Parametric Modeling Using Sensitivity-Based Adjoint Neuro-Transfer Functions for Microwave Passive Components

This paper proposes a sensitivity-based adjoint neuro-transfer function (neuro-TF) model for parametric modeling of microwave passive components. In the proposed technique, not only the inputoutput behavior of the modeling problem but also the sensitivity analysis information generated from electromagnetic (EM) simulators are used in the model development. Compared to the previous neuro-TF modeling method, the proposed technique can obtain accurate and parametric models with less training data. Once trained, the proposed models provide accurate and fast prediction of EM responses and derivatives used for high-level design with geometrical parameters as design variables. Two EM examples are illustrated to demonstrate the validity of this technique.