difference between the of Fig. 2. A phonon transmission function T (E)
hot and cold sides. The thermal conductance of ma- is calculated using a Green’s function method and
terials was suppressed using a underlying principle the results are shown in Fig. 3. It is found that the
to prohibit phonon propagation through devices as transmission function is reduced by one order of
much as possible. This is because most thermal en- magnitude, compared to the Si bulk value (dotted
ergy is carried by phonons in semiconductors. There line), when a PtSi silicide layer is incorporated.
have been efforts to suppress thermal conductance Since there is a lack of acoustic modes at 20 ∼
in a variety of ways in nanowires(1D), phononic- 30 meV and of optical modes above 50 meV in
crystals(2D), and three-dimensional heterojunctions a PtSi layer, phonons in this energy range are
(3D).[1], [2] completely blocked. As we increase the number
In this work we examine the suppression of of PtSi layers, the calculated thermal conductance
phonon transport in another example, Si and SiPt becomes smaller. In Fig. 4, we show the suppression
heterostructures (3D). This heterostrucure is be- of the thermal conduction. It is found that thermal
lieved to have the benefit that the electrical con- conductance is rapidly suppressed by the insertion
ductance can be kept high while the phonon prop- of a PtSi layer, however, eventually saturated over
agation is suppressed due to the large acoustic about 3 layers. In the case of T = 300 K with three
impedance mismatch between Si and PtSi.[3] PtSi layers, the thermal conductance is suppressed
by about 30-times, compared to the value for Si
M ETHOD AND D ISCUSSION bulk.
In order to obtain a quantitative value for the ACKNOWLEDGMENT
phonon suppression, we use a method based on
This work was supported by the Basic Sci-
atomic vibrations. To do this, we first perform
ence Research Program through the NRF of Korea
DFT calculations for phonon dispersion of Si and
funded by the Ministry of Education, Science and
PtSi and find Harrison’s potential parameters of Si
Technology(2012R1A1A2039886) and ETRI.
and PtSi by fitting to them (Fig. 1). To generate
atomic positions around the interfaces between Si R EFERENCES
and PtSi, we resort a molecular dynamic method [1] D. Li, Y. Wu, P. Kim, L. Shi, P. Yang, and A. Majumdar,
based on the Stillinger-Weber potential while atoms Appl. Phys. Lett. 83, 2934 (2003).
in PtSi layers are fixed. An example of a unit block [2] S.-M. Lee, D. G. Cahill, and R. Venkatasubramanian, Appl.
Phys. Lett. 70, 2957 (1997).
for heterostructures is illustrated in Fig. 2 after [3] W. Choi, Y. S. Park, Y. Hyun, T. Zyung, J. Kim, H. Jeon,
several hundred simulations with different initial M. Shin, and M. G. Jang, J. Nanosci. Nanotechno. 13,
conditions. It is found that the redistribution of (2013).
ISBN 978-0-692-50554-0 101

TDOS
1.6
50 Pt
N=0 (a)
N=1
Si 1.2
T(E) (nm-2)
40
0.8
Energy (meV)
N=2
30
0.4
20 N=4
0.0
10
0 20 40 60
E(meV)
0 20
ΓX SY ΓZU RT Z 0 2 4 6 8
(b)
DOS E=12meV
55
15
T0(E)/T(E)
50 E=40meV
10
40 E=6meV
5
E(meV)
30
0
0 1 2 3 4 5 6
20 PtSi layer, N
10 Fig. 3. (a) Calculated transmissions as a function of energy for

different numbers of PtSi layers. (b) Normalized transmission
functions for three different energies plotted against the number
0
0 2 4 6 8
0
of PtSi layers. The thickness of the PtSi layer is 8.89nm (15

Γ X S YΓ ZU R T Z
0 800
cells) and their intervals are 8.18nm (15 Si cells), with a cross-
DOS sectional area of 0.546nm×1.092nm.
Fig. 1. Phonon dispersion calculated with an ab initio method 0.08
(upper panel) and a Harrison-type potential (lower panel). An N=0 (a)
κ (109 W/m2 K)
N=1
orthorhombic unit cell of size 0.592nm×0.560nm×0.385nm 0.06
is used and there are four Si and four Pt atoms within it.
[010]
0.04 N=2
y N=3
Si PtSi x [100] Si N=5
3
0.02
0.00
0 200 400 600 800 1000
T(K)
3 3
Number of atoms
Number of atoms
-1
left right
-8 8
2 2 (b)
30 T=500K
1 1
T=300K
κSi/κ(N)
0
0.2 0.3 0.4
0
0.2 0.3 0.4 20 T=100K
Bond length (nm) Bond length (nm)
Fig. 2. (upper panel)Schematic representation of a Si/PtSi 10

heterostructure. Thin (thick) circles are Si (Pt) atoms where 0 1 2 3 4 5 6
the size of circles denotes their positions on the z-axis. The PtSi layer, N
x-axis ([100]) is a phonon propagation direction and the cross-
sectional area normal to that is aSi ×2aSi . (two lower panels) In Fig. 4. (a) Calculated thermal conductance is plotted as a
a unit cell just left (right) to the PtSi layer, the average number function of temperature with the transmission function in Fig. 3.
of neighboring atoms for each Si is plotted as a function of (b) Ratios of thermal conductance between bulk Si and PtSi/Si
bond √ length. The dotted line denotes the Si bond length in heterostructures are compared as we increase the number of
bulk ( 3/4aSi ). PtSi layers for three different temperatures.
102
Thermal conductivity tensor of
InxGa1−xAs/Iny Al1−y As superlattices and
application to quantum cascade lasers
S. Mei,∗ Z. Aksamija,† and I. Knezevic∗
∗ Electrical and Computer Engineering, University of Wisconsin-Madison, Madison, WI 53706, USA
† Electrical and Computer Engineering, University of Massachusetts-Amherst, Amherst, MA 01003, USA
An InGaAs/InAlAs superlattice (SL) on an InP terial mismatch separately. The roughness scatter-
substrate is the mainstream material system for mid- ing is incorporated in the scattering time τb (q)
IR quantum cascade lasers (QCL). The thermal con- and the extra interface resistance caused by mis-
ductivity tensor of SLs is critical for energy-efficient match is added when calculating the cross-plane
performance of QCLs; understanding the relative SL thermal conductivity. We consider a SL with
importance of different factors that influence heat 2n layers in a single stage, which is formed by
flow in these systems is essential in device model- alternating materials 1 and 2 with layer thick-
ing and optimization [1]. It is known that thermal nesses L11 , L12 , ..., L1n ; L21 , L22 , ..., L2n . The in-
conduction in SLs is extremely anisotropic, with the plane and cross-plane thermal conductivities are
cross-plane thermal conductivity much lower than calculated as:
the in-plane conductivity. Unfortunately, it is very n xx xx
κin−plane = (L
i=1
1i κ1 +L2i κ2 )
; (2)
difficult to experimentally measure the full thermal n
(L1i +L2i )
i=1
conductivity tensor, and systematic modeling of SL n
(L1i +L2i )
systems is also missing.
κcross−plane = n
L1 i
i=1
L i
1 1
.(3)
i=1 κ
yy + κyy
2
+n G1→2
+G
In this work, we calculate the full thermal con- 1 2 2→1
ductivity tensor in the InGaAs/InAlAs SL sys- Here, G1→2 and G2→1 are the thermal boundary
tem at a specified lattice temperature, considering conductances from 1 to 2 and from 2 to 1, re-
yy
full phonon dispersions, accurate phonon scattering spectively. For materials 1 and 2, κxx1,2 and κ1,2 are
rates, and the thermal interface resistance caused by the in-plane and cross-plane thermal conductivity,
both interface roughness and material mismatch. respectively, and they include τb (q).
We calculate the thermal conductivity tensor Thermal boundary conductance (TBC) is the
based on the phonon Boltzmann transport equation dominant factor in the anisotropy of thermal con-
[2]: ductivity in SLs. TBC has been extensively studied
by molecular dynamics (MD), but very few exper-
καβ = τb (q)Cb,T (q)vbα (q)vbβ (q), (1) iments have been conducted due to difficulties in
b,
q
sample preparations [3]. Neither does a model exist
where τb (q) is the that explains the experimental or the MD data over a
total phonon relaxation time and
∂ h̄ω(
q) wide range of temperatures and surface conditions.
kB T
Cb,T = ∂T is the phonon heat capacity for There are two primary models describing phonon
mode b at temperature T . Both the group velocity transmission through the interface: the acoustic mis-
and the heat capacity are calculated from the full match model (AMM) and the diffusive mismatch
dispersion relation using the adiabatic bond charge model (DMM) [4]. In the AMM, an essential as-
model (Figs. 1,2). sumption is that phonons are governed by contin-
We have previously shown [2] that we can uum acoustics and can therefore reflect or transmit
treat the influence of surface roughness and ma- at the boundary, following Snell’s law. In the DMM,
ISBN 978-0-692-50554-0 103

a phonon is destroyed on one side of the interface
and a new phonon, with the same energy and a
completely random momentum, is created on the
other side; effectively, the DMM describes fully
diffuse elastic scattering. Both experiment and MD
show that real transport is somewhere between the
two limits.
Here, we introduce a parameter that helps inter-
polate between the AMM and DMM transmission
rates and get a better description of the TBC. With
the SL thermal conductivity and the TBC model,
we calculate the thermal conductivity tensor of
InGaAs/InAlAs SLs (Figs. 2, 3). The results also in-
cludes strain effects, which are often associated with
III-V superlattices and very important in QCLs. The
Fig. 1. Phonon dispersion of In0.53 Ga0.47 As calculated from
calculated cross-plane thermal conductivity values the adiabatic bond charge model.
agree very well the recent experimental results (Fig.
3) [1].
This work was supported by the NSF award
1201311 (S.M.) and by the DOE award DE-
SC0008712 (I.K.).
R EFERENCES
[1] A. Sood, J. A. Rowlette, C. G. Caneau, E. Bozorg-Grayeli,
M. Asheghi, and K. E. Goodson, Thermal conduction
in lattice-matched superlattices of InGaAs/InAlAs. Appl.
Phys. Lett. 105, 051909 (2014).
[2] Z. Aksamija and I. Knezevic, Thermal conductivity of
Si1−x Gex /Si1−y Gey superlattices: Competition between
interfacial and internal scattering. Phys. Rev. B 88, 155318
(2013).
[3] D. G. Cahill, P. V. Braun, G. Chen, D. R. Clarke, S. Fan,
K. E. Goodson, P. Keblinski, W. P. King, G. D. Mahan, A.
Majumdar, H. J. Maris, S. R. Phillpot, E. Pop, and L. Shi,
Nanoscale thermal transport. II. 2003-2012. Appl. Phys.
Rev. 1, 011305 (2014). Fig. 2. Calculated bulk thermal conductivity of InGaAs and
[4] E. T. Swartz and R. O. Pohl, Thermal boundary resistance. InAlAs as a function of temperature.
Rev. Mod. Phys. 61, 605 (1989).
Fig. 3. Comparison of the calculated cross-plane thermal con-

ductivity for different InGaAs/InAlAs SLs with experimental
results from Sood et al. [1].
104
Multi-Scale Modeling of Self-Heating Effects in
Nano-Devices
S. Qazi1, A. Shaik1, A. Laturia1, R. Daugherty1, X. Guo2, E. Bury3, B. Kaczer3, K. Raleva4 and D.Vasileska1
1
School of Electrical Computer and Energy Engineering,
2
School for Engineering of Matter, Transport and Energy
Arizona State University, Tempe, Arizona, 85287-5706, USA
3
IMEC, Kapeldreef 75, 3001 Leuven, Belgium
4
Faculty of Electrical Engineering and Information Technologies,
Ss. Cyril and Methodius University, Skopje, Republic of Macedonia
This paper discusses a multi-scale device modeling methods have to be employed. One such method is
scheme for analyzing self-heating effects in nanoscale the heater-sensor experimental structure (IMEC)
silicon devices. A 2D/3D particle-based device simulator is
self-consistently coupled to an energy balance solver for
[3]; the heater is the device under test (DUT) and
the acoustic and optical phonon bath. This simulator is the sensor determines the local temperature. This
used to analyze the hot-spot temperature and location in configuration uses the proximity of the heater to
various SOI devices, dual gate structures and nanowire the sensor and the temperature dependence of the
transistors. This device simulator has been coupled to a subthreshold slope of the sensor to infer the hot-
SILVACO simulation tool which solves for heat transport
spot temperature in the system.
in interconnects at the circuit level. The proposed multi-
scale simulation scheme allows for analysis of thermal To simulate the conditions in IMEC experi-
effects in an integrated circuit. Simulation results obtained ments, two significant changes were made to the
with this simulator are in agreement with experimental existing simulator: (1) introduction of multiple
measurements from IMEC using specialized heater-sensor terminals to model two transistors in common-
test structures in common-source and common-drain source and common-drain configuration, and (2)
configurations.
coupling the thermal solver to a circuit level simu-
Since operating voltages do not scale propor- lator to account for heat propagation in the inter-
tionally to device size [1], scaling semiconductor connects. Important SILVACO modules that treat
devices towards the nanometer regime leads to a these self-heating effects are THERMAL3D and
variety of unwanted phenomena. Among them, GIGA3D. This coupling, through a global Gum-
self-heating is of particular interest as it can demel loop is schematically illustrated in Fig. 3. In
grade device performance. The mechanism for this arrangement, the SILVACO module simulates
self-heating can be understood through the energy thermal transport characteristics at the interconnect
flow diagram in Fig. 1. Through the application of level, providing temperature boundary conditions
drain bias, electrons gain energy from the applied for the device-level simulation. The device level
lateral electric field and interact with both optical simulator then solves for the lattice temperature
and acoustic phonons. The majority of the electron profile and the temperature of the hot-spot. The
energy is transferred to optical phonons. The zone convergence of the global Gummel scheme is
center optical phonons practically have zero group shown in Fig. 2. Joule heating power and inferred
velocity and the transfer of heat to acoustic pho- temperature of the hot spot are in agreement with
nons via anharmonic decay is very slow. This experimental findings from IMEC.
causes hot-spots to form. The elevated temperature
[1] K. Raleva, D. Vasileska, S. M. Goodnick, and M.
in the active region of the device leads to enhanced Nedjalkov, “Modeling thermal effects in nanodevices,”
phonon scattering, and therefore, mobility and IEEE Trans. Electron Devices, vol. 55, no. 6, pp. 1306–
current reduction. This effect is called self-heating 1316, Jun. 2008.
[2] A. Majumdar, K. Fushinobu, K. Hijikata, “Effect of hate
[2]. voltage on hot electron and hot phonon interaction and
Direct measurement of the temperature of the transport in a submicrometer transistor”, J. Appl. Phys.,
hot-spot is an impossible task; therefore, indirect vol. 77, pp. 6686-6694, (1995)
[3]. E. Bury and B. Kaczer, IMEC, private communication
ISBN 978-0-692-50554-0 105

Fig. 1. This diagram shows the basic mechanism by which
Fig. 2. Convergence of the global Gummel loop. Inset shows
energy is exchanged between high energy electrons and the
converged lattice temperature profile.
phonon bath.
VS=0
VG1=1.66V VG2=0.7V
VD1=1.66V VD2=0
Device setup
heater sensor
VD=1.66V
VG1=1.66V VG2=2.2V
VS1=0 VS2=1.45V
heater sensor
Silvaco GIGA3D
Joule heating
Temperature BC MATLAB
MATLAB Thermal Solver
Joule heating
Lattice temperature
no Converged? yes
Fig. 3. Multiscale model implemented in this work.
106
Enhanced Seebeck coefficient due to quantum
confinement in gated 2d nanomembranes
A. Kommini and Z. Aksamija
Electrical and Computer Engineering, University of Massachusetts Amherst, MA 01003, USA
E-mail: [email protected]
The effective control on electron and phonon of both charge and energy and applied voltage
transport due to confinement, interface and quantum controls the subband seperation. The subband ener-
effects has made nanostrutures as a good way to gies are calculated by self-consistently solving the
achieve high figure-of merit (ZT ). Maximizing the Poisson and 1-D Schrödinger equation. The S of a
power factor S 2 σ (S is seebeck coefficient and σ is nanostructure is given [2] by L(1) /L(0) where the
electrical conductivity) and restricting the phonon transport integrals are
transport can improve ZT . Mahan and Sofo [1] ∞ ∂fF D α
proposed that a delta-shaped Transport Distribution L(α) = −e2 b
∂E σ(E)(E + En − EF ) dE
n,b 0
Function (TDF) σ(E) = τ (E)g(E)v 2 (E) can sig-
nificantly improve S thereby improving the overall and Enb is energy of the n’th subband, gnb (E) is den-
ZT through electron filtering. Achieving a delta- sity of states, τnb (E) is energy dependent relaxation
shaped TDF, however, proved difficult as the sharp time, and vx (E) is velocity along the x direction.
features in the density-of-states (DOS) g(E) are The electron relaxation times are calculated by con-
readily cancelled by related features in the scattering sidering surface roughness, elastic acoustic phonon
rate τ −1 (E). scattering, and inelastic optical phonon (both intra
In this work, we studied ways to achieve such and intersubband) scattering mechanisms. We com-
a delta-shaped TDF by gate-tuning the subband puted S and observed a significant enhancement,
spacing and scattering mechanisms in a gated 2- especially at low temperatures when optical phonon
dimensional silicon nanomembrane. The 2-d nature scattering is the dominant mechanism (restricting
of the system leads to a step-like electronic DOS the other scattering mechanisms) as shown in Fig. 1.
gnb (E) = m∗n /(πh̄2 )Θ(E − Enb ) with one step In a confined nanostructure, the discrete energy
contributed by each subband n in each ladder b. bands limits the states available for electron to
Inelastic scattering rates are proportional to the occupy which in turn limits phonon energies that
DOS; however, inelastic and intersubband scattering can participate in scattering. Fig. 2 shows optical
can only begin to occur when the carrier accumu- phonon energies that can provide enhancement in
−1
lates enough energy so τinel. (E) ∝ g(E ± ∆E) S at low temperatures. Enhancement occurs when
where ∆E is the energy being exchanged–either the optical phonon energy is comparable to the
the difference in subband energies for intersubband subband energy difference. This filters out only
scattering or the optical phonon energy h̄ωop for a narrow energy window between two successive
optical transitions. Hence these mechanisms, when subbands, resulting a higher Seebeck coefficient as
inelastic scattering is dominant, delay the onset of seen in Fig. 2. We thus produce a delta shaped
the step in the scattering rate relative to the DOS Transport Distribution Function (TDF) as shown in
and thus lead to a delta-shaped TDF. Fig. 3 with a peak at lower subbands and decreas-
To explore the practical possibility of achieving ing rapidly towards the high energies. Tuning the
the delta shaped Transport distribution function a energy difference between the subbands to match
back gated undoped silicon nanoribbon (SiNR) on the optical phonon energies by varying the gate
oxide is simulated. The oxide acts as capacitor voltage is demonstrated in Fig. 4. The observed
which induces charges that participates in transport enhancement is more evident at low temperatures
ISBN 978-0-692-50554-0 107

because of the suppression of phonon absorption.
The enhancement in S can be further increased
by reducing the effective mass, as it is inversely
proportional to step height in the 2-d density of
states, increasing the height of the peak in the TDP
and giving a higher S as shown in Fig. 5.
In summary, we simulated gated 2-d systems
and showed that enhanced thermoelectric properties
are achieved by tuning the subband separation to
match the optical phonon energy and creating a
delta shaped transport distribution function.
R EFERENCES
[1] G. D. Mahan, and J. O. Sofo, The best thermoelectric, Proc.
Natl. Acad. Sci. USA, Vol 93, 15 (1996). Fig. 3. Delta-shaped Transport Distribution Function (TDF)
[2] H. J. Ryu et.al., Quantitative determination of contributions formed at T=100K, Vgate =5V and optical phonon energies
to the thermoelectric power factor in si nanostructures, comparable to the subband energy gap.
Phys. Rev. Lett., Vol 105, 256601 (2010).
Fig. 1. S at several temperatures with a Vgate = 5V for

all scattering mechanisms, reduced acoustic phonon scattering, Fig. 4. The shift in the phonon energies due to the change in
with and without surface roughness scattering. the energy gap between subbands being introduced by applying
a higher gate voltage Vgate = 7V at low temperatures.
Fig. 2. S variation with optical phonon energies showing the

peak being positioned near the valley bottom subband energy
difference with Vgate =5 V Fig. 5. Variation in S with the longitudinal effective mass(m∗l )
at T=100K with Eoptical =0.043eV and Vgate =5V.
108
Electrically Doped 2D Material Tunnel Transistors
H. Ilatikhameneh, F. W. Chen, J. Appenzeller, R. Rahman, and G. Klimeck
Department of Electrical and Computer Engineering, Purdue University, West Lafayette, IN, 47907 USA
e-mail: [email protected].
INTRODUCTION RESULTS
Tunnel field effect transistors (TFETs) have Contradictory to the chemically doped devices
attracted a lot of attention as promising candidates where electric field (E) at the tunnel junction
for low power transistors [1, 2]. It is recently shown depends on EOT [7], in electrically doped BLG and
that 2D material TFETs such as transition metal WTe2 TFETs, electric field depends on the oxide
dichalcogenide (TMD) based TFETs can provide a thickness as shown in Figs. 3 and 4.
high ON-current while keeping the subthreshold
swing (SS) less than conventional Boltzmann limit
 V1  V2
E (1)
of 60 mV/dec at room temperature [3]. These high 2 Ttot
ON-current values in 2D TFETs are due to the tight Here V1 and V2 are the potential levels at two sides
gate control which translates into a high electric of the tunnel junction. This finding shows that the
field at the tunnel junction. Despite the fact that a new analytic model for electrically doped devices
high chemical doping of the source contact is [8] can be applied to small band gap materials too.
crucial for having a high performance TFET [3], Figures 5 and 6 show that the impact of oxide
this doping method has several drawbacks such as thickness is much larger on the transfer
introducing the defect states in the bandgap which characteristics of the electrically doped TFETs
can degrade device OFF-state performance [4]. compared to the impact of their dielectric constant.
Electrical doping on the other hand, avoiding these
problems, is currently the predominant method in CONCLUSION
the realization of 2D TFETs. It is critical for device In summary, atomistic transport simulations
designers to understand how different parameters unveil that in the novel class of electrically doped
affect the performance of electrically doped TFETs (even those made of small band gap
junctions. Equivalent oxide thickness (EOT) has materials) the key parameter for transistor
been generally accepted as the key performance performance is oxide thickness instead of EOT.
factor in these devices. However, in this work, it is
shown through atomistic simulations that in ACKNOWLEDGMENT
electrically gated junctions the thickness of the This work was supported in part by the Center
oxide rather than EOT plays the main role. Two for Low Energy Systems Technology (LEAST),
small band gap 2D materials Bilayer graphene one of six centers of STARnet, a Semiconductor
(BLG) and monolayer WTe2 are studied for high Research Corporation program sponsored by
ON-current TFET application. MARCO and DARPA.
MODEL REFERENCES
Two different tight-binding (TB) models are [1] J. Appenzeller et al., Phys. Rev. Lett. 93, 196805 (2004).
[2] A. M. Ionescu et al., Nature 479, 329337 (2011).
employed here; WTe2 Hamiltonian is represented [3] H. Ilatikhameneh et al., “Tunnel Field-Effect Transistors in
by a sp3d5 2nd nearest neighbor with spin-orbit 2D Transition Metal Dichalcogenide Materials,” accepted for
coupling while BLG Hamiltonian considers only pz publication in IEEE JxCDC (2015) DOI:
orbital. The atomistic quantum transport 10.1109/JXCDC.2015.2423096, arXiv:1502.01760.
[4] S. Agarwal et al., IEEE tran. on elec. dev. 61, 1488 (2014).
simulations use a self-consistent Poisson-NEGF
[5] S. Steiger et al., IEEE tran. on nano. 10, 6, 1464 (2011).
method. The structures of electrically doped WTe2 [6] J. E. Fonseca et al., Journal of Comp. Elec. 12, 592 (2013).
and BLG TFETs are shown in Figs. 1 and 2, [7] K. Suzuki et al., IEEE tran. on elec. dev., 40, 2326 (1993).
respectively. The total thickness of the device is [8] H. Ilatikhameneh et al., “Scaling Theory of Electrically
shown as Ttot in Fig. 1. All of the transport Doped 2D Transistors,” http://arxiv.org/abs/1504.03387.
simulations have been performed with our
simulation tool NEMO5 [5, 6].
ISBN 978-0-692-50554-0 109

Fig. 1. Physical structure of an electrically doped WTe2 TFET Fig. 2. Physical structure of an electrically doped BLG pn
with the oxide lengths of 13nm. junction with the oxide lengths of 20nm.
Fig. 3. Conduction band profiles of WTe2 TFET with different Fig. 4. Conduction band profiles of BLG pn junction with
oxide thicknesses while EOT is fixed. Analytic results are different oxide thicknesses while EOT is fixed. Analytic results
obtained following the new scaling theory [8]. are obtained following the new scaling theory [8].
Fig. 5. Transfer characteristics of WTe2 TFET with different Fig. 6. Transfer characteristics of WTe2 TFET with different
oxide thickness (ϵ=20). oxide dielectric constants. (𝑇𝑇𝑡𝑡𝑡𝑡𝑡𝑡 = 4.4𝑛𝑛𝑛𝑛).
110
Performance Degradation of superlattice
transistors by scattering
Pengyu Long, Mark Rodwell*, Gerhard Klimeck and Michael Povolotskyi
Purdue Univeristy, West Lafayette, Indiana, USA 47906
*
University of California Santa Barbara, Santa Barbara, California, USA 93106
transport throughout the device, the other divides

INTRODUCTION
the device into two reservoirs with scattering and a
As supply voltage is scaling down with device ballistic channel. Constant Fermi levels are
dimension, steep subthreshold slope (SS) assumed in the reservoirs, which consists of high
transistors have been studied intensively because doping contacts and superlattice region.
they can operate at low supply voltage and with To include the effect of scattering in the
fast switching speed. reservoirs, a small imaginary scattering potential is
One approach to obtain a SS steeper than added to the diagonal elements of Hamiltonian
60mV/dec is to use an energy filter to filter out matrix. 5meV and 10meV are used for the
high energy electrons. Tunneling FETs use band- imaginary potential, similar to values used in the
to-band tunneling as an energy filter and have literature [4].
steep SS, but suffered from small Ion due to the
limit of tunneling probability [1] . RESULTS
Superlattice MOSFETs use the miniband and As the imaginary scattering potential increased
minigap of superlattice as energy filter, and from 0 to 10meV, the transmission probability in
ballistic modeling has shown SS of ~ 30 mV/dec the minigap increased from 10-3 to 2x10-1 (Fig. 2).
and on current of ~ 350 uA/um [2]. However there This indicates that the OFF-current will increase.
is a doubt whether scattering will degrade the steep The slope of decay at upper edge of 1st miniband
Subthreshold swing and ON/OFF ratio of the also degraded, indicating a degradation of SS. in
device. addition the transmission in the miniband gets
A full self-consistent quantum transport smoother. This indicates that the ripples in the
simulation including explicit scattering process miniband will hurt the Ion less than it did in the
can describe the superlattice MOSFET accurately, ballistic scenario.
but high computation load is required. Here we Under ballistic assumption, there is a clear
will use a simple model that is as fast as ballistic distinction between miniband and minigap in
simulation but capable of describing different energy resolved current (Fig. 3). With the
intensities of scattering. imaginary scattering potential introduced, the
distinction between miniband and minigap
MODEL
becomes blurred, so more high energy electrons
The device is based on a double-gate are getting into the channel.
MOSFETs (Fig.1) with a 3.3 nm thick undoped Using 0.1 uA/um as the off current(Ioff), the
In0.53Ga0.47As channel, 20 nm gate length, and a transfer characteristics is simulated at a drain bias
2.56 nm thick gate dielectric having εr=20. The of 0.2V. The SS degraded from 32mV/dec to
superlattice has 12 monolayers(ml) thick wells and 57mV/dec as imaginary scattering potential
4 ml, 6 ml, 6 ml, 4 ml barriers. increased from 0 to 10meV. The on current
The potential is obtained by solving the Poisson degraded from ~350 uA/um to 160 uA/um. A
Equation and Quantum transmitting boundary conventional MOSFET with the identical structure
method equations self-consistently [3]. Given the but without the superlattice has 110 A/m on
same potential, transport results from two models current (Fig. 4), so the superlattice MOSFET still
are compared: one calculates ballistic quantum
ISBN 978-0-692-50554-0 111

has better SS and ON/OFF ratio even with
considered scattering.
CONCLUSION
In this work a model that uses an imaginary
scattering potential to describe the effects of
scattering in the reservoirs is applied to simulate
transport properties of superlattice MOSFETs and
results are compared with ballistic simulations. As
scattering intensities increase, the Subthreshold
Swing and ON/OFF ratio both degraded, but
superlattice MOSFETs can still outperform Fig. 2. Left: Transmission probability of a superlattice
conventional MOSFETs at low supply voltage. MOSFET under ballistic assumption, 5meV broadening and
ACKNOWLEDGEMENT 10meV broadening. Right: corresponding band diagram
indicating three regions in the device
The use of nanoHUB.org computational
resources operated by the Network for
Computational Nanotechnology funded by the US
National Science Foundation under Grant Nos.
EEC-0228390, EEC-1227110, EEC-0228390,
EEC-0634750, OCI-0438246, OCI-0832623 and
OCI-0721680 is gratefully acknowledged. This
material is based upon work supported by the
National Science Foundation under Grant Number
(1125017). NEMO5 developments were critically
supported by an NSF Peta-Apps award OCI-
Fig. 3. energy resolved current density for ballistic simulation,
REFERENCES 5meV imaginary scattering potential and 10meV. The figure
[1] Seabaugh A C, Zhang Q. Low-voltage tunnel transistors is in log scale.
for beyond CMOS logic. Proceedings of the IEEE, 2010,
98(12): 2095-2110.
[2] Long P, Povolotskyi M, Novakovic B, et al. Design and
Simulation of Two-Dimensional Superlattice Steep
Transistors 2014
[3] Luisier M, Schenk A, Fichtner W, et al. Atomistic
simulation of nanowires in the s p 3 d 5 s* tight-binding
formalism: From boundary conditions to strain
calculations Physical Review B, 2006, 74(20): 205323.
[4] Klimeck G, Lake R, Bowen R C, et al. Quantum device
simulation with a generalized tunneling formula. Applied
physics letters, 1995, 67(17): 2539-2541.
Fig. 4. Transfer characteristics calculated under ballistic

assumption, with 5meV broadening and 10meV broadening
compared with a MOSFET with identical structure but no
superlattice in the source. The dashed line indicates the
60mV/dec limit.
Fig. 1. Simulation geometry: cross-section of a Double Gate

MOSFET with a superlattice in the source
112
Hybrid states and bandgap in zigzag
Graphene/BN heterostructures
V. T. Tran, J. Saint-Martin, and P. Dollfus
IEF, Université Paris-sud, CNRS, UMR 8622, Bât 220, 91405 Orsay, France
e-mail: [email protected] and [email protected]
(Fig 2b). These states with high velocity are

INTRODUCTION
specific of the B-C interface. We name them
It is well known that in the band structure of “hybrid states”.
zigzag graphene nanoribbons (ZGNRs) edge states TWO INTERFACE STRUCTURE
appear as nearly flat bands, i.e. with nearly zero
Hybrid states at N-C interface have been also
group velocity (Fig. 1c), and thus they cannot
studied. Fig. 3 shows a structure with two BN
contribute significantly to electron transport.[1]
regions attached to a graphene ribbon forming
However, studies of topological insulators (TIs)
both B-C and N-C interfaces. Band structure and
have revealed the existence of states with very
wave functions are analyzed in Fig 4. The flat
high group velocity localized at the surface or edge
bands previously evidenced in B-C structure are
of samples.[2] Taking advantage of the fact that it
no longer present in this B-C--C-N system. New
is now possible to grow in-plane heterostructures
upward bended curves with states localized at the
of hexagonal BN (h-BN) and graphene on the
N-C interface emerge. They are hybrid states
same monolayer,[3] by means of atomistic
specific of the N-C interface.
simulation we have evidenced the emergence of
interface states that look like edge states in ZGNRs
but with high group velocity, as it can be observed HYBRID STATES: SPECIFIC PROPERTIES
in TIs. A comparison of group velocities in different
structures is shown in Fig 5a. The maximum group
MODEL velocity reaches 4.3×105 m/s at B-C interfaces and
Zigzag Graphene/BN heterostructures with even 7.4×105 m/s at N-C interfaces. At given
different configurations of interfaces have been energy, hybrid states in B--C-C-B, N-C--C-N (not
investigated. The notation “B-C” (“N-C”) refers to shown) propagate at both interfaces as the ones in
interfaces where only Boron (Nitrogen) atoms are TIs, while transport interface can be selected in B-
bonded with Carbon atoms. Fig. 1b shows a “one- C--C-N structure as a function of energy.
interface” system with B-C interface, while a Additionally, a bandgap of 207 meV can be open
“two-interface” system, noted as B-C--C-N, is in B-C--C-N structure in the case of sub-ribbon
schematized in Fig. 3. Hamiltonians were widths of 5 nm (Fig 5b).
developed within a Tight Binding (TB) model up The role of B(N) atoms in the formation of
to second neighbor interactions. Parameters were hybrid states can be intuitively understood by
taken from [4]. The properties of interface states applying on each side of a pure zigzag graphene
were investigated in terms of localization of wave ribbon two different effective potentials, as
functions, group velocity and bandgap. sketched in Fig 6a. These potentials are equal to
ONE INTERFACE STRUCTURE
1.394 eV and -0.75 eV, respectively, i.e. the actual
energy values of the hybrid states at the Brillouin
Fig 2a shows the energy bands of the B-C zone edge in Fig. 4(a). This approach makes it
structure. It is clear that the flat bands in the first possible to mimic the dispersion and hybrid edge
valence band of pure graphene ribbon in Fig.1c are states in graphene/BN heterostructures, as shown
now replaced by downward bended bands (red in Fig 6c.
lines) localized at the interface of Graphene/BN
ISBN 978-0-692-50554-0 113

CONCLUSION
We have evidenced the emergence of
dispersive edge states at the interfaces of zigzag
graphene/BN in-plane heterostructures. Their
specific properties suggest new ways to engineer
and control the bandgap and carrier transport
properties in graphene nanostructures.
REFERENCES Fig. 3. Graphene/BN zigzag heterostructure with 2 interfaces.
[1] A. Cresti, G. Grosso, and G. Parravicini, Electronic states

and magnetotransport in unipolar and bipolar graphene
ribbons, Phys. Rev. B 77, 115408 (2008). a) 1.5 b) 0.35
N-C N-C
BN1 graphene BN2
1 0.3
[2] B. A. Bernevig, T. L. Hughes, and S.-C. Zhang, Quantum
0.25
spin Hall effect and topological phase transition in HgTe
(arb. unit)
0.5 (1)
quantum wells, Science 314, 1757 (2006). 0.2
E (eV)
0 (2)
0.15 (2) (1)
[3] Y. Gao, Y. Zhang, P. Chen, Y. Li, M. Liu, T. Gao, D. Ma,
2
-0.5

0.1
Y. Chen, Z. Cheng, X. Qiu, W. Duan, and Z. Liu, Toward B-C B-C
-1
Single-Layer Uniform Hexagonal Boron Nitride-Graphene 0.05
Patchworks with Zigzag Linking Edges, Nano Lett. 13, 10 -1.5 0
-1 0 1 0 30 60 90 120 150
(2013). k a / atom position along y
x x
[4] G. Seol and J. Guo, Bandgap opening in boron nitride Fig. 4. a) Band structure and b) Wave functions in B-C--C-N
confined armchair graphene nanoribbon, Appl. Phys. Lett. heterostructure.
98, 143107 (2011).
a) a) 0.8 b)
c) 1.5 B-C--C-N (upper band)
0.7
0.7 B-C--C-B MBN=2
B-C--C-N (lower band) 0.6
1 0.6 Zigzag graphene MBN=3
0.5
vgroup (106 m/s)
MBN=5
0.5
0.5
Egap (eV)
0.4 0.4 MBN=10
E(eV)
0 0.3
b) 0.3
0.2
-0.5 0.2
0.1
-1 0.1
0
-0.1 0
-1.5 -13 -12 -11 -10 -9 0 10 20 30 40 50 60
-1 0 1
k x ax /  kx (nm-1) M
CC
Fig. 1. a) A pure zigzag graphene ribbon. b) Zigzag Fig. 5. a) Group velocity in different structures. b) Bandgap
graphene/BN with B-C interface. c) Band structure of zigzag in B-C--C-N structures as a function of the graphene width.
graphene ribbon for MCC = 50 chain lines.
a) c
2
a) 1.5 b) 0.6
)1.5
1
1 0.5 BN graphene 0.5
E (eV)
0
0.5 b)
 (arb. unit)
0.4 0.9 0.5 -0.5

