DR-CircuitGNN: Training Acceleration of Heterogeneous Circuit Graph Neural Network on GPUs

Keyphrases

• Electronic Design Automation 100%
• Circuit Graph 100%
• Graph Neural Network 100%
• Training Acceleration 100%
• GPU 100%
• Heterogeneous Circuits 100%
• Message Passing 40%
• Heterogeneous Graph Neural Network 40%
• Automation Design 40%
• GPU Kernel 40%
• Mini 20%
• Parallel Design 20%
• Performance Enhancement 20%
• Error Rate 20%
• Integrated Circuit Design 20%
• Small Media 20%
• Optimization Strategy 20%
• Computational Complexity 20%
• Processing Cost 20%
• Concurrency 20%
• Complexity Costs 20%
• Full-size 20%
• Sparsity-aware 20%
• Performance Bottleneck 20%
• Relationship Characteristics 20%
• Circuit Analysis 20%
• Topological Relations 20%
• Graph Datasets 20%
• Geometric Features 20%
• CPU-GPU 20%
• Computational Processing 20%
• Serial Modules 20%
• Backward Propagation 20%
• Forward Propagation 20%
• Correlation Score 20%
• Kernel Design 20%
• SpMM 20%
• CUSPARSE 20%
• Parallel Optimization 20%
• Dynamic ReLU 20%
• Graph Neural Network Training 20%
• Correlation Error 20%

Computer Science

• Electronic Design Automation 100%
• Graph Neural Network 100%
• Graphics Processing Unit 100%
• Message Passing 40%
• Heterogeneous Graph 40%
• Sparsity 20%
• Integrated Circuit Design 20%
• Computational Complexity 20%
• Speed-up 20%
• Subgraphs 20%
• Concurrency 20%
• Performance Bottleneck 20%
• Topological Relationship 20%
• Enhance Performance 20%
• Kernel Execution 20%
• Neural Network Training 20%
• Optimization Strategy 20%
• Processing Cost 20%