Structural Descriptors Controlling Ionic Motion in Solid Electrolytes from Automated Atomistic Computations

Modern battery technology is limited by the use of organic liquid electrolytes, which are known to have considerable safety and stability issues [1, 2]. Fast solid-state inorganic conductors offer a path toward safer batteries with high energy density, but apart from a few material classes [3-5], the inorganic solid-state space remains mostly unexplored. Computational approaches using density functional theory (DFT) have been proven to be successful for the screening and discovery of electrode materials, but have not been used for screening solid electrolytes. Unlike electronic conductivity, which can be estimated from the electronic structure, the physiochemical factors that regulate ionic conductivity are poorly understood. We present relationships between the ionic conductivity and several potential structural descriptors, such as the size and dimensionality of ion-conducting pathways, void fraction, Li-concentration, sensitivity to volume, etc. We use these relationships, obtained from automated ab-initio molecular dynamics simulations (AIMD), to develop strategies for rapid screening and deploy them on a dataset of 1500 distinct crystal structures.

Our results indicate that there exists a sharp threshold channel size (see Fig. 1), below which there is no motion of Li-ions. Another important feature is the dimensionality of the channels present in the material. We have found that almost all materials that show conductivity have a three dimensional channel of the threshold size. Conductivity has also been found to be sensitive in some cases to slight changes in volume. A 10% reduction in volume leads to a decrease in conductivity in a large fraction of materials and a similar volume increase causes numerous non-conducting materials to become conducting. We also show that it is possible to change the conductivity by altering the concentration of Li ions present in the material. Using these results, we have developed a pre-screening methodology that dramatically reduces the set of candidate materials by a factor of five. By accepting only three-dimensional conductors, and using conservative thresholds on the band gap and the channel size, we can reduce the size of dataset by 50% before performing a simulation. Short AIMD simulations constitute the next level of screening, and though the obtained conductivity is not converged, it is sufficient to differentiate between conductors and non-conductors. We test the non-conducting materials by varying volume and Li-concentration, and eliminate those that are immobile. The resulting dataset can now be used for more detailed investigations of transport mechanisms. We show that this approach can successfully rediscover known ionic conductors. We expect that these results are a vital first step towards accelerating the discovery of new solid-state electrolyte materials.

This work was performed using an award of computer time provided by the Innovative and Novel Computational Impact on Theory and Experiment (INCITE) program at OLCF.

References

[1] Aurbach, D., Journal of Power Sources, 89, 206-218 (2000).

[2] Goodenough, J.B., Kim, Y., Chemistry of Materials, 22, 567-603 (2010).

[3] Knauth, P., Solid State Ionics, 180, 911-916 (2009).

[4] Kamaya, N., Nature Materials, 10, 682-686 (2011).

[5] Mo, Y., Ong, S.P., Ceder, G., Chemistry of Materials, 24, 15-17 (2012).

Figure 1. Apparent conductivity vs. channel diameter for potential solid electrolyte candidates. A clear threshold channel size can be seen below which there is no conductivity. Almost all conducting materials have a three dimensional channel of the threshold diameter. Note: Conductivity values have been obtained from short ab-initio molecular dynamic simulations and are not fully converged.