Artificial Neural Networks and Bayesian Techniques for Flip-Chip Package Thermo-Mechanical Analysis

In this paper, we present a predictive model created using Artificial Neural Networks (ANN). ANN is used to develop approximate models that mimic the overall package mechanical behavior. Furthermore, the approximate ANN models can be efficiently coupled with probabilistic Bayesian Approaches which would normally take several hours to implement using the traditional FEA modeling methodology. Detailed implementation of a robust ANN model on flip-chip electronic packages is presented. The forward and backward propagation steps of the neural network model are programmed using MATLAB, with a cost function that minimizes the error between the predicted and input training data. The training set for ANN is generated using an FEA thermo-mechanical model. The FEA input data set encompasses effects of package geometry and material parameters on various package mechanical risk factors such as package warpage, delamination, and thermal degradation. We present results that show the methodology to co-optimize ANN hyperparameters and minimize the ANN inference error. The ANN surrogate models are then coupled with Bayesian inferencing techniques and applied for performing package mechanical integrity analysis.