Research map / Reinforcement learning / Visual and representation RL
MaskPlace: Fast Chip Placement via Reinforced Visual Representation Learning
Yao Lai, Yao Mu, Ping Luo
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 |
|---|---|---|---|
| Use pixel-level visual masks and MaskPlace's policy instead of DREAMPlace | Macro placement · adaptec1 | HPWL: 15.81±1.64 -> 6.38±0.35 | Improved |
| Use pixel-level visual masks and MaskPlace's policy instead of DREAMPlace | Macro placement · adaptec4 | HPWL: 37.41±0.87 -> 79.21±0.65 | Hurt |
| Use pixel-level visual masks and MaskPlace's policy instead of DREAMPlace | Macro placement · bigblue1 | HPWL: 2.44±0.06 -> 2.39±0.05 | Improved |
| Relax the occupied-grid calculation using rounding instead of ceiling (ablation) | MaskPlace · ariane | HPWL: 0.1025±0.0015 -> 0.0879±0.0012 | Improved |
| Relax the occupied-grid calculation using rounding instead of ceiling (ablation) | MaskPlace · ariane | overlap area ratio: 0.00±0.00 -> 3.33±0.79 | Hurt |
| Add curriculum learning pretraining on 1/3 of macros for 30 epochs (ablation) | MaskPlace · adaptec1 | HPWL | Improved |
| Remove the one-step lookahead input M^(t+1) (ablation) | MaskPlace · adaptec1 | HPWL | Hurt |
| Remove the number-of-nets feature from placement ordering (ablation) | MaskPlace · adaptec1 | HPWL | Hurt |
| Replace 1x1 convolutions with 7x7 kernels in local feature fusion (ablation) | MaskPlace · adaptec1 | HPWL | Hurt |
| Use sparse reward only after all macros are placed (ablation) | MaskPlace · adaptec1 | HPWL | Hurt |
| Remove the view mask input (ablation) | MaskPlace · adaptec1 | HPWL | Hurt |
| Remove the wire mask input (ablation) | MaskPlace · adaptec1 | HPWL | Hurt |
Later papers that compared with this method
4 later accepted papers report MaskPlace (3k) as a baseline in their results tables; 4 of them report a better number on most of the results they share with it.
- RSPlace: Rotation Sensing Macro Placement via Bidirectional Tree Expansion (AAAI 2026 · better on 8 of 8 shared results)
- RulePlanner: All-in-One Reinforcement Learner for Unifying Design Rules in 3D Floorplanning (ICML 2026 · better on 2 of 2 shared results)
- Expertise Can Be Helpful for Reinforcement Learning-based Macro Placement (ICLR 2026 · better on 1 of 1 shared result)
- ChiPFormer: Transferable Chip Placement via Offline Decision Transformer (ICML 2023 · better on 3 of 3 shared results)
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
- Chip Placement with Diffusion Models (ICML 2025 · similarity 0.90)
- See it to Place it: Evolving Macro Placements with Vision Language Models (NeurIPS 2026 · similarity 0.89)
- DeepLayout: Learning Neural Representations of Circuit Placement Layout (ICML 2025 · similarity 0.89)
- Reinforcement Learning Policy as Macro Regulator Rather than Macro Placer (NeurIPS 2024 · similarity 0.88)
- Reinforcement Learning within Tree Search for Fast Macro Placement (ICML 2024 · similarity 0.88)
- EGPlace: An Efficient Macro Placement Method via Evolutionary Search with Greedy Repositioning Guided Mutation (ICML 2025 · similarity 0.87)
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.
