Circuit partitioning with logic perturbation

TL;DR: This work proposes an efficient method that is able to preserve a local optimal solution in the graph domain while a different graph, representing the same circuit, is generated.

Abstract: Traditionally, the circuit partitioning problem is done by first modeling a circuit as a graph and then partitioning is performed on the modeling graph. Using the concept of alternative wires, we propose an efficient method that is able to preserve a local optimal solution in the graph domain while a different graph, representing the same circuit, is generated. When a conventional graph partitioning technique reaches a local optimal solution, our proposed technique generates a different graph that is logically equivalent to the original circuit, and that has equal or better partitioning solution. Faced with a different graph which is newly generated, together with a currently good partitioning solution, a conventional graph partitioning technique may then escape from the local optimum and continue searching for better solutions in a different graph domain. The proposed technique can be combined with almost any graph partitioner. Experiments show encouraging results.