The study of flip-chip Cu stud bump's reliability based on PCA-BP neural networks

Abstract

In this paper, the model of Cu stud bump is established by ANSYS/LS-DYNA, and under different parameters such as: the chopper, copper wire diameter and pad size and so on, simulate the process of forming dynamically. Collect the relevant data as the BP neural network training and test samples. In accordance with the problems that BP neural network which has overabundance input factors exists the slow convergence and the low forecasting precision, the method of PCA is established to decompose the input factors, reduce the input factor dimension and eliminate the input factor linear correlation, and then based on the implementation of the improved BP algorithm by adding the item of the momentum, the model is established. The simulation result shows that: this model has a faster training speed and higher rate of accuracy and can better predict the post-welding Cu stud bump's quality, so as to optimize the process parameters and provide an effective way to improve the Cu stud bump's reliability.