High Throughput Optical Measurements and Machine Learning Analysis of Hybrid Perovskites

Hybrid organic-inorganic perovskites are an emerging class of materials for energy-efficient devices. Intriguingly, their high density of defects has not prevented this material from presenting nearly optimal optical properties for light-absorbing and -emitting application. A primary challenge to commercialize optoelectronics based on hybrid perovskites lies on the lack of material stability. To advance the understanding of the role of environmental stressors in the optical behavior of perovskites, we implement high-throughput steady-state environmental photoluminescence (PL) while exposing the samples to distinct levels of temperature and relative humidity. Then, we use these measurements to compare the performance of linear regression, Echo State Network (ESN), and Auto-Regressive Integrated Moving Average with eXogenous regressors (ARIMAX) machine learning algorithms when predicting the PL response of perovskites with variable chemical composition. While linear regression is not adequate to forecast PL (with average normalized root mean square error, NRMSE, >20%), both ESN and ARIMAX enable reliable predictions, with NRMSE < 16% and 8%, respectively. Overall, our results indicate the promise of machine learning to accelerate the development of stable perovskites. Further, the paradigm presented here can be extended to other perovskites families.