High-Throughput Analysis of Combinatorial Metal Halide Perovskite Libraries, a FAIR Dataset in NOMAD

Machine learning and artificial intelligence claims to offer a paradigm shift in how experimental materials research is performed. However, these methods often require uniformly labeled and very large datasets, something that is not traditionally available in experimental materials research. One way to begin obtaining the necessary data volumes is by applying high-throughput technologies.<br/>In this work, we present a high-throughput investigation of metal halide perovskites for the use in optoelectronics. Combinatorial libraries of Cs<i>_y</i>Pb_<i>1-y</i>(Br_<i>x</i>I<i>_1-x</i>)<i>_2-y</i> perovskites were fabricated by thermal co-evaporation and investigated using contact-less high-throughput characterization such as hyperspectral photoluminescence imaging, time-resolved photoluminescence mapping and grazing-incidence wide-angle X-ray scattering mapping. The high-dimensional datasets of each sample were reduced and combined to zero-dimensional descriptors using automated analysis. These descriptors were used to show how the theoretical photovoltaic power conversion efficiency depends on the Br to I and Cs to Pb ratios in a previously unexplored range of the Cs<i>_y</i>Pb_<i>1-y</i>(Br_<i>x</i>I<i>_1-x</i>)<i>_2-y</i> solid solution. A generalized data schema for combinatorial thin films was developed in order to disseminate the dataset in a Findable, Accessible, Interoperable and Reusable (FAIR) way within the Novel Materials Discovery (NOMAD) repository. Through NOMAD, the labeled data of the 3456 individual samples of this dataset can be combined with others to provide the volume of experimental data needed for applications in machine learning and artificial intelligence.