Abstract

This paper is motivated by the need to generate classes of circuits that are closely related. Ideally, such circuits not only have identical size and distributions of cells and nets but also resemble each other closely in terms of cell clusters and interconnect patterns. Such classes are essential for welldefined performance evaluation of layout algorithms. For example, if the heuristic in the algorithm is being optimized for multiplier structures, then all circuits in the class should have a multiplierlike structure. Such tight control of circuit class generation has not been demonstrated with recent techniques, including the generation of clone and mutant classes. We introduce an approach to generating circuit classes called siblings. Siblings retain all significant characteristics of their parent circuit. We demonstrate that siblings inherit the essential characteristics of the parent only if we control, for each connected component, the diameter of the clusters and the arrangement ...

Citations