Towards Stable Hybrid Perovskites: High-Throughput Characterization and a Machine-Learning-Assisted Analysis

<br/>Hybrid organic-inorganic perovskites (HOIPs) represent a promising class of material for next-generation optoelectronics. Yet, their long-term stability must be established prior to commercialization. Edisonian, traditional trial-and-error methods for material screening, development, and stability testing are slow considering the vast hyper-parameter space entailing chemical composition and the potential influence of environmental stressors when acted in combination. To overcome this bottleneck towards the identification of stable HOIPs, we realize automated experimentation and machine learning (ML) to gain physical understanding of how these materials’ behavior is affected by environmental stressors, such as relative humidity and temperature. We apply ML models to analyze high throughput, <i>in situ</i> steady-state photoluminescence and predict the changes in Cs<i>_y</i>FA_1−<i>y</i>Pb(Br<i>_x</i>I_1−<i>x</i>)₃ perovskites while exposed to relative humidity cycles. We compare linear regression, echo state network, and seasonal auto-regressive integrated moving average with eXogenous regressor algorithms, and attain consistent accuracy of &gt;90% for the latter, while following long-term changes for 50 hours. Our accurate time series predictions showcase the promise of ML to mimic non-linear response from a series of hybrid perovskite compositions.