We present an improved partitioning-based placement tool in terms of the half-perimeter wirelength. In contrast to placement tools that exploit white spaces to reduce the routing congestion, our placement tool exploits white spaces to reduce the total wirelength. In particular, we generate white spaces during the legalization step and treat white spaces as pseudo cells during the detailed placement step. On IBM benchmarks, our results are better than those of Feng Shui  and Capo  by 1.8% and 11.1%, respectively, on the average. Our results are comparable to those of Dragon . On the synthetic benchmarks PEKO, our placement tool outperforms all other placement tools except mPL . On another suite of synthetic benchmarks PEKU with global nets, our results are the best among all the existing public academic placement tools.
Date of this Version