GPlaceRL: An Open-Source Graph Reinforcement Learning Framework for Detailed Placement

2026-10-05

AI Summary

This study addresses the scarcity of learning-based methods and the absence of a unified open-source framework for detailed placement optimization by proposing the first open-source graph reinforcement learning framework tailored to this task. The method models legalized placements as graph structures, employing the Proximal Policy Optimization (PPO) algorithm with a compact Graph Attention Network (GAT) encoder. Furthermore, it provides a modular environment to systematically investigate the design of encoders, policies, and reward mechanisms. Our analysis reveals the critical roles of the compact GAT architecture and a flexible local action space. Evaluated across five benchmarks, the proposed approach achieves substantial improvements in Half-Perimeter Wire Length (HPWL), ranging from 3.27% to 32.87%.

Abstract

Reinforcement learning (RL) has emerged as a promising approach for placement optimization, particularly when combined with graph neural networks (GNNs) that capture circuit connectivity. However, most learning-based placement approaches focus on floorplanning, macro placement, or global placement, while detailed placement refinement remains relatively unexplored. In this paper, we present GPlaceRL, an open-source graph reinforcement learning framework for detailed placement refinement. GPlaceRL represents legalized placements as graphs and provides a modular environment for studying graph encoders, policy architectures, reward formulations, and local placement actions. To demonstrate the capabilities of GPlaceRL, we conduct a systematic evaluation of proximal policy optimization (PPO) policies with graph attention network (GAT) encoders in a per-design optimization setting. Across five placement benchmarks, the best greedy evaluation results achieve HPWL improvements ranging from $3.27\%$ to $32.87\%$. The results highlight the importance of compact GAT architectures and flexible local action spaces for placement optimization. Overall, GPlaceRL provides a reproducible and extensible framework for systematic research on RL-based detailed placement refinement.

Problem

• Detailed Placement
• Reinforcement Learning
• Placement Optimization
• Graph Neural Networks

Innovation

• Graph Reinforcement Learning
• Detailed Placement
• Graph Attention Network
• Proximal Policy Optimization
• Open-Source Framework