The Fine-Grained Chip Placement with Hybrid Action Spaces and Feature Fusion

Mingyue Wang

Primary Area: reinforcement learning

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Keywords: Deep Reinforcement Learning, Chip Placement, hybrid action space, feature fusion

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.

Abstract: Chip placement is an essential and time-consuming step in the physical design process. Deep reinforcement learning, as an emerging field, has gained significant attention due to its ability to replace weeks of expert model design. We devise a fusion-based reinforcement learning framework to address the limited representation problem of both graph networks and CNN networks. Furthermore,the structure of PDQN in the hybrid action space allows for precise coordinate placement, compared to other RL-based structures in placement. The experimental results can demonstrate the effectiveness of our model.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors' identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 8602