Delving into Macro Placement with Reinforcement Learning
Machine Learning
2021-09-07 v1 Artificial Intelligence
Abstract
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 (Mirhoseini et al., 2020). 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.
Cite
@article{arxiv.2109.02587,
title = {Delving into Macro Placement with Reinforcement Learning},
author = {Zixuan Jiang and Ebrahim Songhori and Shen Wang and Anna Goldie and Azalia Mirhoseini and Joe Jiang and Young-Joon Lee and David Z. Pan},
journal= {arXiv preprint arXiv:2109.02587},
year = {2021}
}
Comments
Accepted at 3rd ACM/IEEE Workshop on Machine Learning for CAD (MLCAD)