Exploration of New Lithium-Ion Conductors Based on Structural Analogy Using Crystallographic Site-Fingerprints

In the exploration of solid-state electrolytes (SSEs), researchers have gradually realized the significance of framework structure for lithium ion migration, and have applied refined descriptors based on framework structures, achieving considerable suceesee.^1,2 Recently, Local structure order parameters (LSOPs) have been reported as reliable procedures for both traditional and machine learning approaches, from the perspective of local environment information, aiming to exploit structure with predicting material properties.³ This new theory could provide the a new perspective of framework structure for next-generation SSEs. Herein, a new semi-supervised learning was attempted, with simplified LSOPs as the material descriptors for exploration of SSEs for the first time. 3835 raw data after preprocessing were converted to simplification-descriptors corresponding to four framework structures, and 171 experimental conductivity data was used as the conductivity to evaluate the clustering quality during the semi-supervised learning. Li-sublattice (LSOP_M) simplification-descriptor was chosen in this work since it performed minimal conductivity variance. In addition, when considering cluster number, cluster number of 11 was used for exploration of common feature within the same cluster; while it was cluster number of 228 that was consequently utilized in practical screening. In the cluster depth of 11, famous lithium conductors (Li₁₀GeP₂S₁₂, Garnet and Argyrodite-type) were mainly in the cluster #6 and #7, with the characteristic of disordered structures, specifically, compounds in cluster #6 are mainly characterized by a preference of coordination number 2 and 4, while 1 and 2-fold coordination in cluster #7. However, it would be more than 1000 compounds in both clusters, therefore, cluster number of 228 was employed in practical screening, and 147 compounds ultimately were chosen for molecular dynamics (MD) simulation after semi-supervised learning. Li₃LaP₂S₈ was selected for experiment validation and obtained the conductivity of 3.21 × 10^-7 S cm^-1 for pristine compound and 1.07 × 10^-6 S cm^-1 for Ge doping ones at room temperature. This study underscores the potential of using local coordination environments within various structural frameworks to discover promising lithium-ion conductors.<br/><br/>Reference<br/>1. Jun, K. <i>et al.</i> Lithium superionic conductors with corner-sharing frameworks. <i>Nat. Mater.</i> <b>21</b>, 924–931 (2022).<br/>2. Zhang, Y. <i>et al.</i> Unsupervised discovery of solid-state lithium ion conductors. <i>Nat Commun</i> <b>10</b>, 5260 (2019).<br/>3. Zimmermann, N. E. R. & Jain, A. Local structure order parameters and site fingerprints for quantification of coordination environment and crystal structure similarity. <i>RSC Adv.</i> <b>10</b>, 6063–6081 (2020).