EECS 298-11: CAD Seminar Wednesday, April 19, 5pm 531 Cory Hall, Hogan Room How Matrix-Free Iterative Methods Have Changed Computer Simulation of Circuits, Devices, Interconnect, and Micro-Electro-Mechanical Systems Prof. Jacob White MIT Iterative matrix solution methods, like relaxation, have been used in computer simulation programs for several decades and their use has greatly expanded in the last several years. This expanded use is primarily due to the development of robustly converging Krylov-subspace methods for general nonsymmetric problems, such as preconditioned GMRES and QMR. Perhaps surprisinglyf, the most important practical aspect of using Krylov-subspace methods is not that they are faster than direct factorization for well-conditioned problems. What has proven to be much more important is that Krylov-subspace methods only require the ability to compute a matrix-vector product and do NOT require an explicit representation of the matrix. In this talk, we will show how this simple "matrix-free" idea can be exploited to develop fast algorithms for periodic steady-state analysis of large nonlinear circuits and parallel algorithms for transient simulation of semiconductor devices. We will then show how the matrix-free idea can be combined with sparsification techniques to perform very fast electromagnetic analysis of complex 3-D structures. Finally, we will show how the matrix-free idea can be used as a software engineering tool, by demonstrating its use as a technique for coupling existing simulation programs to perform coupled electromechanical simulation of micro-electro-mechanical systems. Future Seminars: April 26 - Tom Parks, UC Berkeley "Dataflow Process Networks" May 3 - Antun Domic, Cadence "Some problems in submicron technology and their effect on design methods and CAD tools" May 5 - Georges Gielen, KU-Leuven (Special seminar, Friday, 2PM) "Research Developments in Analog CAD at KU-Leuven" May 10 - Stan Liao, MIT "Code Generation and Optimization for Embedded DSP Processors" End of semester