To provide a better understanding of physical design algorithms and the underlying circuit architecture they are targeting, we need to exercise the algorithms and architectures with many benchmark circuits. The lack of existing benchmarks makes us consider the automatic generation of netlists. In this paper, we formally define the equivalence class of circuit "clones" of a given seed circuits, based upon physical properties of the seed circuit's netlist graph. Using these equivalence classes of circuits, a given seed circuit can be used to generate many similar circuits. A more finely grained statistical analysis of algorithm behaviour can then be obtained from using the multiple benchmarks than would be available from using the seed benchmark alone.
Michael D. Hutton, Jonathan Rose