A Reduced-Order Decoupling Method Applied to Efficient Junction Temperature Simulation Model in Multichip Devices

Abstract

As the power density of power modules increases, the cross-heating effects in multichip modules become severe, which intensifies the local overheating stress. To accurately predict the dynamic thermal characteristics of the multichip modules, this paper establishes the mathematical relationship between the resistance-capacitance (RC) parameters and the thermal diffusion angle to obtain accurate RC parameters. The decoupling of the self-heating and cross-heating effects is realized according to the dynamic power flow in the RC thermal network, reducing the transient thermal impedance matrix to the steady-state thermal resistance matrix. In this way, the 3-D cross-heating effect on the junction temperature is characterized by the 1-D RC thermal network. Then, an efficient junction temperature simulation model based on the reduced-order decoupling method is proposed. This article also identifies the variation law of RC parameters with boundary conditions. Compared with existing models that need transient information, the model parameter extraction process requires only the steady-state information of the finite element method (FEM) and fewer RC parameters, which improves the parameter extraction efficiency by more than 87 times. In addition, compared with the FEM model, the transient junction temperature prediction efficiency is improved by more than 994 times. Finally, based on a 1200V/50A half-bridge module, an electrothermal coupling model is constructed and applied to a three-phase two-stage inverter circuit. The results verify the accuracy and effectiveness of the proposed model, demonstrating that the junction temperature error between the proposed model and the experimental result is less than 5%.