Machine learning assisted synthetic acceleration of Ruddlesden-Popper and Dion-Jacobson 2D lead halide perovskites

Two-dimensional lead halide perovskites (2D LHPs) can be generally categorized into the Ruddlesden-Popper (RP) and Dion-Jacobson (DJ) phases, with unbiased and biased structural packing manners, respectively. To distinguish the two different phases during materials discovery, the synthesis has heavily relied on the traditional trial-and-error methods which severely delay the rapid development of these emerging 2D materials. In this work, we applied machine learning (ML) to accelerate the synthetic development of (100)-oriented 2D A2BX4 and ABX4 LHPs (A = organic amine cation, B = Pb, X = halide). The experimentally available 264 crystal structures were used as training data for our ML model to identify the descriptors influencing the formation of the RP- and DJ-phases. The number of nitrogen atoms able to be protonated, the heteroatom type, and the nitrogen content of the organic amine are identified as the most influential factors. To validate our model, three organic amines were tested by growing single crystals and two of the three new structures are consistent with our ML results. This study takes one step further toward the rational synthesis of RP- and DJ-phase hybrid 2D LHPs via an approach of ML.