The role of phosphorus in the solid electrolyte interphase of argyrodite solid electrolytes

The transition away from fossil fuels is driving an ever-growing demand for batteries¹. One key area of demand is transportation, where increased energy density is crucial for improving vehicle range. Lithium metal negative electrode batteries hold significant promise in this regard, with the highest energy density achieved in a zero-lithium-excess (anodeless) configuration². However, non-uniform lithium plating during charging compromises cell lifespan and safety when used with a liquid electrolyte³. In theory, this issue can be mitigated by applying a homogeneous stack pressure with a solid electrolyte (SE) with sufficient mechanical strength⁴.

Several Li-ion conducting SEs have been discovered⁵. Oxides exhibit the lowest theoretical reduction potentials and, therefore, the greatest stability against Li metal negative electrodes⁶. However, their ionic conductivities have so far been too low to be used as the positive electrode electrolyte in practical applications⁵, whilst the oxide's poor solid-solid contacts between the Li metal and a rigid oxide solid electrolyte hinder its use as a separator⁷. In contrast, softer sulphides have demonstrated ionic conductivities exceeding those of liquid electrolytes but suffer from limited electrochemical stability windows and poor compatibility with metallic lithium⁵.

Among sulphide solid electrolytes, argyrodites–composed of or closely related to Li₆PS₅Cl–are considered among the most promising due to their enhanced stability compared to other sulphides⁸. Despite the theoretical reduction potential of Li₆PS₅Cl being 1.71 V vs. Li₊/Li⁶, the solid electrolyte interphase (SEI) formed between Li₆PS₅Cl and lithium was considered passivating, with a thickness on the order of nanometres⁹. Work by the Janek group challenged this assumption, reporting continuous SEI growth, which they described using the Wagner model for diffusion-controlled solid-state reactions^10–12. Yet the underlying diffusion mechanism was not understood. They also estimated the SEI thickness to be on the order of hundreds of nanometres using Time-of-Flight Secondary Ion Mass Spectrometry (ToF-SIMS)¹³, consistent with measurements from our group obtained via in situ X-ray photoelectron spectroscopy (XPS)¹⁴.

The practical implications of this finding are critical for the viability of solid-state batteries, as continuous SEI growth leads to increased cell impedance¹⁰, electrolyte and lithium consumption, and exacerbation of the chemo-mechanical degradation of the solid electrolyte, ultimately limiting battery life¹⁵. It is therefore imperative to gain a deeper understanding of the SEI and the mechanism that underpins its growth.

This work investigates the mechanism behind the continual growth of the Li-argyrodite SEI, linking its behaviour to SEI composition. The conductivity of the SEI is assessed through a combination of coulometric titration time analysis (CTTA) and three-electrode potentiostatic electrochemical impedance spectroscopy (PEIS). The evolution of the SEI's chemical composition is characterised using virtual electrode plating X-ray photoelectron spectroscopy (VEP-XPS), while non-destructive depth profiling of the SEI is performed via soft and hard X-ray photoelectron spectroscopy (SOXPES and HAXPES). Finally, the electrochemical stability of the SEI is examined using cyclic voltammetry (CV).

Impedance measurements (sinus amplitude: 10 mV, frequency range: 400 kHz to 10 Hz) were taken once all the plated lithium had been consumed-i.e., when the 50 mV vs. Li⁺/Li cutoff was reached-but before the next lithium plating step (Fig. 1a). The SEI impedance was fitted to a modified Randles circuit, where the semi-infinite Warburg impedance was replaced with a finite diffusion model featuring a reflective boundary (Fig. 1c, inset). This approach is equivalent to previous SEI fitting models^14,17,18, except that the finite diffusion transmission line used here better represents the multi-component nature of the SEI¹⁹. The M_a element is used to approximate the transmission line element and calculate the ionic resistance of the SEI (R_SEI) using Equation S10 shown in Supplementary Note 3. The reflective boundary arises from the finite thickness of the SEI and the fact that the electrode remains above the Li plating potential. In total, 32 plating steps of 1.56 μA h cm⁻² each were performed. Figure 1d shows a progressive increase in resistance from finite diffusion (R_SEI) with increasing charge passed (see Supplementary Fig. S3 for the trend over time). However, this trend does not follow a linear relationship, as would be expected if the SEI grew uniformly in terms of composition and morphology. Instead, the resistance curve deviates to lower values. One possible explanation for this behaviour is the emergence of highly, yet not fully, lithiated Li_xP (where x > 1.5). This is reported to have a higher lithium diffusivity compared to Li₃P (Supplementary Fig. S4)²⁰. If the initial 20 μA h cm⁻² of R_SEI is fitted linearly through the origin-assuming negligible electronic conductivity and a fully dense SEI composed solely of Li₃P, Li₂S, and LiCl (where 1 μA h cm⁻² ~ 9 nm)¹¹,-an ionic conductivity of 204 (±8) nS cm⁻¹ is obtained. This value is directly comparable to the recently reported ionic conductivity of a synthetic SEI formed from Li₆PS₅Cl (134 nS cm⁻¹)¹².

