Stable Lithium Ion Conducting Thiophosphate Solid Electrolytes Li_x(PS₄)_yX_z (X = Cl, Br, I)

In the search for new solid electrolytes with favorable combinations of fast ionic conductivity and sufficient structural and electrochemical stability, we here explore lithium compounds containing multiple anions, in particular the combination of thiophosphate and halide anions. On the basis of computational screening using our bond valence site approach and DFT studies several lithium thiophosphate halides including Li4PS4I, Li5PS4Cl2, Li7P2S8I, and the new Li15(PS4)X3 (X = Cl, Br) have been explored. Their properties are discussed based on both static bond valence site energy pathway models and empirical molecular dynamics simulations using a BVSE-based force field, introducing also a new algorithm to predict absolute ionic conductivities from the static BVSE pathway analysis. MD simulations show that both Li4PS4I and Li5(PS4)Cl2 is found to undergo an order–disorder phase transition and thus Li5(PS4)Cl2 should, contrasting to earlier predictions, not be a fast Li+ ion conductor at room temperature. Similarly, Li4PS4I the ionic conductivity should not reach the previously predicted extraordinary room temperature values. The newly predicted thermodynamically stable cubic solid electrolyte Li15(PS4)4Cl3 was successfully prepared and characterized for the first time. Li15(PS4)4Cl3 crystallizes in space group I4̅3d. Aliovalent doping of Mg2+ on Li+ sites enhances ionic conductivity only slightly from 4 × 10–8 S cm–1 to 2 × 10–7 S cm–1. Though the conductivity of Li15(PS4)4Cl3 does not reach a superionic level, it demonstrates that our computational approach can successfully predict a completely new classes of solid electrolytes. Beyond the specific class of compounds we also discuss the fundamental challenge that the high probability of order–disorder phase transitions in intrinsic ionic conductors poses to the predictability of room temperature ionic conductivity by extrapolations from high temperature simulations. Empirical simulations although less precise can provide substantial additional insight, as they can cover both more complex structure models and wider temperature ranges to eliminate the need for extrapolations.