Nature of the Cathode–Electrolyte Interface in Highly Concentrated Electrolytes Used in Graphite Dual-Ion Batteries

Dual-ion batteries (DIBs) generally operate beyond 4.7 V vs Li+/Li0 and rely on the intercalation of both cations and anions in graphite electrodes. Major challenges facing the development of DIBs are linked to electrolyte decomposition at the cathode–electrolyte interface (CEI), graphite exfoliation, and corrosion of Al current collectors. In this work, X-ray photoelectron spectroscopy (XPS) is employed to gain a broad understanding of the nature and dynamics of the CEI built on anion-intercalated graphite cycled both in highly concentrated electrolytes (HCEs) of common lithium salts (LiPF6, LiFSI, and LiTFSI) in carbonate solvents and in a typical ionic liquid. Though Al metal current collectors were adequately stable in all HCEs, the Coulombic efficiency was substantially higher for HCEs based on LiFSI and LiTFSI salts. Specific capacities ranging from 80 to 100 mAh g–1 were achieved with a Coulombic efficiency above 90% over extended cycling, but cells with LiPF6-based electrolytes were characterized by <70% Coulombic efficiency and specific capacities of merely ca. 60 mAh g–1. The poor performance in LiPF6-containing electrolytes is indicative of the continual buildup of decomposition products at the interface due to oxidation, forming a thick interfacial layer rich in LixPFy, POxFy, LixPOyFz, and organic carbonates as evidenced by XPS. In contrast, insights from XPS analyses suggested that anion intercalation and deintercalation processes in the range from 3 to 5.1 V give rise to scant or extremely thin surface layers on graphite electrodes cycled in LiFSI- and LiTFSI-containing HCEs, even allowing for probing anions intercalated in the near-surface bulk. In addition, ex situ Raman, SEM and TEM characterizations revealed the presence of a thick coating on graphite particles cycled in LiPF6-based electrolytes regardless of salt concentration, while hardly any surface film was observed in the case of concentrated LiFSI and LiTFSI electrolytes.


S1.1 Al current collector stability
: CVs recorded at a scan rate of 0.050 mV s -1 , showing the stability of the Al current collector in the electrolytes of interest, within the potential region of 2.8-5.2 V vs Li + /Li 0 . The stability test was conducted in a 3-electrode cell, using metallic Li as both the reference and counter electrodes and Al as the working electrode. The CVs in (a), (c) and (e) show the electrochemical response in the 1 M LiPF6, 1 M LiFSI and 1 M LiTFSI in EC:DEC electrolytes, respectively. Plots (b), (d) and (f) represent the CVs from the corresponding 4 M electrolytes. In (g), the CV of the 1 M LiFSI in Pyr14FSI is shown. From this, it is deduced that an efficient passivation of the current collector is achieved in the LiPF6 containing systems, as well as in the highly concentrated (4 M) LiFSI/ LiTFSI and ionic liquid electrolytes. Note the changes in the scale of the y-axis. S1.2 Impact of electrolyte concentration on the electrochemistry Figure S2: In (a) and (b), voltammograms of the KS6 cathode cycled vs Li (in a 2-electrode configuration) with 1M LiFSI in EC:DEC and 2M LiFSI in EC:DEC as the electrolyte, respectively. In (c) and (d), voltammograms of the cells cycles in the LiTFSI analogue electrolytes. All voltammograms were recorded with a scan rate of 0.050 mV s -1 . Note the changes in the scale of the y-axis. This experiment highlights the benefits of using highly concentrated electrolytes in the context of dual-ion batteries. In both cases, an increase of the salt concentration leads to (1) a lower background current due to parasitic reactions (i.e. solvent decomposition and current collector corrosion) and (2) a decrease in the required overpotential for anion intercalation to take place. Figure 2 Figure S3: Galvanostatic charge-discharge profiles of KS6 cycled in 1 M LiPF6 in EC:DEC electrolyte, at 10 mA g -1 . Even though cycling is possible in this dilute electrolyte, the Coulombic efficiency remains low throughout the course of cycling (~ 60 %) and the discharge capacity drops quickly to values that would not be of practical use in commercial devices (from 46 mAh g -1 in the first cycle down to 31 mAh g -1 after the 45 th cycle). The important point remains however: the extreme decomposition observed for the dilute, 1 M LiTFSI in EC:DEC and 1 M LiFSI in EC:DEC electrolytes, is not visible for 1 M LiPF6 in EC:DEC. In this case, the degradation of the PF6anion and its subsequent reaction with EC result in the build-up of a passivating layer on both the graphite cathode and the current collector. Figure S4: First charge of KS6 cycled in 4 M LiFSI in EC:DEC electrolyte, recorded at a charging current of 10 mA g -1 and