Circuit Routing Using Monte Carlo Tree Search and Deep Reinforcement Learning
Circuit Routing Using Monte Carlo Tree Search and Deep Reinforcement Learning
复制标题
DOI：
10.1109/vlsi-dat54769.2022.9768074
复制
发表时间：
2022-04
期刊：
影响因子：
--
通讯作者：
Youbiao He;Hebi Li;Jin Tian;F. Bao
中科院分区：
文献类型：
--
作者：
Youbiao He;Hebi Li;Jin Tian;F. Bao
我们提出了一种新的方法，电路布线建模为一个顺序决策问题，并解决它在MCTS与DRL引导推出。与传统的路由算法，无论是手动设计的领域知识或定制特定的设计规则相比，我们的方法可以重新配置几乎任何路由约束和目标，而不改变算法本身，因为AI代理探索解决方案在一个通用的搜索策略。随机生成电路和流行的开源硬件项目上的实验结果表明，我们的方法比传统的基于A * 的方法提高了33.3%的成功率。
We propose a new approach to circuit routing by modeling it as a sequential decision problem and solving it in MCTS with DRL-guided rollout. Compared with conventional routing algorithms that are either manually designed with domain knowledge or tailored to specific design rules, our approach can be reconfigured for nearly any routing constraints and goals without changing the algorithm itself because the AI agent explores solutions in a general search strategy. Experimental results on both randomly generated circuits and popular open-source hardware projects show that our method achieves 33.3% higher success rate than traditional A *-based approach.