Resynthesis of multi-level circuits under tight constraints using symbolic optimization

Abstract

We apply recently introduced constructive multi-level synthesis in the resynthesis loop targeting convergence of industrial designs. The incremental ability of the resynthesis approach allows more predictable circuit implementations while allowing their aggressive optimization. The approach is based on a very general symbolic decomposition template for logic synthesis that uses information-theoretical properties of a function to infer its decomposition patterns (rather than more conventional measures such as literal counts). Using this template the decomposition is done in a Boolean domain unrestricted by the representation of a function, enabling superior implementation choices driven by additional technological constraints. The symbolic optimization is applied in resynthesis of industrial circuits which have tight timing constraints yielding their much improved timing properties.