Data-Driven Discovery of Broadband Cr3+ Phosphors Emitting in the Second Near-Infrared Window

Broadband near-infrared (NIR) phosphors based on LED-converted Cr3+-substituted inorganic materials are of interest for biomedical imaging, night vision, and spectroscopy. Extending emission into the NIR-II window (>1000 nm) would enable deeper tissue penetration with reduced scattering, but such emission is exceedingly rare, with only six materials previously reported. Finding new Cr3+ phosphors is challenging because the emission maximum (λem) and full width at half-maximum (fwhm) are highly sensitive to host composition, crystal structure, and local coordination environment, making them hard to predict in advance. Here, we present a data-driven framework for identifying Cr3+ phosphors and apply it to find two new NIR-II-emitting materials, LiInW2O8 and NaScW2O8, with broadband emission centered at 1085 and 1100 nm, respectively, and a third NIR-I emitter, Ba2ScTaO6, centered at 890 nm. This effort required building a data set of 208 experimentally reported Cr3+ phosphor spectra using an automated literature-extraction pipeline and training two CatBoost regression models to independently predict λem and fwhm. Examining 4062 oxide and fluoride hosts highlighted candidates with desired emission characteristics. Synthesis and optical characterization show that measured λem values agree with model predictions to within 5%, and fwhm values agree within 20% across all three compounds. SHAP analysis of the trained models provides physically meaningful design principles from the data and recovers expectations from crystal-field theory without explicitly encoding them. These results extend Cr3+-based broadband NIR emission deeper into the NIR-II window than previously demonstrated by data-driven discovery and establish a transferable framework for predicting spectroscopic properties of transition-metal-substituted NIR phosphors governed by d-d electronic transitions.