Machine learning optimal ordering in global routing problems in semiconductors

2024-12-28

Scientific Reports

AI Summary

Net ordering optimization in semiconductor global routing is an NP-hard problem critical to routing quality. Method: This paper proposes the first end-to-end differentiable sequence generation framework that integrates graph neural networks (GNNs) with policy gradient reinforcement learning to jointly model and optimize net ordering. Innovatively, it combines routing graph-structure encoding, a differentiable ranking loss function, and adaptive policy learning—overcoming the representational limitations of traditional heuristic rules. Results: Evaluated on ISCAS-85 and ITC-99 benchmark circuits, the method achieves average reductions of 23.6% in total wirelength and 17.4% in via count, while accelerating runtime by 5.8×. This work establishes the first deep reinforcement learning–based differentiable net ordering paradigm for EDA, significantly advancing automated routing algorithms toward data-driven, learnable methodologies.

Technology Category

• Search and Optimization: Learning to Search
• Planning, Routing, and Scheduling: Planning/Scheduling and Learning
• Natural Language Processing: Learning & Optimization for NLP

Application Category

• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Economics, Online Markets and Human Computation: Incentives in network design for Web infrastructures and ecosystems

Problem

• Semiconductor Industry
• Optimization
• Global Path Selection

Innovation

• Machine Learning
• Deep Learning
• Semiconductor Packaging Optimization