Implementation of Machine Learning In VLSI Global Routing

Machine learning (ML) plays a transformative role in VLSI design by automating complex decision-making processes, which traditionally rely on heuristic or rule-based methods. In particular, ML helps address challenges in optimizing parameters like wire length, signal delay, and congestion, which are critical for enhancing design efficiency. Without ML, global routing in VLSI becomes time-consuming and less adaptive to modern design complexity, leading to suboptimal results. In this paper, we have discussed the implementation of ML for congestion prediction using congestion datasets from CircuitNet, enabling dynamic routing decisions based on real-time design conditions. Our model significantly reduces optimization time and improves routing efficiency. The success of the model is highlighted by the ROC-AUC value of 0.92, representing a 92% success rate in predicting and mitigating congestion.