Data Driven High Quantum Yield Halide Perovskite Phosphors Design and Fabrication

Abstract: The outstanding emission of halide perovskites make them ideal candidates for white emission lightemitting diodes (LEDs) for lighting applications. However, many perovskites contain toxic or scarce elements and have unsatisfactory stability. Here, we report a target-driven approach, based on active learning (AL) techniques, to discover halide perovskites suitable for commercial LED applications. Based on the similarity between halide and oxide perovskites, a model trained on an oxide perovskite dataset plus six AL-selected halide perovskites exhibited excellent performance for photoluminescence quantum yield (PLQY) predictions of oxide and halide perovskites. The model proposed a strong relationship between ionic radii and PLQY, postulated to be due to the self-trap excitons derived from the Jahn-Teller deformation. A novel halide perovskite phosphor, Cs4Zn(Bi0.85Sb0.15)2Cl12:0.01Mn, was designed and synthesized with the aid of the model. It exhibited an 88 % PLQY and outstanding thermal and luminescent stability. A simple white LED was fabricated from this material, exemplifying its commercial potential. This study demonstrates how machine learning techniques can accelerate discovery of next-generation phosphors for high performance single emitter-based white-light emitting devices.