Integration of Artificial Neural Network and Finite Element Simulation for Package Warpage Prediction

Many design parameters affect the warpage behavior of the advanced micro electronics package, such as different type of die stack up, material selection, substrate thickness and mold compound height etc. Thermal Shadow Moire (TSM) test is used to estimate warpage behavior of the advanced package warpage. However, optimizing, the design parameters through TSM method is time consuming and needs more resources, reducing the number of experiments becomes a critical issue. Currently, researchers and engineers are adopted finite element design-on-simulation (DOS) technology for the risk assessment of package warpage. DOS technology can effectively shorten the design cycle, reduces resource, and effectively optimize the package structure. However, TSM simulation analysis requires more number of iterations to do the risk assessment of package warpage which leads to more consumption of time. Artificial Intelligence (AI) can help researchers avoid the above shorting and it can give the quick risk assessment of package warpage. Therefore, this study is mainly focused on AI-assisted DOS technology by combining artificial intelligence and simulation technology to predict the package warpage. In order to ensure warpage prediction accuracy, the simulation model was validated with the TSM experiment prior to establishment of design on experiment (DOE) matrix and generate the AI training database. An Artificial Neural Network (ANN) machine learning algorithm combined with AI training database to build the AI training model. Once, the AI model built with proper validation, we can immediately predict the package warpage by using the critical design parameters. The results of this investigation demonstrate good accuracy, with a difference between AI prediction warpage and finite element simulation of less than 8%, and a reduction in solving time from a few days to a few minutes.