Romain Gautier1,Florian Massuyeau1
CNRS-IMN1
Romain Gautier1,Florian Massuyeau1
CNRS-IMN1
Hybrid lead halides of perovskite type have recently shown a great potential in optoelectronic applications. For this reason, many research groups are currently exploring this chemical system to discover new low dimensional hybrid perovskites. However, discovering such materials is challenging as the necessary structure determination by X-ray crystallography is time consuming and non-perovskite compounds are very often synthesized instead of the more interesting perovskites. In this context, we developed a deep learning approach, which automatically and accurately assign the structure type from the X-ray diffraction patterns of new hybrid lead halides.<sup>[1]</sup> The models could automatically identify new hybrid perovskites with an accuracy as high as 92%. Interestingly, we were able to identify and explain the key features in the diffraction patterns, which allow the machine learning algorithms to discriminate between perovskites and non perovskites. From this information, the scientists’ ability in discriminating the different structure types is largely augmented and such algorithms could be included in autonomous materials discovery cycles in the future.<br/>[1] F. Massuyeau, T. Broux, F. Coulet, A. Demessence, A. Mesbah, R. Gautier, <i>Advanced Materials</i> <b>2022</b>, <i>Accepted</i>, 2203879.