A Fusion Method for Error Detection and Classification in Regression Testing

In the field of integrated circuit (IC) design, regression testing is a critical step to verify the correctness of design modifications. However, the complexity of massive waveform data makes traditional manual analysis inefficient and prone to subjective bias. This paper proposed an automated framework for error detection and classification in regression testing forms. By combining multi-dimensional feature extraction with machine learning techniques, the framework achieved efficient error localization. A Support Vector Machine (SVM) was used as the core model, optimized with grid search and random search to determine the regularization parameter (C = 0.0746) and kernel function (linear kernel). Additionally, a semi-supervised self-training algorithm was introduced to leverage high-confidence unlabeled data, addressing the performance bottleneck caused by limited labeled data. Experiments conducted on the datasets from Track 9 of the 2024 China Postgraduate IC Innovation Competition•EDA Elite Challenge Contest [1] showed that the proposed fusion method achieved an F1 score of 99.62% in error classification, significantly outperforming traditional CNN (84.21%) and Random Forest (80.65%).