Modelling wetting angle of solder on substrate using machine learning approach

The solder wettability is a key factor for assessing joint reliability. However, the extensive range of solder composition, substrate material and reflow temperature make it challenging to modulate wettability through experimental methods. This study utilizes the machine learning method to develop a surrogate model fitted to experimental wetting angle data, aiming to predict the solder wettability as a function of composition, substrate material, and reflow temperature. The model full-fit (five-fold cross validation) root-mean-square error was 4.27 (6.73 ± 1.42). The model performance was characterized by using the cross-plot method, and the input feature contribution was determined through SHapley Additive exPlanations analysis. The model extrapolation to unknown alloys, which were not shown in the training data set showed a good agreement with experimental data.