Design Space Exploration of Reinforcement Learning Configurations for Post-Legalization Placement

Foteini Oikonomou; Maria Pantazi-Kypraiou; George Goudroumanis; Pavlos Stoikos; Athanasios Tziouvaras; Christos Anagnostopoulos; Georgios Karakonstantis; George Floros

cs.AIOct 14, 2026

Design Space Exploration of Reinforcement Learning Configurations for Post-Legalization Placement

Authors: Foteini Oikonomou, Maria Pantazi-Kypraiou, George Goudroumanis, Pavlos Stoikos, Athanasios Tziouvaras, Christos Anagnostopoulos, Georgios Karakonstantis, George Floros

Organizations: Department of Electrical and Computer Engineering, University of Thessaly, Volos, Greece · School of Computing Science, University of Glasgow, UK · Department of Electronic and Electrical Engineering, Trinity College Dublin, Ireland

Abstract

Graph Neural Networks (GNNs) and Reinforcement Learning (RL) have recently been explored for VLSI placement optimization, but their effectiveness for post-legalization placement refinement remains unclear. This paper presents a systematic design space exploration of a GNN-Proximal Policy Optimization (GNN-PPO) framework across three ISPD2015 benchmarks. We evaluate 18 configurations covering reward functions, GNN architectures, action spaces, candidate-set sizes, and training strategies. We analyse the results we obtain to identify pitfalls and to assess the prospects and opportunities of the RL methods for post-legalization placement. On top of that we propose a design classification approach to guide the design decisions of RL models.

Figures & tables

Design	$N$ (cells)	Baseline HPWL ( $\mu$ m)
mgc_fft_1	32K	260,172
mgc_pci_bridge32_a	${\sim}$ 30K	548,355 B
mgc_matrix_mult_1	155K	1,518,463

TABLE I: The ISPD2015 benchmarks used to perform the design space exploration.

Config.	Arch.	Action	Reward	$K$	Partit.	Train $\Delta$ %	Eval $\Delta$ %
mgc_fft_1
C1	GAT-2L	1-site	Local	5K	No	$0.44$	$0.20$
C2	GAT-2L	1-site	Normalized	5K	No	$0.41$	$0.20$
C3	GCN	1-site	Local	5K	No	$0.46$	$0.21$
C4	SAGE	1-site	Local	5K	No	$\mathbf{0.53}$	$\mathbf{0.20}$
C5	GAT-2L	3-site	Local	5K	No	$0.26$	$0.22$

TABLE II: Design space exploration configurations and results, grouped per benchmark.

Design	$S_{p}$	$R_{m}$	Class	Train $\Delta$ %	Eval $\Delta$ %
mgc_fft_1	$1.14{\times}10^{-3}$	9.27	A	$\mathbf{0.53}$	$0.20$
mgc_pci_bridge32_a	$1.96{\times}10^{-3}$	0.97	B	$\mathbf{5.85}$	$0.43$
mgc_matrix_mult_1	$1.41{\times}10^{-4}$	4.77	C	$\mathbf{0.04}$	$0.03$

TABLE III: Evaluation of the proposed design classification approach over the ISPD2015 benchmarks.

RL for Combinatorial Optimization

Reinforcement learning for combinatorial optimization learns sequential decisions and search strategies for difficult discrete problems.

See all 106 papers

GPlaceRL: An Open-Source Graph Reinforcement Learning Framework for Detailed Placement
Oct 5, 2026Pavlos Stoikos, Foteini Oikonomou, Christos Poulos +5Reinforcement Learning Electronic Design Automation
Department of Electrical and Computer Engineering, University of Thessaly, Volos, Greece · School of Computing Science, University of Glasgow, UK · Department of Electronic and Electrical Engineering, Trinity College Dublin, Ireland
How Can Reinforcement Learning Achieve Expert-level Placement?
Apr 28, 2026Ruo-Tong Chen, Ke Xue, Chengrui Gao +7Reinforcement Learning Electronic Design Automation
State Key Laboratory of Novel Software Technology, Nanjing University, China · School of Artificial Intelligence, Nanjing University, China · Huawei Noah’s Ark Lab, China
Component Centric Placement Using Deep Reinforcement Learning
Feb 26, 2026Kart Leong LimElectronic Design Automation RL for Combinatorial Optimization
Institute of Microelectronics (IME), Agency for Science, Technology and Research (A*STAR), 2 Fusionopolis Way, Innovis #08-02, Singapore 138634, Republic of Singapore

Oct 5, 2026J/K move · Enter open · S save

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

Pavlos Stoikos, Foteini Oikonomou, Christos Poulos, Maria Pantazi-Kypriou+4

Department of Electrical and Computer Engineering, University of Thessaly, Volos, Greece · School of Computing Science, University of Glasgow, UK · Department of Electronic and Electrical Engineering, Trinity College Dublin, Ireland

Reinforcement Learning Electronic Design Automation RL for Combinatorial Optimization

Open paper AskPDF arXiv

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\%

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.

Abstract

RL for Combinatorial Optimization

Explore similar work

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

How Can Reinforcement Learning Achieve Expert-level Placement?

Component Centric Placement Using Deep Reinforcement Learning