Prediction of ABO3-type Perovskite Structures Using Machine Learning for Developing Microwave Dielectric Ceramics

Microwave dielectric materials have extensive applications in wireless communication devices, from satellite communication to GPS and telecommunications. To meet the current and future system requirements, new microwave components using specialised dielectric materials and innovative designs must be developed. Recent advancements in microwave telecommunications have increased the interest in device miniaturisation by developing dielectric resonators (DRs) with high dielectric constants and low dissipation factors. Oxide perovskite ceramics have attracted interest as promising DR materials owing to their tunable structures that satisfy the required microwave characteristics. This study employed a novel predictive modelling approach using machine learning wherein the properties, namely the dielectric constant and quality factor of existing perovskites, reported in the literature, were utilised to predict the structure of a new perovskite ceramic having the desired properties. Using Extreme Gradient Boost (XGBoost) and Categorical Boost (CatBoost) algorithms, the radii of sites A and B in ABO3 perovskites were predicted to match the desired dielectric constant and quality factor. A separate code was developed to predict the stoichiometric ratio of the cations occupying each site. Two sample sets of dielectric constants (60,90) and quality factor value (105) for prospective DR applications were selected, and the model predicted two new perovskite structures: (Ca0.5Sr0.3Ba0.2)(Zr0.2Ti0.8)O3 and Si(Sr0.3Ti0.7)O3. The phase formation of the predicted oxide perovskite structures was experimentally validated using solid-state sintering and XRD analysis. The developed model is useful for selecting DR materials with dielectric constants of 60–90 for various microwave applications.