Lithium-sulfur (Li-S) batteries could be game-changers in many ways: a theoretical specific capacity among the highest of all batteries combined with the low cost and sustainability of sulfur. However, intrinsic obstacles are imposed by the sulfur-to-lithium sulfide (Li2S) conversion mechanism via soluble polysulfides. First, polysulfides primarily cause irreversible side reactions and, thus, poor cycle life. Second, large amounts of liquid electrolyte are necessary to dissolve polysulfides during the solid-liquid-solid conversion. This tears down Li-S batteries' hoped-for energy density advantage over Li-ion batteries. Thus, realizing high-performance Li-S batteries requires foremost a fundamental understanding of electrochemical conversion. We need methods that can quantify the complex multiphase structural evolution at nanoscopic length scales during electrochemical cycling.
Here we present operando small- and wide-angle X-ray scattering (SAXS / WAXS) and operando small-angle neutron scattering (SANS) to track the growth and dissolution of solid deposits at nanometer length scales during charging and discharging Li-S battery cathodes. [1] Machine-learning-assisted stochastic modelling allows fitting the SANS data and quantifying the chemical phase evolution in real space. [2, 3] Neural networks significantly accelerate the SANS model fit via Plurigaussian random fields and improve the general validity of the fit. Complimentary local structural information obtained from cryo transmission electron microscopy verifies the model fits of the operando SANS data. Interestingly, we found that the deposit nanostructure in Li-S batteries consists of the known nanocrystalline Li2S and a second solid Li2Sx phase. The example on Li-S batteries demonstrates that structural information on nanoscopic length scales from 1 – 1000 nm is critical to understanding complex transformations such as the electrodeposition and stripping of insulating materials in beyond-intercalation-type battery cathodes. Operando SAXS/SANS, cryo electron microscopy, and machine-learning-assisted stochastic modelling combine the advantages of integral time-resolved structural information, local element-specific microscopy, and quantitative data analysis.

References:
[1] C. Prehal et al. Nature Communications, 2022, 13, 6326
[2] C. Prehal et al. PNAS, 2021, 118, e2021893118
[3] C. Prehal et al. Nature Communications, 2020, 11, 4838