Hybrid Data-Driven and Mechanistic Modeling Approach for Power Module Rapid Thermal Analysis

The safe operation and lifetime of the power module are heavily dependent on the temperature distribution, making it imperative to optimize the thermal performance of the layout at the design stage. During the module design phase, a significant number of layouts are assessed. It is important to constrain computational complexity while achieving high accuracy. Mechanistic modeling and data-driven modeling have limitations in terms of complexity and generalizability respectively. Therefore, this article proposes a hybrid data-driven and mechanistic modeling approach. This approach incorporates the theoretical analysis of mechanistic modeling and the high efficiency of data-driven modeling to accurately and efficiently compute the temperature distribution of a given layout. This approach initially acquires the dataset from simple structures and conducts a dimensionality reduction to obtain the isothermal point set (IPS). Later, IPS of complex structures is developed based on mechanistic modeling. Finally, the IPS is converted to a field to obtain a temperature distribution of the given layout. The accuracy of the approach is verified by a steady-state thermal experiment. With the proposed approach, an accurate assessment of surface thermal distribution of different layouts can be quickly finished. The thermal characteristics evaluation efficiency can be improved roughly 300 times, which provides the necessary technical basis for layout optimization.