Ternary halide perovskites for possible optoelectronic applications revealed by Artificial Intelligence and DFT calculations

In this article, we present 134 AMX3 compounds as potential candidates to adopt the perovskite structure. These compounds, where X = F, Cl, Br or I, and M = an alkali or earth-alkali element, were found as feasible by a previously informed Artificial Neural Network (ANN). None of these compounds have been reported as synthesized up to now. The ANN assesses the probability of a set of atoms to crystallize with the perovskite structure using structural factors as input data. In addition, we implemented geometry optimization calculations with periodic boundary conditions to test the stability of the perovskite compounds predicted by the ANN. These calculations were performed within the Generalized Gradient Approach using the functional PBEsol. After these calculations, 96 of the 134 compounds preserved the most symmetric form of the perovskite structure. The rest of the compounds had a distortion within the octahedral framework. As a consequence of the DFT calculations, we found that the 134 AMX3 perovskite compounds spanned band-gap energies up to 7.88 eV. In this manner, the here reported compounds may find an application as new solar cells or transparent semiconductors, for instance. [Display omitted] •An Artificial Neural Network (ANN) was used to probe new AMX3 perovskite compounds.•The ANN used to probe the new compounds is available via GitHub.•Quantum calculations validated the 134 compounds predicted by the ANN.•The predicted compounds spanned band gap energies up to 7.88 eV.•The band gap energies of the predicted compounds are informed for the first time.