E (eV)
(1) 0.7 layer1 V =1.394 eV

0 0.3 (1) 0.5 0.4
layer2 -1 eff
B-C hybrid states
(2) (2') (2)
(3) -1.5 V =-0.75 eV
2 (arb. unit)
eff
0.3
2
N-C hybrid states

-0.5 0.2 0.1
0.3
-2
E (eV)
-0.1 -1 -0.9 -0.8 -0.7

0.2
-1 0.1 -0.3
V = -0.1 eV
eff
V = 0.1 eV
eff
V = 0.3 eV
eff
V = 0.5 eV
eff k a /
(3) xx
-0.5 0.1 40 45 50
-1.5 0 -0.7
-1 0 1 0 20 40 60 80 100
k a / atom position along y -0.9
-1 -0.75 -0.5 -1 -0.75 -0.5 -1 -0.75 -0.5 -1 -0.75 -0.5
0
0 25 50
x x
kxax/ k x ax /  kxax/ kxax/ atom position along y
Fig. 2. a) Band structure of B-C structure for MCC = MBN = 50
chain lines. b) Wave function of states (1), (2) and (3). Fig. 6. a) & b) Effective model with a potential Veff applied
only on edge atoms of ZGNR. c) Comparison of dispersion of
edge states in the effective model with actual dispersions.
114
Quantum Dot Lab: An Online Platform for
Quantum Dot Simulations
Prasad Sarangapani1,* , Daniel Mejia1 , James Charles1 , Hesameddin Ilatikhameneh1 , Tarek Ameen1 ,
Andrew Roche2 , James Fonseca1 , and Gerhard Klimeck1
1
Network for Computational Nanotechnology, Purdue University, West Lafayette, IN 47907
2
University of Louisiana, Lafayette, LA 70504
*
email: [email protected]
I NTRODUCTION the TB Hamiltonian to get a physically reasonable

Self-assembled quantum dots have improved the answer for the eigenvalues. With the eigenstates
performance of many optoelectronic devices such and eigenvalues obtained, the optical absorption
as infrared photodetectors and intermediate band spectrum is calculated using Fermi’s golden rule.
solar cells [1], [2]. They offer the ability to tune The tool provides users the ability to vary different
device parameters by carefully adjusting device physical parameters such as temperature, Fermi
dimensions. Designing an optimal quantum dot level and polarization angles which affect the final
for a specific purpose requires intensive simulation absorption spectrum. The users can view/download
beforehand to get an idea on effect of various results such as real space eigenfunctions, strain
materials and dimensions on optical properties such components, eigen energies, local bandstructure and
as position of eigenstates and absorption spectrum. absorption spectrum.
Quantum Dot Lab deployed on nanoHUB [3] pro- The tool runs within a Rappture interface [7]
vides such a simulation platform for the research and uses the NEMO5 [8] simulation tool in the
community. The tool has the ability to simulate background to perform serial and parallel de-
quantum dots ranging from simplistic particle in a vice calculations. Typical device dimensions that
box in effective mass approximation (EMA) till self- are simulated range from 10nm×10nm×10nm till
assembled quantum dots in 10-band sp3d5s* tight 50nm×50nm×50nm run on a maximum of 256
binding (TB) basis. cores.
T OOL D ESCRIPTION C ONCLUSION

Quantum Dot Lab provides users the ability to An online platform for simulating quantum dots
choose between simulating a simple particle in a has been shown. The tool has been deployed on
box problem in EMA, multilayered quantum dots nanoHUB and can simulate quantum dots with
comprising of a substrate, wetting layer, dot and varying degrees of complexity acting as a research
a capping layer topology in both EMA and 10 and learning tool for the scientific community.
band sp3d5s* TB basis with spin-orbit coupling.
Different dimensions and materials for each region ACKNOWLEDGEMENT
can be chosen as well. The tool solves Schrödinger’s The use of nanoHUB.org computational re-
equation using the Lanczos method [4] to com- sources operated by the Network for Computational
pute the eigenstates of the system. In multilayered Nanotechnology funded by the US National Sci-
device, in addition to Schrödinger’s equation, a ence Foundation under Grant Nos. EEC-0228390,
valence force field based strain model [5] is solved EEC-1227110, EEC-0228390, EEC-0634750, OCI-
using PETSc Newton solver [6]. The relaxed atom 0438246, OCI-0832623 and OCI-0721680 is grate-
positions are then calculated and incorporated into fully acknowledged.
ISBN 978-0-692-50554-0 115

Fig. 1: Typical quantum dot structure simulated with Fig. 4: Local bandstructure output showing the
the tool. effect of strain on conduction and valence band
edges.
Fig. 2: Visualization of real space electronic wave-

function in the quantum dot structure. Fig. 5: Absorption spectrum plot. θ and φ corre-
spond to the orientation of the polarization vector
with respect to the quantum dot structure.
R EFERENCES
[1] Liu, H. C., et al. ”Quantum dot infrared photodetectors.”
Applied Physics Letters 78.1 (2001): 79-81.
[2] M. Razeghi, Technology of Quantum Devices. Springer,
2010.
[3] https://nanohub.org
[4] Heath, M. ”Computing: An Introductory Survey.” (1998).
[5] Lee, Seungwon, et al. ”Effect of wetting layers on the strain
and electronic structure of InAs self-assembled quantum
dots.” Physical Review B 70.12 (2004): 125307.
[6] http://www.mcs.anl.gov/petsc/
[7] McLennan, M. ”The rappture toolkit.” (2004)
Fig. 3: Visualization of xz strain component in the [8] Steiger, Sebastian, et al. ”Nemo5: a parallel multiscale
quantum dot structure. nanoelectronics modeling tool.” Nanotechnology, IEEE
Transactions on 10.6 (2011): 1464-1474.
116
Engineering The Optical Transitions of
Self-Assembled Quantum Dots
Tarek A. Ameen∗† , Hesameddin Ilatikhameneh∗† , Daniel Valencia∗ , Rajib Rahman∗ , and Gerhard Klimeck∗
∗ Network for Computational Nanotechnology, Department of Electrical and Computer Engineering,
Purdue University, West Lafayette, IN 47907, USA. e-mail: [email protected]
† These authors contributed equally to this work.
I NTRODUCTION spring forces connecting the atoms together. The

In this paper, we report a fast effective mass relaxed atom positions are obtained by minimizing
model for accurately calculating the bound states the energy of the whole structure with respect to
and optical transitions of self-assembled quantum individual atom positions. The dimensions of the
dots. The model include the atomistic strain effects, simulated quantum dot systems are 60 nm x 60 nm x
namely, the strain deformation of the band edges, 60 nm and the quantum dot is dome shaped InAs in
and strain modification of the effective masses. The the middle of GaAs substrate. The strain simulation
explicit inclusion of strain effects in the picture contains around 10 Million atoms and an atomistic
has significantly improved the effective mass model grid is shown in Fig 1. Such large systems are
results. For strain calculations, we have found that computationally expensive to run strain simulations
atomistic strain depend solely on aspect ratio of the and require highly scalable computational codes.
quantum dot, and it has been calculated and reported The code that has been used for our simulations
here for a wide range of quantum dot aspect ratios. is the Nano Electronic MOdeling tool version 5
The deformation theory has been used to include NEMO5 [4].
the strain deformation of the band edges. Density
function theory has been used to study the effect of
R ESULTS
strain on the electron and hole effective masses.
T HEORETICAL M ODEL Atomistic strain simulations indicate that lattice

The quantum dot is modeled as a 3-dimensional strain in the quantum dot depends only on the
dome shaped finite potential well, the depth of the aspect ratio not on the individual dimensions. As
well is the band offset that can be calculated using shown in Fig. 2, the magnitude of hydrostatic strain
deformation potential theory [1], decreases with increasing the diameter to height
ratio (D/h), while the magnitude of biaxial strain
∆Ec = CBO − ac εH , increases with increasing (D/h). The electron, light
b hole, and heavy hole effective masses calculated
∆EvHH = V BO + av εH + εB , at the different strain components corresponding
2
to the different aspect ratios as indicated in Fig.
b 3 . To show the impact of including the strain in
∆EvLH = V BO + av εH − εB ,
2 the effective mass models, we have calculated the
The electron and hole effective masses under strain optical transition of a dome shaped quantum dot
are calculated using density function theory (DFT) with height of 5 nm and diameter of 20 nm and
using AtomistixToolKit [2]. compared against experimental measurements [5].
For the atomistic strain simulations, harmonic The error in calculating the optical transition has
Keating model has been used [3] . The harmonic been reduced from 25% to 3.7% as shown in Table
Keating model treats the interatomic forces as I.
ISBN 978-0-692-50554-0 117

Fig. 2. The atomistic hydrostatic and biaxial strain at different
diameter to height ratio (D/H). Strain depend on aspect ratio
(D/H) not on individual dimensions. The magnitude of hydro-
static strain decreases with increasing D/h while the magnitude
of biaxial strain increases with increasing D/h.
Fig. 1. The atomistic grid of one of the simulated InAs/GaAs
quantum dot systems. The quantum dot is a dome with base
diameter of 20 nm and height of 5 nm.
ACKNOWLEDGMENT
The use of nanoHUB.org computational resources operated by the
Network for Computational Nanotechnology funded by the US National
Science Foundation under grant EEC-1227110, EEC-0228390, EEC-
0634750, OCI-0438246, and OCI-0721680 is gratefully acknowledged.
This research is also part of the Blue Waters sustained-petascale com-
puting project, which is supported by the National Science Foundation
(award number ACI 1238993) and the state of Illinois. NEMO5 devel-
opments were critically supported by an NSF Peta-Apps award OCI-
0749140 and by Intel Corp. Any opinions, findings, and conclusions or
recommendations expressed in this material are those of the authors and
do not necessarily reflect the views of the National Science Foundation.
R EFERENCES
[1] G. L. L. Bir, G. E. Pikus, P. Shelnitz, and D. Louvish,
Symmetry and strain-induced effects in semiconductors,
vol. 624. Wiley New York, 1974.
[2] “Atomistixtoolkit version 2014.2, quantumwise a/s Fig. 3. The electron and light hole effective masses calculated
(www.quantumwise.com)..” using DFT. The heavy hole effective masses do not change
[3] P. Keating, “Effect of invariance requirements on the elastic significantly with strain, it is around 0.26 for [100] and 0.32
strain energy of crystals with application to the diamond for [110].
structure,” Physical Review, vol. 145, no. 2, p. 637, 1966.
[4] S. Steiger, M. Povolotskyi, H.-H. Park, T. Kubis, and
G. Klimeck, “Nemo5: a parallel multiscale nanoelectronics
modeling tool,” Nanotechnology, IEEE Transactions on,
Without Strain Including Strain Experimental
vol. 10, no. 6, pp. 1464–1474, 2011.
Optical transition 0.7169 eV 1.012 eV 0.976 - 0.93 eV
[5] J. Tatebayashi, M. Nishioka, and Y. Arakawa, “Over 1.5
Error 25% 3.7 %
µm light emission from inas quantum dots embedded in in-
gaas strain-reducing layer grown by metalorganic chemical TABLE I
vapor deposition,” Applied Physics Letters, vol. 78, no. 22, C ALCULATED OPTICAL TRANSITION VS EXPERIMENTAL
pp. 3469–3471, 2001.
MEASUREMENT.
118
Theoretical study of excitonic energy transfer in
carbon-nanotube-based photovoltaics
A. H. Davoody, A. J. Gabourie, and I. Knezevic
1415 Engineering Dr, Madison, Wisconsin 53706, USA
Email: [email protected], [email protected]
Carbon nanotubes (CNTs) are quasi-one- in Fermi’s golden rule. In the molecular aggregates
dimensional materials with a very unique set of where the sizes of molecules are small relative to the
optical, electronic, and mechanical properties. Their separation between them, the Coulomb interaction
tunable optical band-gap and high optical density is effectively equal to the interaction between the
have drawn the attention of researchers in the optics transition dipoles of the molecules; this is known
community. While a majority of research efforts as the transition-dipole approximation (TDA). In the
have focused on understanding the photophysics original work by Förster [2], it was shown that in
of individual CNTs, understanding the coupling this limit the EET rate is dependent on the overlap
of CNTs to the surrounding environment and the of the emission and absorption spectra of molecules,
nearby CNTs have shown to be of vital importance both of which depend on the bright excited states.
in understanding the photoexcitation dynamics in However, our study shows the breakdown of this
these systems [1]. approximation from different aspects.
Excited states of the electronic system in CNTs First, the EET rate has an exponential decay with
are the bound states between an electron in the increasing spacing between the donor and acceptor
conduction band and a hole in the valence band, tubes, as opposed to the R−6 dependence calculated
known as excitons. The excitonic energy transfer through the TDA (Figure 1a). Second, the EET rate
(EET) from an initial CNT (donor) to a final CNT is significant when the donor and acceptor tubes
(acceptor) is the focus of this study. We show are perpendicular to each other (Figure 1b). This
the unconventional behavior of the transfer rate is in sharp contrast with the result of the TDA
with respect to chirality of the CNTs and their which predicts zero transfer rate at perpendicular
orientation. We also show, for the first time, that the orientation. Third, we show that, in addition to
excitons that cannot be excited by light (dark exci- bright excitons, one type of dark exciton (singlet
tons) contribute significantly to the energy transfer and symmetric) contributes to the EET due to the
process. large size of the CNTs (Figures 1b and 1c). This
The excitonic states are best described by the is a nonintuitive result which underlines the non-
Bethe-Salpeter equation (BSE). Here, we use the radiative nature of the EET.
tight-binding single-particle wave functions as a Furthermore, we show that the EET between sim-
basis to solve the BSE and calculate the excitonic ilar chirality CNTs is about 3 orders of magnitude
eigenfunction and energies. The exciton states are larger than the transfer rate between donor and
divided into singlets and triplets according to spin, acceptor CNTs with dissimilar chiralities (Figure
and can have symmetric or antisymmetric wave 1a). This is due to the momentum conservation
function over the relative coordinate. Among the in the parallel orientation. The difference in the
four types of excitons, only the spin singlet and transfer rate rapidly diminishes with rotating the
antisymmetric excitons are bright and the rest are CNTs with respect to each other.
dark [3]. It has been shown that the environmental screen-
The EET rate is calculated by treating the ing of the Coulomb interaction between electrons
Coulomb coupling between CNTs as a perturbation and holes changes the exciton energy dispersion
ISBN 978-0-692-50554-0 119

×10 12
(7,5) to (7,5)
10 12
6
Transfer rate [s-1 ] Symmetric singlet exciton
(7,5) to (8,6) 4
10 10 (8,6) to (7,5) 1 2 3 4 5
(7,5) to (7,5)
11
10 8
×10
Symmetric (7,5) to (8,7)

6
Transfer rate [s-1 ]

10 6 Symmetric singlet exciton
4
1 2 3 4 5
2 4 6 8 10
11
Center-to-center distance [nm] ×10
(a) 2.5 (8,7) to (8,7)
2 Antisymmetric singlet exciton

×10 11
1.5
3.5 1 2 3 4 5
10
3 15
×10
Antisymmetric singlet exciton
2.5 10
Cκ = 1 (7,5) to (8,7)
5
2 center to center distance = 1.2 nm
1 2 3 4 5
1.5 (7,5) to (7,5)
(7,5) to (8,7) Dielectric constant
1 (8,7) to (7,5)
(8,7) to (8,7)
0.5 Fig. 2. The EET rate of symmetric and antisymmetric singlet
excited states between donor and acceptor CNTs of various
0 0.5 1 1.5 chirality as a function of environment dielectric constant.
θ [rad]
(b)
curve, as well as the exciton wavefunction. Here, we
×10 12
show that the change is not uniform for CNTs with
10 different chirality; As a result, the EET rate tends
to decrease with increasing environment dielectric
8
constant. This dependence is more pronounced for
Cκ = 1
the bright exciton transfer process between CNTs
6 center to center distance = 1.2 nm
of dissimilar chirality (Figure 2).
4
This work was primarily funded by the U.S.
(7,5) to (7,5) Department of Energy, DOE award No. DE-
(7,5) to (8,7)
2 (8,7) to (7,5) SC0008712. Some preliminary efforts were funded
(8,7) to (8,7)
by UW MRSEC, IRG2 (NSF award No. 1121288).
0 0.5 1 1.5
AJG was funded by the REU supplement to NSF
θ [rad] award No. 1201311.
(c)
R EFERENCES
Fig. 1. (a) The EET rate of antisymmetric singlet excited
states between parallel donor and acceptor CNTs of various [1] M. S. Arnold, J. L. Blackburn, J. J. Crochet, S. K. Doorn,
chirality as a function of tube separation. (b) The EET rate of J. G. Duque, A. Mohite, and H. Telg, Phys. Chem. Phys.
antisymmetric singlet excited states donor and acceptor CNTs 15, 14896 (2013).
of various chirality as a function of the relative orientation. [2] Th. Förster, Annalen der Physik 437, 55 (1948).
(c) The EET rate of symmetric singlet excited states between [3] J. Jiang, R. Saito, Ge. G. Samsonidze, A. Jorio, S. G. Chou,
donor and acceptor CNTs of various chirality as a function the G. Dresselhaus, and M. S. Dresselhaus, Phys. Rev. B 75,
relative orientation. 035407 (2007).
120
Effect of confinement in III-V nanowire field
effect transistors
A. Price and A. Martinez
College of Engineering, Swansea University
ABSTRACT RESULTS
Using quantum transport simulations the The effect of changing the cross-section on the
effect of confinement in III-V gate-all-around relative position of the valleys can be observed by
nanowire field effect transistors (NWFETs) of comparing Fig. 2. and Fig. 3, which show the
different dimensions has been investigated. The energy-resolved current spectra for a 2.2 x 2.2 nm2
Non-Equilibrium Green's Function (NEGF) and 4.2 x 4.2 nm2 GaAs NWFET respectively. For
formalism in the effective mass approximation the 2.2 x 2.2 nm2 NWFET the low mass Γ-valley is
(EMA) has been used, and both ballistic and elevated in energy such that it is higher than the
dissipative transport have been considered. The heavier L and X-valleys. This results in greatly
effective masses have been extracted from tight- reduced current in the Γ-valley for the 2.2 x 2.2
binding (TB) simulations. nm2 device in comparison to the 4.2 x 4.2 nm2
INTRODUCTION device. The Id-Vg characteristics for each valley
for both cross-sections can be seen in Fig. 4. and 5.
NWFETS are one of the promising candidates The transverse potential energy profile and the
being considered towards the end of the electrostatic potential energy is shown in Fig. 6 .
International Technology Roadmap for We have also calculated the percentage tunnelling
Semiconductors (ITRS). [1] III-V nanowires are
and current reduction. For a 2.2 x 2.2 nm2, 6 nm
also being considered because of their high channel GaAs scattering caused a 41% increase in
mobility compared to Si. In this work, GaAs [2] tunnelling and a 72% reduction in the on-current at
and InGaAs NWFETs of cross-section 2.2 x 2.2
low drain.
nm2 and 4.2 x 4.2 nm2, and channel length 6 nm,
10 nm and 20 nm are considered. An example of CONCLUSION
the structure of the devices is given in Fig. 1. In this work, III-V NWFETs of different
MODEL dimensions have been investigated. It was found
that the strong quantisation caused the relative
The NEGF formalism is a widely used tool to position of the valleys to vary with the cross-
study dissipative quantum transport (i.e. inelastic section, with the low mass Γ-valley becoming
electron-phonon transport) in nanotransistors. The
elevated in energy for the 2.2 x 2.2 nm2 cross-
self-consistent Born approximation (SCBA) with
section devices. This resulted in very low current
local self-energies has been deployed. The EMA
in the 2.2 x 2.2 nm2 NWFETs.
has been used, decomposing the 3D problem and
allowing phonon scattering to be implemented REFERENCES
efficiently. All scattering parameters have been [1] http://www.itrs.net.
taken from [3], acoustic, optical and polar optical [2] A. Price and A. Martinez, “Investigation on phonon
phonon scattering have been considered. The scattering in a GaAs nanowire field effect transistor using
the Non-Equilibrium Green's Function formalism,” J.
effective masses have been extracted form TB Appl. Phys. (2015 awaiting publication).
simulations. [4] The EM Hamiltonian is given by [3] M. Lundstrom, “Fundamentals of carrier transport,”
Eq. (1). Cambridge University Press (2000).
[4] Y. M. Niquet, A. Lherbier et al. Phys. Rev. B 73, 165319
  2 ⎛ 1 ⎞  (1)
Hν (r ) = − ∇ i ⎜ ⎟ ∇ j + V (r ) (2006).
2 ⎝ mν ⎠ij
ISBN 978-0-692-50554-0 121

7
x 10
1
8 valley ball
10 valley scatt
Logarithmic Drain Current (A)

Lvalley ball
0.8
Linear Drain Current (A)

10
10 Lvalley scatt
Xvalley ball
12 Xvalley scatt
10 0.6
14
10
0.4
16
10
18 0.2
10
20
10 0
0.4 0.5 0.6 0.7 0.8 0.9 1
Gate Voltage (V)
Fig. 1. Example of device structure for a 2.2 x 2.2 nm2 cross- Fig. 4. Id-Vg characteristics for each valley for a GaAs core,
section NWFET. 2.2 x 2.2 nm2 cross-section, 20 nm channel length NWFET at
low drain bias, VD = 1 mV, with and without scattering.
7
x 10
2
valley ball
valley scatt
Logarithmic Drain Current (A)

8
10 Lvalley ball
Linear Drain Current (A)

Lvalley scatt
10
Xvalley ball
10 Xvalley scatt
12
10 1
14
10
16
10
18
10 0
0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Gate Voltage (V)
Fig. 2. Energy-resolved current spectrum for a 2.2 x 2.2 nm2 Fig. 5. Id-Vg characteristics for each valley for a GaAs core,
cross-section, 20 nm channel length, GaAs NWFET with 4.2 x 4.2 nm2 cross-section, 20 nm channel length NWFET at
scattering, VG = 0.8 V and VD = 1 mV. low drain bias, VD = 1 mV, with and without scattering.
Fig. 3. Energy-resolved current spectrum for a 4.2 x 4.2 nm2 Fig. 6. Transverse potential energy profile with 0.2 nm oxide
cross-section, 20 nm channel length, GaAs NWFET with penetration, and electrostatic potential energy up to the
scattering, VG = 0.8 V and VD = 1 mV boundary of the metal at low gate, VG = 0.3 V
122
Impact of Discrete Dopants in ultrascale
FinFETs and the Effect of XC on Dopant
Clustering
A. Martinez and R. Valin and JR Barker*
College of Engineering, Swansea University, United Kingdom

*
School of Engineering, University of Glasgow, United Kingdom
differ in the realization of disorder, have been

ABSTRACT
simulated [2]. The electron concentration of one of
Non-equilibrium Green’s Function simulations the devices is shown in Fig. 1. The Non-
of an ultra-scale FinFET in the ballistic regime equilibrium Green’s Function formalism combined
have been carried at low/high drain bias. We with a coupled mode space approach has been
have calculated variability due to random deployed. The Hamiltonian is written in the
dopants located in the source/drain regions of effective mass approximation. The ID-VG
the transistor. The channel length of the scaled characteristics for the ensemble of devices at
transistor is under 10 nm and therefore VD=0.05/0.6 V are shown in Fig. 2. As a
substantial tunneling is expected. We have comparison, the characteristics of the devices with
calculated the tunneling as a function of the continuous doping are also shown in the figure. At
gate bias in two dopant configurations with the low gate, the fluctuation in the channel length is
lowest and highest drain current. We have also due to the location of the dopants, which produce
computed the threshold voltage, sub-threshold fluctuation in the off current and in the sub-
slope and off current variability. We have threshold slope. At high gate bias the coherent
studied the effect of the exchange correlation on scattering of dopants decreases the current with
the simulation of a cluster of dopants in the respect to the corresponding current of the
channel of the transistor. continuous doping device. Fig. 3 shows the ID-VG
of the device with maximum and minimum current
3D Field Effect Transistors (FETs), such as of the whole ensemble (VD=50mV). The
FinFETs, are already being fabricated. The corresponding tunneling currents are also shown.
channel length of transistors currently in Fig 4 shows the percentages of tunneling currents
production is approximately under 20 nm. At the showed in Fig. 3. We have also studied the effect
end of the International Technology Roadmap for of exchange-correlation (XC)[3] on a device with
Semiconductors (ITRS) [1] a further reduction in a cluster of dopants in the middle of the channel.
channel length into the sub-10 nm regime is The current spectrum is shown in Fig. 5. As the
envisaged. Source to drain tunneling is expected to electron concentration in the channel increases
degrade the transistor on/off ratio at these with the increasing of the gate bias, the impact of
dimensions. Variability induced by the discrete the XC also increases (with the gate bias). This can
nature of the dopants is also expected to be be seen in Fig. 6, which shows the ID-VG of the
detrimental. Semi-classical descriptions of the device with/without XC.
electron transport lack tunneling and coherence,
and from a physical point of view are unsuitable REFERENCES
for describing transport at sub-10 nanometer scale. [1] http://www.itrs.net.
In this work, a Si FinFet of 10.6/4.2 nm high/width [2] R. Valin. A. Martinez and J. R. Barker, J. Appl. Phys.
(2015 awaiting publication).
and 6nm channel length has been considered. The [3] L. Hedin et al J. Phys. C: Solid St. Phys. 4, 2064-83
discrete dopants are distributed in 6nm region in (1971).
the source and drain extension. 35 devices, which
ISBN 978-0-692-50554-0 123

80
70 Min Ioff
Tunneling Current (%)

Max Ioff
60
50
40
30
20
10
0
0 0.2 0.4 0.6 0.8
Gate Potential (V)
Fig. 1: Electron density of a device with discrete random
dopants. Fig. 4 Percentage of tunneling current as a function of the gate
bias.
0.3
0.2
Energy (eV)
0.1
−0.1
−0.2
−0.3
−0.4
5 10 15 20 25
Distance (nm)
Fig. 2: ID-VG characteristics of 35 discrete dopant devices at
Fig. 5. Energy-resolved current spectrum for the device with
high and low gate bias.
discrete random dopants and the cluster of dopants in the
middle of the channel (VD = 0.3 V).
−5
x 10
0
Current (µA)
10 8
XC
Current (A)
Min Ioff 6
Max Ioff
Tun Min Ioff
no XC
−2
Tun Max Ioff 4
10
0.2 0.4 0.6

0
VG (V) 0 0.2 0.4 0.6 0.8
Fig. 3. ID-VG characteristics of the devices with maximum Gate Voltage (V)
and minimum off-current. The corresponding tunneling
currents are shown for comparison.
Fig. 6. ID-VG characteristics of the device with the cluster of
dopants in the channel (VD = 0.3 V).
124
Analysis of Vacancy defects in Hybrid
Graphene-Boron Nitride Armchair Nanoribbon
based n-MOSFET at Ballistic Limit
Anuja Chanana† , Amretashis Sengupta∗ , and Santanu Mahapatra†
† Department of Electronics Systems Engineering, Indian Institute of Science, Bangalore-560012, India
∗ School of VLSI Technology, Indian Institute of Engineering Science and Technology,
Shibpur, Howrah-711103, India
A BSTRACT separately in Fig1(b)-(d).The bandgap and effective

Here, we report the performance of hybrid mass are calculated using DFT-LDA simulations
Graphene-Boron Nitride armchair nanoribbon on Quantumwise ATK [4]. Monkhorst-Pack k-grid
(a-GNR-BN) n-MOSFET at its ballistic limit mesh of 1x1x15 is used for the simulations.Further
using Non Equilibrium Green’s Function (NEGF) they are used in our in-house ballistic NEGF
methodology with different vacancy defects. The simulator [2] to evaluate the device characteristics.
band gap and effective mass show a decrease which The channel length is 10 nm(Fig 1(e)),hence the
affects the device parameters showing an increase transport is assumed to be purely ballistic.
in ON-current (ION ) and Sub Threshold Slope (SS).
D ISCUSSION
I NTRODUCTION The band structure(Fig2(a)) for ring defect
Embedding Graphene with Boron Nitride (nearly shows a larger decease as compared to mono
same lattice constant with graphene) is one of and di vacancy defects confirmed by Density of
the effective ways of opening a band gap in States in Fig2(b). The band gap and effective
gapless graphene and such atomic layers are mass values show a decrease with an increase
synthesized experimentally [1].Ballistic device in number of defects shown in Fig 2(c)-(d) and
performance of a-GNR-BN based n-MOSFET the decrement is largest for ring vacancy defect.
and effect of Stone Wales(SW) defects using For number of defects 3, ION from pure to
NEGF formalism is earlier investigated [2].Here, defected device varies in the range of 276-293
we analyze the performance of a-GNR-BN with µA (Fig 3(a)-(b)). For a ring defect density of
vacancy defects monovacancy(single B atom 6.67%-20%(1, 2 and 3 ring defects out of 15 at the
removal), divacancy(one B and one N atom interface) ION ranges from 277-293 A(Fig 4(a)-(b)).
removal) and ring vacancy(removal of 2 B, 2 N
and 2 C atoms) formed at the interface of graphene C ONCLUSION
and BN nanoribbon.The defects are distributed From the above analysis, we see that the band
randomly at the Graphene-Boron Nitride interface gap, effective mass and device characteristics illus-
and are likely to be formed here [3]. trate a minimal change when the defect is mono or
divacancy, but the change is more considerable for
S IMULATION T ECHNIQUE ring vacancy.
We examine a super cell formed by repeating the R EFERENCES
3p configuration of hybrid a-GNR-BN consisting of [1] Lijie Ci et.al, , Nature Materials 9, 430 (2010).
30GNR atoms and 12BN atoms(6 on either side). [2] A. Chanana et.al, , J. Appl. Phys 115, 034501 (2014).
[3] J. M. Pruneda , Phys. Rev. B 85, 045422 (2012).
The supercell of hybrid a-GNR-BN with 3 ring [4] ATK 13.8, http://www.quantumwise.com/
defects is shown in Fig1(a) with rest of the defects
ISBN 978-0-692-50554-0 125

Fig. 1. (a) Supercell used in the present study. It is formed
Fig. 2. (a)-(b) Comparison of band structures and density of
by repeating 8 times the unit cell 42 a-GNRBN. (b),(c) and (d)
states of pure and vacancy affected supercell. (c) - (d) Band gap
showing mono vacancy, di vacancy and ring defects all formed
and Effective mass comparison with the increase in number of
at the interface. Hydrogen passivation is done to minimize
defects
contribution from edge states (e) Transistor schematic used for
the device calculations.
TABLE I
BANDGAP AND E FFECTIVE M ASS OF PURE AND VACANCY
AFFECTED SUPERCELL EVALUATED USING DFT (N UMBER
OF DEFECT IS 3 FOR EACH TYPE OF VACANCY )
Type of Vacancy defect Band Gap (eV) Effective Mass

(m∗ /m0 )
Pure 0.369 0.0429
Mono vacancy 0.354 0.0413
Di vacancy 0.339 0.0407
Ring vacancy 0.157 0.022
Fig. 3. (a) ID -VD and (b) ID -VG characteristics of pure and
defected supercell. Defect density for each type of vacancy is
3.
TABLE II
N -MOSFET DEVICE PARAMETERS OF PURE AND VACANCY
AFFECTED SUPERCELL CALCULATED USING NEGF
FORMALISM (N O . OF DEFECT IS 3 FOR EACH TYPE OF
VACANCY )
Type of Vacancy defect ION (µA) Subthreshold Slope

(mV/decade)
Pure 98.14 62.096
Mono vacancy 98.38 62.138
Di vacancy 98.5 62.189
Ring vacancy 101.6 62.622
Fig. 4. (a) ID -VD and (b) ID -VG characteristics ring defects

increasing from 1 to 3.
126
Simulation of Current-Voltage Curves for CdS
Cylinders Embedded in P3HT
C. Huang, S. Poduri, and M.A. Stroscio
Dept. of Electrical and Computer Engineering, University of Illinois at Chicago, Chicago, IL 60607, USA
could be presented by an activation energy (EA) in

INTRODUCTION
the I-V equation [4].
In this paper, the current-voltage (I-V) curve of
the heterojunction formed by cadmium sulphide (1)
(CdS) cylinder arrays embedded in the conductive
polymer poly (3-hexylthiophene-2, 5-diyl) (P3HT) (2)
is modeled in light of its possible use in is the saturation current and A is area of the
photovoltaic applications. CdS cylinders were heterojunction which is 0.25 cm2. and are
grown by DC electrodeposition on a nanoporous the initial minority carrier concentrations which
template which is made by anodized aluminum are calculated by minority carrier equations.
layer on Indium Tin Oxide (ITO) coated glass
substrate, then P3HT solution was put on the CdS (3)
nanowires template to form the CdS/P3HT
(4)
heterojunction [1]. The Ag contact on top was
and,
formed by silver paint, and Ag/P3HT/CdS/ITO
heterojunction was formed as shown in Figure 1. (5)
THEORY (6)
Some polymers are conductive since carriers
can transport in overlap bonds. The overlapping The diffusion constants - and - were
carbon p orbitals in the successive units in a calculated by Einstein equation.
conjugated polymer produce two split energy (7)
levels. One is π bonding orbital and another is π*
anti-bonding orbital. A series of π bonds form a π where is mobility of minority carriers.
energy band which is called the highest occupied The diffusion length of electrons in P3HT (Ln) is
molecular orbital (HOMO), and a series of π* 10-6 cm [5].
bonds form a π* energy band which is called the The diffusion length of holes in CdS is:
lowest unoccupied molecular orbital (LUMO) in (8)
the energy band of the polymer. These two energy where is lifetime of holes in CdS.
bands are similar to the valence band and
conduction band of inorganic semiconductors DISCUSSION
[2][3]. The band diagrams of CdS/P3HT are The I-V curve was computed iteratively until
shown in Figure 2. CdS is n-type semiconductor the value of EA provided for a simulation closest to
and P3HT is a p-type material. The equation of measurement; this yielded the value of EA to be
heterogeneous pn junction is used to build the 0.3*VA1.88 as shown in figure 4. If the causes of the
simulation model. The I-V curve was measured defects could be figured out and minimized, it is
from -2 eV to +2 eV and compared with helpful for CdS/P3HT to be used in forming
simulation as shown in figure 3. The deviation is photovoltaic and other devices.
caused by lots of reasons such as defects from
ACKNOWLEDGMENT
growth of CdS, polymer properties…etc. The one
main cause is that defects cause some carriers Work supported, in part, under AFOSR Grant
recombine at the interface. The recombination FA9550-11-1-0271.
ISBN 978-0-692-50554-0 127

REFERENCES
[1] Shripriya Poduri, Mitra Dutta, and M. Stroscio,
“Characterization of CdS Nanowires Self-Assembled in a
Nanoporous Alumina Template,” J Electron Mater. vol
43, pp. 3979-3983, (2014).
[2] Coropceanu, V., Cornil, J., da Silva, D.A., Olivier, Y.,
Silbey, R., and Bredas, J.L.: Charge transport in organic
semiconductors. Chem. Rev., 107: 926-952, 2007.
[3] Heeger, A.J., Kivelson, S., Schrieffer, J.R., and Su, W.P.:
Solitons in Conducting Polymers. Rev. Mod Phys., 60:
781-850, 1988.
[4] Wilhelm, Helena, Hans-Werner Schock, and Roland
Scheer. "Interface recombination in heterojunction solar
cells: Influence of buffer layer thickness."Journal of
Applied Physics vol. 109 pp. 084514 (2011).
[5] Ayzner, Alexander L., et al. "Reappraising the need for
bulk heterojunctions in polymer− fullerene photovoltaics:
the role of carrier transport in all-solution-processed
P3HT/PCBM bilayer solar cells." The Journal of Physical
Chemistry C 113.46 (2009): 20050-20060.
Fig.2. (a) Lineup band structure of CdS and P3HT before

contact. (b) The band diagram of CdS/P3HT after
contacted. The vacuum level of P3HT is lowered by
0.2534eV.
Fig.1. Brief schematic of CdS/P3HT circuit.
Fig.3. Comparison of measurement (blue) and simulation (red)

