关键词: 三维片上网络, 布图算法, B*-tree, 离散量子粒子群算法, 模拟退火算法, 粒子群算法

Abstract: The performance of three dimensional network-on-chip is much better than that of two dimensional network-on-chip in many aspects, so that it has become a hot research topic. The floorplanning algorithm directly affects the chip size, wiring length, and becomes the significant direction of the optimization design in three dimensional network-on-chip. This paper proposes a floorplanning optimization algorithm based on discrete quantum-behaved particle swarm algorithm. Compared with the simulated annealing algorithm commonly used in the previous research, this algorithm initializes the random population and uses the evolutionary way, instead of using a local single solution perturbation method to get solution space, so it has better search ability and faster convergence speed. Simulation results show that using this algorithm can select floorplanning scheme which can reduce flit latency and save CPU computing time. It has significant effect especially in test cases which has more IP cores and high injection rate. In ami49 experiment with 100% of the injection rate, compared with the simulated annealing algorithm, the average flit latency of this algorithm reduces 20.63%; while the average CPU time of this algorithm reduces 69.40%.

Key words: three dimensional network-on-chip, floorplanning algorithm, B*-tree, discrete quantum-behaved particle swarm optimization, simulated annealing, particle swarm optimization