Towards Practical Sulfolane Based Electrolytes: Choice of Li salt for Graphite Electrode Operation

: Sulfolane (tetramethylene sulfone, SL) is known for leading to Li-ion electrolytes with high anodic stability. However, the operation of graphite electrodes in alternative electrolytes is usually challenging, especially when ethylene carbonate (EC) is not used as co-solvent. Thus, we study here the influence of the lithium salt on the physico-chemical and electrochemical properties of EC-free SL-based electrolytes and on the performance of graphite electrodes based on carboxymethyl cellulose (CMC). SL mixed with dimethyl carbonate (DMC) leads to electrolytes as conductive as state-of-the-art alkyl carbonate-based electrolytes with wide electrochemical stability windows. The compatibility with graphite electrodes depends on the Li salt used and, even though cycling is possible with most salts, lithium difluoro-oxalato borate (LiDFOB) is especially interesting for graphite operation. LiDFOB electrolytes are conductive at room temperature ( ca. 6 mS cm -1 ) with an anodic stability slightly below 5 V vs. Li/Li + on particulate carbon black electrodes. In addition, it allows cycling graphite electrodes with steady capacity and high coulombic efficiency without any additive. The testing of graphite electrodes in half-cells is, however, problematic with SL:DMC mixtures and, by switching the Li metal counter electrode for LiFePO 4 , the graphite electrode achieves better practical performance in terms of rate capability.


Introduction
Current lithium-ion batteries (LIBs) electrolytes are mostly made of lithium hexafluorophosphate (LiPF6) dissolved in mixtures of ethylene carbonate (EC) and linear alkyl carbonates such as dimethyl carbonate (DMC), diethyl carbonate (DEC) or ethylmethyl carbonate (EMC) [1][2][3]. The use of EC was once dictated by the use of graphite electrodes, for which EC possesses excellent Solid Electrolyte Interphase (SEI) [4,5] forming ability. More recently, SEI additives, such as vinylene carbonate (VC) or fluoroethylene carbonate (FEC) have allowed the operation of graphite anodes even when mixed with electrolytes with poor SEI forming ability such as propylene carbonate (PC) [6], alklynitriles [7], linear sulfones [8] or TFSI-based ionic liquids [9]. However, for best graphite performance, mixtures of EC and linear alkyl carbonates are still mainly used, in combination with additives.
In the last years, many attempts have been made to increase further the energy density of LIBs and cathode materials operating above 4.5 V versus Li/Li + have been proposed in the late 90's, such as LiNi0.5Mn1.5O4 [10,11], LiCoPO4 [12,13] or, more recently, LiCoPO4F [14,15]. However, the so-called '5 V' LIB is still not a reality, due to the failure of EC/linear alkyl carbonate mixtures to operate above 4.3 V.
Indeed, if the electronic structure (HOMO) of each single electrolyte components plays a role in the anodic stability of electrolyte mixtures [16], the presence of Li salt [17] often leads to inferior anodic stabilities than those derived from HOMO calculation for the solvent alone. In particular, it explains why EC, despite having among the lowest calculated HOMO energy of all alkyl carbonate solvents, leads to electrolytes with lower practical anodic stabilities as compared to it intrinsic stability. In fact, Borodin et al. noted the detrimental effects of BF4, PF6 and B(CN)4 on alkyl carbonate solvents, while, on the other hand, they have less influence of sulfone-based solvents such as sulfolane (SL) (Figure 1a), which explains their better performance at high voltage [18,19]. According to these studies, however, DMC anodic stability was found to be also significantly lowered by the presence of Li salts, whereas mixtures of SL and other solvents, such as EMC [20] or ethyl acetate [21] are reported having similar anodic stability as SL alone. The reason would be linked to the double layer structure where the electro-oxidation occurs [22], which also explains the stronger decomposition of EC vs. DMC in EC/DMC mixtures [23].
SL is a by-product of the oil industry, thus cheap and produced by tons [24]. Its high potential of oxidation, as well as those of linear sulfones [25], above 5 V vs. Li/Li + has recently attracted interest as a high voltage electrolyte solvent [26]. Moreover, concerning the practical stability in reduction (linked to the ability to form an effective SEI onto graphite and Li metal electrodes), SL was used as early as 1985 in Li metal studies [27] and a first report of a 'high EMF' electrochemical cell based on Li metal electrodeposited onto graphite from SL electrolytes was issued in 1971 [28]. More recently, SL has shown promising SEI forming ability on graphite at 90 °C [29], which makes it a good candidate for Li-ion application, while linear alkyl sulfones necessitate SEI forming additives for graphite operation [8].
Despite these interesting properties and studies focusing on cathodes performance [30,31], only few reports deal with carbon-based electrodes [32] and even less [33] with EC-free electrolytes able to operate a full Li-ion cell including graphite. Recently, Dahn's group reported on full Li-ion using VC and other additives in SL-based electrolytes [34] and we reported that, by use of carboxymethyl cellulose (CMC)/styrene-butadiene rubber (SBR)-based graphite anodes, efficient cycling of graphite is possible in a 1 M LiPF6 SL:DMC (1:1, wt.) electrolyte without any SEI forming additives [35]. As it is known that the Li salt can have a strong influence on the SEI formation on graphite [36][37][38], we examine here the effect of various lithium salts, (shown in Figure 1a), on the physico-chemical properties of SL:DMC-based electrolytes and assess their suitability for graphite electrode operation.  Swagelok™ cells placed in a temperature-controlled chamber at 20 °C. Conductivity was measured using a BioLogic MCS10, impedance-based conductimeter, from 60 °C to -40 °C with 30 min stabilization every 5 °C, followed by a temperature ramp from -40 °C to 60 °C at the same rate, using hermetical cells sealed in a Mbraun glove box under argon atmosphere with O2 and H2O content below 1 ppm.

