One of society's grand challenges is to reduce energy usage in ways that are cost-effective, sustainable, and environmentally benign. Replacing incandescent and compact fluorescent light bulbs with energy-efficient, solid-state white lighting is one of the easiest and most promising solutions. Eu3+-substituted inorganic oxide phosphors are one class of materials that can serve as the red component in these new light bulbs, allowing the creation of warm white light. Unfortunately, the emission intensity in most of these materials cannot be reliably maintained at elevated temperatures. There is therefore a need to discover entirely novel phosphor materials that are thermally robust; however, this is generally a prolonged and expensive process requiring extensive synthetic effort. In this work, we develop a machine-learning regression algorithm based on 134 experimentally measured temperature-dependent Eu3+ emission data points to rapidly estimate the thermal quenching temperature (T50), which is defined as the temperature when the emission intensity is half of the initial value. The T50 was then predicted for more than 1000 potential oxide Eu3+ phosphor hosts using this model. Five compounds with predicted thermal quenching temperatures >423 K were subsequently selected and synthesized for validation of this approach. The phosphors, Sr2ScO3F, Cs2MgSi5O12, Ba2P2O7, LiBaB9O15, and Y3Al5O12, all exhibit good thermal stability when substituted with Eu3+, suggesting the success of our methodology.
Keywords: inorganic phosphors; machine learning; photoluminescence; support vector regression; thermal quenching.