Over this period, the OCV was observed to increase, notably mirroring the delithiation profile of red phosphorus (Supplementary Fig. S5)²⁰. Simultaneously, R_SEI was seen to increase throughout the OCV period (Fig. 2b), indicating either SEI growth, a compositional change, or a combination of both.

To further investigate the cause of the change in OCV, VEP-XPS was performed¹⁴. Lithium (~ 0.01 mA h cm⁻²) was plated through a 5 mm Li₆PS₅Cl pellet for 1 hour using an electron beam current of 2.5 μA (~ 0.01 mA cm²), as previously reported¹⁴.

During lithiation, the S 2p Li₂S peak grows in relative intensity while the Li₆PS₅Cl peak reduces in intensity, indicative of electrolyte reduction. The proportion of Li₂S increases rapidly from 4 at.% S prior to lithiation to 65 at.% S after 6.7 μAh, reaching 87 at.% S after 51.7 μAh (Fig. 4b). At the same time, the point of maximum intensity in the P 2p region shift to lower binding energy and peak fitting after 51.7 μAh reveals three additional species at 127.7, 126.9 and 125.8 eV, attributed to Li_xP, Li_yP and Li₃P, respectively, where 0 < x < y < 3. The evolution of these phosphorus-containing species in Fig. 4c reveals that lithiation progresses through P⁰ → Li_xP → Li_yP → Li₃P, and the proportions of the most lithiated phases increase with increasing capacity. Interestingly, significant electrolyte reduction was observed in the S 2p spectra before Li_yP and Li₃P were detected, suggesting that even the least lithiated Li_xP species can reduce Li₆PS₅Cl.

After lithiation, the sample was left to rest at OCV within the XPS analysis chamber. Figure 4a and b reveal negligible changes in the sulphur-containing species during this period, suggesting Li₂S is stable. On the other hand, during rest, the maximum intensity in the P 2p spectra shifts back to higher binding energy, and the breakdown of phosphorus-containing species in Fig. 4c reveals that spontaneous delithiation occurs according to Li₃P → Li_yP → Li_xP → P⁰. The final oxidation of Li_xP to P⁰ must be accompanied by an additional reduction reaction, suggesting this delithiation process will lead to further electrolyte decomposition.

These observed trends in the P 2p spectra are mirrored in the Li 1s spectra in Supplementary Fig. 8, while no changes are observed in the Cl 2p spectra. In the O 1s spectra, it is evident that a Li₂O peak at 528.4 eV emerges during lithiation and a peak at 531 eV grows during rest, indicative of the accumulation of surface oxygen species. While the P − O peak in Fig. 4a also grows in intensity during rest, it still remains only a minor phosphorus-containing species (Fig. 4c), so side reactions with gases inside the XPS chamber are not expected to have a significant impact on the observed phase evolution.

To delve deeper into the composition of the SEI, soft and hard X-ray photoelectron spectroscopy (SOXPES and HAXPES) measurements were performed. Photoelectrons emitted during SOXPES had a kinetic energy of 315 eV, and HAXPES was performed with incident photon energies of 2.2 keV and 6.6 keV, resulting in electron inelastic mean free paths (IMFPs) of ~1, ~5 and ~14 nm, respectively (Supplementary Fig. S9). SOXPES and HAXPES were first performed on a pristine Li₆PS₅Cl pellet (Supplementary Fig. S10 and Supplementary Table 9), revealing only minor Li₂S, Li₂SO₄, Li₂CO₃ and LiOH impurities.

