Machine Learning a Phosphor's Excitation Band Position

TL;DR

This paper develops a machine learning model to predict the excitation wavelength of phosphors, specifically Ce$^{3+}$ doped materials, to accelerate the discovery of efficient blue-excited phosphors for LED lighting.

Contribution

It introduces an extreme gradient boosting machine learning method for predicting phosphor excitation wavelengths, validated by synthesizing a novel green-emitting phosphor matching predictions.

Findings

Model accurately predicts excitation wavelengths for Ce$^{3+}$ phosphors.

Successful synthesis of a new blue-excited, green-emitting phosphor.

Model's predictions align well with experimental excitation spectra.

Abstract

Creating superior lanthanide-activated inorganic phosphors is pivotal for advancing energy-efficient LED lighting and backlit flat panel displays. The most fundamental property these luminescent materials must possess is effective absorption/excitation by a blue InGaN LED for practical conversion into white light. The 5$d_1$ excited state energy level of lanthanides, which determines the excitation peak position, is influenced by the inorganic host structure, including the local environment, crystal structure, and composition, making it challenging to predict in advance. This study introduces a new extreme gradient boosting machine learning method that quantitatively determines a phosphor's longest (lowest energy) excitation wavelength. We focus on the Ce$^{3+}$ 4$f$ $\rightarrow$ 5$d$ transition due to its well-defined 5$d_1$ energy level observed in excitation and diffuse reflectance spectra. The model was trained on experimental data for 357 Ce$^{3+}$ cation substitution sites sourced from literature and in-house measurements and ultimately experimentally validated through the successful synthesis of a novel, blue-excited, green-emitting phosphor: Ca$_2$SrSc$_6$O$_{12}$:Ce$^{3+}$. This compound's excitation under commercial blue LED wavelength aligned remarkably well with the model's predictions. These results highlight the transformative potential of data-driven approaches in expediting the discovery of blue-absorbing phosphors for next-generation LED lighting.