Improving Phase Stability of α-CsPbI3 through Combined Strain-Doping Engineering: Insights from First-Principles Calculations and Machine Learning

The all-inorganic perovskite of cubic CsPbI₃ (α-CsPbI₃) exhibited high thermal stability, owing to the absence of volatile organic cations. However, the poor phase stability at room temperature limits its practical applications. In this work, the phase transition mechanism of CsPbI₃ from the cubic phase (α) to the tetragonal phase (β) was studied by first-principles calculations. The critical pathway defined as α-TS1-MS1 with an energy barrier of 42.36 meV/atom was identified. In order to improve the phase stability of α-CsPbI₃, a combined strategy of compressive strain (0-5%) and B-site doping (including Be, Mg, Ca, Sr, and Ba) was proposed to suppress the α-β phase transition. Through systematic calculations, seven candidate systems were obtained, which exhibit higher phase transition barriers and similar band gaps relative to those of α-CsPbI₃. Furthermore, key descriptors, such as strain, B-site doping concentration, etc., are employed to train machine learning models for predicting the α-β phase transition barriers and band gaps of α-CsPbI₃ under different control conditions. Among them, the Extra Trees Regression (ETR) model demonstrates the highest accuracy and predicts a series of efficient and stable candidate systems for experimental validation. These results provide theoretical insights for enhancing the phase stability of α-CsPbI₃ and improving the performance of related optoelectronic devices.