CoRE 2D-HOIP DB: Computation-ready, experimental database of two-dimensional hybrid organic–inorganic perovskites

In silico materials design has become a widely accepted supplement to the experimental trial-and-error approach. Computation-ready, experimental (CoRE) crystal structures are commonly the foundation for creating high-throughput screening workflows to identify optimal compounds, with metal–organic frameworks and covalent organic frameworks being prominent examples. At the same time, data-driven studies devoted to two-dimensional (2D) hybrid organic–inorganic perovskites (HOIPs)—emerging photovoltaics materials—are hindered by the lack of consistently curated datasets. Here we present the CoRE 2D-HOIP database, a collection of 2D HOIP crystal structures that are readily available for atomistic simulations and machine learning. In addition, density functional theory calculations were carried out to obtain thermodynamic and electronic properties, including formation energy, energy above the convex hull, band gap, and electron effective mass. We also implemented a series of graph neural networks to approximate computational and experimental quantities, whereas machine learning interatomic potential for HOIP modeling was developed by finetuning an equivariant neural network originally trained on inorganic compounds. The publicly shared data and models constituting the CoRE 2D-HOIP database are meant to advance the rational design of 2D HOIPs via establishing structure–property relationships and benchmarking machine learning algorithms.