The compositions Li₂Fe_0.9Co_0.1SO, Li₂Fe_0.9Mn_0.1SO and the Li₂FeSO reference material were synthesized by a solid-state reaction between stoichiometric amounts of Li₂O (Alfa Aesar, 99.5%), elemental sulfur (Sigma Aldrich, ≥99.9%) and transition metals Co, Fe, and Mn (all of them from Alfa Aesar, >99.9%). The mixture of reagents (total mass around 0.5 g) was thoroughly grinded in an agate mortar, pressed into a pellet and put into a corundum crucible, which was placed into a quartz tube. All of the described manipulations were done in an Ar-filled glovebox to avoid water adsorption on the surface of initial reagents. The quartz tube was subsequently evacuated to 10^–5 mBar and then filled with argon to 0.5 Bar before melt-sealing it. The volume of the ampule was around 20 cm³. Afterward, the ampule with the pellet was put in an oven, heated up to 750°C with a rate of 50°C/h, kept there for 3 h, and immediately cooled down to room temperature by quenching it in melting ice. In general, the synthesis route was taken from the previous reports on antiperovskites (Lai et al., 2018; Gorbunov et al., 2020); however, some parameters were varied to optimize the quality of the obtained materials.

To check the phase purity of the materials and to calculate the crystallographic unit cell parameters, powder X-ray diffraction studies were done on laboratory X-ray diffractometer STOE Stadi P, with MoK_α1 radiation and Ge (111) monochromator in Debye-Scherrer geometry. All preparations for the measurements were done in an Ar-filled glovebox to avoid decomposition of the materials in air. JANA 2006 program (Petricek et al., 2014) was used in order to perform structural analysis of the obtained X-ray diffractograms using Rietveld refinement (Rietveld, 1969).

Transmission Electron Microscopy (TEM) studies were conducted on a FEI Titan 300–80 TEM with third-order spherical aberration correction. The Li₂Fe_0.9Mn_0.1SO and Li₂Fe_0.9Co_0.1SO powders were loaded by direct contact onto Cu 300 mesh TEM grids (Agar) and holey carbon film on Cu 300 mesh TEM grids (Plano). High-resolution imaging was done at 300 kV.

All electrochemical tests were performed on a multichannel potentiostat VMP3 (BioLogic) at room temperature in two- or three-electrode Swagelok cells. The electrolyte consisted of 1 M lithium bis(trifluoromethanesulfonyl)imide, LiTFSI (99.95%, Sigma-Aldrich), dissolved in a mixture of DME (dymethoxyethane, Sigma Aldrich, 99.5%) and DOL (1,3-dioxolane, Sigma Aldrich, 99.8%) 1:1 by volume. Metallic lithium (Alfa Aesar, 99.8%) served as a counter electrode, and in case of the three-electrode setup, also as a reference electrode. The working electrode represented a mixture of the antiperovskite material, carbon black or carbon nanotubes as a conductive additive, and Polytetrafluoroethylene (PTFE, Sigma Aldrich) binder in 80:10:10% mass ratio, respectively, which was pressed on a steel or aluminum mesh.

Galvanostatic cycling with potential limitation was carried out in the two-electrode set-up in order to estimate the rate capabilities of the materials and their stability during long-term cycling. Various current densities were applied throughout 50 cycles. The value of 1C corresponds to the extraction (insertion) of 1 Li⁺ ion from(into) the Li₂Fe_0.9M_0.1SO formula unit within an hour. Long-term cycling was performed at a 1C current density. A mass loading was around 5 mg/cm².

A low scan rate cyclic voltammetry technique was applied in the two-electrode set-up to preliminarily estimate redox processes during extraction and insertion of Li⁺ ions. The voltage sweeping rate was 0.05 mV/s.

Galvanostatic intermittent titration measurements were done for kinetical characterization of lithium extraction/insertion from/into the Li₂Fe_0.9M_0.1SO structure in a three-electrode set-up. The duration of the current pulse with 0.1C was 10 min, and the relaxation time was 8 h. The experimental parameters for calculations were as following: the surface area of the electrode material with the mass of about 3.28 mg was 1.130 cm², the crystallographic density and the volume of the material were 3.281 g/cm³, and 10^–3 cm³, respectively. The values of lithium diffusion coefficients at different potentials were calculated according to work (Weppner and Huggins, 1977).

