Machine learning methods for fast evaluation of static IR drop effect
Abstract
This paper describes the approach of applying a machine learning model to analyze the static IR drop effect. The problem of static IR drop analysis was solved within the framework of ICCAD Contest 2023(ProblemC). The paper discusses the methodology of ML model training for predicting static IR drop. Various methods for generating a dataset from the SPICE description of circuits for ML model training are proposed. The provided solution to the static IR drop effect problem using machine learning techniques was ranked among the top 3 solutions in the ICCADContest 2023. Two scores, MAE and F1 score, were used to evaluate the obtained results. Compared to the results of static IR drop analysis performed by SPICE modeling, the main score that was used to evaluate the prediction results does not exceed V(MAE). The F1 score has high values up to 0.87.