CdS/P3HT I-V curves.
Fig.4. Comparison of measurement (blue) and simulation (red)

CdS/P3HT I-V curves after EA was added in simulation.
128
Nanosized-Metal-Grain-Induced Characteristic
Fluctuation in Gate-All-Around Si Nanowire
Metal-Oxide-Semiconductor Devices
Chun-Ning Lai1,2, Chien-Yang Chen1,2, and Yiming Li1,2,3,*
1
Parallel and Scientific Computing Laboratory; 2Institute of Communications Engineering; 3Department of
Electrical and Computer Engineering, National Chiao Tung University, 1001 Ta-Hsueh Road, Hsinchu
300, Taiwan; *e-mail: [email protected] (Y. Li)
INTRODUCTION is equal to 4 nm because the electron ground state

Gate-all-around (GAA) nanowire structure is energies are dispersed due to different metal
an important role for MOS devices in sub-16-nm pattern. Figs. 3(a)-(b) show the averaged value and
technologies. Grain orientation of metals on small the standard deviation of the electron ground state
gate areas is uncontrollable during fabrication energy. The averaged ground state energies with 6-
processes. Such random orientation will result in nm channel width are higher than other channels
different workfunction for each grain locally. and keep relatively small standard deviation. Figs.
Workfunction fluctuation (WKF) is thus severe 4(a)-(c) show the fluctuated electron’s ground state
and has received significant attention recently [1- wave functions (left plot), the potentials (middle
3]. Simulation studies on WKF of FinFET and one), and the electron densities (right plot) for the
nanowire MOS devices were reported by using device with respect to different grain sizes and
different quantum-mechanically corrected models. positions. The random grains induce different local
It will be an interesting study for us to investigate potentials on the surface of channel. Then, the
WKF of GAA Si nanowire MOS devices by electron’s wave functions are affected to different
solving full quantum mechanical model. extents by local confinement. Such WKF-induced
In this work, we numerically solve a set of 2D random confinement levels twist the distributions
Schrödinger-Poisson (SP) equations [4-5], where of wave functions and electron density. When the
the random grains are statistically incorporated gate voltage increases, the electron density
into the metal gate area of GAA nanowire MOS increases and the deviation of electron density due
devices. The impact of localized WKF on channel to random nanosized metal grain is suppressed
quantization is examined with respect to different from more than 100% variation to 10%.
grain size. The random orientations of metal grains CONCLUSIONS
result in different workfunction locally and affect In summary, the impact of WKF on GAA Si
the boundary condition of the 2D SP equations. nanowire MOS device was studied by solving 2D
STATISTICAL SIMULATION SP equations statistically. Non-uniform potentials
To study WKF on the GAA Si nanowire MOS at channel surface affect the distributions of wave
devices, 2D SP simulation is performed, as shown functions and electron densities.
in Fig. 1(a), statistically. We directly partition the ACKNOWLEDGMENT
This work was supported in part by the Ministry of Science and
region of the devices’ metal gate into many Technology, Taiwan, under Contract No. MOST 103-2221-E-009-180.
subregions according to experimentally observed
REFERENCES
grain sizes: 1, 2, and 4 nm. Then, we randomly [1] K. Nayak, S. Agarwal, M. Bajaj, P. J. Oldiges, K. V. R. M.
generate WKs for each subregion according to the Murali, and V. R. Rao, Metal-Gate Granularity-Induced
material properties and map them into the devices’ Threshold Voltage Variability and Mismatch in Si Gate-All-
Around Nanowire n-MOSFETs, IEEE Trans. Electron Devices,
gate for the 2D SP simulation. The settings are vol. 61, no. 11, pp. 3892–3895 (2014).
listed in Fig. 1(b) and the 2D SP equations and [2] Y. Li, H.-W. Cheng, and C.-H. Hwang, Threshold Voltage
Fluctuation in 16-nm-Gate FinFETs Induced by Random Work
solution steps are shown in Fig. 1(c). We generate Function of Nanosized Metal Grain, J. Nanosci. Nanotechnol.,
more than 200 statistically random devices to vol. 12, no. 6, pp. 4485-4488 (2012).
study random WK-induced fluctuation. [3] Y. Li, C.-Y. Chen, and Y.-Y. Chen, Random-Work-Function-
Induced Characteristic Fluctuation in 16-nm-Gate Bulk and SOI
RESULTS AND DISCUSSION FinFETs, Int. J. Nanotechnol., vol. 11, no. 12, pp. 1029-1038
Fig. 2 shows the distribution of electron ground (2014).
[4] A. Trellakis, A. T. Galick, A. Pcaelli, and U. Ravaioli, Iteration
state energy for the GAA nanowire MOS device scheme for the solution of the two-dimensional Schrödinger-
under thermal equilibrium with respect to different Poisson equations in quantum structures, J. Appl. Phys., vol. 81,
metal grain sizes. The low-frequency peaks no. 12, pp. 7880-7884 (1997).
[5] A. Trellakis and U. Ravaioli, Computational issues in the
increase, as shown in Fig. 2(a), when the grain size simulation of semiconductor quantum wires, Comput. Methods
Appl. Mech. Engrg., vol. 181, pp. 437-449 (2000).
ISBN 978-0-692-50554-0 129

Workfunction
NA 1e16 (1/cm-3)
fluctuation
Metal Gate EOT 1 (nm)
M e t a l G r a i n S iz e = 4 n m
40% <111> C h a n n e l W id th = 1 0 n m
Insulator Orientation
WK=4.4 eV Width 6 / 10 / 14 (nm)
-3 )
5e+18
S ta te
Source 0 .0 2 0
y (c m
Drain
Grain 1 / 2 / 4 (nm)
E le c t r v e F u n c t io n
4e+18
60% <200> 0 .0 1 5
ound
n s it
Silicon Orientation 0 .0 1 0 10
3e+18
10
on De
2e+18
on Gr
WK = 4.6 eV 8 8
0 .0 0 5 1e+18 6
6
Wa
Generate random
)
E le c t r
nm
0 .0 0 0 0 4
4
)
grain distribution.
y (
nm
2 2
40
4 2 4 2
y (
6 6
35 Mean = 14.4 x (n x (n 8 0
m 8 0 m ) 10
Corresponding )
Frequency (number)
30 10
2D structure 25
20
Metal Gate 15 M e ta l G r a in S iz e = 2 n m
10 C h a n n e l W id th = 1 0 n m
Insulator 5
S ta te
6 8 10 12 14 16 18 20 22 0 .0 2 0
Silicon 5e+18
-3 )
Number of TiN <200> Grain
E le c t r v e F u n c t i o n
y (c m
Setup problem 0 .0 1 5 4e+18
ound
Width boundary 0 .0 1 0 10 3e+18 10
n s it
on Gr
(0, 0) 8 2e+18 8
on De
0 .0 0 5
6 1e+18
Wa
Y 6
EOT 0 .0 0 0 0
4
)
4
m)
E le c t r
nm
2 2
n
4 2
y(
4 2
y (
6 6
…
X x (n
m) 8 0 x (n
m )
8 0
10 10
M e t a l G r a in S i z e = 1 n m
C h a n n e l W id th = 1 0 n m
-3 )
S ta te
0 .0 3 0 5e+18
y (c m
E l e c t r e F u n c t io n
0 .0 2 5 4e+18
ound
0 .0 2 0
D e n s it
10 3e+18 10
0 .0 1 5
on Gr
0 .0 1 0 8 2e+18 8
W av
c tro n
0 .0 0 5 6 1e+18 6
m)
m)
0 .0 0 0 4 0 4
n
y (
E le
2
y (
2 2
4 4 2
6 6
x (n 8 0 x (n 8 0
m) 10 m ) 10
8 8 8
Metal Grain Size = 4 nm Metal Grain Size = 2 nm Metal Grain Size = 1 nm
Channel Width = 10 nm Channel Width = 10nm Channel Width = 10 nm
6 6 6
Frequency
Frequency
Frequency
4 4 4
2 2 2
0 0 0
45 50 55 60 65 70 45 50 55 60 65 70 45 50 55 60 65 70
Ground State Energy (meV) Ground State Energy (meV) Ground State Energy (meV)
1e+19
(meV)
Metal Grain Size = 2nm

per Unit Channel Area (cm )
-3
Channel Width = 10nm

6 1.59E17
Total Electron Density
100
Energy (meV)
83.74 83.35
State Energy
4.17
83.14
tion of
5 4.77 4.01
ctron
80
3.05
54.08 54.80 4
Standard Devia
3.67 2.91
Averaged Ele
60 56.42
49.92
51.50 52.37
3 2.09
2.79 2.31
Ground State
40
Electron Ground
2 1e+18
20 9.87E17
)
6 1 6
)
m
m
(n
(n
0 10 0 10
th
h
0.00 0.25 0.50 0.75 1.00

t
id
id
4 14 Meta4 14
W
W
2 VG (V)
Meta l Gra 2
l
el
1 in S
ne
l Gra ize (n1

nn
in Siz
an
e (nm m)
a
)
Ch
Ch
130
Conductance fluctuations in graphene
Bobo Liu, Richard Akis, and David K. Ferry
School of Electrical, Computer, and Energy Engineering, Arizona State University, Tempe, AZ 85287
email: [email protected]
to any symmetry breaking process (such as that

ABSTRACT
due to the breaking of time reversal symmetry or
Conductance fluctuations in graphene which the lifting of spin or valley degeneracy), since such
arise from a long-range disorder potential induced mechanisms are not included in our calculations.
by random impurities are investigated with an Indeed, spin degeneracy should be lifted also when
atomic tight-binding lattice. The screened the magnetic field is applied in the plane of the
impurities lead to a slow variation of the graphene, but no such evidence of such symmetry
background potential and this varies the overall breaking was found in the study of ref. 2. In
potential landscape as the Fermi energy or an particular, an in-plane magnetic field shows no
applied magnetic field is varied. We find effect of the magnetic field on the amplitude of the
qualitative agreement with recent experiments, and fluctuations. So, there must be another
it appears that the reduction of the fluctuations in a explanation for the magnetic-field induced
magnetic field arises from a field induced reduction in the amplitude of the conductance
smoothing of the conductance landscape. fluctuations. Reflection from impurities is reduced
INTRODUCTION as edge states form in the magnetic field. The
Conductance fluctuations of a semiconductor at result is to suppress the amplitude of the
low temperature have been studied for many years fluctuations in the conductance. A similar
reduction in the amplitude of the fluctuations with
[1]. In such systems, the conductance will exhibit
such fluctuations as the Fermi energy, or an density has been seen for both gate voltage sweeps
applied magnetic field, is varied. It is presumed and magnetic field sweeps.
that this arises from the variations in the local CONCLUSION
potential landscape and changes in the electron The idea conductance fluctuations in disordered
wave interference as these variations are material is described as both ergodic and universal
introduced. Recently, these fluctuations have been in nature is not applicable. We suggest that a new
observed in monolayer and bilayer sheets of understanding of universality should be adopted,
graphene, and it was recently observed that these one based upon the Anderson theory, and which
fluctuations are small in a magnetic field and at we call Anderson universality. We will
higher densities [2]. demonstrate how this brings the graphene data into
We use an atomic-basis, tight-binding model to line with similar observations in GaAs and other
study the transport. The atoms are treated in materials.
slices, which contain two atomic rows in order to
preserve the slice symmetry [3]. Disorder is
REFERENCES
introduced through the appearance of an on-site
[1] D. K. Ferry, S. M. Goodnick, and J. P. Bird, Transport in
potential at each atom. In this work, a number of
Nanostructures, 2nd Ed. (Cambridge Univ. Press,
donor and acceptor impurities are randomly sited Cambridge, 2009).
on a plane spaced a small distance from the [2] G. Bohra et al., Nonergodicity and microscopic symmetry
graphene sheet itself. breaking of the conductance fluctuations in disordered
mesoscopic graphene, Phys. Rev. B 86, 161405 (2012).
RESULTS [3] B. Liu, R. Akis, and D. K. Ferry, Conductance
fluctuations in graphene nanoribbons, J. Comp. Electron.
We find a significant magnetic field-induced 13, 950 (2014).
reduction in the magnitude of the fluctuations
computed for Fermi energy sweep. This is not due
ISBN 978-0-692-50554-0 131

0.2
0.15
Experiment (X14)
(4e /h)
2
0.1
rms
δG
Theory
0.05
0
12 12 12 12 12 12
0 1 10 2 10 3 10 4 10 5 10 6 10
-2
Electron Density (cm )
Fig. 1. Potential that arises from the random set of impurities.
A different ensemble is used for each simulation run. Fig. 4. Variation of the amplitude of the fluctuations for
magnetic field sweeps with density. The data from Ref. 2 is
Magnetic Field (T) also plotted for comparison.
0 5 10 15
2 0.4
1.5 0.2 0.4
0.35
1 0
δG (4e /h)
δG (4e /h)
0.3
2
(4e /h)
0.5 -0.2
2
0.25
0 -0.4
rms
0.2
δG
-0.5 -0.6 0.15
-1 -0.8 0.1
0.05 0.1 0.15 0.2 0.25
Fermi Energy (eV) 0.05
Fig. 2. Typical traces of the fluctuating part of the 0

0 2 4 6 8 10 12
conductance (a background fit has been subtracted) for
Magnetic Field (T)
magnetic field and Fermi energy sweeps.
0.4
Fig. 5. Amplitude of fluctuations for a Fermi energy sweep
when a short range potential is used (blue circles). The red
0.35
Theory diamonds are data from Ref. 2 for an in-plane magnetic field.
0.3
(4e /h)
2
0.25
rms
0.2
δG
Experiment (X14)
0.15
0.1
0.05
0
0 2 4 6 8 10
Magnetic Field (T)
Fig. 3. Comparison of the magnetic field dependence of the

fluctuations observed for Fermi energy sweeps with the data
of Ref. 2.
132
Reformulation of quantum noise: when
indistinguishable becomes distinguishable?
E. Colomés, D. Marian, and X. Oriols
Departament d’Enginyeria Electrònica, Universitat Autònoma de Barcelona, Bellaterra 08193, Spain
I NTRODUCTION At low frequencies, when the displacement cur-

rent is neglected, current and noise can be computed
The study of noise effects on the performance of
from the knowledge of the number N of transmitted
quantum devices is difficult either from a compu-
particles through the barrier during the time td :
tational or experimental point of view. Quite often
noise predictions are obtained using DC parameters N td
I(E) = limtd →∞ q , (1)
(like single-particle transmission coefficient) into td
simple noise expressions, rather than by a direct
evaluation of the time-dependent fluctuations. In N 2 td − N 2td
this conference, we show that an explicit computa- S(E) = limtd →∞ 2q 2 , (2)
td
tion of the quantum noise in quite simple scenarios
provides surprising and unexpected results. The probabilities P (N ) are computed from the
direct solution of the two-particle Schrödinger equa-
R EFORMULATION OF QUANTUM tion with exchange interaction and Nsummarized in
NOISE
=∞
TABLE I. We define N td = P (N )N
In a recent work [1], we proved that two N =∞ N =−∞
(fermions) electrons impinging in a potential bar- and N 2 td = N =−∞ P (N )N . As seen in

2
rier can be detected at the same place when the TABLE I, the new P (N ), related to PLR , PLL
transmitted and reflected components become dif- and PRR have a direct impact on the current and
ferent. Then, indistinguishable electrons behave as noise. The usual LBF expressions are obtained if
distinguishable. The only requirement is to con- we consider PRR = PLL = 0.
sider (normalizable) wave packets, instead of the
E XAMPLE : RESONANT TUNNELING DIODE
infinite-spatially extended scattering states used in
the Landauer-Büttiker formalism (LBF) [2]. In Fig. We apply expressions (1) and (2) with the TABLE
1 we schematically show these new possibilities of I for a simple resonant tunneling diode (RTD).
detecting two electrons at the left c) or at the right In Fig. 3 we plot the current and Fano factor
d), with probability PLL and PRR respectively. for our results, LBF and distinguishable particles.
These new two-particle scattering possibilities must The differences in the Fano factor are explained
be added to the usual ones considered by the LBF. because the exchange correlation is reduced in our
i.e. Fig.1 a) and b). Some relevant and unexplained results when indistinguishable electrons behave as
experiments [3] and [4] are in agreement with these distinguishable ones.
new probabilities. Depending on the size of the We have considered a 1D two terminal quan-
initial wave packet and on its central momentum, tum wire device where only one or two electrons
PRR and PLL have two limits: the probabilities are simultaneously injected from the reservoirs. At
associated with the LBF (equal to zero) and the high-temperature, when one-particle or two-particle
probabilities associated with distinguishable parti- processes are relevant, the extension to many-
cles (for more details see [1] and [5]). In Fig. 2 we particle scattering probabilities with self-consistent
report the non-zero probability PLL that provides an potentials can be done straightforwardly but will
increment of the noise level above LBF predictions. not modify the present qualitative discussion. The
ISBN 978-0-692-50554-0 133

typical super-poissonian noise in the negative dif-
 



ferential resistance does not appear because we did   
not consider Coulomb correlations, only exchange 
 
correlations [6]. At low temperature, many-particle  
probabilities tend to recover the LBF results, satis- 


  

fying the fluctuation-dissipation theorem [7]. 


 
C ONCLUSIONS
When studying scattering events with (normaliz-  
able) wave packets, new probabilities P (N ) appear,    
 
allowing the location of two electrons at the same

 
place. See Fig.1 c) and d). Therefore quantum Fig. 1: Scattering of two particles in a potential
noise is increased over previous predictions. The barrier system. (a) and (b) are the normally accepted
importance of these new probabilities for the ac- final scenarios. (c) and (d) are now also possible
curate computation of the quantum noise at high final scenarios, both electrons are detected at the
temperatures is discussed by computing the current same position with probability PLL and PRR .
and the Fano factor in a simple RTD.
R EFERENCES
[1] D. Marian, E. Colomés and X. Oriols, J. Phys. Condens. 0.2
Matter (in press). arXiv:1408.1990v2 [quant-ph]
0.15
[2] M. Blanter and M. Büttiker, Phys. Rep. 336, 1 (2000).
[3] R. C. Liu, B.Odom, Y. Yamamoto and S. Tarucha, Nature
PLL
0.1
391, 263 (1988).
[4] E. Bocquillon, V. Freulon, J. M. Berroir, P. Degiovanni, 0.05
B. Plaçais, A. Cavanna, Y. Jin and G. Fève, Science 339,
1054 (2013). 0
4 0.2
[5] http://europe.uab.es/xoriols/two identical electrons.mov 3 0.15
2
[6] X. Oriols, APL, 85 3596 (2004). x 10
−7
1 0.05
0.1
[7] Johnson J, Phys. Rev. 32 97 (1928). 0 0 Wave packet central

2FWHM (m)
energy (eV)
ACKNOWLEDGMENT
This work has been partially supported by the Fig. 2: Probability of finding both electrons (de-
“Ministerio de Ciencia e Innovación” through the scribed by a Gaussian wave function) at the left side
Spanish Project TEC2012-31330 and by the Grant of a double barrier potential system when there is no
agreement no: 604391 of the Flagship initiative applied bias. We see how the zero LBF probability
“Graphene-Based Revolutions in ICT and Beyond”. is recovered for large and far from the resonant
energy wave packets.
4
0,6 Distinguishable results
a Injection No injection LBF results
b Transmitted Reflected 0,5
Our results
PLR 3
Transmitted fa fb PLL fa fb Tb (1 − fa )fb
Injection
2 Current
0 -1 -1 0,4
Fano Factor
Current (µA)
PLR
Reflected PRR fa fb 2
fa fb Rb (1 − fa )fb Fano factor
2
1 0 0 0,3
No injection Ta fa (1 − fb ) Ra fa (1 − fb ) (1 − fa )(1 − fb ) 0,2

1 0 0 1
0,1
TABLE I: Probability P (N ) (upper) that N (lower) 0,0 0
electrons are transmitted from left to right reservoir 0,00 0,05 0,10 0,15
Applied bias (eV)
0,20 0,25 0,30
during the time interval td . fi is the Fermi distri-

bution (i = a, b) and Ti and Ri the i-wave packet Fig. 3: Current and Fano factor from 3 approaches
single particle transmission and reflection. for a RTD (0.8nm/5.6nm/0.8nm and 0.4 eV height).
We set the Fermi level = 0.1eV and T = 300K .
134
Coupled nano-rings: strain and magnetic field
effects
E. R. Hedin, M. B. Orvis, and Y. S. Joe
Center for Computational Nanoscience, Department of Physics and Astronomy,
Ball State University, Muncie, IN 47306, USA
INTRODUCTION SELECTED RESULTS

The effect of a magnetic field on the electron
Over the past two decades, remarkable
transport is considered via the AB effect and
advances have been made in the field of quantum
Zeeman splitting of the QD quasi-bound states.
transport in mesoscopic systems. Much study has
The effects of longitudinal strain on the
been put into the different geometries of quantum
nanoribbon are investigated through a relative
wires and their effect on electron propagation. Of
variation in the inter-ring couplings, Vm. As shown
particular note are the transmission properties of
in Fig. 2, increasing Vm relative to VQD results in
coupled molecular and mesoscopic rings [1], and
closing the transmission bandgap, thereby
the electron-spin transport in semiconductor
producing a semiconductor to metallic transition.
nanostructures and nanoscale electronic devices
In a graphene nanoribbon, the ratio of the coupling
[2]. Graphene nanoribbons are an example of a
strengths, VQD/Vm is a function of strain [3]. With
system composed of coupled molecular rings
the electron energy set equal to εQD, the resonant
which have demonstrated interesting electronic
transmission, T(εQD), is varied as a function of Vm,
properties [3].
the system-leads coupling, Vleads, and the number
MODEL of rings in series, N. In Fig. 3, Vleads=0.775 and
T(εQD) is plotted vs. N for 6 different values of V m.
We investigate a series of coupled nano-rings Exponential decay of the T(εQD) occurs as N
with a total of six quantum dot (QD) sites around increases, except for when Vm=0.6, or twice
each ring. Two of the six sites couple to nearest- VQD=0.3. At Vm=0.6 and Vleads=0.775, the
neighbor rings or to the source/drain leads. The transmission remains at 100%, independent of N.
number of rings in series is a variable parameter in Resonance saturation (Fig. 4) occurs for specific
our model and the system represents a single- values of inter-ring and system-leads coupling
molecule graphene nanoribbon of arbitrary length, constants. A non-zero magnetic flux through the
as depicted in Fig. 1. The hopping integrals plane of the rings splits the transmission bands and
between QD sites and the QD site energy values produces a shifting and narrowing of the
are parameters which can be varied to study the resonances along the energy axis (Fig. 5), until
effects of physical strain and site defects on the complete destructive interference occurs at half a
system. We also incorporate a variable external flux quantum. Zeeman effects are shown in Fig. 6.
magnetic field into the computational system in
order to study the effects of energy-level REFERENCES
separation via the Zeeman effect and resonance [1] G. Cohen, O. Hod, and E. Rabani, “Resonant Transport
interference modulation via the Aharonov-Bohm Properties of Tight-Binding Mesoscopic Rings,” Phys
Rev. B 76, 235120 (2007).
(AB) effect. Using the tight-binding approximation [2] J. Zhou, M.W. Wu, and M.Q. Weng, “Spin-Dependent
to the Schrödinger equation (1), we solve for the Hole Quantum Transport in Aharonov–Bohm Ring
transmission through the entire system for Structure: Possible Schemes for Spin Filter,” Phys. Lett.
arbitrary numbers of rings in series. A 349, 393 (2006).
[3] D. A. Bahamon and V. M. Pereira, “Conductance across
 Vn,m m   n n  En (1) Junctions in Graphene Nanoribbons,” Phys Rev. B 88,
m 195416 (2013).
ISBN 978-0-692-50554-0 135

Fig. 1. Schematic of a 4-ring system comprised of 6-QD rings coupled so as to represent a graphene nanoribbon in the armchair
configuration. Coupling constants (Vi) and QD site energy values (εi) are illustrated as labeled. In subsequent results, V u = Vd ≡
VQD, and εu = εd ≡ εQD.
Fig. 2. Transmission through N=8 rings, with εQD = 0.0, VL=VR=VQD=0.5, for different inter-ring couplings: (a) Vm = 0.5, and (b)
Vm = 1.0.
Fig. 3. Resonant transmission at E=εQD vs. number of rings in Fig. 4. Resonance saturation occurs for Vm = 0.60 and
series for 6 different values of inter-ring coupling. Vm = 0.60, Vleads=0.775, giving nearly flat 100% transmission across
0.63, 0.65, 0.7, 0.8, and 1.0 (top to bottom curves). V leads=0.775. the energy range. Shown above for N=10 rings.
Fig. 5. Variation in transmission as a function of magnetic Fig. 6. Variation in transmission as a function of Zeeman
flux. Vleads=VQD=Vm=0.3. splitting of the QD energy levels.
136
Nanothermocouple Characterization Platform:
Simulation and Experiment
G. P. Szakmany, A. O. Orlov, G. H. Bernstein, and W. Porod
Center for Nano Science and Technology, Department of Electrical Engineering
University of Notre Dame, IN 46556, USA
INTRODUCTION greatest at the center of the thermometer and

decreases towards the terminals. The thermometer
We study the heat distribution by simulations in
measures the average temperature between its
nanoscale structures for nanothermocouple (NTC)
terminals, which is indicated by the dashed line.
characterization. The relative Seebeck coefficient
Although the thermometer and the hot junction are
(RSC) of NTCs is reduced compared to the bulk
located equidistant from the heater in this initial
values [1]. Direct measurement of the RSC of
design, their simulated increases in temperature
NTCs requires an accurate measurement of the
are, unfortunately, not the same. The difference in
temperature difference between the hot and cold
temperature increases above ambient between the
junctions. An accurate measurement of the
hot junction and the thermometer is 45%.
temperature at the hot junction is determined from
The design of the platform was modified to
the resistance of the thermometer, assuming that it
compensate for this temperature difference. The
is at the same temperature as the hot junction.
improved layout features symmetric geometries of
Therefore, we have designed, simulated, and
the hot junction and the thermometer. Fig. 1b
fabricated a characterization platform that allows
shows the simulated temperature increase on the
the simultaneous measurement of the temperature
surface of the improved characterization platform,
at the hot junction and the induced open-circuit
and Fig. 2b shows its simulated temperatures. The
voltage of the NTC. A spatially confined heat
difference between the average temperature on the
source is used to raise the temperature of only the
thermometer and the temperature of the hot
hot junction, while the cold junction remains at the
junction is now reduced to 2%. However, the error
ambient temperature. Here we discuss the
is geometry-dependent, which limits the accuracy
simulation of the characterization platform to
for single-metal NTCs [2].
optimize it for accurate temperature
For accurate RSC measurement of single-metal
measurements.
NTCs, the design was further optimized to
SIMULATIONS eliminate the geometry-dependent error (Fig. 3) by
The characterization platform is constructed placing the hot junction on top of the heater. Figs.
from a heater, a resistive thermometer, and an 4 and 5 show the simulated surface temperature,
NTC (Fig. 1a). The layout of the platform was temperature distribution, and hot junction
simulated using the electric currents and the heat temperature. The difference between the average
transfer module of COMSOL to investigate the temperature of the thermometer and the
temperature increase in the thermometer and the temperature of the hot junction is reduced to 0.5%.
hot junction. Simulations predict that placing the The difference is independent of the dimensions of
hot junction and the thermometer equidistant on the wire widths between 50 nm and 300 nm. Fig. 6
either side of the heater wire does not guarantee shows that the simulated and measured
their equal temperature increase. Figure 1a shows temperatures are in excellent agreement.
the simulated temperature increase at the surface REFERENCES
for the initial design of the characterization [1] S. Sadat, A. Tan, Y. J. Chua, and P. Reddy, Nanoscale
platform. Fig. 2a shows the temperature increase Thermometry Using Point Contact Thermocouples, Nano
along the thermometer and at the hot junction. Lett., 10, 2613 (2010).
Note that the temperature increase on the [2] G. P. Szakmany, A. O. Orlov, G. H. Bernstein, and W.
Porod, Single-Metal Nanoscale Thermocouples, IEEE
thermometer is not uniform. The temperature is Trans. Nanotechnol., 13, 1234 (2014).
ISBN 978-0-692-50554-0 137

(a) (b)
heater heater
50
thermocouple
40
thermocouple
thermometer
D
thermometer
A I
C F 30
B
E 20
G
H 10
10 20 30 40 50 Fig. 4. Simulated surface temperature of the finalized design.
Fig. 1. Simulated surface temperatures of (a) asymmetric and

(b) symmetric designs. 0.9
(a) (b) T Hot junction
G
Asymmetric Symmetric 0.8
3.0 3.0 Average T
A D 0.7 T profile
2.5 2.5
T Hot junction T Hot junction
Average T Average T 0.6
2.0 T profile 2.0 T profile
Position (m)
0.5
Position (m)
Position (m)
1.5
C 1.5
F I
0.4
1.0 1.0
0.3
0.5 0.5
0.2
0.0 B 0.0
E
0.25 0.30 0.35 0.40 0.45 0.25 0.30 0.35 0.40 0.45 0.1
Temperature increase (K) Temperature increase (K)
H
Fig. 2. Simulated temperature increase along the thermometer 0.0
0.125 0.126 0.127 0.128
(between points A-B, and D-E) and the temperature at the hot
junction (points C and F), when 300 μA flows through the Temperature increase (K)
heater. The average temperature of the thermometer is Fig. 5. Simulated temperature increase along the thermometer
indicated by the dashed line. and hot junction when 300 μA flows through the heater.
Heater 1
Thermometer 2
0.20
TC junctions Measurements
0.15 Simulation
Temperature increase (K)
Thermocouple
Thermometer 1
0.10
0.05
Heater 2 0.00
0 100 200 300

Fig. 3. Scanning electron micrograph of the finalized Heater current (A)
structure. The heater is electrically insulated from the TC and Fig. 6. Comparison of the simulated and measured
the thermometers by a 20-nm-thick AL2O3 layer. temperature increase as a function of heater current.
138
Low temperature electrostatic simulations of
realistic large-scale structures
H. Sahasrabudhe1, Y. Wang2, M. Povolotskyi2, R. Rahman2, and G. Klimeck2
1
Department of Physics and Astronomy, Purdue University, West Lafayette
2
School of of Electrical and Computer Engineering, Purdue University, West Lafayette
We first solve for the semi-classical density and

ABSTRACT
potential using the Thomas-Fermi approximation
We present a tool in the NEMO5[1] nano- to get a good initial guess. This potential is then
electronics simulation software, targeting large used to solve the effective mass Schrödinger and
scale quantum confined structures such as Poisson equations self-consistently. The predictor-
quantum dots, Quantum Point Contacts (QPCs) corrector method is used for faster convergence of
and quantum Hall interferometers. This tool can the Schrödinger-Poisson system.
either do a semi-classical electrostatic or a We employ Dirichlet boundary conditions for
quantum mechanical Schrödinger-Poisson top gates in GaAs and Si-SiO2 hetero-structures.
simulation of the substrate and hetero-structure Top gates in Si-SiO2 structures are treated as
with top-gates. It can be used to obtain potential Schottky or Ohmic contacts. For Schottky
and charge-density (along with the wave- contacts, we set the Dirichlet boundary condition
functions). This information can be used for as
insights into device design, or as an input to the    M  EC  EF
NEMO5 tight-binding and transport solvers to do a dirichlet Vg  S
multi-scale multi-physics simulation of the device. q
where  S and  M are semiconductor and
INTRODUCTION
metal work functions respectively, and Vg is the
Advances in Scanning Tunneling Microscopy
(STM) and Molecular Beam Epitaxy (MBE) gate voltage. The electrons are assumed to be
techniques are allowing fabrication of high quality confined in 3D by the gates.
confined structures[2][3]. Some applications of We use Finite Element Method (FEM) to
these techniques are in the fabrication of quantum discretize Schrödinger and Poisson equations in
dots for quantum computing, QPCs for read-out of 3D. The libraries LibMesh[6], PETSc[7] and
quantum state, Fabry-Pérot interferometers for SLEPc[8] are used for discretizing and solving the
measuring quantum Hall effect etc. Design of equations. The library VTK is to be used for
these devices is not a straightforward task. reading in 3D gate structures, which are converted
Simulation tools are helpful in determining the from SEM images. For now, we create gate
optimal gate configuration and voltages for getting structures using simple shapes.
specific electronic configurations. Tools such as RESULTS
QCAD[4] and nextnano3[5] already exist for We can currently solve semi-classical density
simulating electrostatics in Si and GaAs based using our tool. Fig. 1 shows potential in a Si
devices. This tool targets similar simulations in the Double Quantum Dot[9] (DQD) device from a
NEMO5 framework, which gives it potential to be self-consistent semi-classical simulation. Fig. 2
a powerful bridge between realistic devices and shows potential for the same device simulated
detailed physics simulations including non- using Sentaurus[10]. The potentials in Fig. 1 and
equilibrium quantum transport capabilities. This Fig. 2 differ a bit because of different boundary
tool is also targeted for the nanoHUB users conditions, Fig. 3 shows converged potential for a
making it accessible to everyone. GaAs DQD device.
METHODOLOGY Our next goal is to perform self-consistent
Schrödinger-Poisson simulations of these devices
ISBN 978-0-692-50554-0 139

and calibrate capacitances with experimental
measurements.
ACKNOWLEDGMENT
We would like to express our sincere thanks to
Prof. M. Manfra, S. Fallahi and J. Watson from
Department of Physics and Astronomy for
providing experimental data and insight. The Si
DQD device is obtained from Prof. Dzurak’s
group, University of New South Wales.
REFERENCES
[1] S. Steiger, M. Povolotskyi, H. H. Park, T. Kubis,
and G. Klimeck, “NEMO5: A parallel multiscale
nanoelectronics modeling tool,” IEEE Trans. Nanotechnol.,
Fig. 1. Potential on top surface from a semi-classical
vol. 10, no. 6, pp. 1464–1474, 2011.
simulation of a Si DQD device with top Al gates using
[2] M. Manfra, “Molecular Beam Epitaxy of Ultra-High
NEMO5.
Quality AlGaAs/GaAs Heterostructures: Enabling Physics in
Low-Dimensional Electronic Systems,” arXiv Prepr.
arXiv1309.2717, pp. 347–376, 2013.
[3] F. J. Rueß, W. Pok, T. C. G. Reusch, M. J. Butcher,
K. E. J. Goh, L. Oberbeck, G. Scappucci, A. R. Hamilton, and
M. Y. Simmons, “Realization of atomically controlled dopant
devices in silicon,” Small, vol. 3, no. 4, pp. 563–567, 2007.
[4] X. Gao, E. Nielsen, R. P. Muller, R. W. Young, A.
G. Salinger, N. C. Bishop, M. P. Lilly, and M. S. Carroll,
“QCAD Simulation and Optimization of Semiconductor
Quantum Dots,” no. April, pp. 1–20, Mar. 2014.
[5] S. Birner, S. Hackenbuchner, M. Sabathil, G.
Zandler, J. a. Majewski, T. Andlauer, T. Zibold, R. Morschl,
a. Trellakis, and P. Vogl, “Modeling of semiconductor
nanostructures with nextnano3,” Acta Phys. Pol. A, vol. 110,
no. 2, pp. 111–124, 2006.
[6] B. S. Kirk, J. W. Peterson, R. H. Stogner, and G. F. Fig. 2. Potential on top surface from a semi-classical
Carey, “libMesh : a C++ library for parallel adaptive mesh simulation of a Si DQD device with top Al gates using
refinement/coarsening simulations,” Eng. Comput., vol. 22, Sentaurus for comparison with Fig. 1.
no. 3–4, pp. 237–254, Nov. 2006.
[7] S. Balay, W. D. Gropp, L. C. McInnes, and B. F.
Smith, Efficient Management of Parallelism in Object
Oriented Numerical Software Libraries. Birkhäuser press,
1997.
[8] V. Hernandez, J. E. Roman, and V. Vidal, “SLEPc,”
ACM Trans. Math. Softw., vol. 31, no. 3, pp. 351–362, 2005.
[9] M. Veldhorst, C. H. Yang, J. C. C. Hwang, W.
Huang, J. P. Dehollain, J. T. Muhonen, S. Simmons, A.
Laucht, F. E. Hudson, K. M. Itoh, A. Morello, and A. S.
Dzurak, “A Two Qubit Logic Gate in Silicon,” pp. 1–6, 2014.
[10] “Synopsys Sentaurus Semiconductor TCAD
Software.” East Middle eld Road, Mountain View, CA 94043
USA.
Fig. 4. Potential inside a GaAs spin DQD from a semi-

classical simulation using NEMO5 tool.
140
Calculation of Band-to-band Tunnelling
Currents in Germanium
H.M. Rafferty, R.W. Kelsall and Z. Ikonic
Institute of Microwaves and Photonics, School of Electronic and Electrical Engineering, University of
Leeds, Woodhouse Lane, Leeds, LS2 9JT, UK
INTRODUCTION bulk Ge is continuous, the simulation considers

