Machine learning analysis of microwave dielectric properties for seven structure types: The role of the processing and composition

Abstract During recent decades the rapid growth of wireless communications technologies has resulted in developing new dielectric ceramics for different applications. Materials for the microwave dielectric resonators have to meet three main requirements: controllable permittivity value for possible device miniaturization or fast signal transmission, high quality factor (low dielectric loss) for frequency selectivity and near-zero temperature coefficient of resonant frequency for temperature stability. In this study, the machine learning-based analysis of the experimental data was performed to distinguish the parameters responsible for the microwave dielectric characteristics and to analyze their relative contribution to the functional properties for considered compounds of several structure types: aeschynites, euxenites, columbites, ixiolites, wolframites, fergusonites and scheelites. New descriptors encapsulating the information on the constituent building blocks, their number and the point group symmetry as well as the connectivity of the building blocks have shown good prospects for use as robust parameters describing the structure of the inorganic crystalline compounds. The impact of the synthesis route and composition on the microwave dielectric characteristics was analyzed. The role of the grain boundaries and the recrystallization processes is discussed. For the considered classes of compounds, the values of the microwave dielectric parameters can be predicted with Root Mean Square Error (RMSE) of 4.84 for the dielectric constant and 21.86 for the temperature coefficient of resonant frequency.