Perovskite or Not Perovskite? A Deep‐Learning Approach to Automatically Identify New Hybrid Perovskites from X‐ray Diffraction Patterns
Florian Massuyeau, Thibault Broux, Florent Coulet, Aude Demessence, Adel Mesbah, Romain Gautier
Abstract
Determining the crystal structure is a critical step in the discovery of new functional materials. This process is time consuming and requires extensive human expertise in crystallography. Here, a machine-learning-based approach is developed, which allows it to be determined automatically if an unknown material is of perovskite type from powder X-ray diffraction. After training a deep-learning model on a dataset of known compounds, the structure types of new unknown compounds can be predicted using their experimental powder X-ray diffraction patterns. This strategy is used to distinguish perovskite-type materials in a series of new hybrid lead halides. After validation, this approach is shown to accurately identify perovskites (accuracy of 92% with convolutional neural network). From the identification of the key features of the patterns used to discriminate perovskites versus nonperovskites, crystallographers can learn how to quickly identify low-dimensional perovskites from X-ray diffraction patterns.