An HC and Deep MIM Enhanced PINN SciML Algorithm for the Transient Thermal Analysis of 3-D Packages of Integrated Circuits

In this work, a physics-informed neural network (PINN) scientific machine learning (SciML) algorithm combined with the hard constraint (HC) and the deep mixed residual method (MIM) is proposed for solving transient thermal problems of 3-D Packages of Integrated Circuits (ICs). By incorporating the physics information of heat transfer into the neural network, the network can be trained to accurately predict the temperature changes over the time. In addition, hard constraints are used to learn initial conditions, while deep mixed residual methods are applied to reduce the order of the derivations, resulting in an improved accuracy and lower training complexity. To validate the proposed method, a representative 3-D package of a integrated circuit (IC) is investigated in this work.