Kong Songjia, Matsui Naoki, Hori Satoshi, Hirayama Masaaki, Mori Kazuhiro, Saito Takashi, Kanno Ryoji, Suzuki Kota
Department of Chemical Science and Engineering, School of Materials and Chemical Technology, Institute of Science Tokyo, 4259 Nagatsuta-cho, Midori-ku, Yokohama, Kanagawa 226-8502, Japan.
Research Center for All-Solid-State Battery, Institute of Integrated Research, Institute of Science Tokyo, 4259 Nagatsuta-cho, Midori-ku, Yokohama, Kanagawa 226-8502, Japan.
J Am Chem Soc. 2025 Jul 16;147(28):24336-24346. doi: 10.1021/jacs.5c00856. Epub 2025 Jun 9.
The development of high-performance solid-state electrolytes for Li-ion batteries represents a critical challenge because many potential Li-containing compounds remain unexplored. In order to overcome this challenge, in this study, we utilized a semisupervised learning approach to streamline the discovery of novel Li-ion conductors by focusing on local coordination environments. Herein, we introduced four structure-representation descriptors to represent local coordination and applied agglomerative clustering to a data set of 3,835 Li-containing structures. The clusters were subsequently labeled with available experimentally determined ionic conductivity values to assess the efficacy of these descriptors in identifying promising conductors. After screening the obtained high-conductivity clusters and their neighboring structures, we shortlisted 147 compounds, which were further evaluated by molecular dynamics simulations to identify LiLaPS as a potential candidate. LiLaPS experimentally displayed low conductivity; however, optimizing the lithium content yielded LiLaSrPS, which showed a conductivity of 2.1 × 10 S cm at 298 K. To the best of our knowledge, this is the first reported investigation of LiLaPS as a solid-state electrolyte and highlights the power of semisupervised learning in accelerating the discovery of advanced materials. Our findings provide a valuable methodology for developing next-generation solid-state battery technologies.
开发用于锂离子电池的高性能固态电解质是一项严峻挑战，因为许多潜在的含锂化合物仍未得到探索。为了克服这一挑战，在本研究中，我们采用了一种半监督学习方法，通过关注局部配位环境来简化新型锂离子导体的发现过程。在此，我们引入了四个结构表示描述符来表征局部配位，并将凝聚聚类应用于一个包含3835个含锂结构的数据集。随后，用现有的实验测定离子电导率值对这些聚类进行标记，以评估这些描述符在识别有前景的导体方面的有效性。在筛选出获得的高电导率聚类及其相邻结构后，我们列出了147种化合物，通过分子动力学模拟对其进行进一步评估，以确定LiLaPS为潜在候选物。LiLaPS在实验中显示出低电导率；然而，优化锂含量得到了LiLaSrPS，其在298 K时的电导率为2.1×10 S cm。据我们所知，这是首次报道将LiLaPS作为固态电解质的研究，并突出了半监督学习在加速先进材料发现方面的作用。我们的研究结果为开发下一代固态电池技术提供了一种有价值的方法。
