Efficient Chip Macro Placement via Discrete Soft Actor-Critic with Enhanced Channel Attention

Macro placement is a critical step in modern IC design, where reinforcement learning (RL) has shown promise by treating it as a sequential decision process. However, existing RL-based methods often suffer from low sample efficiency and insufficient exploration. To address these issues, we propose DSPLACE (Discrete Soft Actor-Critic PLACE), a macro placement approach that integrates Discrete Soft Actor-Critic (DSAC) with Efficient Channel Attention (ECA). DSAC enhances exploration through its maximum-entropy framework, while ECA strengthens state representation by adaptively recalibrating features. Experimental results on ISPD-2005 benchmarks show that DSPLACE improves wirelength by up to 14% and converges about 30.7% faster compared to recent RL-based counterparts.