Laboratory discovery of new phosphors assisted by machine learning

Conference by Romain Gautier, CNRS Research Fellow, as part of the “AI and Science” lecture series

Artificial intelligence (AI) and sciences maintain rich and constantly evolving connections, opening up unprecedented opportunities for research. What are the concrete applications of AI for research, and what impact does AI have on scientific practice and on researchers? Conversely, how do the sciences inspire the evolution of AI?

The lecture series dedicated to AI for science explores these questions. These lectures are intended for all researchers and faculty researchers, PhD students, and members of higher education and research institutions. A wide range of topics will be covered, from theoretical foundations to practical use cases.

Romain Gautier, CNRS Research Fellow at the Institut des matériaux de Nantes Jean Rouxel, will give a lecture entitled “Laboratory discovery of new phosphors assisted by machine learning” (in English):

The design of materials with specific characteristics is a complicated task as very small modifications in the chemistry or crystal structure of materials can have drastic effects on the physical properties. For example, the presence of dopants/defects in very low concentrations combined with different phenomena make, in most cases, the properties difficult to predict prior to synthesis and characterization. In this context, we use machine learning approaches to guide the discovery of materials. After a brief introduction on the potential of machine learning in materials science, I will present two families of materials that we recently investigated using these tools: (i) the lanthanides doped silicates, and (ii) the hybrid metal halides (including hybrid perovskites). Such materials can exhibit broad‐ or narrow‐band light emissions and relatively high photoluminescence quantum yields. The machine learning tools were used to identify the key experimental parameters to design phosphors with specific photoemission colors, to optimize the color rendering for phosphors with tunable correlated color temperature, or to automatically identify key structure types from the powder X-ray diffraction patterns of new compounds.