Autonomous stabilization of metastable h-TbFeO3 thin film

Metastable materials have attracted great interest due to their exotic physical properties, which can potentially lead to novel semiconductor devices. However, the realization of such materials is often hindered by thermodynamic constraints, requiring extensive trial and error to prevent decomposition or relaxation into more stable phases. In this study, we successfully stabilized hexagonal TbFeO₃ (h-TbFeO₃) through a fully autonomous synthesis approach. By combining in situ RHEED diffraction patterns with a real-time, deep-learning–based analysis pipeline, we developed a robust metric to characterize the pulsed laser deposition (PLD) growth process. A Gaussian process (GP)–based Bayesian optimization framework was then employed to maximize this metric, leading to the discovery of optimal conditions for growing high-quality h-TbFeO₃ films with smooth surfaces suitable for subsequent device fabrication. The stability of h-TbFeO₃ under various conditions was extracted from the posterior mean of the GP model and further confirmed through a series of post-optimization deposition experiments. By comparing our results with previous literature, we found that the autonomous workflow independently reproduced the optimal conditions reported in earlier studies, while also revealing a rarely reported phase transformation occurring at elevated temperatures above 900 °C. We believe that this workflow can be generalized to other material systems and provide researchers with a powerful foundation for understanding and optimizing PLD-grown thin films.