Study on an Artificial Intelligence Based Kernel Ridge Regression Algorithm for Wafer Level Package Reliability Prediction

The use of artificial intelligence (AI) to optimize product performance and increase product productivity has been adopted in many fields. Kernel Ridge Regression (KRR) is an AI algorithm that can be used to effectively predict the reliability life of wafer level packaging (WLP). Compared with the familiar artificial neural network (ANN), KRR's AI model can complete the data training of a huge data set in a short time, and has good generalization ability. In electronic packaging structure, there are many design parameters that will affects its reliability performance and using experiment way to obtain the reliability result will take lot of time and huge cost. Therefore, greatly shortening the development time and development cost has become a key issue for newly developed electronic packaging structures. The focus of this research is to combine artificial intelligence method and simulation technology to predict the reliability life of wafer-level packaging; this new concept is so called AI-assisted design-on-simulation technology. In order to overcome the time-consuming problem of experiment work, a validated finite element simulationwas used instead of experiment work to generate reliability life cycle databases of different WLP structures for AI training, this huge database will be used as a training set for AI training. After the completion of the AI training, a simple AI prediction model was obtained. Designers only need to input the geometry of each component of the newly designed wafer level package to immediately obtain the reliability life cycle of this specific WLP. Kernel Ridge Regression is a machine learning algorithm. In this research, it is used to analyze the database of the relationship between the geometric size input of each component of WLP and the reliability life output, and a regression model is developed after training. Because of its better generalization performance and faster computing time, KRR can immediately predict the reliability life of WLP. In addition, this study successfully established the accuracy and CPU time analysis of the KRR regression model for the reliability life prediction of wafer-level packaging structures. KRR preliminary results show that the average percentage difference between artificial intelligence predictions and finite element simulation results is 1.27% and maximum percentage of difference is 5.81% for 576 training dataset whereas, the percentage of average difference is 0.53% and maximum percentage of difference is 2.53% for 1,296 train dataset. Similarly, the CPU time required for 576 and 1,296 train dataset are 0.09 seconds and 0.42 seconds to predict the reliability life of WLP structure. KRR performance will also be compared with the results of other regression models, such as Artificial Neural Network (ANN) and Support Vector Regression (SVR) algorithms.