Cell Spreading Optimization for Force-Directed Global Placers

Xueyan Wang,Yici Cai,Qiang Zhou
DOI: https://doi.org/10.1109/iscas.2017.8050572
2017-01-01
Abstract:Wirelength is a traditional optimization objective in global placement algorithms. To eliminate cell overlaps, spreading forces need to be added to pull cells away from highly congested areas. At the same time, to optimize wirelength, the quadratic nature should be maintained. In this paper, several techniques are proposed to optimize spreading force orientation and modulation. Specifically, a percentage-driven method is proposed to cluster overfilled bins, followed by a center-uniformization algorithm to demarcate the expand region for the cluster. Finally, cells are distributed evenly within each expand region while maintaining relative cell positions and minimizing cell displacements. Experimental results show that the global placer that integrated with the proposed strategies achieves 13.0% and 2.1% less wirelength compared with Capo10.5 and Aplace3, respectively.
What problem does this paper attempt to address?