Machine-learning guided engineering of Mo4+ activated halide near-infrared phosphors for AI-augmented medical imaging.

Developing highly efficient lead-free near-infrared (NIR) phosphors with strong thermal stability is a key challenge in material design and optoelectronics applications. Here, a machine-learning (ML) guided co-doping strategy to construct a broadband NIR-emitting phosphor, Cs2Zr(Cl0.46Br0.54)6:12%Mo4+/3.6%Sb3+ (SM-CZCB) is reported, achieving record-high internal and external quantum efficiencies of 92.4% and 65.9% at 920 nm, respectively. Guided by ML, Sb3+ and Br- were selected to co-dope and synergistically enhance energy transfer through the spin-orbit coupling, d-d correlation, and lattice distortion to enhance NIR emission of Mo4+. Notably, a [SbCl6]3+-[ZrCl6]2--[MoCl6]2- sequential energy transfer chain form a near-resonant configuration to reach the emission centers. The fabricated NIR light-emitting diode using SM-CZCB exhibits a record-high power conversion efficiency of 27.07% with an operational T50 exceeding 4000 hours at 450 nm excitation. Moreover, the AI-enhanced biomedical imaging was demonstrated using NIR light with high-resolution. This marks the integration of AI-guided material design with practical AI-enhanced medical imaging.