Non-equilibrium crystallization and spatial fingerprints in flash IR-annealed halide perovskite films.

Metal-halide perovskite solar cells combine high power-conversion efficiencies with solution processability, yet scalable fabrication remains limited by incomplete control over crystallization pathways and the resulting film heterogeneity. Under flash infrared annealing (FIRA), millisecond photonic pulses drive strongly non-equilibrium nucleation and growth, producing spherulitic microstructures whose final geometry stores measurable comparative signatures of the underlying crystallization pathway. Here, we establish a segmentation-based framework that converts bright-field microscopy of FIRA-processed films into quantitative comparative descriptors of grain morphology, video-anchored effective kinetics, and spatial microstructural fingerprints, providing a practical route to analyze crystallization under manufacturing-relevant rapid-processing conditions. Using time-resolved crystallization videos of pristine FAPI and FAPI-TEMPO together with a larger static microscopy dataset of roughly 3000 images from about 100 processed films, we quantify how additive chemistry reorganizes crystallization across both dynamic and end-state image populations. The workflow combines semi-supervised instance segmentation and mask-quality classification with grain-level morphology extraction and video-anchored kinetic reconstruction, with the video data providing the kinetic anchor and the static dataset providing the principal statistical support. From a curated library of more than 420 000 validated spherulites (180 545 for FAPI and 241 619 for FAPI-TEMPO), we derive effective growth-rate distributions, transformed-fraction curves, empirical kinetic descriptors, and spatial signatures based on texture entropy, defect loading, shape regularity, radial profiles, and crowding metrics. We find that TEMPO delays and narrows the dominant crystallization burst, reduces grain-size dispersion (median area reduced by 34%,