Mining Solid-State Electrolytes from Metal–Organic Framework Databases through Large Language Models and Representation Clustering

Metal–organic frameworks (MOFs) are attracting
increasing
attention as solid-state electrolytes (SSEs) due to their three-dimensional
porous diffusion paths for Li+ migration. However, their
development is hindered by their inherent complexity and the absence
of design guidelines. Large language models (LLMs) and machine learning,
as emerging artificial intelligence (AI) technologies, can significantly
accelerate the development of MOF SSEs by analyzing data and identifying
potential materials. Herein, we use LLMs and representation clustering
to intelligently mine MOF SSEs from 11,393 candidate MOF materials,
subsequently verified by physicochemical characterizations and electrochemical
demonstration. Specifically, we adopt an interactive iteration text
mining framework based on LLMs to extract information on MOF SSEs,
constructing a specialized data set for structural and electrochemical
properties of MOF SSEs, with high precision and recall. Each property
is projected into a representation space, and representation clustering
is performed on samples to mine promising MOF SSEs from a candidate
MOF data set. As a typical result, NOTT-400 is successfully identified
through the clustering analysis, exhibiting high Li+ conductivity
(2.23 × 10–4 S cm–1) and
a wide electrochemical stability window (0–4.79 V), confirming
both the material feasibility and the reliability of the entire AI-driven
approach. The AI-assisted mining of novel MOF SSEs, along with their
design principles, creates a new paradigm for accelerating materials
discovery.