SHF:Small:Graph Sparsification Approach to Scalable Parallel SPICE-Accurate Simulation of Post-layout Integrated Circuits

点击翻译按钮获取中文摘要

Unlike traditional fast SPICE simulation techniques that rely on a variety of approximation approaches to trade off simulation accuracy for greater speed, SPICE-accurate integrated circuit (IC) simulations can truthfully predict circuit electrical behaviors, and therefore become indispensable for design and verification of nanoscale ICs. However, for post-layout nanoscale circuits, using traditional SPICE-accurate simulation techniques to encapsulate multi-million or even multi-billion devices coupled through complex parasitics can be prohibitively expensive, and thus not applicable to large IC designs, since the runtime and memory cost for solving large sparse matrix problems using direct solution methods will increase quickly with the growing circuit sizes and parasitics densities. To achieve greater simulation efficiency and capacity during post-layout simulations, preconditioned iterative solution techniques have been recently proposed to substitute the direct solution methods. However, existing preconditioned methods for post-layout circuit simulations are typically designed with various assumptions and constraints on the circuit and systems to be analyzed, which therefore cannot be effectively and reliably applied to general-purpose SPICE-accurate circuit simulations. In this research project, the PI will study efficient yet robust circuit-oriented preconditioning approaches for scalable SPICE-accurate post-layout IC simulations by leveraging recent graph sparsification research. By systematically sparsifying linear/nonlinear dynamic networks originated from dense parasitics components and complex device elements of post-layout circuits, scalable, and more importantly, parallelizable preconditioned iterative algorithms will be investigated and developed by the PI to enable much greater speed and capacity for SPICE-accurate IC simulations in both time and frequency domains.The successful completion of this work will immediately benefit the semiconductor industries. The algorithms and methodologies to be developed through this project will be integrated into undergraduate/graduate level VLSI design/CAD courses, while the research results will be broadly disseminated to major semiconductor and EDA companies for potential industrial applications. The CAD tools developed under this research plan will also be exchanged with collaborating industrial partners. The acquired experience in the proposed research plan is also likely to contribute to computing advances in other science and engineering fields, impacting broader research areas that are related to large/complex system modeling and simulation.