Results and discussion
Six different lithium salts were used to prepare 1 M electrolytes in SL:DMC (1:1, wt.). The salts and solvents and their acronyms are shown in Figure 1a. Their practical temperature ranges (liquid range) were evaluated by DSC. Viscosity, density, and conductivity were measured as well to better understand the respective influence of dissociation and viscosity on the conductivity behavior. The size and molecular weight of the anions are also reported in Table 1, as these parameters likely influence the electrolyte properties.

Density, viscosity and conductivity
The variation of densities of the SL:DMC (1:1, wt.) electrolytes are shown in Figure 2a. Overall, the densities are below that of LP30 and increase with the size of the anion, from LiBF4 to LiPF6 and from LiFSI to LiTFSI. LiTDI leads to the lowest measured density despite having, a size similar to that of TFSI and a similar weight as FSI. DFOB also induces a rather low density. TDI and DFOB are rigid anions as both incorporate a planar moiety, whereas FSI [39] and TFSI [40] have high conformational flexibility. It probably results, in addition to their own 'intrinsic' lower density, in a denser packing around solvated Li + . In fact, TFSI, in acetonitrile, tends to form mostly solvent separated ion pairs (SSIP) and contact ion pairs (CIP) at mole fractions below 0.1 (which is the case here at 1 M). In SSIPs, Li + is coordinated by four solvent molecules and the anion is uncoordinated, for CIPs, by three solvent molecules and one anion that does not interact with other Li + [41]. Moreover, LiFSI was found to be even more dissocated [42]. In fact, molecular dynamics simulation found that the probability of forming a solvate incorporating n ions decreases exponentially with n [43]. TDI, on the other hand, interacts by both its -CN groups and the nitrogen atoms on the aromatic ring. Thus, it easily forms polymeric chains between several Li + centers bridged by TDI anions.
Such structures are found in various LiTDI/solvent crystalline structures [44] and it was found that, in ether mixtures including LiTDI, larger polysulfide clusters are formed as compared with equivalent LiTFSI electrolytes [45]. DFOB, on the other hand, mostly interacts with Li + via its carbonyl groups, and only to a lesser extent, via its fluorine atom in the liquid phase and is coordinated similarly to LiTFSI [46].
The variations with temperature are linear and can be fitted with equation 1 and the parameters ρ0 and a are reported in Table 1.
(1) If anion size explains the differences for the salts that lead to smaller solvates structures (i.e. LiFSI, LiTFSI and LiDFOB), the LiTDI electrolyte exhibits a relatively high viscosity which contrasts with a previous report in EC:DMC (1:1) [48]. The high viscosity of the LiTDI electrolyte probably proceeds from the same reasons as the low density of the electrolytes, with a reduced mobility of species due to the presence of larger aggregates. LiBF4 which is usually less dissociated [2], leads to a more viscous electrolyte than the larger LiDFOB and LiFSI salts.

