Developing new battery materials requires a deep understanding of the chemistry of materials, as well as how atomic combinations influence physical properties. The number of the unexplored possible atomic combinations is such that, to test all of them to find new functional materials would take hundreds of years. However, a group of researchers has now found a way to significantly reduce the time needed, by using machine learning.

The collaborators from the University of Liverpool developed a machine learning model that was able to use knowledge of chemical interactions to suggest as yet unexplored elemental combinations that are likely to produce new materials.

Their model highlighted the combination of Li-Sn-S-Cl as being likely to form a stable material. Using this as a starting point, they were able to successfully synthesise a new material: Li_3.3SnS_3.3Cl_0.7. They then used an Xpress measurement on the Polaris instrument at ISIS, and I11 at Diamond Light Source, to characterise it.

Although having a similar composition to Li₄SnS₄, the addition of chlorine as a second anion caused the structure to be very different. Their material has a lower symmetry, with more lithium vacancies. They also found that the material was a good lithium-ion conductor, and that it has good chemical compatibility with lithium metal when used as an electrolyte, as it would be used in a battery.

Their study, published in Nature Communications, shows that machine learning can successfully support scientists when they are developing new materials. By suggesting compounds with an unusual combination of atoms, the model encourages scientists to explore avenues they may not have traditionally gone down. 

Further information

The full paper can be found at DOI: 10.1038/s41467-021-25343-7