Accelerating Parameter Extraction of Power MOSFET Models Using Automatic Differentiation

The extraction of the model parameters is as important development compact itself because simulation accuracy fully determined by used. This article proposes an efficient model-parameter method for models power metal-oxide semiconductor field-effect transistors ( mosfet s). proposed employs automatic differentiation (AD), which extensively used training artificial neural networks. In AD-based parameter extraction, gradient all analytically calculated forming a graph that facilitates backward propagation errors. Based on gradient, computationally intensive numerical eliminated and are efficiently optimized. Experiments conducted to fit current capacitance characteristics commercially available silicon carbide using having 13 parameters. Results demonstrated could successfully derive 3.50× faster than conventional numerical-differentiation while achieving equal accuracy.