Language models for accelerated materials design of multi-component oxides

Research question

The discovery of new materials often hinges on exploring a vast composition space in search of optimal compounds. One class of materials in particular, multi-component oxides, can, in theory, be finely tuned through precise control of their composition and topology to meet specifications in temperature stability, corrosion resistance, and catalytic activity. While this composition space, typically spanning 10-15 elements for multi-component oxides, offers incredible opportunities for precision engineering materials, it also presents a significant challenge: the sheer number of possible combinations far exceeds the capacity of traditional experimental and computational methods to identify optimal compounds. Artificial intelligence (AI) provides a possible solution to navigate the chemical space of multi-component oxides in a data-driven way, enabling the more efficient exploration of this complex landscape through techniques like generative AI and active learning. In this pilot project, we will work to develop computational methods based on highly successful strategies from natural language processing to search the composition space of multicomponent oxides more efficiently than current trial-and-error methods.

Sustainability aspects

Due to their thermal stability and ion conductivity, multi-component oxides have been used in solid oxide fuel cells. Aditionally, multi-component oxides have also been shown to be efficient catalysts for oxygen reduction and evolution, which are key processes in fuel cells and electrolysis. As these applications suggest, multi-component oxides hold great promise as sustainable materials, offering valuable solutions to advance our transition toward a more sustainable society. The ability to design multi-component oxides in a de novo fashion would have a transformative impact in sustainable materials engineering and their use in environmental applications.