Network for Computational Nanotechnology (NCN) NEMO5 on Blue Waters - A Flexible Package for Nanoelectronics Modeling Problems Jim Fonseca Network for Computational Nanotechnology PRAC - Accelerating Nano-scale Transistor Innovation PI: Gerhard Klimeck Blue Waters Symposium May 2015
NEMO5 - Bridging the Scales From Ab-Initio to Realistic Devices Ab-Initio TCAD Approach: Goal: • Device performance with realistic • Ab-initio: • Bulk constituents extent, heterostructures, fields, etc. • Small ideal superlattices for new / unknown materials • Map ab-initio to tight binding Problems: • Need ab-initio to explore new (binaries and superlattices) • Current flow in ideal structures material properties • Study devices perturbed by: • Ab-initio cannot model non- • Large applied biases equilibrium. • Disorder • TCAD uses quantum corrections • Phonons 2
NEMO5 – A multiscale simulation tool for nanoelectronic modelling • Multiscale/multiphysics • Empirical tight binding • NEGF, DD, QTBM, EM • Electron core, k.p, mode space • Ohmic and Schottky contacts • Scattering optical and acoustic • Phonons • Strain models-VFF, Keating, Lazarenkova • External magnetic fields • Solves • Atomistic strain • Electronic band structures • Charge density • Potential 30nm • Current • 4-level MPI parallelization • bias, energy, momentum, space 3
NEMO5 and nanoHUB • Distribution and Support Group on nanoHUB.org » https://nanohub.org/groups/nemo5distribution » Source code, example, discussion forum, run NEMO5 on Purdue Resources • nanoHUB.org » 330,000 annual users » 4,200 resources (video lectures, presentations, tutorials, etc.) » 330 simulation tools » Nanoelectronics, nanophotonics, materials science, molecular electronics, carbon-based systems, Microelectromechanical systems » 4,200 resources (video lectures, presentations, tutorials, etc.) » NEMO5 T ools Quantum Dot Lab Crystal Viewer Bandstructure Lab 4 …
Network for Computational Nanotechnology (NCN) Non-equilibrium Green's functions method: Non-trivial and disordered leads Yu He, Yu Wang, Tillmann Kubis, Gerhard Klimeck
Problem: assumption of periodic contacts in NEGF contradicts experiment semi-infinite periodic contacts. Common self- Source Drain energy methods Sancho Rubio, Non-periodic geometries transfer matrix But in the real world… http://www.electroiq.com/articles/sst /2010/12/iedm-reflections_.html Random alloy Roughness Periodic assumption contradicts realistic contacts S. Koenig et al, Appl. Phys. Lett, Vol. 104, pp. 103106, 2014 How to solve non-periodic contacts? Q. Liu, et al, IEDM p.229 2013 6
General Lead Method: problem & idea Problem: No exact solution for semi-infinite systems unless periodicity assumed Approximate solution Physically correct Numerically solvable for arbitrary contact structures CAP CAP J. Driscoll et al, Phys. Rev. B. Vol. 78, pp. 245118, 2008 Idea: extend complex absorbing potential (CAP) method Non-periodic contact : Hamiltonian for explicit contact segments; CAP serves as scattering : physical assumption of contacts; Efficient, memory thin : converge within finite iterations; 7
Example: SiGe random alloy Si Si 0.5 Ge 0.5 Si device length Si 0.5 Ge 0.5 Si 0.5 Ge 0.5 Si 0.5 Ge 0.5 Example: 3x3nm Si 0.5 Ge 0.5 nanowire in sp3d5s* tight binding Device length 20nm and 6nm Results averaged over 50 samples Justification: With same effective alloyed disorder in contacts, expected transmission has weak dependence of device length Yu He, Yu Wang, Gerhard Klimeck, Tillmann Kubis, "Non-equilibrium Green's functions method: 8 Non-trivial and disordered leads” Appl . Phys. Lett. 105, 213502 (2014)
