Research on TSV Void Defects Based on Machine Learning

Abstract

With the rapid development of 3D TSV (through silicon via) technology, it is particularly important to improve the yield for TSV fault detection. Aiming at TSV void defects, the paper adopts supervised machine learning method to train S parameters in TSV model with void faults, and carries out classification processing, then predicts the size of void faults through stimulus signal and S parameters. The results show that for spherical void defects detection, the classification accuracy of ELM (Extreme Learning Machine) algorithm and KNN (K-Nearest Neighbor) algorithm is above 85%, while for TSV cylindrical void defects detection, the classification accuracy of ELM algorithm is 96%.