Ge-on-Si single-photon avalanche detectors discretized sections of material with a step-size
(SPADs) are useful in number of photonics determined by convergence calibration; each step
systems. Their principal mode of operation is set at a potential offset relative to the previous
involves electron-hole pair generation in the Ge step. Matching between states in the nth and
‘absorption’ region, and carrier drift into the high- n+1th step is given by an interface matrix, I:
field Si ‘multiplication’ region. Whilst the electric
field in the Si layer must be sufficiently high to
promote impact ionisation, a high field in the Ge
[ ]
a n+1
bn+1
a
[]
=I ( n , n+ 1 ) n
bn
and together with the exponential propagation
(1)
leads to increased noise and dark current. One

cause of field-induced dark counts in Ge is Zener matrix across a layer, this is used to construct the
(1), or band-to-band tunnelling. In this work, band- scattering matrix S for the whole structure
to-band tunnelling currents in bulk germanium
have been calculated.
MODEL
[]
an
b0
a
[]
=S ( 0, n ) 0
bn
(2)
and the transmission and reflection co-efficients
Band-to-band tunnelling currents were determined from it. Following the calculation of
calculated using an empirical pseudopotential one-dimensional transmission co-efficients across
method and a complex bandstructure matching a range of energies in the band structure, and
technique for a multilayer structure. across a grid of in-plane wave vectors, the total
Complex band structure describes all (or a tunnelling current may be calculated (3).
number of) electron states in bulk, which behave
exponentially in a particular direction, with a real Tunnelling currents have been calculated for a
wave vector of states (propagating states, some of range of bias conditions. The dependence of the
which have energies in the allowed bands), or with tunnelling current on the thickness of the
a complex wave vector (evanescent states, e.g. all germanium has been investigated, and the relative
those with energies in the band gap). When a contributions of different interband transitions
propagating state comes to a multilayer structure, (heavy or light hole to Γ or X conduction band
it ‘excites’ various propagating and evanescent states) have been identified.
states inside the next layer, and the process
continues until some number of propagating ACKNOWLEDGMENT
state(s) at the exit form a tunnelling current. In the This work was carried out within the EPSRC-
plane-wave representation, the wavefunction funded UK Silicon Photonics Project.
matching at interfaces between layers is done by REFERENCES
using the continuity of each in-plane harmonic of
[1] C. Zener, H.H. Wills A Theory of Electrical Breakdown of
the full wavefunction. Solid Dielectrics Proc. R. Soc. A 145 855 (1934)
For the one-dimensional case considered [2] David Yuk Kei Ko, J.C. Inkson A Matrix Method for
here, states are classified as forward or backward, Tunneling in Heterostructures: Resonant Tunnelling In
depending on their current (if propagating) or the multilayer systems, Phys. Rev. B, 38, 14 (1988)
[3] A Di Carlo, P Vogl, W Pötz Theory of Zener tunneling
direction of decay (if evanescent), with co- and Wannier-Stark states in semiconductors. Phys Rev B, 50
efficients in these directions given by a and b, 8358 (1994)
respectively. Though the real potential in a biased
ISBN 978-0-692-50554-0 141

Fig. 1. Band-to-band tunnelling current for a 0.2μm section of bulk Ge, under a range of bias voltages
Fig. 2. Band-to-band tunnelling current for varying thicknesses of bulk Ge, under a 1.5V bias.
142
Parallelization of Collision Integrals on DG
solvers for Full Band Boltzmann-Poisson
models of Electron Transport
Jose A. Morales Escalante∗ , Irene M. Gamba∗ ,
∗ ICES, The University of Texas at Austin, USA. e-mail: [email protected]
I NTRODUCTION with δ(·) the Dirac Delta distribution due to

The dynamics of electron transport in modern the Fermi’s golden rule, wp the phonon frequency
semiconductor devices can be described by the (assumed constant), c1 = (nq +1)c, c−1 = nq c, with
semiclassical Boltzmann-Poisson (BP) model: c and c0 constants related to optical and acoustic
∂f 1 q electron-phonon scattering in silicon respectively,
+ ∇k ε(k)·∇x f − E( x, t)·∇k f = Q(f ) (1) and nq = (exp(h̄wp /KB TL ) − 1)−1 the occupa-
∂t h̄ h̄
= −∇x V tion number of phonons, with KB the Boltzmann
∇x · ( ∇x V ) = q [ρ(x, t) − N (x)] , E
constant and TL the lattice temperature.
(2)
The specific form of the collisional operator for
f (x, k, t) is the probability density function (pdf) silicon devices is, in this case:
over the phase space (x, k) of a carrier in the
conduction band being in position x, with crystal
1

momentum h̄k at time t. The collision operator
Q(f ) = ci δ(ε(k) − ε(k ) + ih̄wp )f dk
Q(f ) models different scattering mechanisms; for i=−1 Ωk
example, in the case of silicon devices, it can model 1
x, t) is the electric
electron - phonon collisions. E( − f ci δ(ε(k ) − ε(k) + ih̄wp ) dk

field, ε(k) is the conduction energy band surface, i=−1 Ωk

the electron density ρ(x, t) = Ωk f dk is given by
Deterministic solvers for the BP system using
the k -average of f , where Ωk is the k -space domain,
Discontinuous Galerkin (DG) FEM have been pro-
and N (x) is the doping profile.
posed in [1], [2] to model electron transport along
Q(f ) is related to integral (non-local) operators
the conduction band for 1D diodes and 2D double
in k -space, given by the difference of a gain minus
gate MOSFET devices. In [1], the energy band ε(k)
a loss operator:
model used was the nonparabolic Kane band model.
These solvers are shown to be competitive with
Q(f ) = S(k → k)f (x, k , t) dk Direct Simulation Monte Carlo methods [1]. The
Ωk
energy band models used in [2] were the Kane and
− f (x, k, t) S(k → k ) dk Brunetti, ε(|k|) analytical models, but implemented
Ωk numerically for benchmark tests.
The form of the collision kernel S(k → k )
depends on the particular device considered in the P ROPOSED M ETHOD
modeling. For example, for the case of a silicon de-
When using deterministic solvers to model the
vice, the main collisional mechanisms are acoustic
BP system, most of the computational cost of the
and optical electron - phonon scattering, so
numerical scheme is related to the collision integral.
1 In the case of DG solvers for BP, the functions under

S(k → k ) = ci δ(ε(k ) − ε(k) + ih̄wp ), (3) study are projected in a space of piecewise contin-
i=−1 uous polynomials across the cells of the domain.
ISBN 978-0-692-50554-0 143

In this work we will study the reduction of
the computational cost associated to the collision
integrals by means of parallelization of the compu-
tation of the collisional sums approximating these
operators.
We work with a coordinate description of
k(k1 , k2 , k3 ) in which the energy is one of the
coordinates, say, k1 = ε. We could have, for
example, k2 = kx /|k| the polar component and
k3 = ϕ the azimuthal angle, but we could choose
other coordinates k2 , k3 too. We decompose the k -
domain in cells Kkmn = [k1 − −
k , k1 k ] × [k2 m , k2 m ] ×
+ +
−
[k3 n , k3 n ]. The collisional sums are then given by:
+
Q(f ) =
1

ci δ(ε − ε + ih̄wp )f dk
i=−1 k,m,n Kkmn Fig. 1. Piecewise Linear Approximation of the Silicon
conduction energy full band structure (ε) color plot in k-space
−f δ(ε − ε + ih̄wp ) dk (1st octant Brillouin Zone) obtained by Local EPM.
Kkmn
Computing analytically the integrals just over the Current - Momentum (I in 10^6/(s m^2) ) vs. (Position, Time) (x,t)
energy variable involving Dirac deltas: 400nm. V=0.3V - EPMravg
+1

Q(f )(x, ε, k2 , k3 ) = ci χ[ε−k ,ε+k ] (ε + ih̄wp )× 2e+22
1.5e+22
1e+22
i=−1 k
I 5e+21
k2 +
m
k3 +
n ∂k

0
× f dk2 dk3 -5e+21
∂(ε , k2 , k3 ) ε =ε+ih̄w

-1e+22
m n k2 −
m k3 −
n -1.5e+22
p
1
− f (x, k(ε, k2 , k3 )) χ[ε−k ,ε+k ] (ε − ih̄wp ) ×

0.9
0.8
0.7
0 0.6
0.5 0.5
k 1 1.5 0.4 x (microns)
 2 2.5 3 0.2
0.3
k2 + k3 + 3.5
∂k t (ps) 4 4.5 5 0
0.1
m n

× dk2 dk3 
m n k2 −
m k3 −
n
∂(ε , k2 , k3 ) ε =ε−ih̄wp
Fig. 2. Current (Momentum) vs. (t, x) for a n+ − n − n+ Si
The computation of these collisional sums will diode with a 400nm n-channel, using an EPM radial average
be parallelized by means of OpenMP and MPI band. From initial time to t = 5.0ps. 0.3 Volts bias.
for two different silicon devices: the benchmark
n+ − n − n+ diode problem and the double gated
MOSFET. We will find the scaling of computational [3] J.A. Morales Escalante, I.M. Gamba, A. Majorana, Y.
Cheng, C.-W. Shu, and J.R. Chelikowsky, Determinis-
time versus the number of compute nodes on TACC tic DG Solvers for EPM-Boltzmann-Poisson Transport,
high performance computing resources. IWCE16 Abstracts, 2013.
R EFERENCES
[1] Y. Cheng, I. M. Gamba, A. Majorana, and C.W. Shu
A Discontinous Galerkin solver for Boltzmann-Poisson
systems in nano-devices, CMAME 198, 3130-3150 (2009).
[2] Y. Cheng, I. M. Gamba, A. Majorana and C.-W. Shu,
A discontinuous Galerkin solver for full-band Boltzmann-
Poisson models, IWCE13 Proceedings (2009).
144
Mathematical Consistency of Time Dependent
Density Functional Theory with Kohn-Sham
Potentials
J.W. Jerome∗
∗ Department of Mathematics, Northwestern University, Evanston, IL 60208 USA
In residence: George Washington University, Washington D.C.
I NTRODUCTION As explained in [3, Sec. 6.5], Φ0 is determined by

We consider the closed quantum systems ana- the initial state of the Kohn-Sham system and the
lyzed in [1]. The TDDFT model dates from the sem- initial state of the interacting reference system with
inal article [2]. If we denote by Ĥ the Hamiltonian the same density and divergence of the charge-
operator of the system, then the state Ψ(t) of the current.
system obeys the nonlinear Schrödinger equation, The Hamiltonian operator then assumes the stan-
dard form,
∂Ψ(t)
ih̄ = ĤΨ(t). (1) h̄2 2
∂t Ĥ = − ∇ + V (x, t) + W ∗ ρ + Φ(x, t, ρ). (3)
Here, Ψ = {ψ1 , . . . , ψN } and the charge density ρ 2m
is defined by Theorem. Under appropriate assumptions, there
N
is a unique weak solution of (1) on any specified

ρ(x, t) = |Ψ(x, t)|2 = |ψ(x, t)|2 . time interval [0,T]. The solution may be character-
k=1 ized as a fixed point of a mapping defined by the
For well-posedness, an initial condition, evolution operator.
Ψ(0) = Ψ0 , (2) Remarks

• The theory developed here does not (yet) ex-
consisting of N ground states, and boundary con-
ditions must be adjoined. We will assume that the tend to ‘time dependent current density func-
particles are confined to a bounded region Ω ⊂ R3 tional theory’ (TDCDFT). This is discussed in
and that homogeneous Dirichlet boundary condi- [3].
• The evolution operators employed are capable
tions hold for the evolving quantum state within a
closed system. In general, Ψ denotes a finite vector of discretization and simulation [1]. The theory
function of space and time. is due to Kato [4, Ch. 6].
S PECIFICATION OF THE H AMILTONIAN R EFERENCES

OPERATOR [1] J.W. Jerome and E. Polizzi, Discretization of time de-
pendent quantum systems: Real time propagation of the
We study potentials which are of the form, evolution operators. Appl. Anal. 93 (2014), 2574–2597.
[2] E. Runge and E.K.U. Gross, Density functional theory
Veff (x, t, ρ) = V (x, t) + W ∗ ρ + Φ(x, t, ρ), for time dependent systems. Physical Review Letters 52
(1984), 997–1000.
where, for W (x) = 1/|x|, the convolution W ∗ρ de-
[3] C. A. Ullrich, Time-Dependent Density-Functional The-
notes the Hartree potential, and where Φ represents ory: Concepts and Applications. Oxford University Press,
a time history of ρ: 2012.
t [4] J.W. Jerome, Approximation of Nonlinear Evolution
Φ(x, t, ρ) = Φ0 (x, 0, ρ) + φ(x, s, ρ) ds. Equations, Academic Press, New York, 1983.
0
ISBN 978-0-692-50554-0 145

146
Bipolar Monte Carlo simulation of hot carriers
in III-N LEDs
Pyry Kivisaari1 , Toufik Sadi2 , Jani Oksanen1 , Jingrui Li3 , Patrick Rinke3 , and Jukka Tulkki1
1 Departmentof Biomedical Engineering and Computational Science,
Aalto University, P.O. Box 12200, FI-00076 Aalto, Finland
2 Device Modeling Group, School of Engineering,
University of Glasgow, Glasgow G12 8LT, Scotland, UK

3 Department of Applied Physics, Aalto University, P.O. Box 11100, FI-00076 Aalto, Finland
I NTRODUCTION T HEORY, R ESULTS & D ISCUSSION
III-Nitride (III-N) light-emitting diodes (LEDs) A coupled MC simulation is performed for elec-
have started a revolution in general lighting and are trons and holes, resulting in device-level character-
on track to replace all the competing lighting tech- istics and distribution functions as a function of the
nologies [1]. Even with the spectacular evolution of bias voltage [8]. The initial values are taken from
III-N LEDs into a very efficient technology, they a standard DD simulation, and the bandstructure is
still suffer from an efficiency droop, whose origin an analytical model fitted to the full band structure
is not fully understood [2]. In addition, the device- shown in Fig. 2.
level understanding of III-N multi-quantum well Figure 3 shows the electron densities from the
(MQW) MQW LEDs is still incomplete, the most DD and MC simulations for the LED structure
obvious indication of which being the difficulty with three QWs at strong injection. The largest
to fit simulated current-voltage characteristics to amount of electrons is located in the rightmost QWs
experiments. closest to the p-side, which is also responsible for
Typically the full device-level models for III-N most of the light emission. The electron densities
LEDs are based on the drift-diffusion (DD) model, from the DD and MC simulations also exhibit a
which assumes quasi-equilibrium carrier distribu- slight difference between each other. This difference
tions [3]. However, Iveland et al. and Binder et is a direct consequence of the fact that the MC
al. have recently measured clear indications of hot simulations do not assume any quasi-equilibrium
electrons and holes in LEDs already at typical distributions for the carriers.
operating conditions [4], [5]. Inspired by these mea- DD simulations typically result in unrealistically
surements, we recently developed a Monte Carlo– high turnon voltages for III-N MQW LEDs. The
drift-diffusion (MCDD) model to specifically study higher turnon voltage in the DD simulation origi-
hot electrons. According to the MCDD simulations, nates from quasi-Fermi losses in the barrier layers
a significant amount of hot carriers is generated by between the QWs [9], which are illustrated by the
Auger recombination, and part of these hot electrons band diagram shown in Fig. 4. The MC simulations
are visible even at the p-contact [6]. At certain for the sample structures of Fig. 1 enable studying
conditions we also detected a significant difference how accounting for hot carriers increases the current
between the MCDD and DD results [7]. density for a given bias voltage. The smaller turnon
In this work, we perform fully bipolar MC sim- voltages given by the MC model clearly indicate
ulations for electrons and holes in realistic III-N that accounting for hot carriers is needed for com-
LED devices taking into account all the relevant plete understanding of the device physics of III-N
scattering and recombination processes and device LEDs.
physics. We acknowledge the Aalto Energy Efficiency
ISBN 978-0-692-50554-0 147

25
2x10
DD
Electron density (/m )

3
MC
25
Fig. 1. The III-Nitride MQW structures simulated in this work. 1x10
0
0.99 1 1.01 1.02 1.03 1.04
Distance (µm)
Fig. 3. Electron densities from the initial DD and final MC
simulations for a structure with three QWs, at a current density
of 400 A/cm2 .
Fig. 2. The full bandstructure calculated from first principles.
Programme (AEF), the Academy of Finland, and

the Nokia Foundation for support.
R EFERENCES 4
Cond. band
[1] S. Nakamura and M. R. Krames, Proc. IEEE 101, 2211– 3.5 Q−F level
Energy (eV)
2220 (2013).
3
[2] J. Piprek, Compound Semiconductor 20, 44–48 (2014).
[3] J. Piprek, J. Römer, and B. Witzigmann, Appl. Phys. Lett. 2.5
106, 101101 (2015).
[4] J. Iveland, M. Piccardo, L. Martinelli, J. Peretti, J. W. Choi, 2
N. Young, S. Nakamura, J. S. Speck, and C. Weisbuch, 1.5
Appl. Phys. Lett. 105, 052103 (2014). −100 −50 0 50 100 150 200
[5] M. Binder, A. Nirschl, R. Zeisel, T. Hager, H.-J. Lugauer, Position (nm)
M. Sabathil, D. Bougeard, J. Wagner, and B. Galler, Appl.
Phys. Lett. 103, 071108 (2013). Fig. 4. Band diagram of the structure with 5 QWs from the DD
[6] T. Sadi, P. Kivisaari, J. Oksanen, and J. Tulkki, Appl. Phys. simulation. The bias voltage is 4.5 V and the current density
Lett. 105, 091106 (2014). is 284 A/cm2 .
[7] P. Kivisaari, J. Oksanen, J. Tulkki, and T. Sadi, J. Comput.
Electron., DOI: 10.1007/s10825-015-0687-z (2015).
[8] P. Kivisaari, T. Sadi, J. Oksanen, and J. Tulkki, Proc. SPIE
9363, 93631S (2015).
[9] P. Kivisaari, J. Oksanen, and J. Tulkki, J. Appl. Phys. 111,
103120 (2012).
148
Modeling of Parallel Electric Field Tunnel FETs
K. Fukuda, Y. Morita, T. Mori, W. Mizubayashi, M. Masahara, T. Yasuda, S. Migita, H. Ota
National Institute of Advanced Industrial Science and Technology (AIST)
1-1-1 Umezono, Tsukuba, Ibaraki 305-8568, JAPAN
additional silicon layer on the top of the source

ABSTRACT
which is considered as the tunnel area.
Tunnel FETs with vertical tunnel paths are
successfully modeled by the nonlocal band to SIMULATIONS
band tunneling model. Although enhancement Device structures of the PP-TFET used in the
of ON currents are obtained by longer source simulation is schematically shown in Fig.3. The
gate overlap lengths, the increase of the ON source concentration is 2×1020cm-3, effective oxide
current is not proportional to the lengths, thickness is 1.3nm, the drain bias is -1V, and the
because of non-uniformity of the band to band epitaxial layer thickness TEPI is 2nm. Fig.4 shows
tunneling generations. It is explained by the the comparison of measured and simulated on
peak generation rates at the source edge. currents at VDS=-1V, VGS=-2.5V depending of the
INTRODUCTION overlap length LOV. It is successfully explained by
the simulation that the on currents do not increase
Tunnel FET is a candidate for low power in proportional to LOV. Fig.5 shows the simulated
transistor alternative to CMOS. The authors have ID-VG characteristics in linear plot. It is clearly
developed TFETs with vertical tunnel path called observed that the currents do not differ up to a
parallel electric field TFET (PE-TFET), and certain gate voltage. This is because that the BTBT
obtained enhancement of drive current by longer generation is mainly at the edge of the source
source-gate overlap lengths [1]. However the junction, not covering whole the overlap region in
enhancement is less than proportional to the these gate bias conditions. Fig.6 shows the BTBT
overlap lengths. The reason of this current generation rates at VDS=-1V, VGS=-2.5V, for the
degradation is investigated by device simulation LOV=200nm case. Even at this bias condition, the
with the nonlocal band to band tunneling (BTBT) peak of BTBT generation rates is at the source
model. edge, which is the reason that the ON currents do
MODEL not increase in proportional to LOV.
Fig. 1 explains our scheme to obtain tunnel SUMMARY
path lengths by tracing energy band profiles in the TFETs with vertical tunnel paths are
device simulator. The nonlocal electric field, successfully modeled by the device simulation
defined as the band gap divided by the obtained with the nonlocal BTBT tunneling model. The on
tunnel path length is used to calculate carrier currents do not increase in proportional to the
generation rate by famous Kane’s formula. source gate overlap length, because of the
Accuracy of the model has been confirmed generation peak at the source edge.
through comparisons to the experimental results of
silicon TFETs [2]. Model parameters are not ACKNOWLEDGMENT
changed from conventional ones for silicon, and This research is granted by JSPS through
implemented into the device simulator [3]. FIRST Program initiated by CSTP.
REFERENCES
PARALLEL FIELD TFETS [1] Y. Morita et al. "Performance evaluation of parallel
Fig. 2 shows the cross section of the PE- TFET. electric field tunnel field-effect transistor by a distributed-
element circuit model." Solid-State Electronics 102
Assumed tunnel directions are shown by arrows in (2014): 82-86.
the figure. The drive current of the TFET is [2] K. Fukuda et al. "On the nonlocal modeling of tunnel-
intentionally increased by tunnel areas increased FETs." SISPAD, 2012.
by longer source-gate overlap lengths. The device [3] HyENEXSS ver5.5.
is fabricated by epitaxial growth technique of
ISBN 978-0-692-50554-0 149

Fig. 1. Trace algorithm of the nonlocal band to band tunnel Fig. 4. Comparison of simulated and measured on-currents at
model, band diagram (left) and 2D trace (right). VDS=-1V, VGS=-2.5V.
Fig. 2. TEM image of the PE-TFET and assumed tunnel Fig. 5. The source and gate overlap length dependence of I-V
directions. curves simulated by the nonlocal BTBT model.
Fig. 3. Schematics of the PE-TFET simulated by device Fig. 6. 2D distribution of the BTBT generation rates for
simulation. LOV=200nm case.
150
Analytic solution of Ando’s surface roughness
model with finite domain distribution functions
K. Moors∗† , B. Sorée†‡ , and W. Magnus†‡
∗ Institute
for Theoretical Physics, KU Leuven, Celestijnenlaan 200-D, B-3001 Heverlee, Belgium
† Imec, Kapeldreef 75 , B-3001 Heverlee, Belgium
‡ Physics Department, Universiteit Antwerpen, Groenenborgerlaan 171, B-2020 Antwerpen, Belgium
In this abstract we present two variants of a proposed by Lizzit et al., for a one subband toy
new approach to simulate surface roughness (SR) model in Fig. 2 [4]. The distribution functions have
scattering with Ando’s model and the multi-subband different disadvantages (fI : negative weights when
Boltzmann transport equation (BTE) [1]. An crucial C > 3/4, fII : Dirac delta peak for S = S , bivariate
step to obtain the SR matrix element is the inte- normal: non-zero probability for SR crossing the op-
gration of the product of wave functions (of the posite wire boundary) but mutually agree quite well,
initial and final state) over the length of the SR while the PN result is incorrect. The finite domain
deviation in the direction normal to the boundary models allow for fast and accurate simulations due to
surface. Typically one expands this integral up to their analytic expression, while the bivariate normal
first order of the SR deviation, usually even taking model requires numerical treatment. Hence with the
the limit of an infinite barrier height, known as the latter approach simulations are much slower (see
Prange-Nee (PN) approximation [2]. The validity Fig. 3) and we were unable to study metal wires
of these approximations can be seriously doubted with larger diameters.
when applied in metal nanowires, as small diameters For larger diameter simulations the first order
lead to substantial wave function penetration in the approximation and finite domain models agree well
barrier while a large number of subbands is still on the diameter scaling trend, but the predicted
present, allowing for wave functions that oscillate resistivity values differ up to a factor of 2 (see
many times over the typical SR deviation length Fig. 4). The PN model works much better than for
(see Fig. 1). However, this type of approximation the one subband model and in certain cases even
allows for a replacement of the SR deviation by better than the first order approximation, however
its auto-correlation function (ACF), leading to an this is believed to be coincidental.
analytical expression of the absolute value squared A large resistivity drop is observed for D ≈
of the matrix element that appears in the collision 3.25 nm for all methods and appears when the
term of the BTE [3]. bottom of all subbands is as far below the Fermi level
The SR integral can still be solved analytically for as possible. In this case back-scattering of the Fermi
the finite potential well with averaging over the SR level states is suppressed and the current is protected.
if an appropriate SR distribution function f (S, S ) The predicted resistivity values for this drop are in
is chosen, with S, S the SR deviations at different good agreement for the finite domain models, while
positions on a wire boundary. We propose to √ use a the approximate methods deviate substantially.
distribution√ functions on a finite domain ( − 3∆ ≤
R EFERENCES
S, S , ≤ 3∆) and give two variants below:
[1] T. Ando, A. B. Fowler, and F. Stern, “Electronic properties
fI S, S of two-dimensional systems,” Reviews of Modern Physics,
vol. 54, pp. 437–672, 1982.
1 4

≡ 1 + C θ SS − θ −SS , [2] R. E. Prange and T.-W. Nee, “Quantum spectroscopy of the
12∆2 3 low-field oscillations in the surface impedance,” Physical
(1 − C) Cδ (S − S ) Review, vol. 168, no. 3, p. 779, 1968.
fII S, S ≡ + √ . [3] K. Moors, B. Sorée, Z. Tőkei, and W. Magnus, “Resistivity
12∆2 12∆ scaling and electron relaxation times in metallic nanowires,”
with ∆ the SR standard deviation, C the correlation Journal of Applied Physics, vol. 116, no. 6, p. 063714, 2014.
of the SR at the different positions, provided by a [4] D. Lizzit, D. Esseni, P. Palestri, and L. Selmi, “A new
formulation for surface roughness limited mobility in bulk
given ACF (here Gaussian) and θ the step function. and ultra-thin-body metal–oxide–semiconductor transistors,”
We compare the resistivity results with the bivari- Journal of Applied Physics, vol. 116, no. 22, p. 223702,
ate normal SR distribution function, that was recently 2014.
ISBN 978-0-692-50554-0 151

(a) Rough boundary position: -0.5 nm
Fig. 3. The computation times (t) for the SR matrix elements

are shown in a logarithmic scale for the different methods: the
Prange-Nee approximation (PN), the first order approximation
(First order), the approach with a finite domain distribution
function using fI (Finite domain I) and fII (Finite domain II),
and the bivariate normal distribution function (Bivariate normal).
The timescale is in milliseconds and corresponds to the one
(b) Rough boundary position: 1 nm subband toy model simulation shown in Fig. 2.
Fig. 1. A product of wave functions for a wire with width of

approximately 2.15 nm and the integral up to the rough boundary
position are shown as a shaded area under the curve. The correct
value in the SR matrix element of Ando’s model should be the
area under the wave function product, but the PN and the first
order approximation replace this area by a rectangular box.
(a) Subbands
(a) One subband toy model
(b) Resistivity
(b) Resistivity
Fig. 4. The resistivity for copper nanowires with square cross-
Fig. 2. In (a) the dispersion relation of a one subband toy model section of equal sides D ranging from 2 to 6 nm (SR standard
is shown with the momentum gap ∆kz indicated. The resistivity deviation of 2aCu ≈ 0.7 nm and correlation length of 5aCu ≈
values obtained with the SR matrix elements of Ando’s model 1.75 nm) are shown in (b), using the different methods for
are shown in (b), using the PN and the first order approximation, Ando’s model that are computationally feasible (same methods
the two newly proposed variants: using fI and fII , and the as in Fig. 3 without Bivariate normal). The subbands dropping
bivariate normal model [4]. below the Fermi level of the D ≈ 2.15 nm wire are shown in
(a).
152
Thermoelectric Transports in Geometry- and
Doping-Controlled Nanostructures
K.H. Park, and U. Ravaioli
Beckman Institute and Department of Electrical and Computer Engineering, University of Illinois,
Urbana, Illinois 61801, USA
which is free from carrier concentration gets

ABSTRACT SUBMISSION
interfered by the structural variation
This work investigates electron and phonon
transports in semiconductor nanowires and METHODS
nanoribbons with geometrical fluctuation and non- Change in the electronic transport due to the
uniform carrier doping concentration, expecting an geometry variation and non-uniform doping is
efficient phonon blockade with a minimal impact examined by computing an energy dependent
on electronic transport, so that enhancement of electrical quantum conductance of the structures
thermoelectric efficiency is expected. This under consideration employing a Recursive Green
comprehensive study examines trade-offs to find an Function approach [4]. The 3D conduction channel
optimal design for thermoelectric applications. is modeled by a 1D tight binding chain, and the
quantum nature is accounted by solving the
INTRODUCTION
Schrödinger equation in a series of 2D cross
Nanostructured semiconductors have attracted sections in the transverse direction. The quantum
considerable attention due to their potential transmissions of the sinusoidal nanowires are
applications in thermoelectric devices [1]. Utilizing compared with those of straight uniform nanowires
their large phonon mean-free-path (MFP) compared in Fig. 3. The nanowire undulations directly affect
to electron MFP, nano-structures have achieved the resistivity particularly at higher carrier energy
improved thermoelectric efficiency dictated by the but have little effect on low energy transport,
dimensionless thermoelectric figure of merit (ZT = leading this structure be potentially good for use of
S2σT/κ) [2]. In such nanowires, the phonon thermoelectric applications. Quantum
confinement starts developing ahead of the electron transmissions of the nanoribbons can be also
confinement and reduces thermal transport with investigated using the same method with the highly
hardly affecting electronic transport [3]. In this doped region brought into the calculation by
study, we investigate thermoelectric properties of solving the Schrödinger equation of the cross
silicon nanowires and nanoribbons with their sections. The Seebeck coefficient and electrical
geometry- and doping-modulation, expecting an conductivity of the structures will be computed
enhancement of the figure of merit ZT. through Landauer formula as a function of quantum
MODEL transmission.
We will consider sinusoidally undulated The lattice thermal conductivity of those non-
nanowires and non-uniformly doped nanoribbons uniform structures can be computed with molecular
with rough edges, as schematically depicted in Fig. dynamics at a constant temperature. We adopt the
1 and 2, respectively. It is expected to possibly reverse non-equilibrium MD method, which
realize an efficient phonon blockade through the imposes a heat flux first and measures the resulting
structural barriers, without degrading appreciably temperature gradient, performed with the Stillinger-
the electronic current. The high doping Weber interatomic potential. The MD simulation
concentration along the center region of the should reveal a decrease in thermal conductivity as
nanoribbon is anticipated to confine electrons structural barriers are imposed to the system. In Fig.
adequately, so that the rough edges have a minimal 4, it is shown that the lattice thermal conductivity of
impact on electron transport while phonon transport the sinusoidal nanowire is lower than that of the
straight nanowire.
ISBN 978-0-692-50554-0 153

CONCLUSION
We will show comprehensive results illustrating
possible trade-offs to optimize the thermoelectric
figure of merits of uniform and non-uniform nano-
structures. As we consider geometrical fluctuation
and non-uniform doping as the sole perturbation to
our system, the ratio of the figure of merits should
provide an insight of an optimum design for
thermoelectric applications.
REFERENCES
[1] A. I. Hochbaum, R. Chen, R. D. Delgado, W. Liang, E. C. Fig. 3. Comparison of quantum transmissions of a sinusoidal
Garnett, M. Najarian, A. Majumdar, and P. Yang, nanowire and a straight uniform nanowire (w = 4.34 nm, L =
Enhanced thermoelectric performance of rough silicon 108.6 nm)
nanowires, Nature 451, 7175 (2008).
[2] K. H. Park, M. Mohamed, Z. Aksamija, and U. Ravaioli,
Phonon scattering due to van der Waals forces in the
lattice thermal conductivity of Bi2Te3 thin films, J. Appl.
Phys. 117, 1 (2015).
[3] P. Martin, Z. Aksamija, E. Pop, and U. Ravaioli, Impact of
Phonon-Surface Roughness Scattering on Thermal
Conductivity of Thin Si Nanowires, Phys. Rev. Lett. 102,
12 (2009).
[4] P. N. Martin and U. Ravaioli, Green function treatment of
electronic transport in narrow rough semiconductor
conduction channels, J. Phys. Conf. Ser. 193, 012009
(2009).
Fig. 4. Comparison of lattice thermal conductivity of a

sinusoidal nanowire and a straight nanowire
Fig. 1. A schematic of a semiconductor nanowire with

sinusoidal undulations
Fig. 2. A schematic of a non-uniformly doped semiconductor

nanoribbon with edge-variation
154
An Advanced Electro-Thermal Simulation
Methodology For Nanoscale Device
L. Wanga, T. Sadia, M. Nedjalkovb, A.R. Brownc, C. Alexanderc, B. Chengc, C. Millarc, A.
Asenova,c
a
School of Engineering, University of Glasgow, Glasgow G12 8LT, U.K.
b
c
Gold Standard Simulations Ltd, 11 Somerset Place, Glasgow, G3 7JT, U.K.
e-mail: Liping.Wang @glasgow.ac.uk
equation in the form of new heat flow equation can

ABSTRACT
be derived, which includes the terms about 𝑇𝑇! and
In this work, we propose an advanced electro- 𝑇𝑇! .
thermal simulation methodology for nanoscale ! !! !!" !! !!!
device based on a macroscopic model for acoustic ∇ 𝑘𝑘 ! ∇𝑇𝑇! = −𝐶𝐶! ! ! − (1)
!!!! ! !!!!
and optical phonon energy transfer, which will be !!"!! !!!! !!! !!!"!! !!" !! !!!
demonstrated on an SOI FinFET example. = −𝐶𝐶! −
!!!!! !! !!!"!!!! ! !!!!
INTRODUCTION where 𝑛𝑛 is the electron density, 𝑣𝑣! is the drift
With the continuous scaling of semiconductor velocity, and 𝐶𝐶! is heat capacities. This can be
transistors down to nanoscale regime, further coupled within Drift-Diffusion framework, which
driven by the emerging novel architecture such as is more computationally efficient. The electrons
FinFETs, thermal transport has become one of the temperature 𝑇𝑇! can be pre-calculated by MC
major concerns for in terms of performance and simulations and a lookup table of 𝑇𝑇! with respect
reliability. Modeling and analysis of self-heating to E can be constructed. Then the coupled Poisson,
effects in nanoscale devices such as FinFETs have CCE, and energy transfer equations can be solved
attracted increasingly more interests [1-3]. iteratively. At each iteration, 𝑇𝑇! will be refreshed
Recently we have developed a thermal simulation as well as recalculation of 𝑇𝑇! . Thermal
module [4], implemented in GSS ‘atomistic’ conductivity in the refined fin region will be
simulator GARAND [5], based on the solution of modeled by position and temperature dependent
the coupled Heat Flow, Poisson, and Current formula. The flow chart of the methodology is
Continuity Equations (CCE). In this work, we are illustrated in Fig. 1.An SOI FinFET example
going to further incorporate an advanced (schematics as sown in Fig. 2) is used as a testbed
methodology into GARAND for electro-thermal for the advanced methodology. It consists of
simulation of FinFETs. complex 3D structure and material composition, as
MODEL AND EXAMPLE shown in Fig. 3. Lattice temperature and potential
distributions obtained from coupled Heat Flow,
The advanced electro-thermal simulation Poisson, and CCE will be used as the initial
methodology for nanoscale device is based on a conditions for the advanced electro-thermal
macroscopic model [6]. The model for acoustic simulation, as shown in Fig. 4 (a) (b).
and optical phonon energy transfer involves the
temperature of electrons, acoustic phonons and CONCLUSION
optical phonons 𝑇𝑇! , 𝑇𝑇! and 𝑇𝑇! , and the relaxation An advanced electro-thermal simulation
times between them 𝜏𝜏!!! , 𝜏𝜏!!! and 𝜏𝜏!!! methodology for nanoscale device based on a
respectively. These can be naturally implemented macroscopic model for acoustic and optical
in Monte Carlo (MC) simulation, however, in the phonon energy transfer is being implemented in
cost of long computational time. If focusing on the GSS ‘atomistic’ simulator GARAND. A full scale
stationary state, these equations can be simplified. simulation on SOI FinFET will be presented
Starting from the energy transfer equations, a new following the complete implementation.
ISBN 978-0-692-50554-0 155

