A Machine Learning Framework for Rapidly Identifying and Predicting Mechanical Properties of Interconnected Layer in SiC Module Considering Surface Metallization Effect Under High-Temperature Aging

Sintered silver (Ag), as the most promising chip interconnect material for high-temperature silicon carbide (SiC) power module packaging, is often subjected to thermal stress during the packaging process and high temperature service conditions, which greatly reduce the thermal mechanical reliability of SiC power module. In this article, the prediction parameters of the interface strength of sintered Ag interconnection layer considering metallized layer and high temperature aging effects were obtained and evaluated by the method of machine learning-assisted experiments and simulations. A multilayered MLP (multilayer artificial neural network)-LSTM (long short-term memory) framework-based model was proposed, which shows higher classification and prediction ability. In this method, MLP was used to extract classification features under different metallized layers effect, and LSTM was used to extract time features for long and short-term prediction of mechanical properties degradation caused by high temperature aging. The presented multilayered MLP-LSTM method has successfully identified the relevant mechanical parameters of the cohesive zone model (CZM) directly from the load-displacement responses of die shear test. The CZM parameters related to interface bonding strength of sintered Ag interconnection layer are predicted and the mechanism of shear strength is also determined rapidly with high accuracy.