Revealing processing science of metal halide perovskites using AI-guided autonomous thin film fabrication

Abstract

Ambient solution-processing of metal halide perovskites remains highly sensitive to subtle variations in environment, formulation, and human technique, often leading to unpredictable outcomes and limited mechanistic understanding. These challenges persist across diverse coating methods, including but not limited to antisolvent dripping, which is especially sensitive to ambient manufacturing and challenging to replicate. In this work, we present a self-driving laboratory platform that integrates closed-loop Bayesian optimization, in situ UV-Vis and photoluminescence spectroscopy, and machine vision to systematically explore how processing dynamics impact perovskite film formation under ambient conditions. Using the benchmark composition FA₀.₈₃Cs₀.₁₇Pb(I₀.₈₃Br₀.₁₇)₃, we identify critical processing parameters governing continuous film formation, nucleation uniformity, and phase evolution. Automated analysis of thousands of in situ spectra using non-negative matrix factorization (NNMF) reveals latent features associated with successful vs failed coatings. In addition to uncovering mechanistic details, our platform provides reusable, interpretable process data for broader application.