ACKNOWLEDGMENT
The research leading to these results has received
funding from the European Union Seventh Framework
Programme (FP7/2007 – 2013) under grant agreement
no. 318458 SUPERTHEME.
Structure,
material, mesh
Fig. 3. Materials and structures of the SOI FinFET’s

Poisson Equation electrical simulation domain, showing 3 quarters.
Current Continuity Equations
J*E
Heat Flow Equations
current converged NO
YES
Poisson Equation
(a)
Current Continuity Equations
MC simulation
Te
Energy Transfer Equations
current converged NO
YES
Output:
I-V, lattice temperature,
potential distributen, etc. (b)
Fig. 3. (a) Lattice temperature distributions and (b) potential
distribution at high drain and high gate biases resulted from
Fig. 1. Flow chart for the advanced electro-thermal simulation coupled Heat Flow, Poisson, and Current Continuity
methodology. Equations, which will be the initial conditions for the
advanced electro-thermal simulation.
DR
AIN
GAT
E REFERENCES
SOU [1] I. N. Adisusilo, K. Kukita, and Y. Kamakura. “Analysis of
RCE er
ac Heat Conduction Property in FinFETs Using Phonon Monte
to Sp
H
x
LG Carlo Simulation, ” SISPAD 2014, pp.17-20, Sep. 2014.
er
W ac
fin
fin Sp [2] S. Kolluri, K. Endo, E. Suzuki, K. Banerjee. “Modelling and

analysis of self-‐heating in FinFET devices for improved
circuit and EOS/ESD performance,” IEDM 2007, Dec 2007.
[3] T. Sadi, R.W. Kelsall and N. J. Pilgrim, “Electrothermal Monte
E Carlo Simulation of Submicrometer Si/SiGe MODFETs”, IEEE
XID
DO Trans. Electron. Dev., Vol. 54, No. 2, pp. 332-339, 2007
RIE
BU [4] L. Wang, A. R. Brown, et al. “3D Coupled Electro-Thermal
FinFET Simulations Including the Fin Shape Dependence of the
TE
TRA Thermal Conductivity,” SISPAD 2014, pp.269-272, Sep. 2014.
BS
SU [5] GARAND Statistical 3D TCAD Simulator [Online]. Available:
http://www.GoldStandardSimulations.com/products/garand/
Fig.2. Schematics of SOI FinFET [6] K. Raleva, D. Vasileska, S. M. Goodnick, M. Nedjalkov.
“Modeling Thermal Effects in Nanodevices,” IEEE Trans.
Electron. Dev., Vol. 55, No. 6, pp.1306-1316, June 2008.
156
Multiscale-modeling of CNTFETs with
non-regular defect pattern
M. Claus∗§ , D. Teich∗ , S. Mothes∗ , G. Seifert∗∗ M. Schröter§‡
∗ Center for Advancing Electronics Dresden, TU Dresden, Germany, [email protected]
§ Chair for Electron Devices and Integrated Circuits, TU Dresden, Germany
∗∗ Physical Chemistry, TU Dresden, Germany
‡ ECE Department, UC San Diego, USA,
I NTRODUCTION defects lead to a cut of the sharp DOS peak as

An important consideration in the design and re- already suggested by the effective band structure
liability of circuits is the role of defects, impurities, shown in Fig. 1.
and parameter fluctuations in affecting the transistor As expected, with increasing number of defects,
characteristics. Here, the impact of vacancies on the equilibrium transmission spectrum decreases. In
CNTFET characteristics is studied by means of Fig. 3, the transmission spectra for two random
a multi-scale modeling and simulation framework. defect pattern with 5 defects (defect density of
Very recently, defect densities of 0.02 % up to 0.4 %) are compared with the average transmission
0.2 % have been reported for different CNT sam- spectra for the same defect density.
ples. Therefore and in contrast to other simulation For the simulation at the device level,
studies [1] at the device level, the impact of defects the effective-mass Schrödinger-Poisson solver
beyond the single defect limit is analyzed. Our described in [4] is employed. In order to consider
atomistic simulation results suggest the developed mono-vacancies, randomly distributed distortions
defect model at the device level to be a reasonable of the bandgap along the channel are considered.
approach. In addition, it has been shown that in Fig. 4 shows a comparison between the pristine
contrast to a single defect, multiple defects lead to DOS and the local DOS averaged along the channel
a larger variability of the device performance. as well as the transmission spectra for different
defect densities. The higher the number of local
R ESULTS AND D ISCUSSION distortions, the stronger is the smoothing of the
Atomistic simulations are performed in order to sharp DOS peak and the lower is the transmission as
analyze the impact of randomly distributed mono- predicted by the atomistic simulation results shown
vacancies for different defect densities. For each in Fig. 2.
defect density, the effective band structure (shown The impact of defects on the CNTFET behavior
in Fig. 1) is extracted by means of the method is shown in Fig. 6 for 1, 3 and 5 defects. According
described in [2]. The fuzzy shape of the band to Fig. 6 the variations of the on-current Ion as well
structure of the defective tube indicates that within as the threshold voltage Vth is larger for a higher
a specific energy range, less states are available number of defects.
compared to the number of states for a pristine CNT. ACKNOWLEDGMENT
In addition, atomistic CNTFET (10 nm long
channel, (14,0)-tube) simulations are performed by The authors acknowledge the Cfaed and the DFG
means of the recursive Green’s function technique projects CL384/2, SCHR95/6 for financial support.
described in [3]. Defects are introduced by changing R EFERENCES
the on-site energies of the p-orbitals. The aver- [1] N. Neophytou, et al., APL, 88(24), 2006.
age impact of randomly distributed defects on the [2] V. Popescu et al., PRL, 10(104), 236403, 2010.
density of states (DOS) and transmission for the [3] C. Lewenkopf et al., J. Comp. Elect., 12(2), 203–231, 2013.
reference transistor is shown in Fig. 2. Obviously, [4] M. Claus, et al., J. Comp. Elect., 13(2), 689–700, 2014.
ISBN 978-0-692-50554-0 157

1.0
1.0 0.75
0.5
0.5
E − EF (eV)
E − Ef (eV)
0.25 0
0.0 0.0 5
Ef 10
-0.25
-0.5 -0.5
-0.75
-1.0
-1.0
-0.3 -0.2 -0.1 0.0 0.1 0.2 0.3 0 1 2
k (nm −1
) DOS (a.u.) T (−)
Fig. 1. Atomistically calculated band structure of a (14,0) tube Fig. 4. Local DOS averaged along the channel and transmis-
with (grey contour plot) and without (red lines) 5 defects. sion spectra for different number of defects calculated at the
device level.
1.0
0.75 10 1
3
0.5
5
E − EF (eV)
0.25 1
0 0
0.0 5 Id (µA)
10
-0.25 10−1
-0.5
-0.75 10−2
-1.0
0 1 2
-1.0 -0.8 -0.6 -0.4 -0.2 0.0 0.2 0.4
DOS (a.u.) T (−) Vgs (V)
Fig. 2. Atomistically calculated average DOS and transmission Fig. 5. Averaged transfer characteristic for different number
spectrum T of a (14,0) CNT with 0, 5 and 10 defects of defects. The error bars indicate the variability of the transfer
curves with respect to different defect pattern.
1.0
15
0.75
1
0.5 3
E − EF (eV)
0.25 5
pristine 10
0.0 pattern 1
#
-0.25 pattern 2
average
-0.5 5
-0.75
-1.0
0 1 2 3 4
0
T (−) -0.5 -0.45 -0.4 -0.35 1.0 1.2 1.4 1.6 1.8
Vth (V) Ion (µA)
Fig. 3. Atomistically calculated transmission spectra T for
0 and 3 defects for different arbitrary defect pattern and the Fig. 6. Distribution of Vth and Ion for different number of
average for 100 arbitrary defect pattern. defects randomly distributed along the channel.
158
Simulation of transport through a cavity defined
in graphene with electrostatic lithography
P. Marconcini∗ , M. Macucci∗ , M. R. Connolly†‡ , and E. D. Herbschleb†
∗ Dipartimento di Ingegneria dell’Informazione, Università di Pisa, Via G. Caruso 16, I-56122 Pisa, Italy
†Cavendish Laboratory, Department of Physics, University of Cambridge, Cambridge, CB3 0HE, UK
‡ National Physical Laboratory, Hampton Road, Teddington TW11 0LW, UK
I NTRODUCTION N UMERICAL RESULTS

The presence of a cavity-shaped potential leads
We present numerical simulations that we have
to Fabry-Pérot resonances in the behavior of the
developed to reproduce and understand the results
resistance as a function of VBG (Fig. 4). We have
of the measurements which we have recently per-
tried different potential profiles at the reference bias
formed [1] on a graphene device (Fig. 1) in which a
point, in order to find the best agreement with the
cavity-shaped potential, orthogonal to the transport
experimental data for the set of voltages VBG at
direction (Fig. 2), has been induced with electro-
which the resonance peaks appear (Fig. 5). The best
static lithography. The resistance of the sample has
fit has been achieved for a Lorentzian shaped cavity
been measured both as a function of the backgate
with a width of about 200 nm. We have then per-
voltage VBG , and of the position of a biased probe
formed a numerical simulation of a scanning probe
scanned at a fixed distance from the flake.
spectroscopy experiment, computing the resistance
of the device as a function of the position of a
N UMERICAL METHOD biased probe. A plot of such resistance as a function
of the probe position is reported in Fig. 6. The
In order to explore a large range of possible measured resistance maps are well reproduced if
potential landscapes and backgate voltage values, we assume the presence of disorder in the potential,
we have used a simplified numerical approach. As- such that the cavity is split into multiple subcavities
suming to know the potential profile at a reference in parallel (Fig. 2). The presence of the biased
bias point, we have determined its variation as a probe, altering the potential in the subcavity under
function of the gate and probe voltages by analyt- it, shifts the value of VBG for which the Fabry-Pérot
ically enforcing the self-consistent solution of two resonances take place in that subcavity and thus
approximate relations [2]. The first one connects the gives rise to the circular features observed in the
local charge density to the graphene potential via the scanning probe maps. The charge accumulation at
local density of states, which, for a slowly spatially- the edges of the flake introduces further oscillations
varying potential, has been expressed shifting the in the behavior of the resistance as a function of
density of states by the local value of the potential. VBG , with a different periodicity due to the stronger
The second one, instead, relates the variations of the capacitive coupling between the graphene and the
charge density, potential, and gate voltages through gate (Fig. 4).
a simple capacitive model (Fig. 3). Under proper
conditions, this set of relations is equivalent to a R EFERENCES
quadratic equation, which can be quickly solved [1] E. D. Herbschleb et al., arXiv:1408.1925.
for each point of the flake [2]. Once the potential [2] P. Marconcini, M. Macucci, IET Circ. Device. Syst. 9, 30
(2015).
profile was obtained, a transport calculation has
[3] D. Logoteta, P. Marconcini, C. Bonati, M. Fagotti, M.
been performed, obtaining the overall resistance Macucci, Phys. Rev. E 89, 063309 (2014).
with a recursive scattering matrix approach [3].
ISBN 978-0-692-50554-0 159

0.4
50
∆ R (Ω )
0
−50
0.35 −10 −8 −6 −4 −2 0 2 4
Probe
Graphene flake VBG (V)
R (K Ω)
PMMA
Silicon oxide 0.3
Doped silicon (backgate)

0.25
1 1.5 2 2.5 3 3.5 4 4.5
VBG (V)
Fig. 1. Sketch of the simulated device.

Fig. 4. Computed behavior of the resistance as a function of
VBG for a single Lorentzian cavity (red solid curve) and for
multiple subcavities in parallel without (black dotted curve) and
with (blue dashed curve) charge accumulation at the edges. The
Potential energy (eV) inset reports the experimental results, in terms of the resistance
variation ∆R due to the cavity.
−0.05
0.25
0.15
0.05
−0.1
0.2
0.1
6
0
4
2
1000 VBG (V) 0
Transport
direction
x (nm)
−2
500
−4
0 −6 Experimental
1000
500
0
2500
2000
1500
Numerical
−8
−10
1 2 3 4 5 6 7 8 9 10
y (nm)
Peak index
Fig. 2. Potential profile in the graphene flake, consisting

Fig. 5. Values of VBG for which the Fabry-Pérot resonances
of a cavity divided by low longitudinal walls into multiple
appear, reported as a function of the peak index (the experi-
subcavities in parallel.
mental and numerical results are compared).
1000 0.36
x (nm)
a) 500
Probe 0.34
0
2500
2000
1500
1000
500
0
0.32
0.03 R (K Ω)
CP (mF/m 2 )
y (nm) 0.3
CP (mF/m 2 )
0.02
0.03
0.02
1000
x (nm)
1000 0.01 0.28

0.01 b) 500
0 500
2500 2000 x (nm) 0
1500 1000
0 0 0.26
500
2500
2000
1500
1000
500
0
y (nm) 0
Graphene flake
y (nm)
Fig. 3. The electrostatic coupling between the probe and each
point of the graphene flake is modeled by means of a spatially- Fig. 6. Scanning probe resistance maps computed (a) for a
varying capacitance. single Lorentzian cavity; (b) for a cavity divided into multiple
subcavities in parallel.
160
Numerical Analysis of Electric Field Enhancement
in ZnO film with Plasmonic Au quantum dots
Min S. Choi, Shripriya Poduri, Mitra Dutta, Michael A. Stroscio*
Department of Electrical and Computer Engineering, University of Illinois at Chicago
Chicago, IL 60607, USA
*e-mail: [email protected]
reference [8]. Two dimensional approximation is

INTRODUCTION
used, and the nanostructures will be treated as
ZnO nanostructures have attracted interests for infinitely long cylinders in z direction, where z=0
broad ranges of fields of study due to their unique plane is the domain of calculations.
optical and electrical properties [1-3]. Especially,
the spontaneous polarization in the nanostructure RESULTS AND DISCUSSION
would produce the permanent strong static electric Fig 2 and Fig 3 show the meshgrid of ZnO film
field that can be applied to many studies. Also, their with and without two Au quantum dots in air,
strong polarizability also make them a suitable respectively. The comparison between maximum
candidate for many applications such as ion channel electric field amplitude of ZnO film with Au
modulation, photodetector and LEDs in UV range. quantum dots and without Au quantum dots versus
In this abstract, the plasmonic enhancement of the photon wavelength (TM mode of incident field) is
induced electric field produced from ZnO shown in Fig 4. In general, the electric field is
nanostructures is studied by numerical analysis. enhanced throughout the entire range. A high peak
ZnO nanostructure of interest is exhibited in Fig 1. in UV range indicates that the Au quantum dots can
significantly enhance the UV radiation from ZnO.
MODEL ACKNOWLEDGMENT
Finite Element Method (FEM) is used for all Partial support from AFOSR and NSF is
numerical calculations done in this abstract, using a acknowledged.
modified version of open-source FEM code,
REFERENCES
JFEM2D [4]. The details on FEM code is described
[1] Gogurla, N. et al. Multifunctional Au-ZnO Plasmonic
in a code author’s article [5]. For ZnO dielectric Nanostructures for Enhanced UV Photodetector and Room
functions, Loudon’s model of uniaxial crystals Temperature NO Sensing Devices, Scientific Reports 4,
gives[6]: 6483 (2014).
𝜔𝜔2 −𝜔𝜔2 ⊥ [2] Bai, S. et al. High-Performance Integrated ZnO Nanowire
𝜖𝜖 ⊥ (𝜔𝜔) = 𝜖𝜖∞
⊥ 𝐿𝐿𝑂𝑂
(1) UV Sensors on Rigid and Flexible Substrates. Adv. Funct.
𝜔𝜔2 −𝜔𝜔2 ⊥
𝑇𝑇𝑂𝑂 Mater. 21, 4464–4469 (2011).
[3] Soci, C. et al. ZnO Nanowire UV Photodetectors with High
⊥
where 𝜖𝜖∞ is the high frequency dielectric constant Internal Gain. Nano Lett. 7, 1003–1009 (2007).
in the direction along the c-axis, 𝜔𝜔𝐿𝐿𝑂𝑂⊥ and 𝜔𝜔 𝑇𝑇𝑂𝑂⊥ [4] J. M. McMahon, JFEM2D, Available from:
<http://www.thecomputationalphysicist.com> (2015).
are the LO and TO phonon frequencies. The [5] McMahon, J. et al, Gold nanoparticle dimer plasmonics:
parameters used in the calculation are summarized finite element method calculations of the electromagnetic
in Table 1. The Lorentz-Drude model of dielectric enhancement to surface-enhanced Raman spectroscopy,
functions is used for Au quantum dots, which is Anal. Bioanal. Chem. 394:1819-1825 (2009).
[6] Stroscio, M. A., Dutta, M., Phonons in Nanostructures,
given by Cambridge: Cambridge UP, 2001. Print.
𝜔𝜔2 2
𝑓𝑓𝑗𝑗 𝜔𝜔𝑝𝑝
𝑝𝑝
𝜖𝜖(𝜔𝜔) = 1 − 𝑓𝑓1 𝜔𝜔2 +𝑖𝑖𝛤𝛤 𝜔𝜔
+ ∑𝑛𝑛𝑗𝑗=2 𝜔𝜔2 2 −𝑖𝑖𝛤𝛤 (2) [7] Hanada, T., Basic Properties of ZnO, GaN, and Related
1 0,𝑗𝑗 −𝜔𝜔 𝑗𝑗 𝜔𝜔 Materials, Advances in Materials Research, Volume 12, 1-
where fj are oscillator strength of jth oscillator, 𝜔𝜔𝑝𝑝 19 (2009)
[8] Zeman, E. J., Schatz, G. C., An accurate electromagnetic
is plasma frequency, Γj are damping constants of jth theory study of surface enhancement factors for silver,
oscillator, 𝜔𝜔0,𝑗𝑗 are resonant frequencies of jth gold, copper, lithium, sodium, aluminum, gallium, indium,
oscillator. The parameters of Au are summarized in zinc, and cadmium, J. Phys. Chem., 91 (3), pp 634–643
(1987)
ISBN 978-0-692-50554-0 161

Fig. 1. Schematic diagram of ZnO thin film assembled by layer-
by-layer technique with PDDA polymer with Au quantum dots.
Fig. 4. Maximum electric field amplitude of ZnO film with Au
quantum dots and without Au quantum dots
Fig. 2. Meshgrid of ZnO film with two Au quantum dots in air.

Red with a “ZnO Film” label indicates ZnO film, Red with
Fig. 5. Scattering, absorption, and extinction spectra of ZnO
“Au” labels are gold quantum dots, and green presents the air.
film with Au quantum dots
ϵ⊥ (0) ϵ⊥ (∞) ωLO ωTO

ZnO 8.49 7.40 72.8/ℏ 51.0/ℏ
Table 1. Dielectric function parameters for ZnO
Fig. 3. Meshgrid of ZnO film in air. Red indicates ZnO film

and green presents the air
162
Numerical Analysis of Scattering Cross Section of
Non-Precious Metallic Plasmonic Nanodisks
Min S. Choi, Mitra Dutta, and Michael A. Stroscio
Department of Electrical and Computer Engineering
University of Illinois at Chicago
Chicago, IL 60607, USA
I NTRODUCTION
n

The recent advancement in plasmonics provided re- f0 ωp2 fj ωp2
(ω) = 1 −
˜ − , (2)
searchers with enormous possibilities. However, the plas- ω(ω − iΓf ) ω 2 − ωj2 + iΓb,j
j=1
monic materials currently being used are mostly precious
metals like gold and silver. In this work, the scattering which is referred as Lorentz-Drude model. [2] The
cross section (SCS) of various sized nanodisks made of parameters of Ag, Al, Ni, and Cu used in this work are
three non-precious metals, aluminum, nickel, and copper, taken from reference [2], and are plotted in Figure 1.
are examined and compared to the result of silver. Over-
B. Numerical computation of scattering cross section
all, all three metals show the same red-shifting pattern as
the diameter increases. Unlike the Ag nanodisk, which One of the most interesting quantities regarding the
tends to present the highest SCS in the ultraviolet range, optical response of metals is the scattering cross section
Al, Ni, and Cu nanodisks exhibit the constantly high SCS (SCS). The SCS is in indicator of how strongly plas-
throughout the visible region. Also the amplitudes of the mons interact with the incident electromagnetic waves,
SCS of the Al, Ni, and Cu nanodisks are comparable especially in the visible range. While it is possible to
with the Ag nanodisk in the visible region. These results calculate the SCS at each wavelength of electromagnetic
suggest of the cost effective replacement of precious waves in the interested range, it is more efficient to
metals in application of plasmonics-based optics. apply a Fourier transform on a response spectrum to
a short pulse to get the broad range spectra. [3] The
I. T HEORY scattered power spectrum of an object through a surface
that encloses it is given by
A. Drude-Lorentz model
The high reflectivity of light from metals is due to the
abundance of free electrons. The collective movement Pscat (ω) = R S ([Eω (r) − Eω,0 (r)]∗
of these free electrons, plasmons, responding to eletro- × [Hω (r) − Hω,0 (r)]) · dA (3)
magnetic fields, can be fully described by the dielectric
The SCS, Cscat , can be obtained by taking the ratio of
function. [1] Since the free electrons play a role as
Pscat (ω) to the incident irradiance, I0 :
significant as bound electrons, the dielectric function of
metals can be described mathematically by Pscat (ω)
Cscat = (4)
I0
˜ (f ree) (ω) + ˜
(ω) = ˜ (bound) (ω), (1)
D ISCUSSION
where ˜(f ree) (ω) is the free electron contribution to Overall, all three metals show the same red-shift pat-
the dielectric function and ˜ (bound) (ω) is the bound- tern as the diameter increases. Unlike the Ag nanodisk,
electron contribution. The free electron contribution can which tends to present the highest SCS in the ultraviolet
be described by Drude model and the bound-electron range, Al, Ni, and Cu nanodisks exibit the constantly
contribution can be described by Lorentz model. As a high SCS throughout the visible region. Also the am-
result, Equation (1) can be written as plitudes of the SCS of the Al, Ni, and Cu nanodisks
ISBN 978-0-692-50554-0 163

(≈ 2.2 × 104 atomic unit for 70-nm-diameter nanodisks)
are comparable with the Ag nanodisk (≈ 3 × 104 atomic
unit). These results can lead to a conclusion that non-
precious metals such as aluminum, nickel, and copper
can be used as an alternative to silver without sacrifying
the efficiency in the visible range.
ACKNOWLEDGEMENT
This material is based upon work supported by, or in
part under AFOSR Grant FA9550-11-1-0271.
R EFERENCES (a) Ag
[1] Ru, E., and Etchegoin, P. ”Principles of surface-enhanced Ra-
man spectroscopy and related plasmonic effects.” Amsterdam:
Elsevier (2009)
[2] Rakic et al., ”Optical properties of metallic films for vertical-
cavity optoelectronic devices”, Applied Optics, Vol. 37, 22, 5271
(1998)
[3] Ardavan F. Oskooi, David Roundy, Mihai Ibanescu, Peter
Bermel, J. D. Joannopoulos, and Steven G. Johnson, MEEP: A
flexible free-software package for electromagnetic simulations
by the FDTD method, Computer Physics Communications, 181,
687702 (2010)
(b) Al
(c) Cu
(d) Ni
Fig. 1. Scattering cross section as a function of light frequency for

(a) Ag, (b) Al, (c) Cu, and (d) Ni nanodisks with diameters of 70 nm,
90 nm, and 110 nm, respectively, from left to right in each material
164
Non-local optical response in silver-silicon-silver
heterostructure system
Min S. Choi
Department of Electrical and Computer Engineering, University of Illinois at Chicago, Chicago, IL 60607, USA
Gautam D. Stroscio
Division of Chemistry and Chemical Engineering, California Institute of Technology, Pasadena, California 91125, USA
I NTRODUCTION of uncertainty in momentum space with high confinement

Surface plasmon polaritons (SPP) in a thin metal film in spatial space. In this abstract, we examine the indirect
has been studied extensively in recent years due to their optical absorption in a silicon nanowire coupled with silver
unique properties such as a huge increase of intensity [1], nanocavity.
non-locality [2], etc. Recently, Trolle et al. [4] suggested a E LECTRIC FIELD IN AG -S I -AG HETEROSTRUCTURE
mechanism of optical excitation in silicon with photon energy
In this abstract, we perform a one dimensional analysis to
below the direct bandgap using the non-locality of SPP in
calculate the optical power absorption rate in the structure
a silver thin film embedded in crystalline silicon. With a
shown in Figure 1. By solving the 2D wave equation assuming
quantum mechanical point of view, this can be explained
E(r, t) = E(r)eiωt , we get
by the uncertainty principle, which asserts a wider spread


 −iA ωs 1
1 0
e−s1 z eiβx
z > h/2
 s
Ex (z) = i A2 ω1 0 2
2− s11 e(−s1 −s2 )h/2 es2 z − s22 + s11 e(−s1 +s2 )h/2 e−s2 z eiβx −h/2 < z < h/2 (1)


 i A 1 s1 2 s2 s1 s2 s1
2 ω0 s2 1 2 − 1 e −s2 h
− 2 + 1 e s2 h
es1 z eiβx z < −h/2
and


 −A ωβ1 0 e−s 1 z iβx
e z > h/2

A 1 β s2 s1 s2 s1
Ez (z) = − 2 ω0 s2 2 − 1 e (−s1 −s2 )h/2 s2 z
e + 2 + 1 e (−s1 +s2 )h/2 −s2 z
e eiβx −h/2 < z < h/2 (2)


 − A β 2 s2
− s1 e−s2 h − s2 + s1 es2 h es1 z eiβx z < −h/2
2 ω0 s2 1 2 1 2 1
where A is incident field intensity, si and β decay constants, The time-averaged optical power absorption rate is given by
i dielectric functions, and ω the energy. With appropriate
ω
boundary conditions in each region, we get a dispersion Pabs = P(r, ω) · E∗ (r, ω)dr (5)
relation 2
s2i = β 2 − k02 i (i = 1, 2), (3) where the polarization density, P, is given by P(r, ω) =
0 χ̃E(r, ω). The non-local optical power absorption can be
and the continuity of Ex at z = −h/2 gives the dispersion computed by Fourier-transforming the equation (5) to get the
equation sum of contributions of SPP over the momentum space.
s 2 1 − s 1 2 e s1 h = 0 (4)
R ESULTS AND D ISCUSSION
which has to be satisfied in order for solutions to exist.
The equations () and () are solved numerically. The frequency- The non-local optical power absorption in silicon and silver
and wavevector- dependent dielectric function of silicon, 2 , is are plotted in Figures (2) and (3), respectively. In silicon,
obtained empirically using time-dependent density functional several distinct peaks are appeared at h̄ω =1.4 eV, 2.35 eV, 2.5
theory [5], and the dielectric function of silver, 1 , is calculated eV, and 2.7 eV. These results overlap well with our previous
using Drude-Lorentz model. [6] calculation of the local density of states of plasmons in silver
ISBN 978-0-692-50554-0 165

[7], and also with an experiment of hot luminescence from a
similiar structure. [8]
R EFERENCES
[1] M. Kerker et al. ”Surface enhanced Raman scattering (SERS) by
molecules absorbed at spherical particles.”, Appl. Optics. Vol 19, No 19,
3373 (1980).
[2] Ru, E., and Etchegoin, P. ”Principles of surface-enhanced Raman spec-
troscopy and related plasmonic effects.” Amsterdam: Elsevier (2009).
[3] S. Maier. Plasmonics: Fundamentals and Applications. New York:
Springer, (2007).
[4] M. Trolle. ”Microscopic Theory of Linear & Nonlinear Optical Response:
Zinc-blende Semiconductors.” Thesis. Aalborg University, (2011).
[5] ”The Elk FP-LAPW Code.” The Elk FP-LAPW Code. Web.
¡http://elk.sourceforge.net/¿. Fig. 2: Power absorption rate in silicon nanowire vs energy of photon
[6] Rakic et al., ”Optical properties of metallic films for vertical-cavity
optoelectronic devices”, Applied Optics, Vol. 37, 22, 5271 (1998)
[7] Choi et al., ”Numerial analysis of local density of state of plasmons
in Si-SiO2-silver nanocavity system” Proceedings of IEEE International
Workshop on Computational Electronics. (2014)
[8] Chang-Hee Cho, Carlos O. Aspetti, Joohee Park and Ritesh Agarwal,
Nature Photonics, 2013.
Fig. 3: Power absorption rate in silver vs energy of photon
Fig. 1: Si-Ag-Si heterostructure
166
Thermoelectric power factor optimization in
nanocomposites by energy filtering using NEGF
M. Thesberg1, M. Pourfath1, N. Neophytou2 and H. Kosina1
1
Institute for Microelectronics, Technical University of Vienna, TU Wien
Gußhausstraße 27-29/E360, A-1040 Vienna, Austria.
2
School of Engineering, University of Warwick, Coventry, CV4 7AL, UK
INTRODUCTION under optimal conditions the transport in the wells

needs to be semi-ballistic, where carriers only lose
The thermoelectric (TE) performance of part of their energy before they reach the next
materials is determined by the figure of merit barrier [4, 5]. In addition, it was also indicated that
ZT=σS2/κ, where σ denotes the electrical ideally the barrier height needs to extend ~kBT
conductivity, S the Seebeck coefficient and κ the above the Fermi level [4, 5]. Thus, in this work we
thermal conductivity. Large improvements in ZT calibrate the geometry, electron-phonon scattering,
have recently been reported in nanoscale materials Fermi level, and barrier height for these optimal
due to drastic reduction in κ [1]. On the other hand, conditions.
much less success has been achieved in improving Once this is done, we proceed by investigating
the power factor (σS2), and ZT still remains low. the performance of energy filtering processes under
Energy filtering in nanocomposite materials with statistical fluctuations in the design parameters.
embedded potential barriers (VB) is a promising way The first parameter we examine is the width of the
to improve σS2 via improvements in the Seebeck barrier W. Figure 2 shows the power factor versus
coefficient [2, 3]. Indeed, the improvement in the W. We can observe that the barriers need to be thick
Seebeck coefficient of nanometers-size layer enough to prevent tunneling (which is detrimental to
superlattices has been demonstrated in several S and could cause up to ~40% degradation in
experimental works. Significant benefits to the performance), but thin enough for reduced
overall power factor, however, were never observed resistivity (so ~2-3nm), since the carrier energy and
in these structures due to large reductions in σ. In momentum can relax on top of the barriers and
this work, we use the Non-Equilibrium Green’s acquire reduced velocities. The next parameter we
Function (NEGF) method to illustrate the design consider is the actual shape of the barrier. In
details under which improvements in σS2 can be practice, an ideal rectangular barrier would not be
achieved by energy filtering. We further achievable, thus we examine the influence of
demonstrate that variation of the design parameters, deviations from the rectangular shape on the
and most importantly in the barrier heights is a performance. Figure 3 shows that the rectangular
strong detrimental mechanism which can take away barriers are ideal (left side), which shows that ~30%
most of the energy filtering benefits. improvement can be achieved compared to the bulk
METHOD AND DISCUSSION TE material case. As we deviate from that shape the
We use the NEGF method in the effective mass power factor drops (approaching the bulk case –
approximation, including both acoustic and optical right side). Finally, the last parameter we examine,
phonon scattering. Figure 1 illustrates the simulated is fluctuations in the height of the potential barriers
1D channel geometry. Figure 1a shows the channel VB. The results are shown in Fig. 4 (black line). In
as a series of potential barriers, Fig. 1b shows the this case, we vary the barrier heights along the
transport path according to a Gaussian distribution.
extracted local density of states LDOS(E,x) from
As the variation increases, a large drop is observed
NEGF, Fig. 1c the charge density in the channel,
and Fig. 1d the current spectrum and how it in the power factor (black line). We perform the
same studies for variations in the barrier position
fluctuates in energy during emission / absorption of
(blue line) and the barrier width (red line), which
optical phonons. Previous works have indicated that
introduce only small power factor reduction.
ISBN 978-0-692-50554-0 167

CONCLUSION
Using the NEGF method we computed the
thermoelectric power factor in nanocomposite
channels in the presence of energy barriers. We
show that ideally, power factor improvements up to
30% can be achieved using energy filtering under
optimal conditions. However, we show that this
achievement is improvement is sensitive to structure
imperfections. Fluctuations in the barrier width and
well size do not affect performance significantly,
but fluctuations in the barrier shape and most
importantly the barrier height (even of the order of Fig. 2. Power factor versus barrier width. The optimal barrier
5meV) could take away most of the power factor width is ~3nm, which is thick enough to prevent tunnelling, but
improvements, and therefore, need to be avoided.
thin enough to keep the electrical resistance from barriers low.
ACKNOWLEDGEMENT
We acknowledge the Vienna Scientific
Computing Cluster for computational resources and
funding from the Austrian Science Fund FWF,
project code P25368-N30.
REFERENCES
[1] R. Venkatasubramanian, et al., Nature, 413, 597, 2001.
[2] D. Vashaee and A. Shakouri, Phys. Rev. Lett. 92,
106103,2004.
[3] N. Neophytou and H. Kosina. J. Appl. Phys. 114, 044315,
2013.
[4] R. Kim and M. S. Lundstrom. J. Appl. Phys. 111, 024508, Fig. 3. Power factor versus barrier shape, defined as an
2012.
exponentially decaying profile, described by a decay length, ξ,
[5] N. Neophytou et al, Nanotechnology, 24, 205402, 2013.
from the top of the barrier. The limit ξ=0 corresponds to a
square barrier (left side), which is found to be the optimal one.
Fig. 1. Sample data for a nanocomposite channel. (a) The

potential profile of the barriers in the channel with width of
4nm and height of 0.16 eV. (b) The local density of states in the Fig. 4. Power factor (σS2) versus statistical variation of the
channel. (c) The charge density. (d) The current density versus barrier placement (blue line), width (red line), and height (black
position (colormap). Superimposed on the image are the line) along the transport path. It is clear that performance is
potential barriers and the carriers energy expectation value <E>. most substantially degraded by barrier height variation.
168
The Impact of the Ring Shaped Valence Band in
Few-Layer III-VI Materials on FET Operation
Protik Das∗ , Gen Yin∗, Somaia Sarwat Sylvia,
Khairul Alam† , Darshana Wickramaratne∗ , and Roger K. Lake∗
∗ Department of Electrical & Computer Engineering, University of California, Riverside, CA 92521
† Department of Electrical & Electronic Engineering, East West University, Dhaka, Bangladesh.
A BSTRACT an order of magnitude larger for a doping density

of 5 × 1020 cm−3 . The drain current versus gate
Mexican hat shaped dispersions are relatively voltage with Vdd = 0.3 V calculated for different
common in few-layered two-dimensional materi- doping densities is shown in Fig. 4.
als. In one to four monolayers of the group-III We have also calculated the screened coulomb
chalcogenides (GaS, GaSe, InS, InSe) the valence scattering time resulting from a Mexican hat disper-
band undergoes a band inversion from parabolic sion and a parabolic dispersion from the Boltzmann
to a Mexican hat dispersion [1]. This Mexican hat transport equation as defined below
dispersion results in a singularity in the density of 2π
1 N e4 √D(ǫ α) v(kα ).v(k)
states at the band edge. This enhances the thermo- τα = h̄ 4ǫ2 0 α 2
. 1 − v2 (k)
q0 +β 2
electric properties, however the effect on field effect

where α is the number of bands crossing k, N is the
transistor performance has not yet been investigated.
number of charged impurity scatterers per area, D
To evaluate the impact of this ring shaped disper- (ǫα ) is the density of states, β is the transition wave
sion on FET performance, we use a top of the bar- vector, and q0 is the screening factor. As shown
rier FET model. The physical gate length, effective by Fig. 5, the room temperature relaxation time
oxide thickness and power supply voltage for the for the Mexican hat dispersion is generally longer
simulated devices are 12.8 nm, 0.68nm, and 0.3V than that resulting from a parabolic dispersion.
respectively, following the low-voltage parameters
This result can be understood from the momentum
described by Nikonov and Young [2]. The simulated
dependence of the matrix element. In the Mexican
device is shown in Fig. 1. To model the electrostatic hat dispersion, larger momentum transfer is required
potential along the channel of the device we solve a for backscattering. From the calculated relaxation
2-D Poisson equation over the simulation domain. times, we will determine the effect of the finite
The density of states and density of modes calcu- mean free paths on the FET performance.
lated from the Mexican hat dispersion described in Acknowledgement: This work is supported in part
Ref. [1] are shown in Figs. 2 and 3, respectively. by FAME, one of six centers of STARnet, a Semi-
The density of modes is used as input for the current
conductor Research Corporation program sponsored
calculation. The performance characteristics of the
by MARCO and DARPA.
devices are benchmarked using the 15nm node low-
voltage criteria defined by Nikonov and Young [2] R EFERENCES
and compared to other devices. [1] D. Wickramaratne, F. Zahid, and R. K. Lake, “Mexican hat
The target ON-current for the low-voltage 15nm and rashba bands in few-layer van der waals materials,”
node of Si CMOS is 2uA/um. For the Mexican arXiv preprint arXiv:1412.2090v1, 2014.
[2] D. E. Nikonov and I. A. Young, “Overview of beyond-cmos
hat shaped dispersion the ON-current for doping devices and a uniform methodology for their benchmark-
density 1 × 1019 cm−3 is 20.4 times larger than the ing,” Proceedings of the IEEE, 2013.
target ON-current. The ON-current is reduced as the
doping density increases but it remains more than
ISBN 978-0-692-50554-0 169

×10 8
Density of Modes (m-1)

15
10
0
0 0.05 0.1 0.15
Energy (eV)
Fig. 3. Density of modes calculated from the analytical model

Fig. 1. Cross section of the device that was used for simulation. of Ref. [1] using an effective mass 0.4 and ǫ0 = 0.1 .
The dimensions used are Lg = 12.8 nm, Lus = 60 nm, Lud =
60 nm. The substrate is SiO2 (tbox = 10 nm) and the gate
dielectric is HfO2 (tox = 4.36 nm) with dielectric constant of 10 2
19 -3
3.9 and 25 respectively. The oxide extension shown is SiO2 1×10 cm
8×1019 cm-3
(tox−ex ) which is 6 nm. The channel is undoped and we have
Current (uA/um)
considered an effective mass of the channel as 0.4 which is the 20 -3
5×10 cm
effective mass of GaS from ab-initio calculation. 10 0
10 -2
10 -4
0 0.1 0.2 0.3
Voltage (V)
×10 19
3 Fig. 4. Current versus gate voltage with Vdd = 0.3 V for
Density of States (eV-1 m-2)
different source and drain doping densities.

