A review of phases, defects and machine learning in hybrid organic-inorganic perovskite solar cells

Abstract

Hybrid organic-inorganic perovskite materials have been a field of keen interest in photovoltaic research community. Its improvement in efficiency over the last fifteen years has been the main motivation. Despite the development it has seen, there are two prevailing fundamental problems that need critical investigation. The first one is related to the formamidinium lead iodide (FAPbI₃) phase stability, whose photoactive phase happens to be thermodynamically metastable at room temperature. Another issue is non-radiative recombination caused by defects at grain boundaries and interfaces which lowers conversion efficiency of the devices. Literature from 2020 to 2026 seem to attempt to solve both the problems. We survey 51 studies that address these issues through tuning composition, integrating multidimensional architectures, additive and interface modification, and complementary DFT and machine learning approaches. Considering the computational side, DFT clarifies the effect of tolerance factor, octahedral connectivity and defects on the phase stability and defect tolerance. ML models accelerate screening of compositions, passivation techniques and device configurations by learning from the experimental and DFT data. The main idea is that phase stability, defects leading to recombination, and the data-driven methods dedicated to address them are not independent topics. This review organizes recent work around a single structure-defect-performance relationship, where experimental, DFT and ML findings converge or diverge. We conclude by discussing how this perspective can help us understand designing of perovskites that are not only efficient in laboratory but also structurally robust for long-term photovoltaic applications.