Machine-Learning-Guided Insights into Solid-Electrolyte Interphase Conductivity: Are Amorphous Lithium Fluorophosphates the Key?

Despite decades of study, the identity of the dominant \ce{Li+}-conducting phase within the inorganic SEI of Li-ion batteries remains unresolved. While the mosaic model describes LiF/\ce{Li2O}/\ce{Li2CO3} nanocrystallites within a disordered matrix, these crystalline phases inherently offer limited ionic conductivity. Growing evidence suggests that interfaces, grain boundaries, and amorphous phases may instead host the primary fast-ion pathways. Using diffusion-based generative structure prediction and machine-learning interatomic potentials (MLIPs), we investigate lithium difluorophosphate (\ce{LiPO2F2}), a key mixed-anion decomposition product of phosphorus- and fluorine-containing electrolytes. We identify a stable crystalline polymorph and demonstrate that the amorphous counterpart is conductive, with projected room-temperature $σ\approx 0.18$ mS cm$^{-1}$ and $E_\mathrm{a} \approx 0.40$ eV. This enhancement stems from structural disorder flattening the Li site-energy landscape and a low formation energy for Li-interstitial defects, which supplies additional mobile carriers. We propose amorphous mixed-anion Li–P–O–F phases as a promising conducting medium in the SEI, offering a specific target for engineering improved battery interfaces.

💡 Research Summary

The paper tackles the long‑standing question of which inorganic component of the solid‑electrolyte interphase (SEI) in lithium‑ion batteries provides the dominant Li⁺ transport pathway. Conventional wisdom attributes the SEI to a “mosaic” of nanocrystalline LiF, Li₂O, and Li₂CO₃ embedded in an amorphous matrix, yet these crystalline phases are intrinsically poor ionic conductors because of high defect formation energies and large migration barriers. Recent experimental observations—especially cryogenic TEM and X‑ray photoelectron spectroscopy—suggest that mixed‑anion fluorophosphate species, particularly LiPO₂F₂, form a substantial fraction of the inorganic SEI, especially when fluorine‑ and phosphorus‑containing electrolyte additives are employed. Moreover, high‑concentration electrolytes tend to produce a homogeneous amorphous inorganic SEI rather than nanocrystalline domains.

To elucidate the role of LiPO₂F₂, the authors combine a diffusion‑based generative crystal‑structure prediction model (CHGGen) with machine‑learning interatomic potentials (MLIPs) to explore both crystalline and amorphous polymorphs. CHGGen operates in two stages: (i) unconditional generation of a plausible atomic arrangement, followed by (ii) temporary removal of Li⁺ ions, refinement of the PO₂F₂ polyanion framework into a higher‑symmetry structure, and finally reinsertion of Li⁺ via an in‑painting step. This approach efficiently navigates the vast configurational space and yields several low‑energy candidates. After rapid screening with a pretrained CHGNet potential, the most stable polymorph is identified as a C2/c crystal with a decomposition energy of –0.017 eV/atom relative to the r²SCAN Materials Project phase diagram. The structure consists of corner‑sharing PO₂F₂ tetrahedra linked by LiO₄ tetrahedra, preserving the molecular motif of the gas‑phase precursor.

For the amorphous phase, the crystalline C2/c structure is melted at 1000 K using the fine‑tuned CHGNet MLIP (trained on a custom DFT dataset for the Li‑P‑O‑F chemistry), then quenched and equilibrated at temperatures ranging from 300 K to 700 K. Large supercells (>10 Å per side) are employed to capture realistic disorder. Radial distribution functions confirm that short‑range Li–O/F coordination is retained while long‑range Li–P correlations disappear, characteristic of an amorphous network. The amorphization energy ΔE_pot (difference between average amorphous and crystalline potential energies) is only 30 meV/atom, substantially lower than for Li₂CO₃ (64 meV/atom), LiF (95 meV/atom), and Li₂O (124 meV/atom), indicating that LiPO₂F₂ can readily become amorphous under SEI formation conditions.

Li⁺ diffusion is quantified from mean‑squared‑displacement trajectories. The crystalline polymorph exhibits a steep Arrhenius slope with an activation energy E_a = 1.13 ± 0.21 eV, implying negligible Li⁺ mobility at room temperature. In stark contrast, the amorphous phase shows E_a = 0.40 ± 0.01 eV, leading to a projected room‑temperature ionic conductivity of σ ≈ 0.18 mS cm⁻¹. A 60 meV reduction in activation energy corresponds to roughly an order‑of‑magnitude increase in diffusivity at 300 K, underscoring the significance of the amorphous state.

To rationalize this enhancement, the authors analyze the density of atomic states (DOAS) for Li site energies. The crystalline phase displays a narrow energy distribution, indicating deep, well‑defined potential wells that hinder hopping. The amorphous phase, however, exhibits a broadened, nearly continuous distribution of site energies, flattening the energy landscape and facilitating thermally activated hops. Additionally, the formation energy of Li interstitials is markedly lower in the amorphous matrix, enabling “Li‑stuffing”—the accommodation of excess Li⁺ near the Li metal anode—which can further increase carrier concentration. This dual mechanism (intrinsic diffusion plus extrinsic defect generation) provides a compelling explanation for the high conductivity observed experimentally in SEI formed with fluorophosphate‑rich electrolytes.

Overall, the study demonstrates a powerful workflow that couples generative AI‑driven crystal‑structure prediction with high‑fidelity ML‑based molecular dynamics to assess both thermodynamic stability and ion transport in complex, mixed‑anion inorganic phases. The key findings are: (1) LiPO₂F₂ possesses a low‑energy crystalline polymorph (C2/c) but its amorphous counterpart is thermodynamically accessible; (2) amorphous LiPO₂F₂ exhibits a dramatically reduced Li⁺ migration barrier and high room‑temperature conductivity; (3) disorder‑induced flattening of the Li site‑energy landscape and facile interstitial defect formation together enable fast Li⁺ transport. These results provide a concrete atomistic rationale for the superior performance of batteries employing phosphorus‑ and fluorine‑containing additives and suggest that engineering SEI to favor amorphous lithium fluorophosphates could be a viable strategy for next‑generation high‑rate lithium‑ion cells.