Linear Regression Model development for Package Warpage

Machine learning is a part of Artificial intelligence (AI) theory, uses large datasets and programming or mathematics to train and learn from the data. It creates predictive models to help engineers make decisions. Many design parameters are affected the warpage behavior of the package structure. Using experiment method to obtain the package warpage will take lot of time and huge cost. This study has introduced the linear regression machine learning algorithm combined with the finite element method (FEM) to develop a regression model for predicting the package warpage. However, linear regression model is very simplified model which is very fast and reduce the cycle time. Given that our warpage datasets exhibit non-linear characteristics, this linear regression model can handle both linear and non-linear datasets without the need for data normalization. Furthermore, this model has the potential to replace FEM simulation methods due to its close alignment with FEM simulation results when design parameters are properly optimized. The linear regression model accurately predicts package warpage, with less than 7% error compared to FEM simulation results. It also reduces cycle time from days to minutes, making it a valuable tool for quick risk assessment in new package designs.