2.5
(a) 5 (b) 0.20
2 MH
Parabolic
MH
Parabolic
4
0.15
1.5
Energy (eV)
3
τ (ps)
2
1 0.10
2
0.5 1 0.05 1 2
1
0
0 0.05 0.1 0.15 0
0.00 0.05 0.10
0.00
0.0 0.1 0.2 0.3
Energy (eV) Energy (eV) k (1/Å)
Fig. 2. Density of states calculated from the analytical model Fig. 5. (a) Momentum relaxation times calculated from
of Ref. [1] using an effective mass 0.4 and ǫ0 = 0.1 . Coulomb impurity scattering and (b) energy-momentum rela-
tions comparing parabolic and Mexican hat dispersions. The
relaxation times for states on the two different branches of
the Mexican hat dispersion are labeled in (a). The density
of impurity and electron density used are 1 × 10−5 Å−2 and
1 × 10−11 cm−2 respectively.
170
The Impact of Lead Geometry and Discrete
Doping on NWFET Operation
S. Berrada, N. Cavassilas, L. Raymond, M. Lannoo and M. Bescond

IM2NP, UMR CNRS 7334, Bât. IRPHE, Technopôle de Château-Gombert, 13384 Marseille, cedex 13, France
Si nanowire MOSFET (NWFET) is one of the curves (Fig. 4) show that subthreshold behavior
most promising device to reach the end of the has not been affected by the discrete aspect of the
technological roadmap. However due to the contin- doping in AR. This result is in line with the fact that
uous reduction of the channel length, resistance of subthreshold behavior is mainly controlled by the
access regions may appear as the main contributor channel characteristics, which remained unchanged
of the global resistance [1]. A physically consistent in both devices.
modeling of nanowire transistors then requires a However, as expected [4], the discrete dopants
realistic description of the open character of the degrade the on-current. Surprisingly, the degrada-
system via access regions. The aim of this work tion depends on the access region geometry since
is to emphasize the importance of considering both current reductions are 19 and 30% respectively for
realistic lead geometry and discrete doping. We use the UN and NUN at VGS =0.49 V.
a 3D real-space non-equilibrium Green’s function To understand the higher degradation in the NUN
(NEGF) approach where electron-phonon scattering device, let’s analyze the LDOS at this bias for both
is treated within the self-consistent Born approxima- transistors. In the UN device (Fig. 5), we can clearly
tion (SCBA) [2]. distinguish the two impurity levels (IL) localized
We consider two lead geometries for NWFETs around the impurity positions. Moreover, the IL are
with LG =8 nm as schematically shown in Fig. 1. close to the Fermi level (EF ) of their respective
We shall refer to the device with the uniform contacts and are strongly coupled to S and D, thus
transverse section in Fig. (1.a) by (UN), and by contributing efficiently to the injection and extrac-
(NUN) to its nonuniform counterpart (Fig. (1.b)). tion of carriers in the device. In NUN device, the IL
We first assumed for both devices a continuous is no longer a quasi-localized state of the coulombic
dopant concentration of 1020 cm−3 in the source (S), potential (Fig. 6). Indeed the larger cross-section
drain (D) and access regions (AR) - represented by reduces the impurity binding energy [5], leading
yellow regions in Fig. 1. ID -VGS curves show that to delocalized IL’s well above the Fermi levels of
both devices exhibit the same threshold voltages the contacts, reducing the injection efficiency and
Vth , subthreshold slopes and ON state currents resulting in a reduction of the ON current.
(Fig.2). This result is in accordance with the results This result shows the importance of consider-
previously obtained for double-gate devices [3]. ing both realistic lead geometry and discreteness
However, a concentration of 1020 cm−3 corre- of the dopants to provide an accurate modeling
sponds on average to less then two dopants in a of NWFETs, where access regions will have an
volume of 2×2×4 nm−3 . As a result, the commonly increasing influence.
accepted assumption of continuous doping in the R EFERENCES
access regions is no longer justified.
[1] Y. He et al., Appl. Phys. Lett. 105, 213502 (2014).
As a sound example we replace the continuous [2] H. Carrillo-Nunez et al., J. Appl. Phys. 116, 164505 (2014).
doping of the AR in both UN and NUN devices by [3] R. Venugopal et al., J. Appl. Phys. 95, 292 (2004).
one dopant exactly localized on the same positions [4] A. Asenov et al., IEEE-Trans. Elec. Dev. 61, 2745 (2014).
along the channel symmetry axis (Fig. 3). ID -VGS [5] M. Diarra et al., Phys. Rev. B 75, 045301 (2007).
ISBN 978-0-692-50554-0 171

Fig. 1. Considered gate-all-around NWFETs (a) UN : with
uniform cross-section of 2×2 nm2 and (b) NUN : non uniform
cross section. Same channel length LCH =8 nm and cross- Fig. 4. Figure 4 : ID − VGS with impurities placed in AR
section of 2×2 nm2 (green) are assumed for both devices. Gate along the channel symmetry axis at x = 5 nm and x=19 nm for
insulator is 1 nm thick. Source and drain are represented in red UN device (squares) and NUN device (stars). The curves have
and are 4 nm long, while the access regions are represented in been shifted to provide the same Vth .
yellow (with a length of 4 nm). Source (drain) cross-section
in NUN device is 3.6 × 3.6 nm2 . For all the simulations
VDS =0.3 V and T=300 K.
Fig. 5. LDOS at VGS =0.49 V for the UN device with

impurities placed in AR along the channel symmetry axis at x =
Fig. 2. ID − VGS with uniformly doped access regions for 5 nm and x=19 nm. White lines corresponds to the eigenvalues
UN (squares) and for NUN (stars) devices. of transverse valleys (mt in the transport direction) and white
dashed lines correspond to the two others. Fermi levels in
source and drain are indicated with red dashed lines.
Fig. 3. (a) Conduction band along transport axis located

at z=2.8 nm (middle of the device) of the NUN device with Fig. 6. Similar physical quantities as in Fig. 5 but for the
impurities placed in AR at x = 5 nm and x=19 nm. (b) NUN device.
corresponding electron density. VGS =0.49 V.
172
Study of transport in Multi-layered High
Electron Mobility Transistors
Samik Mukherjee, Prasad Sarangapani , Michael Povolotskyi and Gerhard Klimeck
School of Electrical and Computer Engineering, Purdue University, West Lafayette, IN, 47906, USA
voltage. The various dielectric layers have relative

ABSTRACT SUBMISSION
permittivity  i and contribute a capacitance
In recent days there is a competition for finding
alternatives to CMOS technology as CMOS  i 0
Ci  (1)
transistor scaling approaches its limits. Among ti
many of these competing new technologies is III- The total insulator capacitance and the gate and
V based HEMTS for logic applications. As semiconductor capacitance are given by :
channel lengths of HEMTs approach ballistic
1 1 1 1
lengths, it is therefore necessary to understand    (2)
such devices with quantum transport formalisms. Cins C1 C2 C3
In this paper we intend to apply Non-Equilibrium d  Qn  S  
Green’s function formalism to study short channel CS  (3)
HEMTs in ballistic regime. We also study C-V d S
characteristics of these devices before moving to 1 1 1
their Id-Vg characteristics   (4)
CG Cins CS
GEOMETRY & MATERIALS
The traditional HEMT structure consists of a
Qn  S   
qN2 D ln 1   nS  off  N 2 D  eq S kBT

single modulation doped junction. For the device (5)
under study we have a layered structure and a The symbol  S is the surface potential and nS the
spacer layer as well to prevent remote scattering
sheet charge density and N 2D is the density of
from dopants. The device structure is shown in fig
1. The 2 –dimensional electron gas is formed in states. The final C-V curves are shown in fig3.We
the InAs channel and is confined by the higher also find good agreement with experimental data.
barrier materials AlInAs(with Al ratio 0.48) and
GaInAs (with Ga ratio 0.47). The total channel b) Transport Calculations with NEGF
length is 30 nm and the region from 10nm to 20nm Here we calculate the Id Vs Vg characteristics for
is controlled by the gate. The respective thickness different sets of applied Vds. In our calculations
of each layer is depicted in fig 1.The gate material we assume transport within the channel is ballistic.
is Ti , although a definite work function cannot is While such a simulation has already been done by
not known at the interface of AlInAs and Ti. [1] , using formalism Quantum Transport through
Barrier Materials , for developing transport with
MODEL & METHODOLOGIES scattering it is necessary to adopt NEGF
formalism. Our aim here is to demonstrate that
a)Static C-V Calculations NEGF can well handle ballistic calculations in
We first begin with studying Capacitance Voltage HEMTs . Results of calculations are shown in fig
characteristics without source drain bias. We build 4.
a simple 1D model calculations with the following
DISCUSSION
set of formulae can help us calculate C-V
characteristics. A diagram in fig 2 illustrates our In our calculations with NEGF we have used an
model in mind and shows how the conduction effective mass approach. For our simulation we
band edge moves with change in applied gate have used the bulk effective mass in each of the
different materials. As pointed out by [1] , we have
ISBN 978-0-692-50554-0 173

to calculate the effective mass for this strained and
confined HEMT structure and then use it in NEGF
calculations. We have already used Valence Force
Field approaches to relax the structure and
calculate the E-k to extract an effective mass of
about 0.05 . The effective mass approximations,
needs to be benchmarked against full band
simulations. Thereafter we can also seek to
understand scattering phenomenon in long channel
HEMTs. As we notice in Fig 4 , the transistor is
already turned on at 0 gate voltage. Therefore the
work function at the gate needs to be better
understood.
CONCLUSION Fig. 1. Device Structure
In this work we have demonstrated that NEGF
formalism can provide means for simulation of
HEMT in ballistic regime. With further
developments we can progress to tackle scattering
and longer channel devices which were not
possible with the framework of QTBM as in [1].
ACKNOWLEDGMENT
This research was funded by Purdue University
and the Semiconductor Research Corporation. The
use of nanoHUB.org computational resources
operated by the Network for Computational
Nanotechnology funded by the US National
Science Foundation under Grant Nos. EEC-
0228390, EEC-1227110, EEC-0228390, EEC-
0634750, OCI-0438246, OCI-0832623 and OCI-
0721680 is gratefully acknowledged. Fig. 2. Bandstructure and charge density along gate to
NEMO5 developments were critically substrate
supported by an NSF Peta-Apps award OCI-
REFERENCES
[1] Mathieu Luisier, Neophytos Neophytou, Neerav Kharche,
and Gerhard Klimeck, “Full-Band and Atomisic
Simulation of Realistic 40 nm InAs HEMT,” IEDM, 2008.
Fig. 3. C-V characteritics
174
Fig. 4. Id Vs Vg Characteristics from NEGF simulation
ISBN 978-0-692-50554-0 175

176
Nanoplasmonics for all-optical control of
devices
Makiko Tsukamoto Quint* and Sayantani Ghosh*
Department of Physics, University of California, Merced, CA 95344, USA
of a densely packed layer of AuNPs to switch

directionality of a nematic LC film deposited on
We have designed and developed an all-optical top of the gold layer.
technique for reversible in-plane and out-of-plane Fig. 1 [2] shows the structure of a typical
switching of liquid crystal (LC) devices by sample. With only white light (WL) illumination,
leveraging the highly localized electric fields the LC transmission is dark (‘OFF’ state, Fig. 1a).
generated in the near-field region of a densely When the sample is additionally illuminated with
packed gold nanoparticle array when excited on- excitation resonant with the LPR of the AuNPs,
resonance with its localized surface plasmon the transmission gets brighter (‘ON’ state, Fig. 1b).
absorption. We observe this switching effect over This effect is caused by re-alignment of the LC
a broad temperature range, starting at room molecules due to the locally generated electric
temperature, and at very low power densities, less fields in the AuNP array, and is completely
than 0.03 W/cm2. Using discrete dipole reversible (Fig. 1c). Fig. 2b is a plot of the
simulations and control measurements, we variation of WL transmission voltage with time, as
establish spectral selectivity in this new and measured by a photo-detector when excitations are
interesting perspective for photonic application turned on and off (dashed lines), and clearly
using low light power. demonstrates the absence of the switching effect
INTRODUCTION when non-resonant excitation is used. The
The electron clouds in metallic nanoparticles simulations [3] in Figs. 2c and 2d support the
(NPs) exhibit coherent oscillations in response to experimental results. Fig. 3 shows the switching
incident time-varying electromagnetic radiation. times observed in our samples. With AuNP
These oscillations are known as localized surface deposited on one side, the switching times are on
plasmons (LSPs) [1], the confined counterparts of the order of seconds, but this is significantly
propagating surface plasmon-polaritons observed reduced when AuNPs are deposited on both
in continuous metallic thin films. LSPs, in turn, surfaces of the sample, to less than half a second.
scatter the incident electric field via dipolar Our findings offer a novel way to develop LC
interactions. At specific spectral ranges of the based devices with reduced ohmic losses by
incident field, the resultant scattered field has a removal of electrical hardware and spectrally-
significantly amplified intensity. This frequency selective response that opens the doors for entirely
(localized plasmon resonance, LPR) is set by a novel applications.
specific relation between the frequency-dependent REFERENCES
NP permittivity and the dielectric constant of the [1] Stefan A. Maier, Plasmonics: Fundamentals and
surrounding medium, known as Frohlich’s Applications (Springer, 2007).
condition [1]. In an ensemble of densely packed [2] Makiko T. Quint, Silverio Delgado, John. H. Paredes,
Zachary S. Nuno, Linda S. Hirst, and Sayantani Ghosh,
NPs, the far-field scattering is strongly suppressed.
All-optical switching of nematic liquid crystal films driven
The result is highly localized scattered fields in the by localized surface plasmons, Opt. Express 23, 6888
space between the NPs, creating what are (2015).
popularly known as “hot spots”. In this study, we [3] Jeremy Smith, Jacob Faucheaux, Sarah White, Abder
use the highly confined scattered field generated Rahman N. Sobh, John Feser, Prashant K. Jain, and Nahil
Sobh, "nanoDDSCAT," (DOI:10.4231/D3G73746M,
during resonant excitation in the interstitial regions (2015). https://nanohub.org/resources/dda
ISBN 978-0-692-50554-0 177

Fig. 1. Schematic of experimental set up demonstrating (a)
‘Off’ configuration with only white light (WL) beam, and (b)
‘On’ configuration with WL and resonant excitation (RE) Fig. 3. Transmission measured as a function of time. The
beams. CS: Cover slip; AuNPs: gold nanoparticles; LC: AuNPs are drop casted on (a) only the cover slip and (c) both
Liquid crystal; GS: glass slide (c) series of polarized the cover slip and the glass slide. The rise times at different
microscopy images of LC sample at 30°C starting with temperatures for (b) a cover slip and (d) both a cover slip and
resonant beam off, then turned on and then off again. glass slide samples. Time = 0 s denotes when resonant
excitation is switched on.
Fig. 2. (a) Extinction spectra of 30 nm AuNPs suspended in

its buffer solvent (blue) and in 5CB (red). (b) Transmission of
WL as measured by photo-detector for resonant (blue squares)
and non-resonant (black circles) excitation. Dashed lines
indicate when the excitation light is turned on and off. Near–
field simulations of scattered E-field intensity when incident
excitation is (c) resonant (532 nm) and (d) non-resonant (750
nm) with the AuNP extinction peak in (a). The incident light
is incident along the y-axis and polarized along the z-axis.
178
VFF-Monte Carlo Framework for Phonon
Transport in Nanostructures
S. Sundaresan1, M. Rashid1, T. Jayasekera2,3 and S. Ahmed1,3
1
Department of Electrical and Computer Engineering, 2Department of Physics, 3Materials Technology Center
Southern Illinois University at Carbondale
1230 Lincoln Drive, Carbondale, IL 62901, USA. (E-mail: [email protected]; [email protected])
ABSTRACT Temporal: to track the phonon drift in time; and 3)

Accurate modeling of non-equilibrium heat Spectral (frequency). After initializing the phonon
transport in nanostructures demands an appropriate ensembles and defining their properties (via the use
description of phonon dispersion relation and proper of various probability distribution functions and
treatment of various anharmonic effects. In this random numbers), the ensembles are allowed to drift
work, we develop and employ a coupled VFF from their respective position towards the colder end
molecular mechanics-Monte Carlo (VFF-MC) using an explicit 1st-order time integration equation.
platform to solve the phonon Boltzmann Transport As the ensembles drift inside the nanostructure, they
Equation (BTE) for modeling thermal conductivity in interact with the structural boundaries, atomic impu-
nanostructures having specified geometry. rities and other phonons.
SIMULATION MODEL RESULTS AND DISCUSSION
Modeling phonon transport via the BTE is adopted Fig. 2 shows the group velocity (vg) profiles for a
when the scattering rates of phonons that determine silicon film (thickness ~4 μm) obtained from the
the thermal conductivity vary significantly within a VFF and the empirical models. At this moment, VFF
distance comparable to their respective mean free is used only for the LA (faster) branch. It is seen that
paths. In this regard, the ability of the particle based the VFF computed vg for the LA branch is smaller
Monte Carlo (MC) method to emulate the phonon for ω < 5×1013 rad/s and then increases for higher
behavior and introduce randomness as close to reality frequencies when compared to the empirical model.
has further favored the use of MC based simulation Fig. 3 shows the thermal conductivity profiles plotted
in the recent years. The Ensemble Monte Carlo using the atomistic VFF and the empirical ap-
(EMC) scheme adopted in this work is similar to the proaches. Looking at the Figure it is found that, as
one described in Refs. [1][2]; yet the current work compared to the empirically fitted model, the thermal
stands out owing to its use of atomistically obtained conductivity calculated using the VFF approach
dispersion relations, which serve as input for the deviates by about 8% (~18 W/m-K) (due to the
aforementioned MC simulation. Fig. 1 illustrates the incorporation of atomistic corrections in the LA
MC schema adopted in this work. The dispersion branch alone). The VFF-MC approach, used in this
relation (and associated parameters such as phonon work, not only allows one to capture the atomicity of
frequency, polarization, group velocity, and density- the lattice but also enables the simulation of
of-states) is obtained via the VFF approach available realistically sized structures containing millions of
in the open source NEMO 3-D toolkit and compared atoms. A corrected VFF (for the TA branch) and how
with an empirically fitted model (against experi- the results obtained from the VFF-MC framework
mental data) as provided by Brockhouse [3]. are used in the multiscale modeling of thermoelectric
cooler units will be the subjects of future study and
 9280 K  2.234  107 K 2 [LA] presented at the Workshop.
*
 5240 K  2.278  107 K 2 [TA] Supported by NSF Grant No. 1218839.
On the other hand, the MC transport kernel is built REFERENCES

upon 3 (three) discretization schemas, namely, 1) [1] D. Lacroix et al., Phys. Rev. B, 72, 6, 064305, 2005.
Spatial: the geometric volume is divided into [2] B. T. Wong et al., Int. J. Heat & Mass Transfer, 54, pp.
multiple cubically stacked spatial bins/cells; 2) 1825, 2011.
[3] B. N. Brockhouse, Phys. Rev. Lett., 2, pp. 256, 1959.
ISBN 978-0-692-50554-0 179

Fig. 1. A high level flowchart showing various steps/modules involved in the MC scheme.
Fig. 2. Phonon group velocity profiles in silicon thin film of Fig. 3. Thermal conductivity computed using the VFF and
thickness ~4 μm using the VFF and the empirical approaches. the empirical approaches.
180
Phonon Engineering of Si/Ge Core-Shell
Nanowires for Thermoelectric Applications
Jaime Bohorquez-Ballen1, Hansika Sirikumara1, Shaikh Ahmed2,3, Thushari Jayasekera1,3
1
Department of Physics
2
3
Materials Technology Center
Southern Illinois University at Carbondale, IL 62901, USA. (E-mail: [email protected])
INTRODUCTION show a quadratic behavior, with respect to the pho-

Quantum confinement in low-dimensional non wave vector q.
nanostructures is a fundamental property that in- The eigenmodes corresponding to phonon modes
creases their thermoelectric conversion efficiency, in the upper frequency region and lower frequency
ZT [1]. Engineering the heat conduction through region are related to distinctly different atomic dis-
manipulation of phonons provides an additional placement patterns. The displacement of the core
route to further enhance their ZT. Parameter-free atoms are responsible for the eigenmodes in the
modeling approach is essential to precisely investi- lower frequency region, and the shell atom move-
gate how the local atomicity influences the phonon ments are more significant in the upper frequency
modes of the nanostructured devices. Using first eigenmodes. This discrepancy is visualized with two
principles Density Functional Theory (DFT) calcu- representative eigenmodes from the upper frequency
lations, we have studied the structural and lattice region and the lower frequency region as shown in
vibrational properties of [111]-oriented Si/Ge core- the Fig. 4.
shell nanowires. The detailed analysis of phonon Phonon dispersion of six core-shelll nanowires
modes shows that thermal conductance due to se- from set B are shown in the Fig. 5. A forbidden
lective phonon modes of Si/Ge core/shell nanowires frequency region is present in all the phonon disper-
can be suppressed by engineering the ratio of sions. The forbidden region divides the type of the
core/shell atoms. egenmodes; core atom displacements are more re-
sponsible for the phonon modes below the forbidden
CALCULATION METHOD AND RESULTS regions, whereas, vibration of the shell atoms are
The calculations were performed within the responsible for the upper frequency region.
framework of the first-principles density functional If we consider two nanowires with the same core
theory (DFT) as implemented in the Quantum Es- atoms to shell atoms ratio, slope of the acoustic
presso Package [2]. We studied three sets of hydro- modes of the Si-shell/Ge-core nanowire is larger
gen-passivated nanowires, each has total number of than that of the Si-core/Ge-shell nanowire. That
atoms (excluding the hydrogen atoms) N=14 (set A), suggests thermal conductivity can be suppressed by
N=38 (set B), and N=74 (set C), as shown in Fig. 1. having heavier shell atoms, than having heavier core
Fig. 2 shows the phonon dispersions of pristine Si atoms. In conclusion, the detailed atomic structure
and Ge nanowires in the frequency range 0-900 cm- plays a major role in determining the thermal con-
1
. The highest frequency modes in the full disper- ductance of the Si/Ge-core-shell nanowires.
sion, those originate from the vibrations of hydrogen REFERENCES
atoms, go upto 2000 cm-1, and are not shown in
[1] L. D. Hicks, M. S. Dresselhaus, Phys. Rev. B., 47, 1663,
these plots.
(1993).
The low-frequency vibrational modes of all wires [2] R. T. Musin, X. Q. Wang, Phys. Rev. B., 74, 65308,
show the typical one-dimensional behavior with (2008).
four phonon modes. The first two modes (Fig. 3-a [3] P. Giannozzi, S. Baroni, N. Bonini et al., Journal of
and b) are the flexural modes. Fig. 3-c and 3-d show Physics: Condensed Matter,21, 395502, (2009).
the torsional and longitudinal modes. At the long-
This work is supported by the SIU Materials Technology Center.
wavelength limit, flexural modes show a linear
behavior, and longitudinal and torsional modes
ISBN 978-0-692-50554-0 181

Fig. 4 Atomic displacements for sample
eigenmodes for pristine set-B Si nanowires:
(a) for the lower frequency region (b) upper
frequency region
Fig. 1. Atomic configurations of all the nanowires considered in

this investigation. The yellow (lighter), red (darker), and blue
spheres show Si, Ge, and H atoms respectively.
Fig 5. Phonons dispersion for all the core-shell nano

wires of the set-B are shown.
The onset of each graph depicts the atomic
configurations in the axial view. The left-most figure is
the phonon dispersion of pristine Ge wires and the
Fig 2. Lower panel shows the phonon dispersions for pristine graphs are arranged with the increasing number of Si
Si (left) and Ge nanowires (right) of different diameters as atoms.
shown in the upper panel.
Fig. 3 Atomic displacements related to the first four lowest-

frequency eigenmodes for the A-(i) Si nanowire: (a) and (b) are
flexural modes (c) torsional mode and (d) longitudinal mode.
182
Diffusive Transport in Monolayer MoS2: Role
of Remote Coulomb Scattering
K. A. Khair and S. S. Ahmed
Southern Illinois University at Carbondale, 1230 Lincoln Drive, Carbondale, IL 62901, USA.
E-mail: [email protected]
INTRODUCTION/BACKGROUND nm (thickness of the MoS2 layer) and the channel

2D MoS2 with a finite bandgap looks promising is undoped. We have used HfO2 as top gate oxide
as channel material for FETs. Recently, MoS2 with an equivalent oxide thickness (EOT) of ~2
transistors operating at gigahertz frequencies (with nm. The MoS2 channel sits on a SiO2 buried oxide
a cutoff frequency of ~6 GHz) have been reported layer. Following scattering mechanisms have been
[1]. In monolayer MoS2, the K-point direct included in the Monte Carlo transport kernel:
bandgap is ~1.8 eV, the electron effective mass at acoustic (TA), acoustic (LA), optical (0th), optical
the K point, me* = 0.48m0 and is isotropic in na- (1st), polar optical, 0th and 1st order surface/remote
ture. Also, the two degenerate valleys at the K optical in top gate oxide and buried oxide layers,
point are almost parabolic [2]. Nevertheless, elec- and remote Coulomb in top gate oxide and buried
tron mobility in monolayer MoS2 is degraded by oxide layers. A quasi-static assumption has been
various scattering mechanisms. The 3-atom unit made for the holes. The Incomplete Lower-Upper
cell gives rise to nine phonon branches (3 acoustic (ILU) decomposition method has been employed
and 6 optical) [3]. Recently, the role of individual for the solution of the Poisson equation.
scattering mechanisms in standalone monolayer RESULTS
MoS2 was evaluated by calculating electron mobil-
The equilibrium simulations were benchmarked
ity (within an RTA framework) and from a
against an experimental device [1] giving rise to a
comparative study it was concluded that mobility
sheet charge density of ~2×1012 cm-2 at a VG = 2V.
is mainly dominated by ionized impurity scattering
Fig. 2 shows various scattering rates as function of
[4][5]. Also, first-principles many-body calcula-
electron energy. It is found that the influence of
tions were done to obtain dielectric constant for
remote Coulomb and phonon scattering diminishes
monolayer MoS2 [6]. For infinite layer separation,
as the free carrier density increases in the channel
it was found that monolayer MoS2 has a static
region (Fig. 3). A pen picture of current degrada-
dielectric constant of 1 whereas the previous stu-
tion due to various scattering processes is depicted
dies reported the value in the range of 4.2-7.6.
in Fig. 4. Finally, the calculated ID-VG and ID-VD
In this work, using the recently reported value
characteristics are shown in Fig. 5. Calculation of
for the dielectric constant, we calculate the scatter-
various scattering rates as a function of carrier
ing rates in monolayer MoS2 due to various a)
density and temperature, fine-tuning the MC
intrinsic phonon, b) remote phonon, and c) remote
simulator for atomicity and contact energetics, and
Coulomb processes. We then study the electron
the optimization of the FET for small off-current
transport in a monolayer MoS2 based FET device
will be the subject of future study.
employing a particle based Monte Carlo device
simulator. Our results show that the total scattering REFERENCES
rate is strongly dominated by remote coulomb [1] D. Krasnozhon et al., Nano Lett., 14, p. 5905−5911, 2014.
scattering, which, when compared to the ballistic [2] E. S. Kadantsev et al., Solid State Commun, 152, 10, pp.
regime, degrades the drain current by ~78%. 909-913, 2012.
[3] Y. Cai et al., Phys. Rev. B, 89, p. 035438, 2014.
MODEL [4] K. Kaasbjerg et al., Phys. Rev. B, 85, 115317, 2012.
[5] N. Ma et al., Phys. Rev. X, 4, 011043, 2014.
The simulated MoS2 FET has a gate length of [6] F. Huser et al., Phys. Rev. B, 88, 245309, 2013.
14 nm, source-drain length of 20 nm and a width
of 10 nm (Fig. 1). The channel thickness is ~0.65
ISBN 978-0-692-50554-0 183

Fig. 1. Simulated MoS2 FET.
Fig. 2. Scattering rates (subcategories not shown).
Fig. 1. Simulated MoS2 FET.
Fig. 3. Role of carrier density on remote Coulomb scattering.

Fig. 4. Current degradation due to various scattering processes.
Fig. 5. The simulated ID-VG (left) and ID-VD (right) characteristics.
184
On the Role of Rashba Spin-Orbit Coupling in
Graphene Nanoribbons
S. Soleimanikahnoj and I. Knezevic
Madison, Wisconsin 53706-1691, USA
I. I NTRODUCTION port in graphene, a first-nearest-neighbor tight-binding

Since the discovery of graphene, this two dimensional method is used to model the band structure of graphene
material with a honeycomb lattice has been the sub- nanoribbons. The hopping parameter between the nearest
ject of extensive research. Owing to its long electronic neighbors, t, is assumed to be −2.7eV. The Hamiltonian
mean free path, high intrinsic carrier mobility, long spin for RSO can be written as [9]:
coherence time, and high breakdown current densities,
graphene has been envisioned as a suitable material for iCi† (U
ij .σ)Cj + H.c. (1)
<ij>
future electronic applications [1], [2], [3].
Recent theoretical studies predicts the conversion of ij is given by:
σ stands for the Pauli matrices and U
graphene into a spin Hall insulator, in which a spin
current flows perpendicular to an electric field. This phe- U ij = − λR ẑ × δij . (2)
2
nomenon, caused by intrinsic spin-orbit coupling (SO),
leads to counter-propagating spin-filtered edge states in λR represents the RSO strength. δij is the vector between
a graphene strip [4], [5]. The flow and manipulation of two atomic sites, i and j .
spin currents has extensive applications in spintronics. Figure 1 shows the probability amplitude of the first
However, weak spin-orbit coupling in graphene and wave function with a positive energy for an AGNR in
fabrication challenges of perfect graphene nanoribbons three different cases. Figure 1a is in absence of RSO and
(GNRs) hinder the experimental feasibility of SO-based Figures 1b and 1c are in presence of RSO for spin up and
spin currents. spin down respectively. As can be seen, in the absence
Another well-known approach to creating a spin cur- of RSO, electrons are most likely to be found in the
rent emerges when an SO-dependent scattering potential middle of the ribbon. However, when RSO is applied,
interacts with electrons. This scattering potential can it breaks the degeneracy between opposite spins. The
be caused by a strong electric field [6]. The effective interaction couples spin up and spin down, leading to a
spin-orbit coupling generated is called the Rashba spin- spatial separation of the two spins along the width and
orbit (RSO) interaction. Recent experiments predict that consequently a spin current.
strong RSO can be realized in graphene, which is con- III. C ONCLUSION
trollable by a giant electric field [7]. The compelling In summary, we studied RSO in graphene nanoribbons
theoretical and experimental evidence of observable RSO within a tight binding model. We showed that the states
in graphene [7], [8], along with the indispensable role of are expected to be strongly affected by the presence
RSO in the field of spintronics, clarifies the necessity to of the RSO interaction. RSO produces localized spin-
investigate this effect in graphene nanoribbons. dependent states along the edge and spin polarization.
II. M ODEL AND D ISCUSSION This finding opens the possibility of making field-effect
spin devices that could manipulate simultaneously both
In this work, the influence of RSO on electronic
the spin and charge of carriers.
transport of graphene nanoribbons has been investigated.
The results are focused on armchair edge nanoribbons IV. ACKNOWLEDGEMENT
because of their substantial energy gaps, which makes This work has been supported by the NSF through the
them suitable for switching applications. Since the elec- University of Wisconsin MRSEC Seed Program (NSF
trons in π bonds are responsible for electronic trans- Grant No. 1121288).
ISBN 978-0-692-50554-0 185

5 5 5
a) b) 0.022 c) 0.022
0.02 0.02
0.02
0.018 0.018
0.016 0.016
0.015 0.014 0.014

