Advanced parametric modeling using neuro-transfer function for EM based multiphysics analysis of microwave passive components

This paper proposes a novel parametric modeling technique for electromagnetic (EM) based multiphysics analysis of microwave passive components. Multiphysics parameters usually affect the EM performance by indirectly influencing the geometrical variables, such as thermal effect causing an expansion in the geometrical parameters, and stress inducing the physical deformation. In the proposed technique, the input classification and correlating mapping is introduced to transform the multiphysics input parameters into geometrical input parameters. Further, the combined neural network and transfer function technique (neuro-TF) is used to model the EM responses w.r.t. the transformed geometrical variables. The model obtained using the proposed technique can achieve good accuracy with low complexity of neural networks, and further can be used in the high-level design. One tunable evanescent mode cavity filter example is used to demonstrate the validity of this technique.