Researchers from the University of Houston have demonstrated the use of machine learning to identify promising phosphor compounds. Rare-earth substituted inorganic phosphors form a critical light color conversion component for solid state lighting. Traditionally, trial and error has been the method for discovering inorganic phosphors.
The researchers used a machine learning algorithm called a vector regression model to predict the Debye temperature of the phosphor host’s crystal structure. In the calculations, they used the Debye temperature as a proxy for the photoluminescent quantum yield. Then, they employed high-throughput density functional theory calculations to evaluate the band gap of the proposed phosphor compound.
Machine Learning Predicted Phosphor Performance
The researchers predicted the Debye temperature of 2071 potential phosphor hosts with the aid of machine learning and correlated with the bandgaps calculated with via the density functional theory. They made a sorting diagram of all of the phosphor compounds they looked with the Debye temperature and the calculated bandgaps. The diagram reportedly helps identify classes of the next generation of inorganic phosphors.
Borates Hold Much Potential Among Phosphor Compounds
The researchers found that phosphor combinations with the borate group hold much potential for finding great performing phosphor compounds. The group found specifically that sodium barium boron 9 oxygen 15 (NaBaB9O15) is among the compound combinations with the highest Debye temperature and largest band gap and, therefore, shows outstanding potential. After the group synthesized the compound, they substituted Eu2+ for a [B3O7]5– polyanionic backbone. Then the group found that NaBaB9O15:Eu2+ has a quantum yield of 95% and boasts excellent thermal stability.
The emission wavelength of the phosphorNaBaB9O15:Eu2+ is 34.5nm and resides in the extreme ultraviolet range.
The researchers detailed their findings in the Oct. 22, 2018, issue of Nature Communications.
The use of machine learning to identify phosphor compounds is not a new idea. A firm called Intematix touted a similar method for phosphor discovery in about 2005 and called it the company’s Discovery Engine.
Reference
Zhuo, Y., Tehrania, A. M., et al. Identifying an efficient, thermally robust inorganic phosphor host via machine learning. Nature Communications, volume 9, Article number: 4377 (2018).
