Rapid discovery of new Eu²⁺-activated phosphors with a designed luminescence color using a data-driven approach

Abstract

Machine learning in conjunction with validation experiments uncovers new Eu2+-activated phosphor materials with a designed green-color luminescence.

GPT-Assisted Learning of Structure–Property Relationships by Graph Neural Networks: Application to Rare-Earth-Doped Phosphors

“…The CGCNN model translates the crystal structures into graphs. In comparison to manually engineered fingerprints of the local chemical environment, , the graphs hopefully serve as a less arbitrary and more lossless representation. An embedding layer transforms a one-hot encoding of elements into a 128-dimensional embedding space for each node, after which graph convolutional, pooling, fully connected, and readout layers are applied.…”

“…6 In addition to trial and error, techniques including combinational chemistry, chemical unit substitution, and a single-particle diagnosis approach 2,7 have been introduced for discovering new phosphor materials. The abundance of available data on the topic has also triggered recent studies employing various machine learning models, 8 such as regularized linear regression, multilayer perceptrons, 9,10 and random forests. 11 Graph neural networks, 12 such as the Crystal Graph Convolutional Neural Network (CGCNN), 13 recently found applications in materials science.…”