Artificial neural network-based solution for PSP MOSFET model card extraction.

A state-of-the-art approach on SPICE model card extraction for the PSP MOSFET model is presented in this work. An Artificial Neural Network (ANN) is used to extract model card parameters from key figures of merit of measured electrical behavior. The proposed method is based on training an ANN with ELDO simulation data. A manual PSP model card extraction can take weeks of work for experienced engineers, while this ANN-based method operates offline and infers in a matter of seconds once the model has been trained. The method has been proven on capacitance, linear and saturation regimes for corner geometries of the PSP compact model, resulting in a 27 parameter-long extracted model card. Moreover, it is a versatile approach that has the potential to be applied to other compact models. All in all, this work provides insight into the complexities of the extraction problem at hand, identifying a future investigation road map and revealing a new horizon for the model card extraction process.