Investigation of Congestion Predictor with Supervised Learning for Routability-Driven Global Routing

Routability in physical design has reached a snag since traditional routers' congestion estimates aren't up to pace with today's advanced routing characteristics. Several approaches based on supervised learning have recently been developed to anticipate rout ability information. However, the characteristics retrieved using these approaches are relatively crude in terms of reflecting actual physical attributes. Furthermore, a lack of global knowledge causes global routing performance to deteriorate. In this work, we present a supervised-learning regression model capable of capturing correct global routing patterns in this article, which is used to train a congestion prediction model to enhance global routing. In comparison to traditional global-routing-based congestion estimates, experimental findings demonstrate that our predictor is at least 10.33 times quicker in execution while retaining an accurate forecast. Furthermore, by including our model into the router, we can obtain not just a better routing topology, but also a higher quality of performance