[nm]
[nm]
[nm]
0.012 0.012
0.01 0.01
0.01
0.008 0.008
0.006 0.006
0.005
0.004 0.004
0.002 0.002
0 0 0
0 [A0] 20 0 [A0] 20 0 [A0] 20
Fig. 1: Probability amplitude of the first wave function with a positive energy in an AGNR with N = 40 (N is the
number of carbon along the edge). Panel (a) is in absence of RSO; (b) and (c) are in presence of RSO, for spin
up and spin, down respectively. k = 0 and λR = 0.2t.
R EFERENCES
[1] K. S. Novoselov, A. K. Geim, S. V. Morozov, D. Jiang, Y. Zhang,
S. V. Dubonos, I. V. Grigorieva, A. A. Firsov, Electric field effect
in atomically thin carbon films, Science 306, 666 (2004).
[2] A. K. Geim and K. S. Novoselov, The rise of graphene, Nat.
Mater. 6, 183 (2007).
[3] G. Balasubramanian, P. Neumann, D. Twitchen, M. Markham, R.
Kolesov, N. Mizuochi, J. Isoya, J. Achard, J. Beck, J. Tissler, V.
Jacques, P. Hemmer, F. Jelezko and J. Wrachtrup, Ultralong spin
coherence time in isotopically engineered diamond, Nat. Mater.
8, 383 (2009).
[4] E. I. Rashba, Graphene with structure-induced spin-orbit cou-
pling: Spin-polarized states, spin zero modes, and quantum Hall
effect, Phys. Rev. B 79, 161409 (2009).
[5] C. L.Kane and E. J. Mele, Quantum spin Hall effect in graphene,
Phys. Rev. Lett 95, 226801 (2005).
[6] Q. Lui, X. Zhang, L. B. Abdalla, A. Fazzio, and A. Zunger,
Switching a Normal Insulator into a Topological Insulator via
Electric Field with Application to Phosphorene, Nano. Lett. 15,
1222 (2015).
[7] Y. S. Dedkov, M. Fonin, U. Rüdiger, and C. Laubschat, Rashba
effect in the graphene/Ni (111) system, Phys. Rev. Lett 100,
107602 (2008).
[8] J. C. Boettger and S. B. Trickey, First-principles calculation of
the spin-orbit splitting in graphene, Phys. Rev. B 75, 121402
(2007).
[9] M. Zarea and N. Sandler, Rashba spin-orbit interaction in
graphene and zigzag nanoribbons, Phys. Rev. B 79, 165442
(2009).
186
Modeling polycrystalline effects on the device
characteristics of CdTe based solar cells
S. Mukherjee, S. Farid, M. Dutta and M.A. Stroscio
Electrical & Computer Engineering Department, University of Illinois at Chicago, Chicago, IL 60607,
USA
1
INTRODUCTION qG( )W (1  e (  (  )W ) ) 
J h ( , V )  [   h (1  e  h )]
Polycrystalline CdTe based solar cells ( h   ( )W )
1
 ( )W
represents one of the most attractive options for W is the width of CdTe layer, α(λ) and G(λ)
low cost photovoltaic module production [1]. represents the wavelength dependent CdTe
However various efforts to grow polycrystalline absorption coefficient and the carrier generation
CdTe based solar cells resulted in lower rate. We have also taken into account the effective
efficiencies. This is due to the fact that the grain mobility due to grain boundary of the
size associated with polycrystalline thin films polycrystalline CdTe films that directly effects the
affects not only carrier lifetime and grain boundary voltage dependent normalized carrier lifetime τ
limited recombination velocity but also enhance and can be written as
interface recombination that deteriorates the 1/ eff  1/ mono   gbl
(2)
device performance significantly [2]. Researchers
where µeff is the effective carrier mobility
have focused primarily on modeling single
dependent on single crystal mobility µmono and
crystalline CdTe cells but no complete model is
grain boundary limited mobility µgbl can be
presented until now taking into consideration the
expressed as
polycrystalline effects on the device J-V
characteristics and the external quantum efficiency q g
 gbl 
(EQE) of CdTe based cells [3]. We have m* (3)
developed a comprehensive model to analyze the where m* is the effective carrier mass and τg is
J-V characteristics and the external quantum expressed in terms of intra grain diffusion length L
efficiency of CdTe cells based on minority carrier and average thermal velocity of carriers vt. The
drift and diffusion processes that takes into effective diffusion length is related to the grain
account the grain size distribution due to the size g by
effects of CdCl2 annealing treatment at elevated L
temperatures thereby affecting the carrier lifetime, Lpoly  (4)
grain boundary limited recombination velocity and 2S gblL2
interface properties. The model has been 1
Dg
thoroughly investigated and matched well with the
published experimental data. where Sgbl is the grain boundary limited
recombination velocity in the CdTe active layer,
THEORY AND MODELING and D is the diffusion coefficient.
The total photocurrent density over all the Further a detailed study to calculate the
incident photon wavelengths λ of the solar spectra effects on varying interface recombination
is first calculated using [4] velocities and CdTe doping on the external
 quantum efficiency (EQE) of the solar cell has
J L (V )   [ J e ( ,V ) J h ( ,V )]d (1) been done. The internal quantum efficiency IQE(λ)
0
is expressed as
where
IQE( )  IQESCR ( ) + IQEQNR ( )
qG( )W (1  e (  (  )W ) ) (  (  )W ) 
1
e
(5)
J e ( , V )  [   e (e (  (  )W )  e )]
( e   ( )W )
1
 ( )W where the IQE within the space charge region
(SCR) region can be calculated as
ISBN 978-0-692-50554-0 187

x1 (
W'
ln(1
x
)) [3] Sen, S., Konkel, W. H., Tighe, S. J., Bland, L. G., Sharma,
eff  'E0 S. R., & Taylor, R. E. (1988). Crystal growth of large-area
IQESCR ( ) =  e e -x W'
dx (6)
single-crystal CdTe and CdZnTe by the computer-
0 controlled vertical modified-Bridgman process. Journal of
where x1 (< W’), W’ is the width of the SCR Crystal Growth, 86(1), 111-117.
region and E0 is the electric field at the junction [4] Anjan, M. S., & Kabir, M. Z. (2011). Modeling of
(x=0). The internal quantum efficiency in the quasi current–voltage characteristics of CdS/CdTe solar
cells. physica status solidi (a), 208(8), 1813-1816.
neutral region (QNR) is calculated as
 W' x1
( ln(1 ))
IQEQNR ( )   e  ( x1  x )  x / L poly
e e
eff  'E0 W'
dx (7)
0
The external quantum efficiency of the device can

be written as
Sf (8)
EQE ( )  e CdS (  ) dCdS [1  R( )](1  ) 1 IQE( )
 eff E0
where Sf is the front interface carrier
recombination velocity. Fig. 1. Energy band diagram of CdS/CdTe solar cell.
RESULTS AND DISCUSSION
The J-V characteristics of the CdS/CdTe
structure as shown in Fig. 1 has been solved by the
iterative improvement method and is plotted in Fig
2. It is observed that with increase in grain size
from ~2 µm to ~5 µm, the short circuit current
density increases by ~2.5 mA/cm2 and the overall
cell efficiency increases by ~3.2% which
demonstrates the significance of the annealing
treatment. From the EQE plot in Fig 3, it is
observed that varying the treatment process, the
interface recombination velocity changes as well
as the effective CdTe doping from ~1016 cm-3 to
low ~1015 cm-3 due to interdiffusion of CdS/CdTe
layers.
The results from this study allow us to
Fig. 2. Modeled J-V characteristics of CdTe/CdS cell with A)
accurately formulate polycrystalline behavior of
12.2% efficiency, grain size ~ 5 µm, B) 10.2% efficiency,
CdTe solar cells and provide recommendations to grain size ~ 2-3 µm, C) 8.8% efficiency, grain size g ~ 2-3
optimally choose parameters for increasing the µm; dotted line shows experimental J-V data with varied
photovoltaic conversion efficiency. annealing CdCl2 treatments
ACKNOWLEDGMENT
We are grateful to Dr. Hye-Son Jung for
performing current-voltage measurements
experimentally. This research was supported in
part by AFOSR Grant FA9550-11-1-0271.
REFERENCES
[1] Farid, S., Mukherjee, S., Jung, H., Stroscio, M. A., &
Dutta, M. (2015). Analysis on the structural, vibrational
and defect states of chlorine treated polycrystalline
cadmium telluride structures grown by e-beam
evaporation. Materials Research Express, 2(2), 025007.
[2] Beach, J. D., & McCandless, B. E. (2007). Materials
challenges for CdTe and CuInSe2 photovoltaics. MRS
Fig. 3. External Quantum Efficiency for varied front interface
bulletin, 32(03), 225-229.
recombination velocity (Sf) and effective CdTe doping (NA).
188
Strong negative differential resistance in
graphene devices with local strain
M. Chung Nguyen1,2, V. Hung Nguyen1,2, H. Viet Nguyen2, J. Saint-Martin1, and P. Dollfus1
1
Institute of Fundamental Electronics, CNRS, Univ. of Paris-Sud, Orsay, France
2
Center for Computational Physics, Institute of Physics, VAST, Hanoi, Vietnam
The effect of negative differential resistance the barrier. When applying strain, the
(NDR) has been widely investigated in devices misalignment (in both the k-space and the energy)
based on conventional semiconductors. This effect of the graphene bandstructure inside (i.e., strain)
is suitable for a wide range of high-frequency and outside (i.e., no strain) the barrier results in a
applications [1]. Recently, many designs of finite energy-gap of transmission at finite bias (see
graphene structures and devices exhibiting an Fig. 3). Because of this effect, the overall current
NDR behavior have been proposed, based on is reduced, especially, the valley current and hence
various physical mechanisms [2]. In most cases, a peak-to-valley ratio (PVR) of NDR behavior is
the opening of a bandgap in the bandstructure of significantly enhanced when a strain is applied
graphene by nanostructuring is used to make the (see Fig. 2). Since the effect mentioned above is
modulation of Klein/interband tunneling possible. strongly dependent on the length of the
Recently, the effects of uniaxial strain on 2D barrier/strained area, it is shown in Fig. 4 that the
unstrained/strained graphene junctions were PVR increases when increasing LS ( ≡ LB) and can
investigated and it was found that a significant reach a value higher than 250 beyond L S = 100 nm
conduction gap of a few hundred meV can be in the ballistic limit.
achieved with a small strain of a few percent [3]. Similar non-linear I-V characteristics can be
This conduction gap is not due to a bandgap achieved in the strained PN diode schematized in
opening in the band structure but to the shift of the Fig. 5, where the strained-induced conduction gap
Dirac cones in the Brillouin zone of the strained allows us to control the interband tunneling. In this
graphene. This effect has been demonstrated to case, even a small strain amplitude of σ = 3% is
strongly improve on/off current ratio in graphene enough to achieve high PVR of a few hundreds.
transistors with local strain in the channel [4]. Additionally, it is shown that the peak current is
Here, we investigate the possibility to use local sensitive to the length of the transition region L T
strain engineering to generate high peak-to-valley separating N and P sections (see Fig. 6), however,
current ratio of NDR behavior in graphene the effect is much weaker than that observed in
devices. We focus on two devices: (i) a single other simple gapped graphene P-N devices [2].
gate-induced potential barrier structure where the In summary, we have shown that local strain
strain is applied on an area of length LS (Fig. 1) engineering is effective to enhance strongly the
and (ii) a PN tunnel diode where the strained area NDR effect in graphene devices thanks to the
covers wide parts of two highly doped regions and conduction gap generated at strained/unstrained
the transition section between them (see the top of interfaces. A PVR of a few hundreds is achievable
Fig..5). Our calculations are based on an atomistic in devices where the current is controlled by a
tight-binding model as in [4], which is solved by gate-induced barrier or a PN junction.
the Green's function formalism in the ballistic
REFERENCES
approximation.
We plot in Fig. 2 the I-V characteristics of the [1] H. Mizuta and T Tanoue, The physics and application of
resonant tunnelling diodes, Cambridge Univ. Press, 1995.
single-barrier structure for different strain [2] V. H. Nguyen et al., J. Comp. Electron. 12, 85-93 (2013).
amplitudes with LS = LB = 40 nm and the Fermi [3] M. C. Nguyen et al., Semicond. Sci. Technol. 29, 115024
energy EF = 0.25 eV. In unstrained device the NDR (2014)
effect is very limited due the high transparency of [4] V. H. Nguyen at al., Nanotechnology 25, 165201 (2014).
ISBN 978-0-692-50554-0 189

Fig. 1. Schematic of single potential barrier structure with
strain is applied locally on an area of length L S and Fig. 4. I-V characteristics for different lengths L S = LB (σ =
gate-controlled potential barrier of length LB. 5%). Inset: peak-to-valley ratio as a function of LS = LB.
Fig. 2. I-V characteristics for different strain amplitudes

(single-barrier structure with LS = LB = 40 nm, U0 = 0.45 eV).
Fig. 5. Top: schematic of graphene p-n junctions. Bottom: I-V

characteristics for different strain amplitudes. (U 0 = 0.5 eV,
LS = 40 nm and LT = 10 nm.).
Fig. 3. (E-ky) map of transmission in strained device for two Fig. 6. I-V characteristics of strained device ( σ = 3%) with
bias voltages (same U0, K0 = unstrained Dirac point position). different lengths LT (LS = 40 nm is fixed, U0 = 0.5 eV).
Inset: peak current as a function of LT.
190
Influence of quantum confinement effects over device
performance in circular and elliptical silicon nanowire
transistors
V. P. Georgiev1, T. Ali1, Y. Wang1,2, L. Gerrer1, S. M. Amoroso3, E. Towie3, Asen Asenov1,3
1
Device Modelling Group, School of Engineering, University of Glasgow, Glasgow, G11 8LT, UK
2
Institute of Microelectronics, Peking University, Beijing 100871, China
*
Gold Standard Simulations Ltd, Glasgow G12 8LT, UK
INTRODUCTION classical DD simulations. This has a significant effect on

the transistor’s performance as the reduction of the gate
Silicon nanowire transistors (NWTs) are considered one capacitance reduces the mobile charge.
of the most promising device architectures for the sub 7- Fig. 3 presents the gate voltage dependence on the
nm CMOS technology [1]. In such ultra-scaled devices the mobile charge in the channel for the discussed NTWs. A
quantum mechanical effects play a significant role that reduction of the gate capacitance, consistent with the
determines device performance [2]. These quantum results from Fig. 2, leads to a reduction of the mobile
confinement effects introduce a threshold voltage shift, charge in the channel. Therefore a reduction in the NWT
simultaneously reducing the gate-to-charge capacitance performance is also expected. Fig. 3 also shows a
and the charge in the channel available for transport [3]. comparison of the 2D charge distribution in the NWTs
Hence, in order to accurately describe the device cross-section based on the QM and classical DD
performance in such ultra-scaled transistors, calculations simulations at VG=0.6V. It is clear that the QM charge is
that consider quantum mechanical effect are essential. In predominantly localised in the middle of the channel while
this paper, taking into account the quantum confinement the DD solution shows increasing of the charge close to the
effects, we establish a link between different cross-sections semiconductor/insulator interface.
of two NWTs and the device performance. Fig. 4 shows the impact of the gate length on DIBL,
METHODOLIGY AND DISSUSSION defined as ΔVT/ΔVD, and on the sub-threshold slope (SS).
Our simulations are based on a recently developed As seen in Fig. 3, there is relatively little difference in the
Poisson-Schrödinger (PS) quantum correction technology electrostatic integrity between the circular and elliptical
introduced in a drift-diffusion (DD) module of the GSS NWTs. Hence, it can be concluded that the QM effects
‘atomistic simulator’ GARAND [4]. The 2D solutions of have no significant impact on DIBL and SS for these
the Schrödinger equation are obtained in each cross- specific shapes of the cross-section for both NWTs.
section along the gate length of the simulated transistor. The impact of the gate length on the leakage (IOFF) and
The 2D Schrödinger equation is in an effective mass drive current (ION) is illustrated in Fig. 5. In this case there
approximation [5]. The charge distribution obtained from is a significant difference between the cylindrical and
the solutions of the 2D Schrödinger equation is used to elliptical nanowire. Overall, the elliptical NWT shows
define the effective quantum-corrected potential. This lower IOFF and higher ION current if compared to the
potential is used as a driving potential in the solution of the cylindrical nanowire.
current-continuity equation, keeping the charge Fig. 6 reveals the 2D charges distribution and the first
distribution in the NWT cross-section identical to the five wave functions with the lowest energy. It should be
charge distribution obtained from the solution of the noted that for the circular nanowire the QM charge
Schrödinger equation. distribution inside of the channel is asymmetric, even
The nanowires simulated in this work have design though the wire has a symmetric cross-section. This is due
parameters and geometry described in Tab. 1 and Fig. 1. to different effective masses in <110> crystal orientation
The simulated NWTs have two cross-section shapes – along the two axes.
cylindrical and elliptical (both have an identical cross- CONCLUSION
section area of 25 nm2). The transport direction is aligned
In conclusion, in this paper we report the effect of
along the crystallographic orientation <110>.
quantum mechanical effects on the electrostatic driven
Fig. 2 shows the capacitance-voltage characteristics (C-
performance of NWTs suitable for the 7-nm CMOS
V) of both NWTs. From this data it is clear that the
technology. Overall, our work shows that the quantum
quantum mechanical (QM) effects have a significant
effects cannot be neglected and that they determine the
impact on the gate capacitance. Overall, the QM gate
available mobile charge in the channel and gate
capacitance is reduced by around 35% if compared to the
capacitance.
ISBN 978-0-692-50554-0 191

Table I. Geometrical dimensions and the main characteristics for all Fig. 1. 3D structure view of the circular (left) and elliptical (right) nanowire
silicon nanowires studied in this work. transistors. Dark blue is the Si channel, yellow is the SiO2 oxideand light blue
is the gate oxide.
7
8e-12 1.5×10
F/cm)
Circular - QM Circular - QM
Elliptical - QM
Electron density (cm )

Elliptical - QM
-1
-11
6e-12 Circular - DD 7 Circular -DD

1.0×10
Gate Capacitance (x10
Elliptical - DD Elliptical - DD
4e-12 Circular -DD

6
5.0×10
2e-12
0 0.0
0 0.1 0.2 0.3 0.4 0.5 0.6 0 0.1 0.2 0.3 0.4 0.5 0.6 Circular - QM
Gate Voltage (V) Gate Voltage (V)
Fig. 2. Gate capacitance vs Gate voltage for the elliptical and cylindrical Fig. 3. Electron density vs Gate voltage for the elliptical and cylindrical
NWTs based on the classical (DD) and quantum mechanical (QM) NWTs based on the classical (DD) and quantum mechanical (QM)
calculations. calculations (left); 2D charge distribution in the NWTs cross-section based
on the QM and classical DD simulations at VG=0.6V (right).
40 Circular
Elliptical
DIBL [mV/V]
30
20
10
6 8 10 12 14
Channel Length [nm]
Fig. 4. Impact of the gate length on DIBL (left) and sub-threshold slope – Fig. 5. Impact of the gate length on IOFF (left) and ION (right) for the elliptical
SS (right) for the elliptical and circular nanowire with a 5x5 and 4.16x5 and circular nanowire with a 5x5 and 4.16x5 nm cross-section.
nm cross-section.
REFERENCES
!
[1] B. Yu, L. Wang, Y. Yan, P. Asbeck, and Y. Taur, “Scalling of nanowire
transistors”, IEEE Trans. Electron Devices, vol. 55, no. 11, pp. 2846-
2858, 2008.
[2] V. P. Georgiev, E. A. Towie, and A. Asenov, “Impact of precisely
positioned dopants on the performance of an ultimate silicon nanowire
transitor: a full three-dimensional NEGF simulation study”, IEEE
Trans. Electron Devices, vol. 60, no. 3, pp. 965-901, 2013.
[3] G. Glindalore, S. A. Hareland, S. Jallepalli, A. F. Tasch, C. M. Maziar,
Carrier
V. K. F. China, and S. Smit, “Experimental determination of threshold
Density
Wavefunctions voltage shift due to quantum mechanical effects in MOS electron and
Fig. 6. 2D cross-sections (at the centre of the channel) of the simulated hole inversion layers”, IEEE Trans. Electron Devices, vol. 206, pp.
electron density and the first five lowest wave functions in the ON-state 206-208,1987.
region. [4] www.goldstandardsimulations.com
[5] B. Bagchi, P. Gorain, C. Quesne, and R. Roychoudhry, “Effective-mass
Schrodinger equation and generation of solvable potential”, Mod. Phys.
Lett. A, vol. 19, pp. 2765, 2004.
192
Algebraic Multigrid Poisson Equation Solver
X. Guo1, S. Qazi2, R. Daugherty2, D. Vasileska2
1
School for Engineering of Matter, Transport and Energy,
School of Electrical Computer and Energy Engineering
2
Arizona State University, Tempe, Arizona, 85287-5706, USA

However, standard Multigrid requires 2n+1 grid

ABSTRACT
points in each direction for aggressive coarsening.
This paper discusses highly efficient and robust It also requires manual setup of coarser grid,
Multigrid numerical solver for the solution of the which becomes difficult for complex geometries.
Poisson equation in device simulators. Fast and To overcome these drawbacks, few variants of
accurate solver is needed when 3D device or the standard Multigrid are implemented and
multi-scale simulations are performed. It is numerically tested [2]. Algebraic Multigrid
important to note that the algorithm complexity of (AMG) is an alternative in the setup stage, which
commonly used iterative methods, such as imposes no requirement for the mesh. Coarser
Successive-Over-Relaxation (SOR) and Conjugate meshes and corresponding coefficients are
Gradients (CG), is typically higher, which means automatically generated based on the finest mesh
time cost of the numerical solver will become requirements. Full Multigrid (FMG) is used to
predominant with increasing number of mesh preprocess initial guess for nonlinear problem to
points. improve the robustness of Multigrid solver.
INTRODUCTION Standard Multigrid can only deal with linear
The complexity of SOR is O(N1.5), where N is system. However, Full Approximation Scheme
the number of total grid points. Compared to this, (FAS) can directly solve nonlinear problem
the Multigrid complexity is on the order of O(N) without linearization.
[1]. Unlike conventional single level iterative APPLICATION ON A 2D MOSFET
methods, Multigrid method uses coarse mesh. The Conventional SOR, AMG, and FAS are tested
original mesh is treated as the finest mesh. The to calculate the equilibrium potential profile of a
coarser grid is a subset of a fine grid. Either 2D MOSFET device shown in Figure 3. Figure 4
computing residuals or approximate solutions are shows that AMG barely has advantages over SOR
transferred back and forth between each grid level because nonlinear Poisson’s equation is linearized
depending on the type of Multigrid. Figure 1 each iteration cycle, which changes the transient
schematically shows how standard Multigrid linear system. Therefore, setup is performed each
works in a 5-level V cycle. iteration. Figure 6 shows that FAS performs much
MULTIGRID AND ITS VARIANTS better than SOR in terms of speed. Figure 5 shows
There are three stages in a complete Multigrid a cutline of converged potential. SOR generates
solver (as shown in Figure 2). Information about different results with different initial guess, while
coarser levels is obtained either manually or the results of FAS are independent from initial
automatically in the setup stage – stage 1. The guess because of FMG as a preprocessor.
second stage is optional. This involves The above results suggest that Multigrid
preprocessing of the initial guess, which is method shows large advantages over SOR to
strongly recommended for nonlinear problems, benefit 3D or multi-scale device simulations.
because the capture zone of a nonlinear problem is REFERENCES
small, i.e. stability of the nonlinear problem is low. [1] J. Demmel, Applied Numerical Linear Algebra. Society
The third stage is the core part of a Multigrid for Industrial and Applied Mathematics, 1997.
solver, which is iterated until a convergence [2] D. Vasileska, S. M. Goodnick, and G. Klimeck,
Computational Electronics: Semiclassical and Quantum
criterion is met. Standard Multigrid is the most Device Modeling and Simulation. CRC Press, 2010.
intuitive with the least overheads in computation.
ISBN 978-0-692-50554-0 193

Fig. 1. Scheme of a Multigrid V Cycle Fig. 4. Convergence of SOR and AMG on 2D MOSFET
Standard Multigrid Algebraic Multigrid

(Manually Aggressive (Automated Standard
Setup Stage
Coarsening, Coarsening, Arbitrary
2N+1 points) Number of points)
Full Multigrid
Initialize Stage
(Preconditioner)
Full Approximation
Standard Multigrid Scheme
(Restrict Residual, (Restrict Approximate
Solve Stage
Interpolate Correction, Solution, Interpolate
For Linear System) Correction, For
Nonlinear System)
Fig. 2. Stages and Components of a Multigrid Solver Fig. 5. Vertical Cutline of Potential Profile
Fig. 3. 2D MOSFET Geometry Fig. 6. Convergence of SOR and Hybrid Multigrid on 2D

MOSFET
194
Phononic Properties for Enhanced Signal-to-
Noise Photodetector
Yi Lan, Chenjie Tang, Junxia (Lucy) Shi, Mitra Dutta, Michael Stroscio
University of Illinois at Chicago
Chicago 60607, United State
[email protected]
ABSTRACT MODEL
A reduced signal-to-noise reduced Setting up regions for this structure, the
photodetector has been designed using single-and dielectric function should be considered first. As
double-quantum well structures as well as phonon known, in system such as AlAs and GaAs,
assisted transitions. To increase electron carriers, (1)
delta-doping is used in the first single well. In this
contribution, the energy levels of the single-well—
double-well structure are prescribed by Since , , and depend primarily on
determining the energy spectrum for the interface the x-value of , n-type doping causes
phonon potentials and their dispersion curves. only small changes in the phononic properties.
Accordingly, only the boundaries defined by the
INTRODUCTION
six interfaces in the active part of the single-well—
According to “Enhanced Signal-to-Noise in double-well device, see Figure 1, need to be used
Photodetectors Interface Phonon-assisted to determine the phonon modes and their
Transitions”[1], photodetector is designed into dispersion curves.
three quantum wells, one single well at the left and The single-double quantum well structure is
one double well at the right side. The depicted as Fig. 1, and by a Schrödinger equation
photodetector absorbs one photon, emits a phonon, calculation of energy levels, it is seen that the
and absorbs another photon with the same energy phonon-assisted transition from the single-well to
the same as the first one. Accordingly, signal-to- the double-well can be driven by emission of a
noise as predicted by the Richardson formula is phonon with an energy of 32.19 meV.
enhanced since the electron is approximately twice The phonon potentials, , in the seven regions,
as deep in its initial state as in the corresponding satisfy the boundary conditions [4],
single-well photodetector.
This series of processes - absorbing a photon, (2)
emitting a phonon, then absorbing a photon -
works efficiently only if excess electrons are kept
in the first single well. That’s because electrons (3)
are needed for absorbing photons. Delta doping
enhances the electron concentration that can Using the phonon normalization condition [4]
absorb photons in the first process, and can be and performing all the calculations for different
accomplished in a way that has little impacts on wave vectors, q, from 0 to , where lattice constant
the energy states and offsets of the full
of GaAs is 5.65325Å, we find the results of
structure.[2,3]
Figures 2-6.
In this contribution, the phononic properties in
this structure will be described, including the CONCLUSION
phonon dispersion curve and phonon potential in Herein, the phononic properties for an
the structure in the single-well—double-well enhanced signal-to-noise photodetector have been
structure. determined. Device design characteristics will be
discussed in the presentation.
ISBN 978-0-692-50554-0 195

REFERENCES
[1] Yi Lan, Nanzhu Zhang, J. L. Shi, M. Dutta, M. A.
Stroscio, Design of a Novel Heterostructure
Photodetectors with Dramatically Enhance Signal-to-
Noise based on Resonant Interface-Phonon-Assisted
Transitions and Engineering of Energy States to Enhance
Transition Rates, 2nd Int’l Conf. and Exhibition on Lasers,
Optics & Photonics, Sept. 2014 Philadelphia.
[2] Paul Harrison, Quantum Wells, Wires and Dot:
Theoretical and Computational Physics, 3rd edition,
(Wiley, New York, 2010).
[3] Michael A. Stroscio, Interface-Phonon--Assisted
Transitions in Quantum Well Lasers, Journal of Applied Fig. 3. Phonon potential along z-axis with q=
Physics, 80, 6864 (1996). [2]
[4] M. Stroscio and M. Dutta, Phonons in Nanostructures
(Cambridge University Press. 2001).
Fig. 4. Phonon potential along z-axis with q=

Fig. 1. Structure of single-double well photodetector along z
direction
Fig. 2. Phonon dispersion curve. Phonon dispersion curves –

as shown as Fig. 2 – form groups near 33.768 meV and
46.692 meV.
196
Electronic Structure Dependence on the Density,
Size and Shape of Ge/Si Quantum Dots Array
Ming-Yi Lee∗, Yi-Chia Tsai∗ , Yiming Li∗ and Seiji Samukawa†
∗ Instituteof Communications Engineering, Department of Electrical and Computer Engineering
National Chiao Tung University, Hsinchu 300, Taiwan (e-mail: [email protected] (Y. Li))
† Institute of Fluid Science and WPI-AIMR, Tohoku University, Sendai 980-8577, Japan
I NTRODUCTION without nodes as the expected s-orbit. The next three states
Among the next-generation ultra-high efficiency photo- show a two-fold degenerate p-orbit due to symmetry of x-
voltaics technologies, a promising candidate is the use of y plane and non-degenerate p-orbit along z-direction. These
semiconductor quantum dots (QDs). The QDs work as the energy distributions and wave functions reveal the symmetry
intermediate band in solar cell and dramatically increase the of the QDs superlattice.
efficiency because of their ability to absorbing sub-bandgap Fig. 4 presents the DoS for QDs superlattice with varied
photons through the two-photon transition [1]. When uniform interdot space from 0.3 nm to 3.3 nm. As the interdot space de-
QDs are closely packed as the superlattice, wavefunction of creasing, i.e. QDs density increasing, QDs interaction between
each QD couples with neighboring QDs to broaden the discrete discrete levels increases and miniband crossing phenomenon
quantum levels to form finite-width miniband, which is a occurs especially for higher excited states. Because the highest
key parameter to determine two-photon transition. Within the excited states are mixed and become continuous energy levels
envelop-function framework, a 3D finite element method that to Si barrier, the effective bandgap of bulk Si decreases.
surmounts theoretical approximations of the multi-dimensional In Fig. 5, the energy level of the ground band becomes lower
Kronig-Penney method has recently been developed to simu- with varied thickness from 2 nm to 8 nm because of the weaker
late miniband structure and density of states (DoS) [2]. In this confinement in z-direction as expected. Meanwhile, the band
work, the electron structure dependence on the density, size width less depends on the thickness because the interaction
and shape of Ge/Si QDs array is studied in further to instruct between QDs is strongly related to the distance between QDs
realistic QDs fabrication and design. in x-y plane. The significant dependence of band structure on
QDs symmetry is revealed for the third excited band E3,k of
T HE C OMPUTATIONAL M ODEL QDs with the thickness 4 nm, which is non-degenerate p-orbit
For the periodic superlattice, the electronic structure is along z-direction in Fig. 2b. In Fig. 6, this p-orbit along z-
numerically solved by Eq. (1) under Bloch theorem as Fig. 1. direction have much weaker confinement and becomes the first
First, a unit cell formed by the primitive vectors is defined. excited band for QDs thickness 8 nm. On the other hand, the
Then, based on the symmetry of superlattice, the k-points energy level goes lower and band width goes smaller with the
space is defined in an IBZ. To get band structure En,k and QDs radius in Fig. 7 because of the weaker confinement and
bloch function uk (r), Eq. (1) is discretized within a unit cell larger distance between QDs in x-y plane.
in real space and solved by a FEM solver for each sampling k- Fig. 8 presents the calculated energy levels and bandwidth
point in IBZ. Finally, the DoS, a key parameter for electronic of ground band as functions of the QDs volume for three
and optical applications of semiconductor system, is calculated different QDs shapes with the same base radius (3 nm). Bigger
numerically using an improved triangle method [3] by Eq. (2). the QDs volume, lower energy level and larger band width.
The most sensitive to the dot volume is the disk-shape and
R ESULTS AND D ISCUSSION the least is the cone-shape. This phenomenon is due to the
Fig. 2 shows the calculated band dispersion relation for QD stronger z-confinement and interaction in x-y plane for disk-
in dimension of the radius 3 nm and the thickness 4 nm [4] shape QDs than that for the other shapes.
and the interdot space 2.3 nm. The energy distribution in IBZ In summary, we have reported the miniband structure de-
for the lowest four bands is shown in Fig. 3. The energy for pendence on the density, size and shape of Ge/Si QDs array.
the ground band is distributed isotropically in k-space and The QDs density has strong impact on the QDs interaction and
minimum in Γ-point. The energy for the first and second miniband width. Meanwhile, the QDs volume and shape affect
excited bands are constant in ky and kx direction respectively the quantum confinement through the geometric symmetry.
while the energy for the third excited band is distributed
R EFERENCES
isotropically in k-space. These distributions are closely related
to their wavefunctions. Fig. 2b shows the isosurface of wave [1] A. Luque et al., Phys. Rev. Lett., vol. 78, p. 5014, 1997.
[2] M.-Y. Lee et al., IEEE Trans. Electron Devices, submitted for publication.
function within Ge QD at Γ-point (k = 0) of IBZ for lowest [3] J.-H. Lee et al., Phys. Rev. B, vol. 66, p. 233102, 2002.
four bounded-states. The ground state has ellipsoid surface [4] M. Igarashi et al., Appl. Phys. Lett., vol. 101, p. 063121, 2012.
ISBN 978-0-692-50554-0 197

0.5
0.4
Energy (eV)
0.3
0.2
0.1
2 nm 4 nm 6 nm 8 nm
0
1 3 1 3 1 3
10 10 10 10 10 10 101 103
DoS (/eV/Unit Cell)
Fig. 5. The density of states for Ge/Si QDs square superlattice with QD
Fig. 1. The simulation flow chart for an in-plane 3D QDs square superlattice radius 3 nm, interdot space 0.3 nm, and varied thickness from 2 nm to 8 nm.
and material parameters used for simulation.
0.5
0.5 0.5
0.460 0.4
Energy (eV)
0.4 0.430
0.3
0.345
0.323
Energy (eV)
0.3 0.303 0.2
0.241 0.1
0.2
0.190 2 nm 4 nm 6 nm 8 nm
0
Γ X M Γ X M Γ X M Γ X M Γ
0.1 0.103
Wave Vector k
Fig. 6. The band structures for Ge/Si QDs square superlattice with QD radius
0
Γ X M Γ Γ X M Γ 3 nm, interdot space 0.3 nm, and varied thickness from 2 nm to 8 nm. The
highlight band with p-orbit symmetry in z-direction (green dash line) shows
Wave Vector k much more dependence on the QDs thickness.
(a) (b)
Fig. 2. (a) The band dispersion relation for 3D in-plane Ge/Si QDs superlattice
with QD’s radius 3 nm, thickness 4 nm and interdot space 2.3 nm. The right 0.5
figure breaks energy axis into several segments in different scale to highlight
the profile of each quantum level. (b) The isosurface of wave function within 0.4
QD at Γ-point of IBZ for lowest four bounded-states: s-orbit E0 , degenerate
Energy (eV)
p-orbit E1 and E2 , and non-degenerate p-orbit E3 . 0.3
0.2
M M M M
0.1033 0.1903 0.1904 0.2417
0.1902 0.1903
0.1902
0.2416 0.1
Energy (eV)
Energy (eV)
Energy (eV)
Energy (eV)
0.1901
2 nm 3 nm 4 nm 5 nm
0.1032 0.1901 0.2415
0.1900
0.1900
0.2414 0
101 103 101 103 101 103 101 103
0.1899 0.1899
0.1031 0.1898 0.1898 0.2413

Γ X Γ X Γ X Γ X
(a) (b) (c) (d) DoS (/eV/Unit Cell)

Fig. 3. The energy distribution in IBZ for (a) ground bounded-band E0,k , Fig. 7. The density of states for Ge/Si QDs square superlattice with QD
(b) first excited bounded-band E1,k , (c) second excited bounded-band E2,k , thickness 4 nm, interdot space 0.3 nm and varied radius from 2 nm to 5 nm.
and (d) third excited bounded-band E3,k .
0.5
0.25
disk
ellipsoid
0.4 cone
0.20
Energy (eV)
Energy (eV)
0.3
0.15
0.2
0.10
0.1
3.3nm 2.3nm 1.3nm 0.3nm 0.05
0 0 50 100 150 200 250
10 1 3
10 1
10 10 3
10 1 3
10 10 1
10 3 Volume (nm3)
DoS (/eV/Unit Cell) (a) (b)

