Machine-learning prediction accelerates the discovery of phase-stable formamidinium-based perovskites

Tan J, Song Q, Peng Z, Zhang J, Hauch J, Lüer L, Du T, Brabec C (2026)

Publication Type: Journal article

Publication year: 2026

Journal

Book Volume: 19

Pages Range: 6188-6195

Journal Issue: 18

DOI: 10.1039/d6ee02719a

Abstract

Formamidinium lead triiodide (FAPbI3) perovskites suffer from degradation from the photoactive α phase into the photoinactive, non-perovskite δ phase. Conventionally, evaluating and optimizing FAPbI3's phase stability requires extensive aging tests, making the process time-consuming and labor-intensive. Here, we develop a predictive model that evaluates FAPbI3's phase stability using a small set of fingerprints extracted from fresh films, thereby shifting stability evaluation from prolonged aging experiments to rapid, fresh-stage characterization. The model is established through a data-driven approach that integrates an automated experimentation platform with machine-learning algorithms, enabling identification of the most informative crystallographic predictors and quantification of their coupled contributions to phase stability. Among all diffraction-derived features, the area of the (221) peak, the full width at half maximum of the (220) peak, and the area of the (111) peak emerge as the three most influential predictors, together explaining 72% of the variance in the degradation rate constant. The results suggest that phase-stable α-FAPbI3 is primarily associated with high crystallinity, while preferred orientation and reduced peak broadening provide additional contributions. Importantly, the predicted thin-film phase stability is reflected in corresponding trends in solar-cell durability under combined moisture, heat, and light stresses. This work redefines stability assessment in perovskite research by replacing prolonged aging tests with early-stage diagnostics, and thus establishes a scalable framework for data-driven materials design.

Authors with CRIS profile

Involved external institutions

Forschungszentrum Jülich / Research Centre Jülich (FZJ) Germany (DE)

How to cite

APA:

Tan, J., Song, Q., Peng, Z., Zhang, J., Hauch, J., Lüer, L.,... Brabec, C. (2026). Machine-learning prediction accelerates the discovery of phase-stable formamidinium-based perovskites. Energy and Environmental Science, 19(18), 6188-6195. https://doi.org/10.1039/d6ee02719a

MLA:

Tan, Jiayi, et al. "Machine-learning prediction accelerates the discovery of phase-stable formamidinium-based perovskites." Energy and Environmental Science 19.18 (2026): 6188-6195.

BibTeX: Download