Efficient Neural-Based Large-Signal and Isothermal Models for the Dual Gate MESFET

This paper presents neural-based large-signal and isothermal models for the dual gate MESFET as efficient alternatives to existing nonlinear models for such a complex device. The developed neural model is a combination of two sub-models; a static model represented by DC current-voltage characteristics and a dynamic model represented by pulsed current-voltage characteristics. The isothermal model is based on pulsed current-voltage measurements to better represent the RF device behavior and to neutralize the effect of channel self-heating on model accuracy. Insights on the discrepancy between model parameter values extracted from pulse and from DC current-voltage measurements are also discussed. The measurement and model data are in very good agreement with global model errors of less than 1%.