Research map / Graph learning / Knowledge and temporal graphs
Versatile Multi-stage Graph Neural Network for Circuit Representation
Shuwen Yang, Zhihao Yang, Dong Li, Yingxue Zhang, Zhanguang Zhang, Guojie Song, Jianye HAO
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 |
|---|---|---|---|
| remove topo-edges from Circuit Graph (ablation) | Circuit GNN · ISPD2011 congestion, placement | grid pearson: 0.697 -> 0.684 | Hurt |
| remove topo-edges from Circuit Graph (ablation) | Circuit GNN · ISPD2011 congestion, placement | cell pearson: 0.887 -> 0.883 | Hurt |
| remove topo-edges from Circuit Graph (ablation) | Circuit GNN · ISPD2011 congestion, placement | grid spearman: 0.770 -> 0.730 | Hurt |
| remove topo-edges from Circuit Graph (ablation) | Circuit GNN · ISPD2011 congestion, placement | grid kendall: 0.577 -> 0.536 | Hurt |
| add geometric edges and cell positions in placement stage vs topology only | Circuit GNN · ISPD2011 congestion, logic synthesis vs placement | cell pearson: 0.779 -> 0.887 | Improved |
| use Circuit GNN instead of CongestionNet | Circuit GNN · ISPD2011 congestion, logic synthesis | grid pearson: 0.277 -> 0.315 | Improved |
| use Circuit GNN instead of MPNN (5x faster) | Circuit GNN · ISPD2011 congestion, logic synthesis | grid pearson: 0.292 -> 0.315 | Improved |
| use Circuit GNN instead of LHNN | Circuit GNN · ISPD2011 congestion, placement | grid pearson: 0.703 -> 0.697 | Mixed |
| use Circuit GNN instead of LHNN | Circuit GNN · DAC2012 net wirelength, placement | MAE: 0.581 -> 0.483 | Improved |
| use Circuit GNN instead of LHNN | Circuit GNN · DAC2012 net wirelength, placement | pearson: 0.801 -> 0.848 | Improved |
| use Circuit GNN features frozen (evaluate) instead of LHNN | Circuit GNN · Transfer congestion to wirelength | pearson: 0.689 -> 0.799 | Improved |
| fine-tune GNN instead of freezing | Circuit GNN · Transfer congestion to wirelength | pearson: 0.799 -> 0.842 | Improved |
Closest papers by meaning
- CircuitFusion: Multimodal Circuit Representation Learning for Agile Chip Design (ICLR 2025 · similarity 0.86)
- Circuit Representation Learning with Masked Gate Modeling and Verilog-AIG Alignment (ICLR 2025 · similarity 0.86)
- DeepGate4: Efficient and Effective Representation Learning for Circuit Design at Scale (ICLR 2025 · similarity 0.86)
- DeepLayout: Learning Neural Representations of Circuit Placement Layout (ICML 2025 · similarity 0.86)
- PreRoutGNN for Timing Prediction with Order Preserving Partition: Global Circuit Pre-training, Local Delay Learning and Attentional Cell Modeling (AAAI 2024 · similarity 0.85)
- Graph Neural Networks and Arithmetic Circuits (NeurIPS 2024 · similarity 0.84)
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.
