基于循环强化学习和模拟退火的布图规划算法OA

Electronic design automation(EDA)involves a series of computationally difficult optimization problems,among which floorplan optimization is a critical step in chip design.Recently,reinforcement learning(RL)-based methods have been successfully applied to deal with various combinatorial optimization problems.For the floorplan problem,a floorplan algorithm based on reinforcement learning(RL)and simulated annealing(SA)is proposed.This algorithm constructed an RL-SA loop framework,adopted sequence pair(SP)to represent floorplan structure,and utilized the ability of RL models to quickly obtain good rough solutions after training and the ability of heuristic algorithms to implement greedy improvement in solutions to obtain good floorplan.Experimental results show that RL can provide SA with a good initial floorplan,thereby generating better floorplan designs.