Detailed Routing Short Violations Prediction Method Using Graph Convolutional Network

Abstract

With the continuous shrink of IC manufacturing process, how to accurately predict the routing violations before detailed routing is becoming more and more important to improve the placement quality. In this paper, we propose a detailed routing short violations prediction model based on the Graph Convolutional Network (GCN). Based on the key features extracted from the placement and detailed routing stage separately, we train a GCN model to build a map relationship between these two stages. Through this model, we can predict the detailed routing short violations at placement stage successfully. Experiments show that the average prediction accuracy of our model is 94% which is higher than existing method based on machine learning.