Knowledge Engineering Issues in VLSI Synthesis

Abstract

This paper explores VLSI synthesis and the role that traditional AI methods can play in solving this problem. VLSI synthesis is hard because interactions among decisions at different levels of abstraction make design choices difficult to identify and evaluate. Our knowledge engineering strategy tackles this problem by organizing knowledge to encourage reasoning about the design through multiple levels of abstraction. We divide design knowledge into three categories: knowledge about modules used to design chips; knowledge used to distinguish and select modules; and knowledge about how to compose new designs from modules. We discuss the uses of procedural and declarative knowledge in each type of knowledge, the types of knowledge useful in each category, and efficient representations for them.