Predicting effective thermal conductivity of thermal interface materials using machine learning

Particle-laden composites have been reported to be widely used thermal interfacial materials (TIM), which is located between integrated heat spreader and chip and can promote the heat dissipation in the electronic packaging area. In this paper, we propose a data-driven algorithm to efficiently predict the thermal performance of the Al2O3/silicone and Al2O3-AlN/silicone composites combining the numerical high-throughput computation and machine learning algorithms, which can further benefit the design of the formula. Specifically, the geometric models of the particulate composites with various compositions are firstly constructed by GeoDict software and the corresponding thermal properties of the models are simulated. Based on the obtained database, various machine learning models are established and evaluated, including Linear Regression, Decision Tree, Random Forest, Support Vector Machine and Artificial Neural Network. Comparing the performance measures of these models such as MSE, MAE and R ² , we conclude that the random forest and artificial neural network have best performance in predicting the thermal properties of the particulate TIMs, showing satisfactory accuracy.