Machine Learning based Routing Guide Generation and its Application to Design Rule Violation Reduction

In the physical design flow, routing and design rule checks are usually performed iteratively to generate a high-quality solution. However, running the whole routing stage to get the design rule violations (DRVs) is time-consuming. Thus, a fast and accurate DRV predictor is urgently needed. In this paper, we apply the machine learning technique to predict the DRVs from the placement and lightweight global routing results. Then, we take the predicted DRV information to generate the routing guides that enable the router to enhance the routing quality. The experimental results are shown to support our contributions.