Interface Reliability Modeling of Coaxial Through Silicon Via Based on WOA-BP Neural Network

Abstract

Abstract Due to the complex structure and thermal mismatch of coaxial through silicon via (TSV), cracks easily occur under thermal load, leading to interface delamination or spalling failure. The reliability issue of coaxial TSV is important for its application in three-dimensional packaging, so it is of great significance to predict the crack trend and evaluate the reliability of coaxial TSV. In this paper, an algorithm model with the combination of whale optimization algorithm (WOA) and back propagation (BP) neural network for the reliability prediction of coaxial TSV is proposed. Based on finite element method (FEM), the training and validation datasets of the energy release rates (ERR) of the crack at the critical interface are calculated to construct the deep learning neural network. Six key structure parameters affecting the reliability of coaxial TSV are selected as the input values of the BP neural network. The maximum relative error of whale optimization algorithm optimized back propagation (WOA-BP) neural network model is 0.88%, which is better than the prediction results of the traditional BP and genetic algorithm (GA) optimized BP models. The WOA-BP neural network model was also compared with BP and GA-BP neural network models with four error metric models. It is verified that WOA-BP neural network model has the best prediction performance. The proposed model can be used to achieve improved prediction accuracy for the interface reliability of coaxial TSV under complex structural conditions since it has higher accuracy and stronger robustness.