An Accurate and Reliable Behavioral Modeling Technique for Fully Printed Vanadium Dioxide RF Switches Using Model Ensembling Approach

This paper develops and showcases a model ensembling-based accurate, reliable and computer-aided design integrable behavioral modeling technique for emerging fully printed Vanadium dioxide (VO2) based Radio Frequency (RF) switches. Initially, separate and independent models are trained using Extreme Gradient Boosting (XGBoost), Categorical Boosting (CatBoost), and Light Gradient Boosting Machine (LightGBM) gradient boosting frameworks. The hyperparameters of the standalone XGBoost, CatBoost, and LightGBM based models are optimized using random search optimization coupled with a cross-validation scheme. Subsequently, weighted ensemble models are constructed by leveraging the optimally trained XGBoost, CatBoost, and LightGBM based models. It is vital to carefully calibrate the ensembling weights; therefore, an optimization algorithm, namely Tuna Swarm Optimization (TSO), is employed. Finally, all the developed models are tried and validated on standard regression tests, including mean relative error across all operating temperature conditions. The proposed weighted ensemble models have achieved remarkable accuracy and efficiency in simulating the behavior of VO2 RF switches.