Application of Neural Networks to Accelerate Thermomechanical Simulations of Power Modules for Lifetime Prediction

Conference: CIPS 2020 - 11th International Conference on Integrated Power Electronics Systems
03/24/2020 - 03/26/2020 at Berlin, Deutschland

Proceedings: ETG-Fb. 161: CIPS 2020

Pages: 6Language: englishTyp: PDF

Acuna, Javier; Afanasenko, Valentyna; Kallfass, Ingmar (Institute of Robust Power Semiconductor Systems, University of Stuttgart, Stuttgart, Germany)
Rupp, Thomas; Sonner, Marcus; Klingler, Markus (Robert Bosch GmbH, Reutlingen, Germany)

This paper presents a methodology to simulate the thermomechanical deformation of die-attach layers of power modules using artificial neural networks. The main benefit of this approach is that the simulation time is much shorter than conventional Finite-Element simulations with only a small loss in accuracy. This method can be applied to simulate the deformation of power modules under realistic loading profiles to predict lifetime in the field. Simulation results are provided for cyclic loading and variable loading where plastic strain is used as damage indicator and the mean relative error of lifetime prediction was under 6 %.