Development, Optimization, and Application of ML based Modeling of Printed VO2 RF Switch

This paper proposes a globally optimized behavioral modeling algorithm using cascaded feed-forward neural network in conjugation with Particle Swarm optimization (PSO) for fully printed V$O_{2}$ based RF switches. The proposed model makes use of varied set of operating conditions (geometric dimensions and operating temperature) over a frequency range of 0.01 to 33 GHz. The modeling algorithm enables flexibility in selecting the optimized hyperparameters such as number of neurons in each hidden layer, and activation function at each layer. The developed model is tested for both interpolation and frequency extrapolation cases up to 40 GHz to establish the validity and robustness of the modeling algorithm. An excellent agreement between the measured and the modeled performance over a broad frequency range demonstrates a good generalization capability and successful model development strategy. The proposed model is then evaluated within a commercial circuit simulator (Keysight’s ADS) to demonstrate usefulness in RF circuit design.