Machine Learning-Based Diagnosis of Defects in Chiplet Interconnects

Integrating Chiplets into advanced packaging technologies presents significant challenges in diagnosing interconnect line defects, primarily due to the immaturity of fabrication processes, reduced interconnect spacing, and increased density. In this paper, a non-destructive interconnect defect diagnosis method is proposed. Firstly, 3D TSV-RDL interconnect channels are simulated, and open and short defects are injected; utilizing S-parameters and group delays as the features, machine learning algorithms are used to realize the classification identification and localization of defects. The results show that the algorithm used can accurately identify open and short defects; in defect localization, the Mean Relative Error (MRE) of localization of the proposed method is less than 8%, and the Maximum Relative Error (MaxRE) does not exceed 13%. Compared with the related algorithms, the localization accuracy is significantly improved, providing a novel perspective for the identification and localization of defects within package interconnect lines.