Delving into Macro Placement with Reinforcement Learning

Delving into Macro Placement with Reinforcement Learning
复制标题

通过强化学习深入研究宏观布局

DOI:
--
复制
发表时间:
2021
期刊:
Workshop on Machine Learning for CAD
影响因子:
--
通讯作者:
David Z. Pan
David Z. Pan
中科院分区:
--
文献类型:
--
作者:
Zixuan Jiang;Ebrahim M. Songhori;Shen Wang;Anna Goldie;Azalia Mirhoseini;J. Jiang;Young;David Z. Pan

文献摘要

被引文献

相似文献

在物理设计中,人类设计师通常通过试错来放置宏,这是一个马尔可夫决策过程。强化学习(RL)方法在宏布局上表现出了超人的性能。在本文中,我们提出了一个扩展到以前的工作[1]。我们首先描述策略和价值网络架构的详细信息。我们用DREAMPlace代替力导向方法,将标准单元放置在RL环境中。我们还比较了我们的改进方法与其他学术配售公共基准。
In physical design, human designers typically place macros via trial and error, which is a Markov decision process. Reinforcement learning (RL) methods have demonstrated superhuman performance on the macro placement. In this paper, we propose an extension to this prior work [1]. We first describe the details of the policy and value network architecture. We replace the force-directed method with DREAMPlace for placing standard cells in the RL environment. We also compare our improved method with other academic placers on public benchmarks.