Prediction on Thermal Cycling Stress of PBGA Package Based on Finite Element Simulation and Machine Learning

Abstract

As the application environment of Plastic Ball Grid Array (PBGA) package becomes more diverse and severe, higher requirements are placed on the thermal cycling reliability of the PBGA package. The thermal cycling reliability of PBGA package has higher standards. The traditional experimental method has the problems of a long cycle and high costs. In this paper, the thermal cycling stress of the PBGA package is studied based on finite element simulation and machine learning methods. Take the typical PBGA package structure solder ball as the research object, and simulate the solder ball stress value under extreme thermal cycling conditions using the finite element software ABAQUS. The dataset is organized based on the finite element analysis results to establish a machine learning prediction model, which aims to predict the thermal cycling stress of PBGA package rapidly. The result shows that the prediction model established by the Random Forest Regression (RFR) algorithm has better prediction accuracy compared with Multiple Linear Regression (MLR) algorithm and Support Vector Regression (SVR) algorithm, and the Mean Absolute Percent Error (MAPE) is 3.01%, which can provide a new way to predict thermal cycling stress of PBGA packages directly.