Predicting Thermal Resistance of Packaging Design by Machine Learning Models

Thermal analysis is an indispensable aspect of semiconductor packaging. Excessive operating temperatures in integrated circuit (IC) packages can degrade component performance and even cause failure. Therefore, thermal resistance and thermal characteristics are critical to the performance and reliability of electronic components. Machine learning modeling offers an effective way to predict the thermal performance of IC packages. In this study, data from finite element analysis (FEA) are utilized by machine learning models to predict thermal resistance during package testing. For two package types, namely the Quad Flat No-lead (QFN) and the Thin Fine-pitch Ball Grid Array (TFBGA), data derived from finite element analysis, are employed to predict thermal resistance. The thermal resistance values include theta JA, theta JB, theta JC, Psi JT, and Psi JB. Five machine learning models, namely the light gradient boosting machine (LGBM), random forest (RF), XGBoost (XGB), support vector regression (SVR), and multilayer perceptron regression (MLP), are applied as forecasting models in this study. Numerical results indicate that the XGBoost model outperforms the other models in terms of forecasting accuracy for almost all cases. Furthermore, the forecasting accuracy achieved by the XGBoost model is highly satisfactory. In conclusion, the XGBoost model shows significant promise as a reliable tool for predicting thermal resistance in packaging design. The application of machine learning techniques for forecasting these parameters could enhance the efficiency and reliability of IC packaging designs.