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

TL;DR

It is proposed that 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, and a stable crystalline polymorph is identified.

Abstract

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.

Figures (3)

• Figure 1: Computational framework for crystal structure prediction and analysis of LiPO2F2. (a) The structural motif from the LiPO2F2 molecule, where Li atoms interconnect to form an inorganic condensed phase. (b) Schematic illustration of reaction product precipitation, inorganic SEI component distribution, and the proposed Li transport mechanism. (c) The computational workflow, including: (i) crystal structure prediction with the CHGGen generative model, (ii) thermodynamic stability evaluation using DFT and the Materials Project phase diagram, (iii) development of a fine-tuned MLIP for molecular dynamics simulations via active learning, and (iv) large-scale MD simulations with a fine-tuned CHGNet to derive Li diffusivities. (d) Schematic of the CHGGen framework for generating LiPO2F2. The process begins with unconditional generation, followed by the removal of Li atoms to create a refined, symmetrized polyanion framework. The framework is then used in an inpainting step to generate the final crystal structure. (e) Thermodynamic stability of the most stable LiPO2F2 polymorphs generated in each space group (blue bars: decomposition energy ($E_d$) predicted by pretrained CHGNet; orange bars: $E_d$ calculated from r$^2$SCAN-DFT).
• Figure 2: Crystal structure and Li transport properties of LiPO2F2. (a) The predicted ground-state structure of LiPO2F2 ($C2/c$), which has a decomposition energy of $E_d = -0.017$ eV/atom relative to the r$^2$SCAN-DFT Materials Project phase diagram. (b, c) Radial distribution functions (RDFs) for the crystalline (orange) and amorphous (blue) phases at 300 K, showing (b) Li--anion (O/F) and (c) Li--P correlations. (d) Arrhenius plots of Li diffusivity from MD simulations for the crystalline (red squares) and amorphous (green dots) phases. Dashed lines are linear fits used to extract the activation energies ($E_a$).
• Figure 3: Li site energy and interstitial defect formation energy distribution in LiPO2F2. (a) Density of atomic states (DOAS) for Li ions in crystalline (blue) and amorphous (orange) LiPO2F2 at 300 K. The distribution for the amorphous phase at 700 K (green) illustrates thermal broadening of the energy landscape. (b) The calculated formation energy for a Li interstitial defect as a function of voltage (vs. Li/Li$^+$) in crystalline (c-) and amorphous (a-) phases. The solid red and green lines represent the c-Li2CO3 and c-LiPO2F2, respectively. The shaded regions, bounded by lines with triangles, represent the distribution of formation energies for an ensemble of defects in the amorphous phases; individual calculations are shown as dashed lines. The histograms on the right illustrate the distributions at 1 V.