A Machine Learning Approach in Identifying the Product Definition Space for the Thermal Impedance Curves of Power Switches

One well-established method for simulating the thermal behavior of power switches in motor drive applications is the use of a Cauer network. This method allows the thermal impedance (Z<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">th</inf>) curve, an important factor in the selection and design of power switches, to be simulated. This helps determine if the power switch will lead to a working application. The product definition (PD) phase establishes key thermal performance parameters, such as on-state-resistance and Z<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">th</inf>, supporting the design phase. Product definition space (PDS) shapes the boundaries of the solution space for the power switch. In this paper, we propose a machine learning (ML) approach to create fast surrogate models as a direct correlation between the Z<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">th</inf> curve and the application requirements. This method allows for optimization at a rate that is six orders of magnitude faster, while maintaining a maximum error of just 2%. This enables the experts to create the PDS by a factor of ten thousand faster and could significantly improve and accelerate the design process.