Automatic generation of synthetic sequential benchmark circuits

Abstract

The design of programmable logic architectures and supporting computer-aided design tools fundamentally requires both a good understanding of the combinatorial nature of netlist graphs and sufficient quantities of realistic examples to evaluate or benchmark the results. In this paper, the authors investigate these two issues. They introduce an abstract model for describing sequential circuits and a collection of statistical parameters for better understanding the nature of circuits. Based upon this model they introduce and formally define the signature of a circuit netlist and the signature equivalence of netlists. They give an algorithm (GEN) for generating sequential benchmark netlists, significantly expanding previous work (Hutton et al, 1998) which generated purely combinational circuits. By comparing synthetic circuits to existing benchmarks and random graphs they show that GEN circuits are significantly more realistic than random graphs. The authors further illustrate the viabilty of the methodology by applying GEN to a case study comparing two partitioning algorithms.