Fast Lithium-Ion Diffusion in Weakly Bound van der Waals Frameworks

We discover that ultrafast Li-ion diffusion can be achieved in materials with van der Waals-bonded layered structures. Using fine-tuned CHGNet, an accurate and efficient universal graph neural network potential model, we predict that the room-temperature ionic conductivity in this framework reaches up to 14.7 mS/cm in the undoped structure, and by introducing excess lithium stuffing, increases up to an order-of-magnitude higher predicted conductivity of 234 mS/cm, which is one of the highest lithium ionic conductivity ever predicted for inorganic materials. While high ionic conductivity often comes at the expense of thermodynamic stability owing to the occupancy of high-energy lithium sites, this material exhibits excellent thermodynamic stability and is a ground state in the phase diagram. The ultrahigh lithium-ion conductivity originates from the combination of the structural advantage originating from van der Waals bonds and its activated diffusion networks via lithium stuffing. We prove that the soft c-directional van der Waals bonding allows both ultrahigh ionic conductivity and exceptionally low defect formation energy to add excess interstitial lithium ions. In addition, lithium stuffing in its diffusion network creates face-sharing octahedral configurations with strong lithium-lithium interactions, corresponding to an activated local environment. The homogeneity of the diffusion network allows the percolation and non-dissipation of the activated local environment to create an activated diffusion network, enhancing the ionic conductivity to a record-high value upon lithium stuffing. Guided by the features of van der Waals structures, we perform high-throughput screening to discover 10 additional novel lithium superionic conductors, verified by long time-scale molecular dynamics simulations using fine-tuned CHGNet. Our work opens an exciting path towards a new class of low-dimensional lithium superionic conductors.