In order to characterize changes in the electronic and local structure of the materials during Li-removal and insertion, operando X-ray absorption spectroscopy studies were performed on the P65 beamline of DESY (Hamburg, Germany) (Welter et al., 2018). The experiments were carried out using coin cells with Kapton windows, which were put into an eightfold coin cell holder (Herklotz et al., 2016), connected to a multichannel potentiostat VMP3 (Biologic), similar to laboratory electrochemical experiments. K-edge spectra of Fe, Mn, and Co were recorded every hour while the cells were charged and discharged at 0.1C current density. Appropriate metallic foils were used for energy calibration, while transition metal oxides or salts were used as reference materials for metal oxidation states. The electrolyte used was LP30, consisting of 1M LiPF₆ in the mixture of ethylene carbonate and dimethyl carbonate (1:1 by volume, BASF). Software packet Demeter (Ravel and Newville, 2005) was used to analyze the obtained data. The spectra were recorded in transmission and fluorescence yield mode.

Operando X-ray diffraction experiments were performed on the BL04–MSPD beamline at ALBA (Barcelona, Spain) (Fauth et al., 2013) and on the P02.1 beamline at DESY (Hamburg, Germany) (Dippel et al., 2015). The measurements were carried out using coin cells with quartz windows and the same eightfold holder. Each diffraction pattern was recorded every 10 min, while the cells were charged and discharged at 0.1C current density. The electrolyte was the same as for operando XAS measurements.

The main challenge of the Li₂MSO synthesis in a closed ampule relates to the high volatility of sulfur at high temperatures and, as a consequence, its removal from the reaction zone and changes in the stoichiometry of the solids. The presence of some amount of sulfur in the gas phase during the reaction time was often confirmed after the synthesis by an off-white scurf-like layer on the wall of the ampule and therefore, a slight amount of impurities was likely to form. When the initial argon pressure inside the tube was increased in comparison to the earlier reported synthetic conditions for Li₂Fe₁−_xM_xSO series (Gorbunov et al., 2020), the scurf disappeared, in accordance with the well-known Le Chatelier’s principle (Le Chatelier and Boudouard, 1898). A value of 0.5 bar Ar in the tube before sealing provided a reasonable balancing between the material purity and the stability of the ampule against the internal gas pressure, increasing with the temperature.

The obtained samples represented well-crystallized and homogeneous dark-gray or brownish powders. Typical experimental powder X-ray diffractograms of Li₂Fe_0.9M_0.1SO are shown in Figure 1 together with their cell parameters calculated from Rietveld refinement.

As one would expect from the values of ionic radii for transition metal cations (Shannon, 1976), partial substitution of high-spin HS–Fe²⁺ (0.78 Å) by HS–Co²⁺ (0.745 Å) must result in a slight shrinkage of the crystallographic unit cell, whereas HS–Mn²⁺ (0.83 Å) enhances the cell parameter. The values of cell metrics for Li₂Fe_0.9Mn_0.1SO and pure Li₂FeSO obtained in this work agree well with the results of the previous research related to these materials (Mikhailova et al., 2018; Gorbunov et al., 2020). For the Co-containing phase as expected, the determined cell parameter of 3.9082(5) Å is slightly smaller than that of unsubstituted Li₂FeSO material (3.914 Å).

Transmission electron microscopy-micrographs of the Li₂Fe_0.9Co_0.1SO material are given in the Supplementary Material (Supplementary Figures 1, 2). From these pictures, one can see the polycrystalline nature of the material and homogeneous distribution of the elements.

At first, cyclic voltammetry and rate capability measurements of substituted materials and the Li₂FeSO reference were carried out, see Figures 2A–D. Cyclic voltammograms show that these three materials undergo a similar and complex mechanism of lithium extraction during the first cycle. Three oxidative peaks are observed for all materials (Figures 2A–C), while during reduction, two redox peaks are present down to 1.5 V vs. Li in the case of Li₂FeSO and Li₂Fe_0.9Co_0.1SO, and an additional peak around 1.55 V is seen for Li₂Fe_0.9Mn_0.1SO. The reduction peak between 2.0 and 2.4 V in all materials represents most probably a combination of two peaks, which are not well-resolved. During the second and the third cycles, the fourth peak appears on each of the oxidation curves, and all peaks are becoming less resolved. The form of the reduction curve changes more in the case of Co- and Mn-substituted materials in comparison to Li₂FeSO. Thus, the reduction peak around 1.55 V disappears almost entirely in Li₂Fe_0.9Mn_0.1SO, while the peak around 2.23 V is shifted to 2.10 V for Li₂Fe_0.9Co_0.1SO. As the differences in maxima positions of oxidation and redox peaks for all materials are significantly higher than the 2.3×RTnF≈0.059Vempirical criterium (Bard and Faulkner, 1980), it is very probable that materials suffer from serious kinetic obstacles during the discharge process. This observation is further quantitatively proved in the section with GITT measurements.

