Rapid progress in manufacturing greatly challenges to the VLSI physical design in both speed and performance. A fast detailed placement algorithm, FAME is presented in this paper, according to these demands. It inhe...Rapid progress in manufacturing greatly challenges to the VLSI physical design in both speed and performance. A fast detailed placement algorithm, FAME is presented in this paper, according to these demands. It inherits the optimal positions of cells given by a global placer and exact position to each cell by local optimization. FM Mincut heuristic and local enumeration are used to optimize the total wirelength in y and x directions respectively, and a two way mixed optimizing flow is adopted to combine the two methods for a better performance. Furthermore, a better enumeration strategy is introduced to speed up the algorithm. An extension dealing with blockages in placement has also been discussed. Experimental results show that FAME runs 4 times faster than RITUAL and achieves a 5% short in total wirelength on average.展开更多
基金Project Supported by National Natural Science Foundation of China( Grant No.697760 2 7) and by National973 Key Projectof China (
文摘Rapid progress in manufacturing greatly challenges to the VLSI physical design in both speed and performance. A fast detailed placement algorithm, FAME is presented in this paper, according to these demands. It inherits the optimal positions of cells given by a global placer and exact position to each cell by local optimization. FM Mincut heuristic and local enumeration are used to optimize the total wirelength in y and x directions respectively, and a two way mixed optimizing flow is adopted to combine the two methods for a better performance. Furthermore, a better enumeration strategy is introduced to speed up the algorithm. An extension dealing with blockages in placement has also been discussed. Experimental results show that FAME runs 4 times faster than RITUAL and achieves a 5% short in total wirelength on average.
