Robustness across Initial Estimates of Optimization Algorithms for Power Semiconductor Model Parameter Extraction

This paper compares optimization algorithms used to automate parameter extraction for SiC power MOSFET models. New methods for evaluating and ranking optimization algorithms are developed, which are relevant for engineering applications. This comparison includes six deterministic and probabilistic optimization algorithms (Trust Region, Least Squares, L-BFGS-B, COBYLA, Differential Evolution, and Bayesian Search) covering a theoretical and two practical examples. The objective of this paper is to demonstrate the dependencies of speed, accuracy, and expensiveness across different initial estimates in optimization problems. This also includes evaluating the necessity of setting initial parameter values in general by including boundary estimates. A robustness score is presented in the standard deviation for metrics to quantify the influence of initial parameter values.