Artificial Neural Network-Based Analysis of (81–86) GHz E-Band RF-MEMS Switches

This paper presents an efficient Artificial Neural Network (ANN)-based framework for the modeling, analysis, and performance prediction of radio-frequency microelectromechanical systems (RF-MEMS) switches operating in the E-band (81-86) GHz frequency spectrum. Traditional electromagnetic simulation methods for millimeter-wave components are highly accurate but computationally expensive, severely limiting the efficiency of iterative design and optimization loops. To address this bottleneck, a multilayer perceptron neural network model is developed to learn the non-linear mapping between the physical geometric parameters of the RF-MEMS switch and its high-frequency response. A dataset comprising various recess dimensions and their corresponding scattering parameters is generated using Ansys HFSS simulations to train and validate the network. The trained ANN model successfully predicts the RF performance, insertion and return loss, across the entire (81-86) GHz band.