Circuit-wise buffer insertion and gate sizing algorithm with scalability

Most existing buffer insertion algorithms, such as van Ginneken's algorithm, consider individual nets and therefore often result in high buffer cost due to lack a global view. Thus, circuit-wise buffering is necessary to reduce buffer cost. Recently, some circuit-wise buffering algorithms are proposed. However, these algorithms are based on heuristics which are not scalable in handling large circuits. In this paper, we present a scalable circuit-wise algorithm with three novel features. (1) A linear modeling of nonlinear delay versus cost tradeoff. Due to the similar nature of buffer insertion and gate sizing, gate sizing is handled in such a manner. (2) A dynamic critical sink selection procedure to solve multiple-sink net. Multiple-sink nets have been problems for previous circuit-wise buffering algorithms. (3) A circuit partition technique to divide the circuit into sub-circuits and apply divide-and-conquer scheme. This technique provides high scalability for the algorithm. Experiments on ISCAS85 circuits show that the new algorithm achieves 17X speedup compared with Sze's path based algorithm. In the meantime, it saves 16.0% buffer cost and 4.9% gate cost without increasing circuit delay. Furthermore, the running time of a testcase in ITC99 with approximate one hundred thousand gates is less than 11 minutes, which demonstrates the scalability of the new algorithm.