Transient electromagnetic modeling using recurrent neural networks

Abstract

A novel technique for modeling the behaviour of two port passive electromagnetic (EM) structures with respect to geometrical and material parameters is introduced. A direct time domain (TD) formulation is proposed that utilizes transient responses of the structure to applied excitation signals as training data for recurrent neural networks (RNN). These EM responses are obtainable from TD EM simulators. Once trained, the RNN macromodel can be inserted into circuit simulators for use in circuit analysis. The RNN macromodel is demonstrated with two examples.