Polynomial Chaos Expansion (PCE) Methods with the Xyce Circuit simulator

Speaker Name
Eric Keiter
Speaker Title
Distinguished Member of Technical Staff, Electrical Models & Simulation Dept.
Speaker Organization
Sandia National Laboratories
Start Time
End Time
Location
Via Zoom Presentation
Organizer
Matthew Guthaus

Abstract:

Accurate analog circuit simulation is an important part of design and verification of electrical and electronic circuits and systems. At modern technology nodes, analog, SPICE-accurate simulation can be a prohibitive development bottleneck. Traditional circuit simulation, such as with SPICE, does not scale well beyond tens of thousands of unknowns, due to a reliance on direct matrix solver methods. This talk contains an overview of the open-source analog circuit simulation tool, Xyce, which was designed from the ground-up to perform large-scale parallel circuit analysis, primarily using message-passing. Current capabilities of the simulation tool will be discussed, including code design philosophy, analysis methods, and parallel algorithm strategies. One new capability that will get extra attention in the second half of the talk will be the recent development of Polynomial Chaos methods (PCE) in Xyce. Uncertainty Quantification (UQ) is an important topic in electronic design automation (EDA), and it is standard practice to apply sampling methods such as Monte Carlo to assess the impact of parametric uncertainties. However, sampling can be prohibitively expensive, especially when applied to a large number of parameters. To mitigate this issue, several versions of the generalized polynomial chaos expansion (PCE) have been developed in Xyce. These methods are applicable to both Gaussian and non-Gaussian random parameters and can be used to perform efficient uncertainty quantification analysis of nonlinear transistor circuits. These methods include the fully intrusive spectral projection PCE method (sometimes called the stochastic Galerkin (SG) method), as well as non-intrusive and semi-intrusive forms of both projection and regression style PCE. Comparisons of these methods, for several circuit UQ problems will be discussed.

Bio:

Eric Keiter is a Distinguished Member of Technical Staff in the Electrical Models & Simulation Department at Sandia National Laboratories. He received his PhD from the University of Wisconsin-Madison 1996, in the field of plasma simulation. After coming to Sandia in 1998 he was one of the original developers of the Xyce parallel circuit simulator. His research focus has been on novel algorithms and numerical simulation techniques for analog, RF and mixed signal systems and most recently uncertainty quantification. Dr. Keiter was a co-recipient of the 2008 R&D 100 Award for the Xyce circuit simulator and has served on the Technical Program Committees of DAC, ICCAD, and ASPDAC. He is current an associate editor of IEEE Transactions for Computer-Aided Design of Circuits and Systems.

Zoom link:  https://ucsc.zoom.us/j/97513249812

Event Type
Event