An Improved Thermal Matrix Model of a Forced-cooled Heat Sink Considering Uneven Cooling Conditions

To predict the thermal performance of the multichip power module, the thermal model with the representation of the thermal impedance matrixes is always used in the power module design and applications. Thermal impedance matrixes consider the impact of thermal coupling effect which improves the junction temperature prediction accuracy. However, the value element of the thermal matrix will be changed with the cooling conditions and chip power losses. This is because the heat transfer from the heatsink to the ambient obeys the law of heat convection but not the law of heat conduction. In this paper, an improved thermal matrix model is established which improved the thermal-coupling matrix of the heat sink layer based on fluid-thermal coupling simulation considering both the power losses of heat sources and the heat transfer coefficient (htc) of the heat transfer surface. Besides, the relationship between the thermal resistance from the heatsink to the ambient and power losses is linear, which is also observed in this paper. Experimental results are given and show high accuracy of the heat sink model with temperature errors below 1°C.

I. Introduction

Nowadays, IGBT power modules are extensively utilized in various fields, particularly in high-power converter applications [1]. The thermal management of power electronic devices is crucial for ensuring reliable performance, especially at high power levels. One of the most important activities in thermal management and reliability improvement is acquiring the junction temperature information of the power module [2], [3]. Due to the difficulty of direct measurement, it is widely shared that the junction temperature can be predicted based on the thermal network models [4]. However, heat flows not only transfer vertically but also in 3D [5], and heat cross-coupling inevitably occurs as shown in Fig. 1. These heat interferences between chips will increase the temperatures rising [6].