Artificial Neural Network Model of SOS-MOSFETs Based on Dynamic Large-Signal Measurements

A measurement-based quasi-static nonlinear field-effect transistor (FET) model relying on an artificial neural network (ANN) approach and using real-time active load-pull (RTALP) measurement data for the model extraction is presented for an SOS-MOSFET. The efficient phase sweeping of the RTALP drastically reduces the number of large-signal measurements needed for the model development and verification while maintaining the same intrinsic voltage coverage as in conventional passive or active load-pull systems. Memory effects associated with the parasitic bipolar junction transistor (BJT) in the SOS-MOSFET are accounted for by using a physical circuit topology together with the simultaneous ANN extraction of: 1) the intrinsic FET current-voltage characteristics; 2) the intrinsic charges of the FET; and 3) the BJT dc characteristics, all from the same modulated large-signal RF data. The verification of the model using load-lines, output power, power efficiency, and load-pull, which is performed using two additional independent RTALP measurements, demonstrates that a reasonably accurate large-signal RF device model accounting for memory effects can be extracted from a single 10.5-ms RTALP measurement with a physically based ANN model.