Example: SiGe random alloy Si Si 0.5 Ge 0.5 Si device length Si 0.5 Ge 0.5 Si 0.5 Ge 0.5 Si 0.5 Ge 0.5 Alloyed contact yield virtually device length independent transmission; DOS of contacts match device better less reflections of electrons; General lead approach works well for contacts with alloy randomness. 9
Example: SiGe random alloy 10 20 10 20 Si Si 0.5 Ge 0.5 Si cm -3 cm -3 device length 20nm Si 0.5 Ge 0.5 Si 0.5 Ge 0.5 Si 0.5 Ge 0.5 gate length 8nm Non-trivial 45% decrease contacts in on-current. critical in transport simulations 10
Network for Computational Nanotechnology (NCN) Bilayer Graphene: a Good candidate for Transistors? Fan Chen, Hesameddin Ilatikhameneh, Rajib Rahman, Gerhard Klimeck
Graphene Transistor ON/OFF < 10 • Graphene has a zero band gap • It has a good high ON current, but it can’t be turned off We need to achieve: • Large ON/OFF ratio needed in transistors (~10 5 ) • Small OFF current -> Low power consumption http://www.jameshedberg.com/img/samples/ 12 12 https://www.kth.se/en/ict/forskning/ickretsar/
Bilayer Graphene Bandgap vs. E-Field Bilayer Graphene Band Gap with D av 300 250 Band Gap [meV] 200 150 100 http://jarilloherrero.mit.edu/research/gated-bilayer-graphene/ 50 0 2 4 6 8 10 D av [V/nm] Bandstructure Small field Large field B and Gap Bilayer Graphene: Create a Band-Gap by Electric Field • Control band-gap by applying vertical electric field 13 13
NEMO5 – realistic atomic approach Tao. Chu, Prof. Zhihong Chen Purdue Daniel Mejia Challenge: • Matrix size ~ 64 million • Inverse, Eigenvalues ① Limitation from fabrication technique, orbitals short channel effect, gate leakage … 3.2 million ② Device size is typically 100nm (thick) x . . . . 200nm (long) x 20nm (wide) 3.2 million . . . . • 3.2 million atoms in simulation . . . . Matrix . . . . 20x20 14 14 http://chemwiki.ucdavis.edu/
Bilayer Graphene: Open band gap IdVg 3 TG 20 10 1 5 I( A/ m) 2 ON/OFF = 100 BG 1 V BG = -1.75 V 0.5 V ds = 0.002 V Image courtesy Gianluca Fiori Fermi = 0 eV 2 Band Gap opens through the change of 0.2 Top Gate -1 -0.5 0 0.5 1 1.5 2 2.5 3 V TG (V) V TG = 3.6V V TG = 1.15V V TG = -1.4V 0.2 0.2 0.2 3 2 Band Edge [eV] 1 Band Edge [eV] Band Edge [eV] 0 0 0 -0.2 -0.2 -0.2 -0.4 -0.4 -0.4 10 20 30 40 50 10 20 30 40 50 10 20 30 40 50 15 15 x[nm] x[nm] x[nm]
Dynamic Band gap Physical structure Dynamic band gap: V TG |V BG | ↑ ⇒ E ↑ ⇒ E g ↑⇒ I ON /I OFF ↑ S D Back oxide V BG Ec V ds = 0.002 V E F,Source E F,Drain Fermi = 0 eV V BG Ev Band Gap modulated by back gate 16 16
NEMO5 – self consistent simulation Construct Hamiltonian 64 million x 64 million matrix Add Potential Determine energy grid V BG 64 M x 64 M Eigenvalue Convergence? Calculate charge density 64 M x 64 M Inverse Calculate Potential Hundreds of data points One I-V data point 17 17
iNEMO Group • PI: Gerhard Klimeck • 3 Research Faculty: Tillmann Kubis, Michael Povolotskyi, Rajib Rahman • Research Scientist: Jim Fonseca • 2 Postdocs: Bozidar Novakovic, Jun Huang • Students: Tarek Ameen, Robert Andrawis, James Charles, Chin-Yi Chen, Fan Chen , Yuanchu (Fabio) Chen, Rifat Ferdous, Jun Zhe Geng, Yu He , Yuling Hsueh, Hesameddin Ilatikhameneh , Zhengping Jiang, Daniel Lemus, Pengyu Long, Daniel Mejia Padilla, Kai • Ryan Mokos Miao, Samik Mukherjee, Harshad • Intel, Samsung, Philips, TSMC Sahasrabudhe , Prasad Sarangapani, Saima Sharmin, Yaohua Tan, Yui Hong (Matthias) Tan, Archana Tankasala , Daniel Valencia Hoyos, Kuang Wang, Yu Wang , Evan Wilson 18 18
Recommend
More recommend