Generalized matching from theory to application

Abstract

We present a novel approach for post-mapping optimization. We exploit the concept of generalised matching, a technique that finds symbolically all possible matching assignments of library cells to a multi-output network specified by a Boolean relation. Several objectives are targeted: area minimization under delay constraints; power minimization under delay constraints; and unconstrained delay minimization. We describe the theory of generalized matching and the algorithmic optimization required for its efficient and robust implementation. A tool based on generalized matching has been implemented and tested on large examples of the MCNC'91 benchmark suite. We obtain sizable improvements in: speed (6% in average, up to 20.7%); area under speed constraints (13.7% an average, up to 29.5%); and power under speed constraints (22.3% in average, up to 38.1%).