Skip to main content - access key m.
Skip to main navigation - access key n.

Monday, March 8, 2010 12:15 p.m.
NSERL 3.204

 

 

 

 

 

 

 

 

 

 

 

 

 

“New Computational Paradigms for Large-Scale Atomistic
Simulations in Nanoelectronics”

Dr. Eric Polizzi, University of Massachusetts Amherst

Abstract 
Atomistic simulations have become critical in nanoelectronics for supplementing the experimental approaches for obtaining detailed electronic structure properties and reliable characterization of emerging materials and devices. The quest for ever higher levels of detail and realism in these simulations presents formidable modeling and computational challenges. The objective of this research is to enhance significantly the role of computational modeling and high-performance computing for addressing the current limitations in large-scale electronic structure and electron transport calculations.

In this presentation we aim at introducing new computational paradigms (changing conceptions of what is computable) by revisiting the traditional ab-initio modeling approaches and by going beyond the use of standard numerical algorithms and library packages. The new numerical framework benefits from well-suited combinations of specific mathematical and numerical methods that fully exploit the potentiality of real-space mesh techniques (finite element method) in achieving linear scaling performances. The significant challenges posed by large-scale simulations have been addressed by the development of two new high-performance computing algorithms and their implementations on parallel architectures: (i) The SPIKE algorithm for solving linear systems, which became an Intel offering and (ii) a highly efficient new algorithm design and library package named FEAST for solving the eigenvalue problem. Both solvers are capable of major broader impacts in science and engineering. We will particularly point out the fundamental implications of the FEAST solver in nanoelectronics modeling and simulations. Time-independent and time-dependent simulation results obtained on carbon nanotubes using these techniques will be presented.

Bio
Dr. Eric Polizzi is an assistant professor of electrical and computer engineering at the University of Massachusetts Amherst. He received a BS and MS in theoretical and computational physics and a PhD in applied mathematics, all from the University of Toulouse France, and he has been a postdoctoral researcher in electrical engineering and a senior research scientist in computer sciences, both at Purdue University. Since 1998 he has been working in the interdisciplinary research field of computational nanoelectronics at the interfaces of the above disciplines. His research combines different fundamental modeling strategies from physics to algorithmic, and it aims at providing a sound basis to face the challenges in large-scale simulations for applications ranging from material sciences and chemistry to nanoelectronics and bio-nanotechnology. He is the developer of the nanoelectronic finite element package NESSIE as well as SPIKE and FEAST. His current research is funded by NSF, Intel and SRC. His awards include an NSF Career Award.