A Machine Learning-Based Method with RA-Net for Static IR Drop Prediction

In this paper, a novel machine learning-based method is proposed for fast and accurate prediction of the static IR drop of integrated circuits. A neural network called RA-Net, deriving from U-Net model, is applied for instance-level IR drop analysis. The ResBlock module and Atrous Spatial Pyramid Pooling (ASPP) are introduced in RA-Net, which improves the capability of feature extraction and leads to a better prediction. A new data preprocessing method is utilized to generate RA-Net inputs and enhance the predictive performance of small-sample designs. The numerical results demonstrate the validity and efficiency of the proposed method.