The results of rate capability tests demonstrate that insertion of even a small amount of manganese into the crystal lattice influences negatively the performance of the material in the Li-battery, especially at high current densities, whereas partial substitution of iron by cobalt provides a significant improvement of the electrochemical properties.

As Li₂Fe_0.9Co_0.1SO shows higher specific capacity values and Li₂Fe_0.9Mn_0.1SO much lower capacities at different current densities in comparison to the Li₂FeSO reference, only Li₂Fe_0.9Co_0.1SO and Li₂FeSO were further subjected to long-term cycling at 1C current density. The results are presented in Figure 3.

As one can see from the corresponding figures, partial substitution of 10% Fe by 10% Co in the Li₂FeSO antiperovskite structure results in a better electrochemical performance in terms of specific capacity, Coulombic efficiency and capacity retention when cycled at 1C current density.

Although it suffered from a significant capacity loss, the Li₂Fe_0.9Co_0.1SO material delivered more than 80% of the initial specific capacity value at 1C even after 100 cycles. The unsubstituted Li₂FeSO material reached 80% of the initial capacity at 75th cycle already, thus showing the advantage of the Co-containing phase as a battery material. The initial values of Coulombic efficiency were around 100% for both materials and decayed slightly after 100 cycles for Li₂Fe_0.9Co_0.1SO, whereas for Li₂FeSO, after 60th cycle a discharge capacity became higher than a charge one.

The results of galvanostatic intermittent titration experiments for the first charge and discharge processes are presented in Supplementary Figure 4. It was observed that partial substitution of iron by other transition metals in the cubic antiperovskite structure generally results in a slight decay of the average Li-diffusion coefficient during the charge. Moreover, the rapid deterioration of the diffusion process at the end of the charge, corresponding to a significant decrease of the diffusion coefficient, occurs earlier for the substituted materials (2.7 V vs. 2.88 V for Li₂FeSO). However, the effect of partial substitution is opposite for the discharge process: the average Li-diffusion coefficient for substituted phases is one order of magnitude higher than for Li₂FeSO (Mikhailova et al., 2018). No direct correlation between the lattice parameter a and the lithium diffusion coefficient was observed. The iron to cobalt substitution results in a slight shrinkage of the unit cell which is opposite to the effect of iron to manganese substitution. Nevertheless, both additional cations improved the kinetics of the discharge process at least during the first cycle by a comparable value. Note, despite a noticeably higher values for the diffusion coefficient during the discharge process compared to unsubstituted Li₂FeSO, the Li₂Fe_0.9Mn_0.1SO material performs worse in terms of rate capability.

In order to evaluate structural changes in Li₂Fe_0.9Mn_0.1SO and Li₂Fe_0.9Co_0.1SO during Li-removal and insertion, operando XRD studies were carried out, with the results presented in Figures 6, 7, respectively. Corresponding galvanostatic charge-discharge curves are shown in the Supplementary Material (Supplementary Figure 5).

The Li₂Fe_0.9Mn_0.1SO material demonstrates a structural behavior with the formation of a second phase (Figure 4), similar to the one reported earlier for Li₂Fe_0.5Mn_0.5SO (Gorbunov et al., 2020). The relative change of the lattice parameter for both Mn-substituted materials during the completed charge is around 0.8%. Both of them exhibit a two-phase delithiation mechanism with an accumulation of the second isostructural phase with a lower cell parameter during each subsequent cycle. The first difference between these two materials includes the amount of lithium extracted from the structure before the second phase appears. In the case of Li₂Fe_0.5Mn_0.5SO, the amount of remaining structural lithium was around 1.7. In case of Li₂Fe_0.9Mn_0.1SO, the second phase appears earlier, when the amount of lithium in the structure is equal to 1.8. The second difference is that in contrast to Li₂Fe_0.5Mn_0.5SO, the phase with the higher cell parameter does not completely transform to the phase with the low cell parameter during delithiation at the first two cycles, even when a large amount of lithium is extracted from the structure of Li₂Fe_0.9Mn_0.1SO.

The structural behavior of the Li₂Fe_0.9Co_0.1SO material (Figure 5) strongly resembles that of Li₂FeSO. No formation of a second isostructural phase was observed, and the change of the lattice parameter is highly non-linear during the charging process, similar to the reported behavior for Li₂FeSO (Mikhailova et al., 2018). The amount of extracted lithium before this non-linearity appears, is different, though. For Li₂FeSO, the extracted Li-amount corresponding to the first minimum in the lattice parameter a is around 0.4, while for Li₂Fe_0.9Co_0.1SO it is almost equal to 0.2. We should mention here that the Co-substituted phase remains crystalline during the whole charge-discharge cycle, in contrast to Li₂FeSO, thus allowing structural analysis during the complete (de)lithiation process.

