Generative AI Discovery of a Novel Li–F Compound as a Solid-State Electrolyte Candidate

The development of advanced solid-state electrolytes is central to achieving safer, high-energy-density lithium-ion batteries. Here we use Microsoft’s MatterGen, a generative AI model conditioned on the chemical system Li–F and a target energy above hull of 0.05eV/atom, to generate 16 novel crystal structures. A pre-trained graph neural network (M3GNet) predicts formation energies and densities. The most stable candidate, Li2F2 (gen_3.cif), has a formation energy of −3.207eV/atom and a density of 2.58g/cm3, indicating excellent thermodynamic stability and low mass. Its crystal structure is distinct from conventional rock-salt LiF, suggesting a new polymorph. This work demonstrates a complete generative AI pipeline for solid-electrolyte discovery and proposes Li2F2 for future experimental synthesis.