Bonding strength of sintered nano silver (Ag) joints has been an important index for evaluating the reliability of power module packages, which has been reported to be influenced by many factors. However, it is hard to evaluate and predict those factors affecting bonding strength. With the help of artificial intelligence (AI), the utilization of AI tools to assist in finding science solutions has become a mainstream consensus. In this paper, we will show how to evaluate and predict those sintered nano Ag-Al bonded joints bonding strength with deep learning (DL) methods. Firstly, a reliable extended dataset was obtained using the conditional formal condition table generation Induction Network (CTGAN) based on the die shear test dataset of sintered Ag-Al interface shear strength with four different metallization layers under different high-temperature aging time. Subsequently, four DL models were adopted to predict the shear strength of Ag-Al interface under different metallization layers to evaluate the interface bonding strength, all with high level of determination coefficient (R2, above 0.99) and classification accuracy (above 85%). Last but not least, factors influencing the shear strength of Ag-Al interface were analyzed and ranked by weight analysis and SHapley Additive exPlanations (SHAP) method. This research could provide a novel perspective on understanding those factors affecting the shear strength of sintered nano Ag interconnect layer in power devices.
Read full abstract