A Generalizable Machine Learning Framework for Identifying Sustainable Multi-Ion Garnet Electrolytes

Lithium-ion (Li-ion) solid-state batteries (SSBs) are
highly regarded
for their exceptional energy density and prolonged operational lifespan.
However, concerns regarding their sustainability have arisen due to
the uneven global distribution of Li resources and Li’s relatively
low abundance in the Earth’s crust. Consequently, significant
interest has shifted toward developing alternative SSBs, such as sodium
(Na), magnesium (Mg) and Aluminum (Al)-ion batteries. A key challenge
in this pursuit is efficiently identifying viable solid-state electrolytes
(SEs) from the vast chemical space, particularly for Na and Mg ions.
This study introduces a generalized framework based on machine learning
for effectively screening high-performance garnet-type SEs. Utilizing
specifically designed chemical descriptors, ML models predict the
thermal stability and electrical conductivity of garnet-type SEs,
achieving predictive accuracies of 94% and 89%, respectively. The
chemical factors influencing stability and conductivity are identified
and validated through interpretability analysis. Leveraging these
models, 1764 garnet-type SEs exhibiting high thermal stability and
wide band gaps were screened from a database of 43,732 compounds.
Furthermore, 44 garnet-type SEs with favorable environmental and economic
advantages were selected, and verified through first-principles calculations
using density functional theory. Given their cost-effectiveness and
high performance, these SEs hold great potential for application in
Na, Mg, and Al ion SSBs. This study provides crucial insights into
developing SSB materials, advances sustainable energy storage, and
offers key perspectives for exploring material systems within specific
space groups.