How circuitnet/circuitnet works
Main Components (Top Level)
a) Data Processing Layer
• Feature Extraction Module
• Graph Building Module
• Data Preprocessing Module
b) Machine Learning Tasks Layer
• Congestion Prediction
• DRC (Design Rule Check) Prediction
• IR Drop Prediction
• Net Delay Prediction
c) Dataset Management Layer
• CircuitNet-N28 Dataset
• CircuitNet-N14 Dataset
• Graph Features Storage
Component Relationships and Data Flow
[Raw Circuit Data (LEF/DEF)]
• ↓
[Feature Extraction Layer]
• ↓ ↓ ↓
[Graph [Circuit [Design
Features] Features] Metrics]
• ↓ ↓ ↓
[ML Models Layer]
• ↓ ↓ ↓
[Prediction Outputs]
Specific Implementation Details
a) Feature Extraction Pipeline:
• LEF/DEF processing
• Netlist analysis
• Graph feature generation
b) Model Architecture Components:
• GPDL (Congestion)
• RouteNet (DRC)
• MAVI (IR Drop)
• GNN (Net Delay)
c) Dataset Organization:
• Training/Testing splits
• N14/N28 technology nodes
• Feature types (congestion, timing, IR drop)
Key Technical Elements
• DGL integration for graph neural networks
• PyTorch-based model implementations
• Data augmentation pipelines
• Evaluation metrics and loss functions
Implementation Notes
• Use hexagons for ML models
• Version information (CircuitNet 2.0)