Research map / Graph learning / GNN architectures
SyncTREE: Fast Timing Analysis for Integrated Circuit Design through a Physics-informed Tree-based Graph Neural Network
Yuting Hu, Jiajie Li, Florian Klemme, Gi-Joon Nam, Tengfei Ma, Hussam Amrouch, Jinjun Xiong
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 |
|---|---|---|---|
| train with TContrast loss instead of default L1 loss (ablation) | SyncTREE (32 layers) · Synthetic dataset | delay MAE (ps): 3.424 -> 3.354 | Improved |
| train with TContrast loss instead of default L1 loss (ablation) | SyncTREE (32 layers) · RISC-V dataset | delay MAE (ps): 0.0106 -> 0.0098 | Improved |
| train with TContrast loss instead of default L1 loss (ablation) | SyncTREE (32 layers) · Synthetic dataset | slew MAE (ps): 12.864 -> 12.608 | Improved |
| train with TContrast loss instead of default L1 loss (ablation) | SyncTREE (4 layers) · RISC-V dataset | slew MAE (ps): 0.0410 -> 0.0406 | Improved |
| replace L1 loss with Circle loss (ablation) | SyncTREE (16 layers) · Synthetic dataset | delay MAE (ps): 3.775 -> 3.912 | Hurt |
| replace L1 loss with Circle loss (ablation) | SyncTREE (8 layers) · RISC-V dataset | delay MAE (ps): 0.0195 -> 0.0223 | Hurt |
| replace L1 loss with Circle loss (ablation) | SyncTREE (4 layers) · Synthetic dataset | slew MAE (ps): 25.672 -> 25.601 | No clear effect |
| replace GAT-based baselines with SyncTREE two-pass tree message passing | SyncTREE (4 layers) · Synthetic dataset | delay MAE (ps): 9.457 -> 8.745 | Improved |
| increase SyncTREE depth from 4 to 32 layers | SyncTREE · Synthetic dataset | delay MAE (ps): 8.745 -> 3.424 | Improved |
| increase SyncTREE depth from 32 to 64 layers | SyncTREE · RISC-V dataset | delay MAE (ps): 0.0106 -> 0.0176 | Hurt |
| increase GCN depth from 4 to 64 layers | GCN · Synthetic dataset | delay MAE (ps): 12.675 -> 13.782 | Hurt |
| transfer pretrained delay representation to slew task with frozen weights and fine-tuned readout (ablation) | SyncTREE (16 layers) · Synthetic dataset | slew MAE change: 2.304 | Improved |
Closest papers by meaning
- Bridging Layout and RTL: Knowledge Distillation based Timing Prediction (ICML 2025 · similarity 0.87)
- AnalogGenie-Lite: Enhancing Scalability and Precision in Circuit Topology Discovery through Lightweight Graph Modeling (ICML 2025 · similarity 0.87)
- PreRoutGNN for Timing Prediction with Order Preserving Partition: Global Circuit Pre-training, Local Delay Learning and Attentional Cell Modeling (AAAI 2024 · similarity 0.86)
- CircuitFusion: Multimodal Circuit Representation Learning for Agile Chip Design (ICLR 2025 · similarity 0.86)
- AnalogGenie: A Generative Engine for Automatic Discovery of Analog Circuit Topologies (ICLR 2025 · similarity 0.85)
- DeepGate4: Efficient and Effective Representation Learning for Circuit Design at Scale (ICLR 2025 · similarity 0.85)
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.