with the upper potential cutoff set to 5.1 V vs Li + /Li 0 . The cycling profile demonstrates that the highly concentrated, 4 M LiFSI in EC:DEC electrolyte becomes unstable after 4.95 V vs Li + /Li 0 , resulting in an extended decomposition plateau. This is the reason why 4.95 V vs Li + /Li 0 is used as the potential cutoff for all experiments in the main manuscript, while a potential cutoff of 5.2 V vs Li + /Li 0 is used for the more stable 4 M LiTFSI in EC:DEC composition. The fact that the decomposition starts at a lower potential for 4 M LiFSI in EC:DEC can be partially explained by the inherently less thermally stable nature of the FSI anion. In addition, the observed instability may also originate in the extent to which the solvent molecules are coordinated to the Li + and FSIions. If the FSIanion coordinates less solvent molecules than TFSI -, this electrolyte will (at the same molarity) be less stable. The later proposition appears reasonable, as the 4 M LiTFSI in EC:DEC seemed to be close to its saturation limit during preparation (required heating for complete dissolution to occur), while 4 M LiFSI in EC:DEC could be prepared easily. Figure S5: In a-e) voltage hysteresis-capacity curves of the cells presented in Figure 2 in the main manuscript. In f), the average charge and discharge voltage is plotted as a function of the cycle number for all electrolyte compositions of interest. The larger hysteresis observed between the charge and discharge voltage for the LiPF6-based electrolytes may be potentially linked to the formation of thicker interphase layers on both the negative and positive electrodes. Figure S6: Differential capacity analysis of the galvanostatic data presented in Figure 2 in the main manuscript. Cycles 1, 2, 10, 20 and 45 are shown for all electrolytes. The graph insets provide a zoomed-in view of the differential capacity curves of cycle 1 and 45, to underline the changes in the discharge potential as a result of extended cycling. Figure S7: To guarantee the reliability of the two-electrode measurements (Li vs graphite), additional measurements were performed in a three-electrode setup with a Li-reference (between 3.0-5.1 V vs Li + /Li 0 and at a specific current of 10 mA g -1 ). The initial cycling is shown in a), where the cell voltage (solid black lines), potential of working electrode (dotted lines in color) and potential of counter electrode (solid lines in color) are plotted as a function of time. In 1b), a zoomed-in view of the Li plating/ stripping overpotential is provided. The Li plating and stripping overpotential was determined to 7.3 mV, 13.5 mV, 1.2 mV, 3.9 mV and 22.8 mV in the 4 M LiPF6 EC-DEC, ionic liquid, 4 M LiTFSI EC-DEC, 4 M LiFSI EC-DEC and 1 M LiPF6 EC-DEC electrolytes, respectively. This result validates the reliability the twoelectrode data, as the overpotential of the counter electrode is not significant. Figure S8: In a), initial cycling of graphite vs Li in the 4 M LiFSI in EC:DEC electrolyte (3.0-5.1 V vs Li + /Li 0 at a specific current of 10 mA g -1 ). Partially lithiated Li4Ti5O12 (LTO) was used as the reference electrode in this case. In b), a zoomed-in view of the Li counter electrode plating/stripping overpotential is shown, which proves to be minimal. Figure S9: Graphite composite electrodes cycled in a three-electrode setup with excess LTO as the counter electrode and Li as the reference (3.0-5.1 V vs Li + /Li 0 at a specific current of 10 mA g -1 ). Since LTO has a redox potential of 1.55 vs Li 0 /Li + and is in large excess, this cell setup enables the study of the irreversibility due the graphite cathode. A comparison to Figure S3 and Figure 2 in the main manuscript reveals that most of the irreversible reactions occur on the positive electrode, since the Coulombic efficiency does not increase significantly here.

S1.6 Influence of adding a constant voltage step (CCCV measurements)
Figure S10: Example of CCCV protocol used to study anion trapping in the graphite electrodes. Charge and discharge were performed with a constant current of 10 mA g -1 , in the range of 3.0-5.1 V vs Li + /Li 0 . A constant voltage step was introduced at the end of the discharge, which lasted until the discharge current decayed below 0.1 μA or until a time limit of 24 h was reached.
Figure S11: Comparison between cycling without a constant voltage and with a constant voltage step by the end of discharge. Additional charge could be recovered in all cases after the introduction of the constant voltage step, but was particularly significant in the cases of the 4 M LiFSI (a) and 4 M LiTFSI (c) in EC:DEC electrolytes. This result indicated that a certain amount of the anions remain trapped in the graphite electrode after discharge, but that these can be extracted given enough time.