During the first reductive sweep, two clear peaks can be observed (labelled a and b in Fig. 6a). Peak a can be ascribed to the reduction of argyrodite. The onset potential of 1.13 V vs. Li⁺/Li is below the computationally predicted value of 1.71 V vs. Li⁺/Li⁶, yet significantly higher than previously thought²³. As reduction initially occurs above the theoretical full reduction potential of phosphorus to Li₃P (0.87 V vs. Li⁺/Li), this peak can be assigned to the reduction of Li₆PS₅Cl to form Li₂S, LiCl and partially lithiated phosphorus species, which can be summarised as Li_xP, where x < 3 (Equation (1)). Peak b can be attributed to the lithiation of Li_xP to form the fully reduced Li₃P (Equation (2)). Indeed, fitting the areas of peaks a and b yields a capacity ratio of ~ 3:1 (Supplementary Fig. S14), which would be consistent with a conversion of LiP to Li₃P (Equations 1 and 2). The total capacity of the reductive sweep is 11.1 mC cm⁻², which would correspond to an SEI thickness of 28 nm (if fully dense and consisting of only Li₂S, LiCl and Li₃P).1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{\rm{Li}}}}_{6}{{{\rm{PS}}}}_{5}{{\rm{Cl}}}+(5+x){{{\rm{Li}}}}^{+}+(5+x){{{\rm{e}}}}^{-}\to {{{\rm{Li}}}}_{x}{{\rm{P}}}+5{{{\rm{Li}}}}_{2}{{\rm{S}}}+{{\rm{LiCl}}}$$\end{document}Li6PS5Cl+(5+x)Li++(5+x)e−→LixP+5Li2S+LiCl2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{\rm{Li}}}}_{x}{{\rm{P}}}+(3-x){{{\rm{Li}}}}^{+}+(3-x){{{\rm{e}}}}^{-}\to {{{\rm{Li}}}}_{3}{{\rm{P}}}$$\end{document}LixP+(3−x)Li++(3−x)e−→Li3P3\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{\rm{Li}}}}_{3}{{\rm{P}}}\to {{{\rm{Li}}}}_{x}{{\rm{P}}}+(3-x){{{\rm{Li}}}}^{+}+(3-x){{{\rm{e}}}}^{-}$$\end{document}Li3P→LixP+(3−x)Li++(3−x)e−

During the oxidative sweep, a peak can be seen (labelled c) at 0.87 V vs. Li⁺/Li, in line with the reoxidation of Li₃P to Li_xP (Equation (3))^6,20,24. Interestingly, this occurs at a potential that is below the observed initial Li₆PS₅Cl reduction potential. On subsequent cycles (Fig. 6b) peak a is no longer clearly visible suggesting that most of the SEI growth occurs on the first reductive sweep. Nonetheless, an increase in the peak current of peak c with cycling is indicative of a continuous growth of the SEI.

The experimental evidence suggests the presence of a gradient of lithiated phosphorus in the SEI. This is in agreement with a recent study by Ren et al. which computationally predicted that the SEI of Li₃PS₄ contains at least two phosphorus regions²⁵. One region with a Li to P coordination number of 11 near the Li electrode, corresponding to Li₃P, and another region with a Li to P coordination number of 6 near the solid electrolyte, corresponding to LiP. However, previous HAXPES research has failed to see this gradient due to an inability to observe phosphorus in the SEI due to a Ni current collector and lithium oxide attenuating the beam²⁶, while previous VEP-XPS measurements have failed to observe the instability of lithiated phosphorus with the solid electrolyte due to either a limited observation time or a different SE being studied^14,27.

These reactions require electrons from the current collector and ions from the solid electrolyte side to proceed. As both are readily available, the initial growth of the SEI is likely kinetically limited. However, as the SEI thickness increases, mass transport becomes the limiting factor. This is consistent with the CV data in Fig. 6a, where peak a corresponds to the kinetically driven decomposition of Li₆PS₅Cl to form Li_xP species, followed by a mass transport-limited regime as the SEI thickens. Determining whether electron or ion transport is the primary limiting factor would require further investigation.

At 0.87 V vs. Li⁺/Li, full lithiation occurs, yielding Li₃P. This is again in agreement with what is observed in Fig. 6a, where peak b corresponds to the lithiation of Li_xP in the SEI to form Li₃P. Li₃P is a mixed electron-ion conductor, and it is therefore likely that at this potential, the SEI is mostly composed of fully reduced Li₃P. This process is reversible, as shown by peak c in Fig. 6a corresponding to the delithation of Li₃P. At this stage, all SEI components-Li₃P, LiCl, and Li₂S-are thermodynamically stable down to the lithium plating potential. Consequently, at 0 V vs. Li⁺/Li, metallic lithium begins to plate onto the current collector.

After charging and while at open-circuit potential, chemically driven decomposition occurs (Fig. 7b). On the solid electrolyte front, Li₃P reacts with fresh argyrodite, forming new SEI while undergoing delithiation. This process establishes a lithium chemical potential gradient within the SEI, driving coupled Li₊/e⁻ diffusion from the lithium front through the SEI to the solid electrolyte front. This would explain the experimentally observed lithium gradient (Li → Li₃P → Li_xP → P) and the diffusion-limited growth of the SEI (see Equations S1-6 for stoichiometric reactions).