Fig. 8. (a) Three kinds of QDs shape used for comparison. (b) The energy
Fig. 4. The density of states for Ge/Si QDs square superlattice with QD
level of ground band versus QDs volume for Ge/Si QDs square superlattice
radius 3 nm, thickness 4 nm, and varied interdot space from 3.3 nm to 0.3
with different QDs shape.
nm.
198
Accelerated 3D Full Band Self-consistent Ensemble
Monte Carlo Device Simulation Utilizing Intel MIC
co-processors on TianHe II
Longxiang Yin1, Minquan Fang2, Lang Zeng3 4, LiLun Zhang2, Gang Du1, Xiaoyan Liu1*
1
Institute of Microelectronics, Peking University, Beijing, China
2
School of Computer, National University of Defence Technology, Changsha, China
3
School of Electrical and Information Engineering, Beihang University, Beijing, China
4
Spintronics Interdisciplinary Center, Beihang University, Beijing, China
INTRODUCTION SOI FinFETs with the structure shown in Fig.4 are

simulated with different grids number to verify the
3D Monte Carlo (MC) simulation method can properly
simulator and evaulate the efficiency. The different grids
simulate high-field non-local carrier transport that makes is shown in Table I. The least grid number N is 44370.
it suitable for nanoscale non-planar device simulations [1].
A. Parallel Efficiency
By appropriately accounting for quantum corrections,
For comparison, the initial condition keeps the same
semi-classical MC simulation can give accurate results
during simulation.
due to including scattering effect. However, both 3D
Firstly we compare one CPU (1CPU) with one CPU
simulation and quantum correction for MC demand huge
and multiple MICs (1CPU+xMIC) on one computing
computational resources [2].
node as shown in Fig.5 and Fig.6.
In recent years, Intel has introduced its Intel ® Xeon
Speedup ratio can be calculated as
PhiTM Many Integrated Core (MIC) co-processor. It’s 𝑇𝑇𝐶𝐶𝐶𝐶𝐶𝐶
famous for its powerful parallel processing capability and Speedup Ratio =
straightforward code portability [3]. We can easily 𝑇𝑇𝐶𝐶𝐶𝐶𝐶𝐶+𝑀𝑀𝑀𝑀𝑀𝑀
acquire high parallel efficiency without rewriting codes Fig.5 shows that for the least grid number N case,
as utilizing GPUs. In our previous work [4], we accelerated “Quantum Correction” can obtain 190-time
developed a 3D parallel Ensemble MC simulator with a speedup with 3-MIC acceleration and the simulation time
max parallel speedup of around 30. In this work, we make is 11.9 mins,which is comparable to DD results by
Sentaurus. Fig.6 shows that, by using one MIC, the
use of Intel MIC co-processors on TianHe II to further
improve parallel efficiency of our simulator. The results simulation time of largest grid number is 19.6 hours and
of the work indicate that MIC can make 3D MC device 3.2 times lower than without using MIC. Fig.7 shows that
simulator more efficient for TCAD application. increasing node number can increase the speedup ratio.
We can get 6.1 speedup when utilizing 5 computing node.
SIMULATION METHOD B. Verification
Our previous 3-D parallel full-band EMC simulator for To verify the new developed MIC+CPU mode 3D MC
Si devices [4,5], which takes into account the major simulator, a 14nm FinFET is simulated and the
scattering mechanisms including phonon, surface parameters is listed in Table II . The I-V curves and the
roughness (SR), impact ionization and ionized impurity corresponding distribution are compared with our
scattering, can reproduce the bulk Si velocity-field original version of 3D MC simulator. Fig.8 shows the I-
characteristic and the inversion channel universal V curves and Fig.9 shows the potential distribution and
effective mobility curves [5]. The bulk Si full-band electron energy distribution along channel direction. The
structure obtained by the local empirical pseudo- results are of almost the same.Fig.10 shows some 3D
potential method is employed [5]. In addition, the electrical characteristic distribution results.
effective potential (EP) method as the quantum correction CONCLUSION
method is used [4]. In 3D self-consistent Monte Carlo
simulation, for each time step, 3D Poisson equation needs We demonstrate that MIC is effective to accelerate the 3D
to be solved and millions of particles are simulated. In MC simulation. MIC co-processors is suitable for parallel
order to accelerate the 3D MC, by simulating a FinFET processing in 3DMC simulation with large grids number.
with different grid number, the simulation time for each ACKNOWLEDGMENT
part of the 3D MC is evaluated and shown in Fig.1. It can We would like to express our sincere thanks to the
be seen that “Quantum Correction” part consumes about staff of TianHe II. This work is supported by 973 Program
half of the CPU time and the time increases with the (2011CBA00600) and NSFC 61421005.
increasing grids number. Hence, following we will
discuss the method to parallel the most time consuming REFERENCES
part by MIC for clearness and simplification. [1] Mark Lundstrom, Fundamentals of carrier transport, second
Fig.2 shows the flow chart of our simulator. Fig.3 edition, Cambridge University Press, 2000, pp 247.
shows a sample of compile statements needed when using [2] H.Haug, A-P.Jauho Quantum Kinetics in Transport and Optics of
Semiconductors, Springer 1998.
MIC and it’s easy and plain. Here, CPU is Intel Ivy [3] Intel® ;HRQ 3KL70 Coprocessor (codename: Knights Corner)
Bridge Xeon processor and MIC is Intel Xeon Phi co- Software Developers Guide Ver.1.04.
processor. [4] W. Zhang, Gang Du, et al, IWCE, 2009
[5] X.Y.Liu, K.L.Wei, Gang Du, et al, ICSICT, 2012
RESULTS AND DISCUSSIONS
ISBN 978-0-692-50554-0 199

Fig.1. Time occupancy of
each component in total
simulation time without
MIC acceleration. Grid
number varies from 1N to
5N in Table I.
Fig.3. Compile statements used on MIC. Fig.4. FinFET structure used in our simulation. Fig.2. Flow chart of our 3D EMC
QuantumCorrection region is indicated with simulator with ”Quantum Correction”
dashed line. part accelerated by MIC.
Convergent Time (min)

QuantumCorrection
Convergent Time (s)
4 The Whole Program

3600 All on CPU 4.5
10
Speedup Ratio
Speedup Ratio
79 2 MIC+CPU
3 149 190 10 3.2
10 2400 3.0
2 823s 8
10 6.3 10
1
3.3 1200
10
1
1 1.5
0
0 10
10
0 1 2 3 1N 2N 3N 4N 5N
Number of MIC Grid Type
Table I. Different grids number in Fig.5. Convergent time and speedup ratio of the Fig.6. The relation of simulation time and
our simulation. whole program and QuantumCorrection when speedup ratio along with different grids number
utilizing 1,2,and 3 MIC co-processors on one com- when utilizing 1CPU and 1CPU+1MIC on one
puting node. Dashed line is the result of DD computing node. The speedup ratio keeps 3.2.
simulation by Sentaurus with one node. Grids is N.
20
Drain Current (100uA/um)
Electron Energy (eV)

All on CPU 6.1 All on CPU
Convergent Time (s)
6000 6
Speedup Ratio
MIC+CPU 1.5 MIC+CPU 0.6

15
Potential (V)
4000 4.4 5.5

10 All on CPU
4 1.0
3.3 CPU+MIC 0.3
2000 3.5 5
2
0 0.5
1 2 3 4 5 0.0 0.2 0.4 0.6 0.8 0 20 40 60 80
Computing Node Number Drain Voltage (V) y (nm)
Fig.7. Convergent time and speedup ratio Fig.8. Comparison of ID-VD output Fig.9. Comparison of electric characteristics
along with node number,1CPU and 1MIC on characteristic curve between original along channel direction between original
each computing node.Grids is N. simulator and MIC+CPU mode simulator. simulator and MIC+CPU mode simulator.
(a) electron density (b) electron velocity (c) electron energy
Fig.10. 3D electric characteristic distributions along y direction in QuantumCorrection region. Table II. structure and operation parameters
Mesh grid number is 130000, particle number is about 2470000. of 14nm FinFET used in our simulation.
200
Crystal Viewer: An Online Platform for
Learning and Understanding Crystalline
Structure
Yuanchen Chu1,*, Daniel F Mejia1, James Fonseca1, Michael Povolotskyi1 and Gerhard Klimeck1
1
Network for Computational Nanotechnology, Purdue University, West Lafayette, IN 47907
*
and position to ‘place’ the plane. Alternatively, the

INTRODUCTION
user can manually set the position by entering the
Most materials found in nature have their atoms cross point on the axis. Although the tool can
arranged in a regular and repeated pattern known visualize a large number of materials, it is difficult
as the crystalline structure. Crystals have a huge to include all existing crystal materials. In order to
number of applications in science with one of them overcome this limitation, a function has been
being the basis for all nanotechnology and developed for creating and viewing user defined
microelectronics. It is very important to crystals. The user can input any basis and lattice
understand the crystalline structure of materials in and the tool can then visualize the input structure.
order to study and predict their properties such as The miller plane can also be drawn along with the
conductivity for electricity, conductivity for heat, user defined crystals.
optical properties and more. As a growing number
of crystals with complex structures are being used The tool runs within the Rappture interface [3]
in modern technology, students and scholars need and uses NEMO5 [4] simulation tool in the
a better way to visualize these crystals than plain background to perform calculations. Usage
text book pictures. The Crystal Viewer [1] statistics [5] shows that more than 7000 users have
deployed on nanoHUB [2] provides such a run over 65,000 simulations.
simulation platform for the research community
CONCLUSION
through the web browser. The tool has the ability
to visualize materials of different crystalline An online platform for simulating crystalline
structures including diamond, zincblende, wurtzite, structures has been shown. The tool has been
etc. It can also visualize the 14 Bravais lattices. deployed on nanoHUB and can visualize different
Besides, it can create user defined crystalline crystal materials along with the miller plane, the
structure with the basis and the lattice input by the 14 Bravais lattices and create user defined
user. crystalline structures. It is acting as a research and
TOOL DESCRIPTION learning tool for the scientific and educational
community.
Crystal Viewer tool is an excellent supplement
to text book pictures of crystal materials. The ACKNOWLEDGEMENT
crystalline structure that can be visualized by the
tool ranges from diamond, zincblend, wurtzite to The use of nanoHUB.org computational
sodium chloride, cesium chloride, fcc, bcc and resources operated by the Network for
carbon meshes. Users can choose to use the Computational Nanotechnology funded by the US
conventional basis or the primitive basis and set National Science Foundation under Grant Nos.
the dimension of the crystal. Up to three miller EEC-0228390, EEC-1227110, EEC-0228390,
planes can be drawn along with the crystal EEC-0634750, OCI-0438246, OCI-0832623 and
material. The tool can analyze the crystalline OCI-0721680 is gratefully acknowledged.
structure and choose an optimal shape of the plane
ISBN 978-0-692-50554-0 201

Fig. 1. Visualization of a NaCl unit cell
Fig. 4. The menu for setting user defined crystals
Fig. 2. Visualization of diamond structure material silicon

with a (111) plane
Fig. 5. Visualization of dichalcogenide structure material

Molybdenum disulfide (MoS2) using the function of user
defined crystal
Fig. 3. Visualization of 3 graphene layers
REFERENCES
[1] https://nanohub.org/tools/crystal_viewer
doi:10.4231/D3VT1GQ8M
[2] https://nanohub.org
[3] McLennan, M. ”The rappture toolkit.” (2004)
[4] Steiger, S.; Povolotskyi, Michael; Hong-Hyun Park;
Kubis, T.; Klimeck, G., ”Nemo5: A Parallel Multiscale
Nanoelectronics Modeling Tool.” Nanotechnology, IEEE
Transactions on 10.6 (2011): 1464-1474.
[5] https://nanohub.org/resources/crystal_viewer/usage
Fig. 4. The hexagonal Bravais Lattice
202
Size Dependence of the Seebeck Coefficient for
Single-Material Thermocouples
A. Kommini† , G.P. Szakmany∗ , A. O. Orlov∗ , G. H. Bernstein∗ , W. Porod∗ , and Z. Aksamija†
† Electrical and Computer Engineering, University of Massachusetts Amherst, MA 01003, USA
∗ Department of Electrical Engineering, University of Notre Dame, IN 46556, USA
I NTRODUCTION M ODEL
In this paper, we present calculations which sup- To model size dependence of single metal ther-
port the size-dependence of the Seebeck coefficient, mocouples, we simulate gallium arsenide (GaAs)
observed by us in recent experiments. These find- under very high doping concentration as a proxy
ings further support the possibility of thermocouples for the metal. Rode s method is used to calculate
formed by a single material using size engineering. the electron mobility by iteratively solving for the
Single-metal nanothermocouples (NTCs) are con- near-equilibrium distribution function in response
structed by joining together two or more segments to a small electric field and a small temperature
from the same metal, but with various cross sections gradient. We include scattering rates for acoustic
as shown schematically in Fig. 1. These devices deformation potential, piezoelectric scattering, polar
exploit the size-dependence of the Seebeck coef- optical, impurity, and boundary roughness scatter-
ficient at the nanoscale. A narrow wire segment ing. The Seebeck coefficient is calculated from S =

−|E|/|∇T | = L(2) /L(1) using transport integrals
is placed between two wider wire segments, thus
forming two junctions a few micron apart. If the L(j) [2]. RSC shown in Fig. 4 is obtained by taking
temperature of one of the junctions is increased, the the difference between Seebeck coefficients at two
resulting open-circuit voltage (VOC) is proportional different sizes.
to the temperature difference between the hot and In un-gated and highly doped nanowires, the
cold junctions and the relative Seebeck coefficient relaxation time due to electron-boundary roughness
between the narrow and wide wire segments. scattering [2] is the given by

In order to directly measure the size-dependent τB (k) = 1−p(k) W
relative Seebeck coefficient (RSC) of single-metal 1+p(k) ν⊥ (k)
NTCs at room temperature, we fabricated a char- where W is size (width or height) of the nanostruc-
acterization platform according to the SEM image ture and ν⊥ (k) is the velocity of carriers normal
shown in Fig. 2. A nanoscale heater, energized to the surface. The specularity parameter p (k) =
by AC currents, is used to selectively increase the exp(−4∆2 k 2 ) accounts for the surface roughness
temperature of the junctions. The resulting VOC (∆) and captures the probability of the electron
is measured with the 2ω method [1]. The four- being reflected specularily at the surface based on
terminal thermometer measures the temperature of the momentum of the electrons. Fig. 4 shows the
the hot junction. The platform was designed by variation in the relative Seebeck coefficient (RSC)
COMSOL to ensure that the temperatures are the as a function of the difference in width between
same at the thermometer and at the hot junction. the wide and narrow segments of the nanowire.
The RSCs were directly obtained using simultane- We observe that the trend in RSC agrees well with
ous measurement of the VOC and the temperature the experimental data in Fig. 2. Finally, to justify
differences for various heater currents. Fig. 3 shows the use of highly doped GaAs as a proxy for a
measured RSC for a constant (50-nm-wide) narrow metal, electrical resistivity is calculated at different
segment cross section and wider segment widths. temperature. For metals, the resistivity should in-
ISBN 978-0-692-50554-0 203

crease linearly with temperature due to increase in
phonon scattering with temperature, which is indeed
observed in GaAs as shown in Fig. 5.
C ONCLUSION
The increase in RSC with increasing difference
between the narrow and wide wire segments is
shown experimentally and numerically, and it leads
to a single-metal NTC functionality.
R EFERENCES
[1] H. H. Roh, J. S. Lee, D. L. Kim, J. Park, K. Kim,
O. Kwon, Novel nanoscale thermal property imaging
technique: The 2ω method. I. Principle and the 2ω signal Fig. 3. Measured relative Seebeck coefficient as a function of
measurement, J. Vac. Sci. Technol., B 24, 2398 (2006). wire width difference between the narrow and wide wires.
[2] Z. Aksamija, I. Knezevic, Thermoelectric properties of
silicon nanostructures, Journal of Computational Electron-
ics,vol 9, 3-4 (2010).
Fig. 1. Schematic of a single-metal nanothermocouple. The red

Fig. 4. Calculated relative Seebeck coefficient with wire width
shading indicates the hot junction and the Th and Tc indicate
difference in a highly donor doped GaAs
the temperatures of the hot and cold junctions, respectively.
Fig. 2. SEM image of a Ni characterization platform. The

hot junction is located on top of the heater separated. They are
thermally connected, but electrically separated by a thin layer
Fig. 5. Resistivity with change in temperature to demonstrate
of alumina.
the metal-like behavior of highly doped GaAs
204
New Computational Perspectives on Scattering and
Transport in III/V Channel Materials
Z. Stanojević, M. Karner, F. Mitterbauer, C. Kernstock
Global TCAD Solutions GmbH., Landhausgasse 4/1a, 1010 Vienna, Austria
Email: {z.stanojevic|m.karner|f.mitterbauer|c.kernstock|h.demel}@globaltcad.com
I. I NTRODUCTION AND METHODOLOGY geometry and long-range effects. We thus propose an extension
III/V materials are considered good candidates for the to the POP scattering model which takes the entire cross-
replacement of silicon as the channel material due to their section geometry into account, with the square modulus of
low effective mass. The combination of non-planar technology the matrix element being
with III/V channel materials shows the most potential since
q2 h̄ωε0 1 1
it benefits from both superior material properties and good |Hn,n ;k,k |2 = 0 3−d N + ±
2L 2 2
electrostatic control [1]. We took the opportunity to review
the methodology for simulating III/V-channels and extend it ∞ 2 1 1
(εr ) − ∇Un,n ;k,k (r)2 dd r, (3)
to non-planar technology. εr∞ εr0

A. Electronic structure and transport U n,n ;k,k (r) = q0 ψn,k
∗
(r )ψn ,k (r )G∞ d
k−k (r, r )d r . (4)
The electronic structure is modeled using an anisotropic Un,n ;k,k (r) is the perturbation matrix element of a single point
effective mass Hamiltonian with non-parabolic correction, charge at r, which is also used for Coulomb scattering. The
h̄ approach is based on the high-frequency electrostatic Green’s
Ekin (1 + αEkin ) = k · m−1 · k, (1)
2 function G∞ k−k
(r, r ), which, when evaluated numerically from
for Γ, L, and X-valleys. This Hamiltonian was shown to be a discretized Poisson equation, takes into account geometry
effective for the modeling of conduction band in III/V nano- and spatial variation of the dielectric constant.
structures [2, 3], as it correctly reproduces the hyperbolic In addition to the POP scattering model, we employ our
dispersion relation at high energies. The non-parabolic correc- surface roughness scattering (SRS) model for non-planar de-
tion is taken into account on three levels: (i) the confinement vices [8] as well as non-polar acoustic, optical intra-valley,
energies, (ii) the density of states, and (iii) the electron group and inter-valley scattering models.
velocity. The non-parabolic distortion is calculated for each The entire modeling chain has been implemented within the
subband individually accounting for the base kinetic energy VSP simulator [9], part of GTS Framework [10].
due to confinement. To obtain the occupancy the Fermi-Dirac II. S IMULATION AND R ESULTS
integrals need to be computed by numerical quadrature, due As a simulation example we used the device architecture
to the non-parabolic correction terms. presented in [1], which consists of an In0.7 Ga0.3 As layer grown
Transport is modeled on top of the subband structure using on top of a In0.52 Al0.48 As buffer, topped by a InP etch stop.
the linearized Boltzmann-transport-equation (LBTE), The fins are cut out from the InGaAs/InP layers and TaSiOx
h̄kξ h̄kξ h̄kξ is deposited isotropically onto the free surfaces.
∑

Sn,n (k, k ) τn,ξ (E)
mn,ξ
− τn ,ξ (E)
mn ,ξ
=
mn,ξ
, (2) We first analyzed the long-channel properties of the InGaAs
n ,k FinFET, by computing the transconductance curve of a 40 µm
where an energy-dependent anisotropic relaxation time tensor long gated fin. The transconductance, shown in Fig. 1, is
τn,ξ is used [4], thus avoiding the shortcomings of the momen- seen to be dominated by POP and surface-roughness. The
tum relaxation time approximation [5]. By solving the LBTE process limited mobilities in Fig. 2 give a more detailed
numerically, channel mobility and conductivity are obtained. picture: POP limits the mobility in the low-density regime,
while scattering off the rough TaSiOx interface limits mobility
B. Scattering models at higher densities.
In contrast to silicon, the most important scattering process To examine the properties of a realistic InGaAs FinFET, we
in III/V semiconductors at room temperature is scattering by combined the electonic structure/transport calculation with the
polar-optical phonons (POP). POP scattering differs from non- device simulator Minimos-NT [11] using self-consistent quan-
polar phonon scattering; unlike non-polar lattice vibrations, tum correction, as shown in Fig. 3. The coupling procedure
which interact with the electrons by altering the band struc- between the two simulators is depicted in Fig. 4. The thus
ture through crystal deformation, polar phonons constitute obtained characteristic and parameters are shown in Fig. 5.
oscillating electric dipoles. Hence, POP scattering constitutes We have demonstrated a framework for simulating novel
an electrostatic perturbation, which is of long-range nature. III/V-based non-planar transistor architectures. The framework
It is unclear whether the bulk perturbation Hamiltonian for is based on physical modeling of electronic structure, scatter-
electron-POP interaction [6, 7] is still applicable as we deal ing, and transport. A combination with a 3D device simulator
with different materials with different dielectric properties. allows to leverage the benefits of physical modeling for the
Also, smaller device feature sizes mean more pronounced simulation of entire devices.
ISBN 978-0-692-50554-0 205

6
Combined
POP-limited
Conductance / µS
5 Roughness-limited
0
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1.1 1.2
Gate voltage / V
Fig. 1. Transconductance curve for a 40 µm long gated fin; Π-gated fin the
has a rectangular shape, 10 nm wide, 20 nm high with a 1 nm InP-layer on top,
and is surrounded by 2 nm-layer of TaSiOx and a metal gate. POP scattering
and SRS dominate transport in the channel. Starting from 0.6 V conductance
becomes severely degraded by SRS. Fig. 3. Electron concentration in an InGaAs FinFET at VG = 1 V and VDS =
0.1 V; fin width and height are 10 nm and 20 nm, respectively, gate length
is 50 nm; the confinement effect due to self-consistent quantum correction is
clearly visible.
Acoustic phonons
106 InP/InGaAs roughness
TaSiOx -interface roughness
Polar-optical phonons
Channel mobility / cm2 /Vs
All processes
ϕ, EFn,p
105
n,p
Minimos-NT Vcorr , µ n,p VSP
104
Fig. 4. Schematic overview of the coupling between Minimos-NT and VSP;

103 The simulation domain is decomposed into slices on each of which Minimos-
106 107 NT invokes a VSP instance (in parallel) and passes the electrostatic potential
Ninv / cm−1 along with the quasi-Fermi-energies. VSP returns the quantum correction
potential and the mobility for the slice which are interpolated back onto the
Fig. 2. Electron mobility plotted against the inversion density in the fin; original 3D simulation domain.
POP scattering and SRS clearly dominate electron transport with SRS on
the TaSiOx -interfaces being far more important than SRS on InGaAs/InP-
intefrace.
103 140
R EFERENCES VDS = 0.1 V
[1] M. Radosavljevic et al., Electron Devices Meeting (IEDM), 102
VDS = 0.5 V 120
2010 IEEE International (2010), pp. 6.1.1–6.1.4. 101
[2] D. Lizzit et al., Electron Devices, IEEE Transactions on 61,
2027 (2014). 100 100
Drain current / µA
[3] O. Baumgartner et al., Book of Abstracts of the 16th

International Workshop on Computational Electronics (IWCE) 10−1 80
(2013), pp. 86–87. 10−2
[4] Z. Stanojević et al., Solid-State Electronics (2015), in press. 60
[5] Z. Stanojevic et al., SISPAD (2014), pp. 181–184. 10−3
[6] H. Fröhlich, Proceedings of the Royal Society A (1937),
10−4 40
Vol. 160, pp. 230–241.
[7] B. Ridley, Quantum Processes in Semiconductors (Oxford 10−5
University Press, 2013). LG = 50 nm 20
[8] Z. Stanojevic et al., Intl. Electron Device Meeting (2013), pp. 10−6
332–335. 10−7 0
[9] O. Baumgartner et al., J. Comput. Electron. 12, 701 (2013). 0.5 0.6 0.7 0.8 0.9 1 1.1 1.2
[10] GTS Framework, http://www.globaltcad.com/framework. Gate voltage / V
[11] Minimos-NT, http://www.globaltcad.com/minimos-nt.
Fig. 5. Transfer characteristic of the InGaAs FinFET shown in Fig. 3 at 0.1 V
and 0.5 V drain bias; the FinFET shows excellent electrostatic control of the
channel with a subthreshold slope of 61.5 mV/dec and a DIBL of 9.19 mV/V.
206
Author Index
Afzalian A. 59 Connolly M. R. 159

Ahmed S. 83, 179, 181, 183 Csaba G. 15, 19
Akis R. 17, 131 Cuniberti G. 49
Aksamija Z. 103, 107, 203 Cusati T. 45
Alam K. 169 Dabral A. 63
Aldegunde M. 89, 91 Das P. 169
Alexander C. 155 Datta S. 57
Ali T. 191 Daugherty R. 105, 193
Alqahtani S. 83 Davoody A. H. 119
Ameen T. 115, 117 Delerue C. 5
Amoroso S. M. 191 Dettmer W. 91
Appenzeller J. 109 Dey H. 19
Aradi B. 39 Dollfus P. 9, 113, 189
Asai Y. 21 Donadio D. 99
Asenov A. 25, 155, 191 Du G. 199
Barker J. 123 Duchemin I. 5
Bermeo J. 55 Dutta M. 75, 161, 163, 187, 195
Bermeo Lopez J. 59 El-Sayed A. M. 97
Bernstein G.H. 19, 137, 203 Ellinghaus P. 93
Berrada S. 67, 171 Elmessary M. A. 89, 91
Bescond M. 67, 79, 171 Fang T. 17
Blawid S. 49 Fang J. 65
Bohorquez-Ballen J. 181 Fang M. 199
Bourdet L. 69 Farid S. 187
Bowden S. 51 Fediai A. 49
Boykin T.B. 41 Ferry D. K. 131
Brinkman D. 17 Fiori G. 45, 73
Brown A. R. 155 Fischetti M. V. 11, 65
Brunetti F. 73 Fonseca J. 115, 201
Bury E. 105 Fortunelli A. 45
Carrillo-Nuñez H. 67 Franckié M. 77
Cavassilas N. 67, 79, 171 Frauenheim T. 39
Chanana A. 125 Frensley W. R. 37, 53
Charles J. 59, 115 Friedman J.S. 13
Chen F. W. 33, 109 Fukuda K. 149
Chen G. H. 39 Gabourie A. J. 119
Chen C. Y. 129 Gamba I. 143
Cheng B. 155 Gelly C. 79
Chimalgi V. 83 Genovese L. 5
Choi M. S. 75, 161, 163, 165 Georgiev V. P. 191
Chu Y. 201 Gerrer L. 25, 191
Chung Nguyen M. 189 Ghosh J. 35
Claus M. 49, 157 Ghosh S. 177
Colomés E. 27, 133 Goes W. 97
Goodnick S. M. 51, 95 Liu X. 199
Grasser T. 97 Liu L. 29
Groeseneken G. 63 Locatelli N. 13
Guo D. 17 Long P. 23, 111
Guo X. 105, 193 Loureiro A. 89, 91
Guo H. 29 Lundstrom M. 85
Guzman D. 55 Maassen J. 85
Hathwar R. 95 Macucci M. 159
Hedin E. R. 135 Magnus W. 63, 151
Herbschleb E. D. 159 Mahapatra S. 125
Huang J. 23, 37, 59 Manfra M. 33
Huang C. 127 Mangold C. 99
Hung Nguyen V. 9, 189 Marconcini P. 159
Iannaccone G. 45, 73 Marian D. 27, 133
Ikonic Z. 141 Markov S. 39
Ilatikhameneh H. 59, 109, 115, 117 Martin P. N. 87
Jang M. 101 Martinez A. 121, 123
Jayasekera T. 179, 181 Masahara M. 149
Jerome J. W. 145 Mei S. 47, 103
Jiang Z. 55 Mejia D. 115, 201
Jirauschek C. 43 Meshik X. 75
Joe Y. S. 135 Michelini F. 79
Jonasson O. 81 Migita S. 149
Kaczer B. 105 Millar C. 155
Kalna K. 89, 91 Miranda H. 5
Karimi F. 3 Mitterbauer F. 205
Karner M. 205 Mizrahi A. 13
Kefayati A. 7 Mizubayashi W. 149
Kelsall R. W. 141 Moors K. 151
Kernstock C. 205 Morales Escalante J. 143
Khair K. A. 183 Mori T. 149
Kivisaari P. 147 Morita Y. 149
Klimeck G. 23, 31, 33, 37, 41, 55, 59, Mothes S. 49, 157
61, 109, 111, 115, 117, Mukherjee S. 173, 187
139, 173, 201 Muralidharan P. 51
Knezevic I. 3, 47, 81, 103, 119, 185 Nagy D. 89, 91
Kommini A. 107, 203 Nakamura H. 21
Kosina H. 7, 167 Narayan A. 21
Kubis T. 33, 41, 59 Nedjalkov M. 93, 155
Lai C. N. 129 Neogi S. 99
Lake R. 169 Neophytou N. 167
Lan Y. 195 Neupane M. 1
Lannoo M. 67, 171 Niquet Y.M. 5, 69
Laturia A. 105 Nishat Md.R. 83
Lee M. Y. 197 Novakovic B. 61
Lemus D. 55, 59 Oh J. H. 101
Li J. 5, 69, 147 Okada Y. 71
Li Y. 129, 197 Oksanen J. 147
Lindberg J. 89, 91 Onofrio N. 55
Liu B. 131 Oriols X. 27, 133
Orlov A. O. 137, 203 Shi Y.B. 47
Orvis M. B. 135 Shi J. 195
Osintsev D. 35 Shi Q. 29
Ota H. 149 Shin M. 101
Pacheco A. 49 Shluger A. L. 97
Paletti P. 73 Sirikumara H. 181
Papp A. 15 Soleimanikahnoj S. 185
Park K. H. 87, 153 Sorée B. 63, 151
Passlack M. 59 Stanojević Z. 205
Pawar R. 73 Strachan A. 55
Pecchia A. 39 Stroscio M.A 75, 127, 161, 163, 187,
Penazzi G. 39 195
Perez S. 55, 59 Sundaresan S. 179
Perić D. 91 Sverdlov V. 35
Poduri S. 127, 161 Sylvia S. 169
Porod W. 15, 19, 137, 203 Szakmany G. P. 137, 203
Pourfath M. 7, 167 Tan Y.P. 41
Povolotskyi M. 23, 31, 41, 55, 59, 111, Tang C. 195
139, 173, 201 Teich D. 157
Price A. 121 Thean A. 63
Qazi S. 105, 193 Thesberg M. 167
Querlioz D. 13 Tominaga J. 21
Quint M. 177 Towie E. 191
Rafferty H. M. 141 Tran V. T. 113
Rahman R. 109, 117, 139 Tsai Y. C. 197
Raleva K. 105 Tulkki J. 147
Rashid M. 179 Ulisse G. 73
Ravaioli U. 87, 153 Valencia D. 31, 117
Raymond L. 171 Valin R. 123
Ringhofer C. 17 Van de Put M.L. 11, 63
Rinke P. 147 Vandenberghe W.G. 11, 65
Roche A. 115 Vasileska D. 17, 51, 105, 193
Rodwell M. 23, 31, 111 Verhulst A.S. 63
Rungger I. 21 Verreck D. 63
Ryndyk D. 49 Viet Nguyen H. 9, 189
Sadi T. 25, 147, 155 Vincent A.F. 13
Sahasrabudhe H. 139 Vodenicarevic D. 13
Saint-Martin J. 9, 113, 189 Wacker A. 77
Samukawa S. 197 Wang L. 25, 155
Sankin I. 17 Wang K. 37
Sano N. 71 Wang Y. 139, 191, 29
Sanvito S. 21 Wickramaratne D. 169
Sarangapani P. 115, 173 Wilson E. 23, 31
Saraniti M. 95 Wimmer Y. 97
Schröter M. 49, 157 Winge D.O. 77
Seifert G. 157 Wirtz L. 5
Selberherr S. 35, 93 Wu Y. 83
Selli D. 99 Xiong S. 99
Sengupta A. 125 Yam C.Y. 39
Shaik A. 105 Yasuda T. 149
Yeo Y.C. 59
Yin G. 169
Yin L. 199
Yoshida K. 71
Zeng L. 199
Zhang L. 199
Zhen Z. 27
Zhu E. 29
ISBN 978-0-692-50554-0
90000>
9 780692 505540
www.nanoHUB.org/groups/IWCE2015

IWCE AbstractsBook Final2015

Uploaded by

Copyright:

Available Formats

IWCE AbstractsBook Final2015

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

IWCE AbstractsBook Final2015

Uploaded by

Copyright:

Available Formats

18th International Workshop on

18th International Workshop on

IWCE 2015 hosted by:

Partial sponsorship kindly provided by:

Welcome to West Lafayette and Purdue University!

Local Organizing Committee

Gerhard Klimeck (Chair) Purdue University

M.P. Anantram University of Washington, USA

Advisory Board Committee

Sheila Aboud Synopsis, USA

7:00 A.M. Breakfast/Registration South Ballroom, Purdue Memorial Union

Session Chair: Yiming Li

Session Chair: Marc Bescond

7:00 A.M. Breakfast/Registration South Ballroom, Purdue Memorial Union

Session 4 - Multiscale Simulations

Session Chair: Irena Knezevic

Session Chair: Robert Kelsall

7:00 A.M Breakfast

Session 7 - Semiclassical Transport

Session Chair: Massimo Fischetti

Session Chair: Massimo Macucci

P1 H. Ilatikhameneh, F. W. Chen, J. Appenzeller, R. Rahman, and G. Klimeck (Purdue University)

P3 V. T. Tran, J. Saint-Martin, and P. Dollfus (IEF; Université Paris-sud; CNRS)

P4 P. Sarangapani, D. Mejia, J. Charles, H. Ilatikhameneh, T. Ameen, A. Roche, J. Fonseca, and

P5 T. A. Ameen, H. Ilatikhameneh, D. Valencia, R. Rahman, and G. Klimeck (Purdue University)

P6 A. H. Davoody, A. J. Gabourie, and I. Knezevic (University of Wisconsin-Madison)

P7 A. Price and A. Martinez (Swansea University)

P8 A. Martinez and R. Valin and J. R. Barker (Swansea University; University of Glasgow)

P10 C. Huang, S. Poduri, and M.A. Stroscio (University of Illinois)

P12 B. Liu, R. Akis, and D. K. Ferry (Arizona State University)

P13 E. Colomés, D. Marian, and X. Oriols (Universitat Autònoma de Barcelona)

P14 E. R. Hedin, M. B. Orvis, and Y. S. Joe (Ball State University)

P17 H.M. Rafferty, R.W. Kelsall and Z. Ikonic (University of Leeds)

P18 J. A. Morales Escalante, I. M. Gamba (University of Texas)

P19 J.W. Jerome (Northwestern University)

P23 K.H. Park and U. Ravaioli (University of Illinois)

P27 M. S. Choi, S. Poduri, M. Dutta, M. A. Stroscio (University of Illinois)

P28 M. S. Choi, M. Dutta, and M. A. Stroscio (University of Illinois)

P33 S. Mukherjee, P. Sarangapani, M. Povolotskyi and G. Klimeck (Purdue University)

P34 M. T. Quint and S. Ghosh (University of California)

P35 S. Sundaresan, M. Rashid, T. Jayasekera, and S. Ahmed (Southern Illinois University)

P36 J. Bohorquez-Ballen, H. Sirikumara, S. Ahmed, T. Jayasekera (Southern Illinois University)

P37 K. A. Khair and S. S. Ahmed (Southern Illinois University)

P38 S. Soleimanikahnoj and I. Knezevic (University of Wisconsin-Madison)

P39 S. Mukherjee, S. Farid, M. Dutta and M.A. Stroscio (University of Illinois)

P42 X. Guo, S. Qazi, R. Daugherty, D. Vasileska (Arizona State University)

P48 Z. Stanojević, M. Karner, F. Mitterbauer, C. Kernstock (Global TCAD Solutions GmbH.)

theoretical and experimental synergetic efforts in

Graphene, a two-dimensional carbon material dielectric function of electronic materials with an

J. Li∗ , H. Miranda† , Y.M. Niquet∗ , L. Genovese∗ , I. Duchemin∗ , L. Wirtz† and C. Delerue‡

102 1.36×1013 cm−2

0.0 0.5 1.0 1.5 2.0

Since the successful experimental isolation of Lch=20nm

graphene in 2004, ultra-thin two-dimensional struc-

Drain Current (mA/µm)

CI scattering effects and boost the mobility [3, 4

remote phonon (RP) scattering [6] . The source of 2

of 9 mA/µm and a high Ion /Ioff ratio of about

Fig. 4. (E-ky)-maps of transmission probability showing the

P1 H. Ilatikhameneh, F. W. Chen, J. Appenzeller, R. Rahman, and G. Klimeck (Purdue University)

P4 P. Sarangapani, D. Mejia, J. Charles, H. Ilatikhameneh, T. Ameen, A. Roche, J. Fonseca, and

P5 T. A. Ameen, H. Ilatikhameneh, D. Valencia, R. Rahman, and G. Klimeck (Purdue University)

P6 A. H. Davoody, A. J. Gabourie, and I. Knezevic (University of Wisconsin-Madison)

P28 M. S. Choi, M. Dutta, and M. A. Stroscio (University of Illinois)

P33 S. Mukherjee, P. Sarangapani, M. Povolotskyi and G. Klimeck (Purdue University)

P34 M. T. Quint and S. Ghosh (University of California)

P36 J. Bohorquez-Ballen, H. Sirikumara, S. Ahmed, T. Jayasekera (Southern Illinois University)

P42 X. Guo, S. Qazi, R. Daugherty, D. Vasileska (Arizona State University)

Fig. 4. Simulated contact resistance for different Φ and