Flexible clustering and floor planning approach to standard cell placement using hierarchical simulated annealing

Abstract

Conventional simulated annealing algorithm, which works on the flattened circuit, has a very large search space and hence requires a huge computational effort. In this research, a hierarchical strategy was established. The placement problem is divided into three sub-problems: clustering, floorplanning and final placement. This research is different from a previous work, which use similar strategy, by allowing clusters to occupy more than one row and have variable aspect-ratio in the stage of floorplanning. In our implementation, the sub-tasks of clustering and floorplanning could be finished with small amount of CPU time. The sub-task of final placement was done by a modified version of the well-known TimberWolfSC program. It was found that, as a consequence of the hierarchical approach, the search-efficiency of the final placement is improved significantly over the conventional non-hierarchical approach.