Whilst the Janek group has interpreted the linear dependence of SEI growth as a function of t^0.5 they observed using a Wagner model for diffusion-controlled solid-state reactions^9,11, our results clearly show a non-linear dependence on t^0.5 during the first 20 h (Fig. 1b and Supplementary Fig. S15). We have therefore explored the Deal-Grove model, originally developed to describe the thermal oxidation of silicon, and previously applied to the chemical growth of the SEI in the liquid state, to interpret our data^28,29. This model includes an initial surface reaction-controlled growth, which follows a linear dependence on time, and a diffusion-limited parabolic growth regime, which dominates at longer times. However, this model does not fully capture the growth mechanism either, as it not only provides a poor fit to the experimental data but also fails to describe the initial SEI growth, which is not linear with time. Nonetheless, fitting the SEI growth at long times predicts that the kinetically limited stage of SEI growth is restricted to < 3 μA cm⁻² (see Supplementary Note 2), which is in good agreement with the charge under peak a in Fig. 6a. We therefore propose that SEI growth is indeed diffusion-limited, but that the diffusivity is not constant. Instead, it is a function of SEI composition, which-as described previously-varies over both time and thickness. This variation in diffusivity could be due to the variation of diffusivity of Li in P with lithiation state²⁰, the change of the phosphorus volume fraction in the SEI through lithiation states (Supplementary Table S1, 2), or an evolving porosity. Further investigations are required to validate this mechanism.

Two important points should be noted. First, this process relies on the presence of a phosphorus percolation pathway throughout the SEI, highlighting the critical role of SEI nanostructure. Indeed, changing the chemical composition of argyrodite without significantly changing the volume fraction of phosphorus in the SEI (Supplementary Table S1–4) showed minimal changes to the kinetics of SEI growth³⁰. While the ability to lithiate phosphorus in the SEI was found to be key to the difference in growth kinetics between Li₁₀GeP₂S₁₂ (LGPS) and Li_1.5Al_0.5Ge_1.5(PO₄)₃ (LAGP)³¹. Meanwhile, the graded structure of the SEI, which ensures that the most reduced phosphorus species are not in physical contact with the solid electrolyte, has been linked to the superior stability of the LiPON SEI³². Second, if a continuous percolation pathway is present, SEI growth will persist as long as metallic lithium is available. Once all the plated lithium is consumed, the remaining Li₃P will begin to delithiate, ultimately leaving an SEI composed solely of LiCl, Li₂S, and P.

Although this work has only looked at the solid electrolyte Li₆PS₅Cl, we believe that this continuous mechanism of SEI formation would be present in any solid electrolyte that can be reduced to contain partially lithiated phosphorus, or potentially any element that can alloy with lithium. Indeed, a loss of Li₃P in SEI at rest has been observed over time with Li₇P₃S₁₁⁹. While this observation was speculated to be due to reactions with trace amounts of oxygen or water in the UHV XPS chamber, our results indicate that it may be exhibiting SEI evolution similar to that of the SEI of Li₆PS₅Cl. More recently, El Kazzi's group has shown multiple degrees of lithiation of phosphorus in the SEI of Li₃PS₄³³, indicating that phosphorus may play the same role in its SEI.

The practical implications of this are quite impactful to the implementation of solid-state batteries, as the SEI growth leads to impedance growth as well as lithium and solid electrolyte consumption. Work by the Janek group has predicted the need to keep SEI resistance below 10 Ω cm^212,34, highlighting the importance of finding a mechanism to halt diffusion-controlled SEI growth. By understanding the diffusion mechanism behind this growth, potential routes to stop it can now be explored. We do not expect metallic interlayers to help, but electronically insulating interlayers could, by introducing a steep decrease in the electrochemical potential of the electrons from the current collector interface to the SE interface, so at the SE interface, the electrons are below the electrolyte decomposition potential. Supplementary Fig. S16 reveals that a Li₂O interphase slows the rate of SEI growth by over a factor of 5, compared to a traditional reduced Li₆PS₅Cl interphase. It shouldn't therefore come as a surprise that the impedance using lithium metal foil negative electrodes seems to grow less rapidly than in the Li-less configuration¹⁷, as lithium metal foils have a native passivation layer. Li-less approaches might be more challenging to implement.

In conclusion, SOXPES and HAXPES studies revealed a phosphorus lithiation gradient in the SEI, and VEP-XPS confirmed that Li₃P in the SEI is not stable and will undergo spontaneous delithiation to Li_xP, whilst voltammetry revealed that Li₆PS₅Cl can be reduced above the oxidation potential of Li₃P. These results show that without a phosphorus-free passivating interlayer between the Li negative electrode and argyrodite, the SEI will continually grow due to the reaction between lithiated phosphorus and Li₆PS₅Cl and subsequent diffusion of lithium through the SEI. The growing SEI simultaneously reduces the coulombic efficiency and energy density of solid-state cells, whilst increasing the overall cell impedance. Assuming a fully dense SEI, the conductivity is seen to grow non-linearly, initially at a rate of ~204 nS cm⁻¹ before increasing.

Supplementary Information Transparent Peer Review file

Source Data