Abigail Hering1,Meghna Srivastava1,Juan Pablo Correa Baena2,Marina Leite1
University of California, Davis1,Georgia Institute of Technology2
Abigail Hering1,Meghna Srivastava1,Juan Pablo Correa Baena2,Marina Leite1
University of California, Davis1,Georgia Institute of Technology2
The compositional tuning of the A- and X-sites in Cs<i><sub>y</sub></i>FA<sub>1−<i>y</i></sub>Pb(Br<i><sub>x</sub></i>I<sub>1−<i>x</i></sub>)<sub>3</sub> (Cs-FA) hybrid perovskites allows bandgap engineering, relevant for light-emitting and -absorbing optoelectronic devices. However, the relationship between chemical composition and the effects of environmental stressors (light, bias, oxygen, temperature, and humidity) is currently not well understood. Thus, we use high-throughput, environmental photoluminescence (PL) to elucidate how the optical response of Cs-FA thin films with variable cation and anion concentrations changes upon materials’ exposure to relative humidity (rH) cycles that emulate accelerated day-night weather variations based on summer days in Northern California. We use machine learning (ML) models to forecast the PL response of the samples. As expected, all samples present PL enhancement with increasing rH, as a direct result of trap states’ occupation. We implement Linear Regression (LR), Echo State Network (ESN), and Seasonal Auto-Regressive Integrated Moving Average with eXogenous Regressors (SARIMAX) algorithms upon splitting the data into 50%-50% for training and testing. The latter algorithm is very suitable for predicting non-linear responses, achieving an average normalized root mean square error (NRMSE) of only 8% over a 50-hour window. Summarizing, our proof-of-concept accurate time series predictions demonstrate how ML can be realized for analyzing large sets of experimental data that can be used in a holistic approach for the further development of stable hybrid perovskites.