Machine vision-enabled octahedral network reconstruction and structural analysis of perovskite quantum dots

Abstract

The structural framework of metal-halide perovskites is defined by corner-sharing PbX6 octahedra, whose tilts, distortions, and connectivity dictate the phase stability, carrier dynamics, and optoelectronic performance. Despite their pivotal role, direct experimental analysis of octahedral configurations in perovskite quantum dots (QDs) remains elusive due to the lack of robust analytical standards. Here, we introduce a machine vision-enabled approach integrating self-supervised denoising (S2SRED) for noise-sensitive datasets, atomic species classification, and automated reconstruction of the PbX6 octahedral network with precise lattice parameter extraction, enabling high-fidelity processing of low-dose scanning transmission electron microscopy (STEM) images. In CsPbI3 QDs, we observe reduced PbX6 octahedral tilting in the outer unit cells, forming an isotropic core–shell feature. In contrast, mixed-halide CsPbI3–xBrx (x = 0.5) QDs show inhomogeneous and anisotropic PbX6 octahedral tilting distributions resulting from dopant segregation and impaired phase stability as corroborated by photoluminescence measurements. By standardizing metrics for octahedral and lattice geometries, this method helps establish atomic-scale structure–property links in perovskite nanomaterials.