As visible from drastically decreased intensities of Bragg reflections combined with their broadening on operando X-ray diffractograms, both Li₂Fe_0.9Mn_0.1SO, and Li₂Fe_0.9Co_0.1SO suffer from a severe degree of structural strains while being charged. However, their crystal symmetry remains constant all the time, and the microstructural effects are regained almost completely after the first cycle.

Analysis of the operando recorded X-ray absorption spectra provided information about redox activities of Fe²⁺ and M²⁺ cations during cycling and allowed to disclose the local environment of the scattering atoms (Newville, 2004) while lithium was reversibly extracted from the lattice.

The results of operando XAS studies of Li₂Fe_0.9Co_0.1SO and Li₂Fe_0.9Mn_0.1SO are presented in Figures 6, 7, respectively. The corresponding galvanostatic charge-discharge curves are available in the Supplementary Material (Supplementary Figure 6). For both materials, two tendencies were observed. Firstly, both Co and Mn are electrochemically active. Secondly, the average oxidation state of iron in materials with Co and Mn, calculated with involving the Fe²⁺ and Fe³⁺ references, changes from +2 to +2.9, and the process is almost reversible (see Supplementary Figure 7). Li₂Fe_0.9Mn_0.1SO, similar to Li₂Fe_0.5Mn_0.5SO with a much higher Mn-substitution grade (Gorbunov et al., 2020), shows a low redox activity of manganese: even when 1.1 Li⁺ per formula unit are extracted from Li₂Fe_0.9Mn_0.1SO, the average oxidation state of Mn does not exceed the value of around +2.2. In contrast, cobalt in Li₂Fe_0.9Co_0.1SO exhibits a noticeably higher redox activity, having nearly the same average value of +2.2 when only 0.65 Li per formula unit is extracted.

Next, we performed an extended X-ray absorption fine structure analysis, which provided local structural information about the first and second coordination spheres of Fe ions in the Li₂Fe_0.9Co_0.1SO and Li₂Fe_0.9Mn_0.1SO materials. An example of the EXAFS fit is given in the Supplementary Material (Supplementary Figure 8). Note, since the cation site in the antiperovskite structure is randomly occupied by lithium, iron and cobalt or manganese, diffraction methods provide information about the average local surrounding for this cation site. In opposite, EXAFS analysis of the Fe K-edge spectra allows us to separately evaluate the local structure of Fe cations.

Comparison of interatomic distances calculated from EXAFS and synchrotron XRD shows that for both substituted antiperovskites, the XRD analysis provides significantly longer metal-oxygen distances and shorter metal-metal distances than the EXAFS analysis, see Table 1. This difference points to the distortion of the local structural surrounding for Fe-cations comparing to the average structure. For example, the Fe–O bond length of 1.85 Å in Li₂Fe_0.9Co_0.1SO is well below the average (Fe, Li, Co)–O distance of 1.955 Å. Moreover, a smaller difference between individual Fe–O and average (Fe, Li, metal)–O distances in Li₂Fe_0.9Mn_0.1SO comparing to Li₂Fe_0.9Co_0.1SO argues for a higher stabilizing effect of Mn²⁺ cations in comparison to Co²⁺ in the initial composition before delithiation.

Next, due to the symmetry reason, metal–sulfur and metal–metal distances should be equal to each other in the averaged crystal structure, giving 2.765 Å in Li₂Fe_0.9Co_0.1SO and 2.773 Å in Li₂Fe_0.9Mn_0.1SO, according to the XRD measurements. However, the EXAFS calculations show a significant discrepancy between Fe–metal and Fe–S distances. For example, the Fe–metal and Fe–S distances in Li₂Fe_0.9Co_0.1SO were calculated as 2.86 and 2.77 Å, respectively. The Li₂Fe_0.9Mn_0.1SO material provides a similar issue having the Fe-metal distance of 2.91 Å and the Fe–S distance of 2.75 Å. Therefore, the antiperovskite structure is probably stabilized by Fe–S bonds, as their lengths are very close to the average value from XRD, while Fe–O bonds bring instability through a substantial deviation from the average value.

Note, the longer Fe–metal distance in comparison to the Fe–S one was also observed for the Li₂FeSO material (Mikhailova et al., 2018). However, in Li₂FeSO, the difference is much smaller, never exceeding the value of around 0.01 Å, and only appearing when around 0.2 Li is extracted.

During delithiation, all interatomic distances decrease for Li₂Fe_0.9Co_0.1SO and Li₂Fe_0.9Mn_0.1SO, thus confirming metal oxidations (Figures 6B, 7B). Interestingly, for both materials the Fe–S bond length decreases stronger than the Fe–O length, thus reflecting higher compressibility of S^2– anion in comparison to O^2–.