Simple yet effective techniques to improve flat multiway circuit partitioning

Partitioning plays an important role in solving large, complicated design problems. For large benchmarks iterative improvement methods based on the conventional Kernighan and Lin heuristic (1970) produce inferior results. These algorithms suffer from a propensity to freeze the movement of large cells that would immediately violate the balance constraint. To enhance the stability and quality of partitioning results a novel algorithm intensifies the capability of escaping from local optimal by first releasing the size constraint temporarily and enforcing it at the end of a run and controlling the migration direction and secondly by adding a dynamic hill climbing capability to avoid the possibility of getting trapped in a local minima. These simple heuristic techniques reduce the total cut on average by 30% and runtime by at least 70-90%.