Rational Control of Multi-level Stochastic Design

Abstract

Search in a complex design space (e.g. in VLSI design) is only feasible when it is factored into smaller ones. A common and useful factoring is the hierarchy of abstraction levels, where each abstraction level has a distinct ontology, evaluation function, and search behavior. Such problems are then attacked by linking together a number of stochastic search programs (e.g., multi-start hill climbers), one for each level. At each point in the design process it is necessary to choose whether to continue looking for a better design at the current level, or working further on some design at a higher or lower level. We present a rational, i.e. utility-based, method for this choice, together with some experiments in the domain of VLSI placement and routing.

Published in the Proceedings of the IASTED International Conference on Modelling and Simulation, May 15-17, 1997, Pittsburgh, PA.

Back to Brian Davison's publications