Research map / Reinforcement learning / Visual and representation RL
Expertise Can Be Helpful for Reinforcement Learning-based Macro Placement
Chengrui Gao, Yunqi Shi, Ke Xue, Ruo-Tong Chen, Siyuan Xu, Mingxuan Yuan, Chao Qian, Zhi-Hua Zhou
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add dataflow guidance (ablation) | EXPlace · ICCAD 2015 ablation (3 cases) | TNS / WNS average rank: TNS rank 4.00 -> 1.33; WNS rank 4.00 -> 2.67 | Improved |
| add periphery biasing (ablation) | EXPlace · ICCAD 2015 ablation (3 cases) | rWL / rOverflowH average rank: rWL rank 4.33 -> 1.00; rOverflowH rank 5.00 -> 3.00 | Improved |
| add macro grouping (ablation) | EXPlace · ICCAD 2015 ablation (3 cases) | rWL / TNS average rank: rWL rank 3.67 -> 1.00; TNS rank 3.67 -> 1.33 | Improved |
| add I/O port keepout (ablation) | EXPlace · ICCAD 2015 ablation (3 cases) | rWL / WNS average rank: rWL rank 2.67 -> 1.00; WNS rank 4.00 -> 2.67 | Improved |
| add timing-driven DPO fine-tuning | EXPlace · ICCAD 2015 (8 cases) | TNS and WNS: TNS 12.29% and WNS 22.68% average improvement | Improved |
| add timing-driven DPO fine-tuning | EXPlace · ICCAD 2015 (8 cases) | global HPWL: 1.77% reduction | No clear effect |
| add timing-driven DPO fine-tuning | EXPlace · ICCAD 2015 superblue1 | preference consistency: 16% -> 23% | Improved |
| EXPlace vs MaskRegulate (strongest RL baseline) | EXPlace · OpenROAD (6 cases) | TNS: 5.97% TNS magnitude reduction | Improved |
| EXPlace vs DREAMPlace | EXPlace · OpenROAD (6 cases) | TNS: 23.06% TNS magnitude reduction | Improved |
| EXPlace vs Hier-RTLMP | EXPlace · OpenROAD (6 cases) | rWL, WNS, TNS, DRC: rWL 5.77%, WNS 4.77%, TNS 20.54%, DRC 84.48% | Improved |
| transfer policy trained on superblue1 zero-shot to other cases | EXPlace · ICCAD 2015 generalization | TNS average rank: TNS rank 1.62 vs 2.12 for LaMPlace | Improved |
| EXPlace vs MaskRegulate training time with parallel environments | EXPlace · ICCAD 2015 | training time (s): 8624.57 vs 12633.31 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- MaskPlace — MaskPlace: Fast Chip Placement via Reinforced Visual Representation Learning (NeurIPS 2022) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Reinforcement Learning Policy as Macro Regulator Rather than Macro Placer (NeurIPS 2024 · similarity 0.93)
- Reinforcement Learning within Tree Search for Fast Macro Placement (ICML 2024 · similarity 0.89)
- Benchmarking End-To-End Performance of AI-Based Chip Placement Algorithms (NeurIPS 2025 · similarity 0.89)
- Chip Placement with Diffusion Models (ICML 2025 · similarity 0.89)
- Towards Automated RISC-V Microarchitecture Design with Reinforcement Learning (AAAI 2024 · similarity 0.88)
- See it to Place it: Evolving Macro Placements with Vision Language Models (NeurIPS 2026 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
