Data-Driven Discovery of Ce3+-Activated Phosphors with Target Excitation Energies for Solid-State Lighting.

Solid-state lighting relies on inorganic phosphors to efficiently convert violet-to-blue light from light-emitting diodes (LEDs) into broadband visible emission. However, discovering phosphors excitable within the technologically relevant wavelength window remains a bottleneck due to complex crystal-chemical interactions on the excitation energy of lanthanide activators. Here, we introduce a data-driven framework for the a priori discovery of Ce3+-activated phosphors with targeted excitation energies. Using a curated data set of 358 experimentally measured Ce3+ substitution sites across 337 host structures, we develop a machine learning model that quantitatively predicts the 5d1 excited-state energy governing the 4f → 5d excitation transition. The model integrates descriptors capturing the local coordination environment, host crystal structure, and chemical composition, achieving a prediction accuracy of ±0.16 eV. Beyond predictive regression, the framework provides mechanistic insight into the structural factors controlling excitation energies and can assist in identifying activator substitution sites in multication hosts. External validation against newly reported phosphors generalizes the model's ability. Leveraging this predictive capability, we perform high-throughput screening of over 10,000 candidate host structures, identifying a subset compatible with commercial LED chips. By linking crystal chemistry to optical functionality, this approach establishes a scalable strategy to accelerate the discovery of next-generation phosphors for solid-state lighting.