S1.7 LiPF6 as an additive for graphite dual-ion batteries
Figure S12: Comparison of the gravimetric discharge capacities and Coulombic efficiencies for the 4 M LiFSI in EC:DEC electrolyte without any additive and a 4 M LiFSI in EC:DEC electrolyte with 2 mol % LiPF6 added to it. The influence of the additive is best seen in the increased initial Coulombic efficiency of the cell with 2 mol % LiPF6 which amounts to 42 % (vs 55 % for the cell with no additive). The cell with 2 mol % LiPF6 retains a higher Coulombic efficiency during the first 20 cycles, before the cell with pure 4 M LiFSI in EC:DEC electrolyte catches up. This is an encouraging result, which suggests that LiPF6 may be successfully employed as an additive in highly concentrated LiFSI electrolytes; resulting in systems that combine adequate passivating properties with good anion intercalation kinetics. However, as seen from the plot, the addition of LiPF6 leads to a lower discharge capacity. If the passivation layer becomes too thick/ dense, it may become less permeable to FSI, leading to decreased intercalation. Hence, the amount of additive is an important parameter which needs to be optimized, a result which will be communicated in a future publication.S3.

S3. Post-mortem transmission electron microscopy
In this section, high-resolution transmission electron micrographs are provided for the pristine KS6 graphite particles and for selected cycled cathodes. The graphite cathodes analyzed with TEM are the ones for which a CEI could not be observed by other means (for example SEM).    A CEI layer could not be discerned in neither of the studied cases ( Figure S21-S23), as the distinct lattice fringes, which could only originate from the graphite structure continued until the outer edge of the particles. This picture is consistent with the XPS findings described in the main manuscript.

S4. RAMAN SPECTROSCOPY
In this section, the Raman spectra of a selection of electrolytes are presented, in which the effect of salt concentration on the solvent coordination is demonstrated. The Raman spectra of 1 M LiFSI in Pyr14FSI was not recorded due to its sensitivity to the beam. As regards the LiPF6 based electrolytes, the fluorescent background did not allow for the extraction of useful information. Similarly to the case of the LiTFSI-based electrolytes, a significant intensity increase is observed in the bands belonging to Li-coordinated EC for the higher salt concentrations. Figure S26: Ex-situ Raman spectra of the pristine and cycled KS6 electrodes. Each spectrum represents an average of four measurements, performed on different spots on the sample. The most prominent features include the disorder-mode (D-band), the in-plane, ring breathing mode (G-band), the splitting of the G-band due to anion intercalation (D´-band) and 1 st overtone of the D-band (2D-band). [4][5] These bands are located at 1351 cm -1 , 1580 cm -1 , 1614 cm -1 and 2712 cm -1 , respectively. In general, the D´ band appears to increase in intensity in all cases during charge, as a result of the anion insertion in graphite. The structural changes induced in graphite upon charge seem to be highly reversible, since the D´ component is less pronounced during discharge. However, as the D´ band does not completely disappear, it is reasonable to assume graphite certain degree of anion trapping in graphite. Comparing the Raman spectra of the FSIintercalated graphite (a-b) to those of TFSI-(c) and 6 − -intercalated graphite (d) suggests that the FSI anion is the least perturbing upon entry in the graphite host. Evidence for this is provided from the low intensity of both the D-and D´-band (compared to the G-band) in the FSI datasets. Figure S27: Ex-situ XRD patterns of pristine graphite and of graphite electrodes cycled in the highly concentrated 4 M LiTFSI in EC:DEC (a) and Figure S28: Electrochemical behavior of HOPG (1 st cycle), cycled galvanostatically between 3.0-5.1 V vs Li + /Li 0 at a charging current of 1 mA g -1 . Figure S29: Experiment where the same piece of HOPG (fully charged, stopped at 5.1 V vs Li + /Li 0 ) was analyzed with the basal plane facing the incoming X-ray (top) and the crosssectional facing the X-ray (middle). In the bottom, spectra of the pristine HOPG, with the crosssectional plane towards the incoming X-ray. The HOPG has been oxidized to a larger extent at the edges (cross-sectional plane), with both the adsorbed and intercalated species residing there. This is expected since the plane edges are a lot more reactive, due to the presence of defects and dangling bonds. This finding is also in line with previous XPS studies on the intercalation of Li + in HOPG. [6][7] Figure S30: Al current collector cycled voltammetrically to 5.1 V at a scan rate of 0.050 mV s -1 in the 4 M LiTFSI in EC:DEC electrolyte. The x-offset is applied as for the salt peaks (N 1s, S 2p, F 1s) to match those of the samples cycled with a graphite electrode. Merely one peak is observed for N 1s and S 2p, which indicates that the peak splitting occurring in graphite samples is indeed due to the differences between the adsorbed and intercalated anion species. The appearance of fluorine peaks