Failure Life Prediction of Wafer Level Packaging using DoS with AI Technology

Abstract

In regard to the design of new electronic packaging structure, e.g., wafer level packaging (WLP), one needs to consider many design factors that could affect the reliability characteristics of electronic package. Before mass production, the electrical packaging has to pass the reliability test. Thermal cycling test (TCT) is one of the standard reliability tests and has been commonly used in electrical packaging industry. To ensure the new products pass the thermal cyclic test is the critical issue in the electronic packaging industry. To react to the rapid growth of electronic components, it is imperative to shorten the development time and cycles of electronic packaging, design-on-experiment (DoE) method is quite time consuming and very costly, finite element method (FEM) based design-on-simulation technology could be used as a feasible development methodology for the reliability assessment and reliability prediction of electronic package. After the FEM model, mechanics theory, simulation procedures and reliability behavior, etc. being verified by experiments, simulation can be treated the same as the experiment if simulation can consistently get the results that similar to experiment. Simulation can build a set of reliability results database of different geometric structures of the WLP and provided this database for machine learning. Machine learning is an automatic analysis method which could learn a regression model from the database for predicting reliability cycles of package instantly. In this paper, the thermal cycling test (TCT) reliability of different geometry structures of Wafer Level Packaging is discussed through verified simulation technology. The database generated by FEM simulation is analyzed by artificial neural network (ANN), the training result shown ANN can predict the reliability of unknown structure of WLCSP in an accurate range.