Delving into Macro Placement with Reinforcement Learning
This is an incremental improvement for chip design automation, potentially enhancing efficiency for engineers.
The paper tackles macro placement in physical design by extending prior reinforcement learning work, replacing a force-directed method with DREAMPlace for standard cells and comparing results on public benchmarks.
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.