at lower binding energies (685.8 eV and 684.5 eV) suggests the partial breakdown of the TFSI molecule on the Al CC, with the formed species possibly corresponding to AlF3˔3H2O and AlF6. Figure S31: Al current collector cycled voltammetrically to 5.1 V at a scan rate of 0.050 mV in the 4 M LiFSI in EC:DEC electrolyte. The x-offset is applied as for the salt peaks (N 1s, S 2p, F 1s) to match those of the samples cycled with a graphite electrode. The anion spectra recorded on the Al CC differ to the spectra recorded on graphite for FSI as well, and no clear peak splitting is observed. However, a couple of smaller peaks are observed in the N 1s (at 401.0 eV) and S 2p spectra (at 163.2 eV). In this case, these may be due to the partial breakdown of the electrolyte, as the 4 M LiFSI in EC:DEC only seemed to be stable up to 4.95 V vs Li + /Li 0 . As regards the F 1s spectra, the additional peaks appearing at lower BEs are reminiscent of the ones observed for the 4 M LiTFSI in EC:DEC electrolyte. Hence, these can be ascribed to the same Al complexes, i.e. to AlF3˔3H2O and AlF6.  O 1s core-level spectra from KS6 electrodes cycled in the sulfonimide-based electrolytes for the 1 st , 2 nd and 10 th cycles at the fully charged and fully discharged states. The appearance of the pristine spectrum may be attributed to carbonates (531.5-532) C doublybonded to O (~531.5-532 eV), C singly-bonded to O (~533 eV) and C-OH (~533-534 eV), functionalities which will all be present in the CMC binder. In the OCV sample, the O 1s spectrum is dominated by the signal at 532 eV, which most probably originates from the carbonate functionalities in EC:DEC and the S=O bond in the anions. The cycled samples exhibit a behavior similar to what observed for the N 1s, S 2p and F 1s spectra, meaning that the O 1s peak shows a shoulder that increases in intensity upon charge and disappears upon discharge. Guiding lines have been added to indicate the approximate position of the component for the surface adsorbed and intercalated anions.   core-level spectra from KS6 electrodes cycled in LP40 for the 1 st , 2 nd and 10 th cycles at the fully charged and fully discharged states. A significant portion of the oxygen signal is found at BEs for double-bonded (531.5-532 eV) and single-bonded (~533 eV) organic O. Combined with evidence from the C 1s and survey spectra, this can be attributed to insoluble surface residues from polymerized EC. Figure S37: Survey spectra from KS6 electrodes cycled in 4 M LiFSI for the 1 st , 2 nd and 10 th cycles at the fully charged and fully discharged states. The major photoelectron peaks detected include the F 1s, O 1s, N 1s, C 1s, S 2s and S 2p signals. The C 1s signal remains particularly well-pronounced during the course of cycling, indicating that if there is any layer build-up on the electrode, it is extremely thin. Figure S38: Survey spectra from KS6 electrodes cycled in 1 M LiFSI in Pyr14FSI for the 1 st , 2 nd and 10 th cycles at the fully charged and fully discharged states. All noticeable photoelectron peaks originate from the electrolyte and graphitic substrate. The KS6 graphite peak retains its high intensity during cycling, in such a way confirming that the layer of insoluble components residing at the electrode surface (if any) is not of substantial thickness. Figure S39: Survey spectra from KS6 electrodes cycled in 4 M LiTFSI for the 1 st , 2 nd and 10 th cycles at the fully charged and fully discharged states. The main photoelectron peaks in the surveys encompass the F 1s, O 1s, N 1s, C 1s, S 2s and S 2p signals. As for the rest of the concentrated electrolytes based on fluorosulfonimide salts, the C 1s signal stays clearly visible during cycling. Once again, this signifies that there is no thick layer of decomposition products covering the electrode surface. In addition, no traces of compounds originating from the Alcurrent collector/ glassfiber separator are detected here, which is indicative of the system's good electrochemical stability. Figure S40: Survey spectra from KS6 electrodes cycled in 1 M LiPF6 in EC:DEC for the 1 st , 2 nd and 10 th cycles at the fully charged and fully discharged states. Here, it is evident that a very thick decomposition layer is build-up in the KS6 electrode, as the C 1s peak disappears almost completely upon prolonged cycling. The F 1s and O 1s peaks exhibit the highest intensities, indicating that the CEI is to a large extent comprised by the decomposition product of LiPF6 and the solvent molecules (mainly EC). At the same time, smaller peaks which can be attributed to the decomposition of the glassfiber separator are present here (i.e. Si 2s and Si 2p). Figure S41: Survey spectra from KS6 electrodes cycled in 1 M LiPF6 in EC:DEC for the 1 st , 2 nd and 10 th cycles at the fully charged and fully discharged states. A similar trend as for the 1 M LiPF6 in EC:DEC is observed; meaning that the dominant signals in the survey signify the presence of a thick decomposition layer. Just as for the dilute composition, the side-reaction forming the CEI appears to involve the degradation of LiPF6 and its subsequent reaction with EC, to form a layer rich in LiF and polymerized EC.