Adaptive IC Design via Stochastic Optimization

Carnegie Mellon University, Pittsburgh PA

Investigators

Abstract

ID: 0702278 Title: Adaptive IC Design via Stochastic Optimization PI: Larry Pileggi Inst: Carnegie Mellon University Abstract The research of this project addresses the robust design of integrated circuits in the presence of large-scale variations from both the manufacturing process and the operating environment. As integrated circuit technologies reach nanoscale feature sizes, the fluctuations in circuit parameters due to manufacturing and operating environment increase significantly. To address this problem for the particularly challenging design of analog and radio frequency integrated circuits, we propose a new adaptive design methodology to dynamically configure and tune the circuit to accommodate all possible variations. Namely, our methodology will attempt to create circuits that offer self-adaptation to external changes and disturbances. The proposed research targets two major problems: (1) how to create new tunable circuit architectures that can adaptively perform self-testing and self-configuration, and (2) how to analyze and optimize such adaptive circuits to explore the tradeoff between performance and cost. We will model both manufacturing and environmental variations as random variables and borrow powerful mathematical methods from statistics to solve these unique problems as posed by our adaptive design methodology. The success of this project will stimulate a paradigm shift in today's electronic system design toward stochastic design. Such a methodology could ultimately enable nano-chips and bio-chips, which are significantly limited by design robustness issues, to reach commercial fruition. In addition, this work will bring adaptive methods into the traditional engineering curriculum, thereby providing the students with concrete application examples for mathematical techniques that are increasingly important for their long term careers.