How Can Reinforcement Learning Achieve Expert-level Placement?
For chip placement practitioners, this work bridges the gap between RL-based methods and expert human performance by addressing reward misspecification.
The paper identifies reward design as the key bottleneck preventing RL-based chip placement from matching expert quality, and proposes learning a reward model directly from expert layouts. The method achieves expert-level placement quality, generalizing from a single design to unseen cases.
Chip placement is a critical step in physical design. While reinforcement learning (RL)-based methods have recently emerged, their training primarily focuses on wirelength optimization, and therefore often fail to achieve expert-quality layouts. We identify the reward design as the primary cause for the performance gap with experts, and instead of formalizing intricate processes, we circumvent this by directly learning from expert layouts to derive a reward model. Our approach starts from the final expert layouts to infer step-by-step expert trajectories. Using these trajectories as demonstrations or preferences, we train a model that captures the latent implicit rewards in expert results. Experiments show that our framework can efficiently learn from even a single design and generalize well to unseen cases.