Accelerated Discovery of Perovskite Solid Solutions Through Automated Materials Synthesis and Characterization

The emergence of laboratory automation has significantly accelerated the materials discovery process. Advances in machine learning (ML) have been applied in various areas, including discovering novel perovskite compositions for photovoltaics, optimizing electronic properties of thin films, and generating databases. Perovskite materials are extensively researched due to their broad functional properties, applicable in wireless communications, tunable antennas, and biosensors. However, traditional discovery and optimization of perovskite solid solutions are hindered by extensive chemical diversity and complex processes of solid state reaction influencing their micro-structure and properties.<br/>Most current ML models rely on computational samples due to the challenge of obtaining substantial experimental data, limiting their accuracy. There is a scarcity of studies that experimentally synthesize and validate ML-predicted materials, highlighting a significant research gap. With the maturation of artificial intelligence (AI) and collaborative robots, more "self-driving laboratories" (SDLs) for material discovery are being developed. However, SDLs are predominantly in their early stages, particularly within the realm of solid solutions due to intricate workflows and the complexities involved in dielectric measurements for 3D materials.<br/>We present a customizable SDL loop concept for the synthesis and characterization of perovskites, combining AI and expert scientific input. Our automated platform for rapid sintering processes and high-throughput dielectric property measurements facilitates fast material screening and characterization. The platform features a central hub on MATLAB for orchestrating lab instruments and data management. The sequence initiates with ML models recommending sample library compositions, followed by automated pellet sintering, synthesizability validation via XRD, temperature tuning, real-time dielectric property measurements, and analysis linking dielectric attributes to the synthesis process. The cycle concludes with the updating of archives.This advancement significantly reduces the time and labor required for generating validation datasets, improving ML models by rapidly revealing correlations between processes and structures. Continuous learning from experimental outcomes allows ML models to propose modifications in composition or processing factors to attain desired dielectric properties.