Fast and accurate neural network GaAs MESFET model for time-domain circuit simulation

Programs for circuit simulation, involving nonlinear device models are very important to obtain fast and accurate results. In general, the model complexity limits the simulation level. In the last few years, nonlinear device models based in artificial neural networks (ANNs) have been used combined with the balance harmonic method, for microwave circuits steady-state analysis. Trained with physical/electromagnetic results, ANNs are precise and efficient. In this paper, a new application is proposed for this neuro-computational technique applied to nonlinear circuit transient analysis. In particular, a multilayer perceptron (MLP) network is trained with numerical 2-D results, for modeling GaAs MESFETs. Obtained results prove the validation of the MLP model applicability for time-domain circuit simulation programs.