Conductivity
The °C) [50], despite a higher viscosity, showing that SL significantly helps dissociation. LiTFSI and LiDFOB also allow conductivities above 5 mS cm -1 at 20 °C. LiBF4 leads to the lowest conductivity values, despite its relatively low induced viscosity. It is in accordance with the low ionic dissociation generally obtained with this salt [2]. On the contrary, the LiTDI electrolyte is more conductive than that incorporating LiBF4 despite its much higher viscosity, which confirms the better dissociation of LiTDI vs LiBF4.

VTF fitting
The viscosity and conductivity curves have been fitted with the VTF equations (2) and (3) to represent the temperature dependency of these electrolytes and the corresponding fitting parameters ( , , B, B' T0, and T0') are reported in Table 1.

Table 1. Van der Waals (VdW) volumes of anions, experimental Tg (taken at the inflection point on the last
heating ramp at 5°C min -1 ) and fitting parameters for the density, conductivity and viscosity.  The values of T0 and T'0, are significantly lower than the experimental glass transition temperatures Tg, as it is usually observed [54]. T0, T'0 and Tg generally follow the same trend as viscosity. For LiTDI, T0 and T'0 are rather low (i.e. in the same range as LiTFSI), but the experimental Tg is more in line with its high viscosity.
The Walden plots of the electrolytes are shown in Figure 2d. It is usually considered that the position of the curves defines the 'ionicity' of solutions lying in the area below the 'ideal KCl line'. It becomes higher as they are closer to the line [55,56]. Therefore, the 'ionicities' of the electrolytes are all lower than that of the LP30 electrolyte and follows the same trend as the conductivity values.   [36]. For the LiDFOB electrolyte, the peak is much larger and starts around 1.8 V vs. Li/Li + , which is also similar to what is seen in carbonate-based electrolytes [36]. In this case, the peak has a symmetrical triangular shape that is characteristic of a reduction reaction controlled by the saturation of surface sites. This first reduction process ends at ca. 1.2 V and do not overlap much with other reduction processes (such as Li + insertion into carbon). The LiTDI electrolyte exhibits a small peak starting at ca. 1.4 V followed by a large symmetrical peak starting at ca. 1.0 V. In this case, however, the peak shape is distorted with another reduction process probably starting at ca. 0.5 V. For this electrolyte, the lithium deposition is significantly shifted to a lower potential, as compared to the other electrolytes, hinting to a highly resistive SEI for Li + transport. The LiFSI electrolyte also exhibits a large peak, starting at a similar voltage as the LiTDI electrolyte and the LiTFSI electrolyte reduction peak starts at ca.

V. This contrasts to what was reported in EC-based electrolytes (with binder-free graphite electrodes),
where no obvious differences were observed with LiPF6 and LiBF4 electrolytes [36]. All electrolytes but that with LiTDI allows lithium deposition at ca. 0 V. Figure 4 shows the rate performance of graphite electrodes (7.5 mg cm -2 ) in half-cell in the SL:DMC electrolytes. With the LiTDI electrolyte (Figure 4b), a rather long reduction plateau is observed at ca. 0.8 V and the ohmic drop at C/10 is very high, even in the first cycle, most likely due to a thick and resistive SEI formed as reported in carbonate mixtures without additives [59] and the capacity drops dramatically with cycling. All the other salts appear to allow graphite electrode operation, although with moderate rate capability. As shown in Figure 4c, for the LiFSI electrolyte, the ohmic drop is not as dramatic as for LiTDI, but the rate performance is very low despite the fact that the electrolyte is the most conductive. The efficiencies are low from the beginning and, as for LiTDI, the capacity decay at C/10 is marked and no capacity can be cycled after the rate-test with this electrolyte. If we compare with LiBF4 (Figure 4e), the less conductive electrolyte, we can see that the LiBF4 electrolyte performs far better with, in particular, well defined insertion plateaus at C/5 and stable capacity, which shows the poor properties of the SEI layers formed in the LiFSI electrolyte (be them on Li metal or graphite), whereas this salt has shown good performance vs. either graphite or Li metal, in either carbonate-based electrolytes [36,60] or in ionic liquidsbased electrolytes [61,62]. LiTFSI (Figure 4d) shows decent initial performance with well-defined plateaus until C/5. However, Li + transport becomes slower over cycling and the capacity at C/2 decays significantly.

Performance of graphite electrodes
LiPF6 (Figure 4f) also shows decent performance up to C/5, as already reported with different graphite electrodes [35]. For this salt as well, the performance decays significantly over cycling, which has been linked with the slow Li + transport at the Li metal counter electrode/electrolyte interface [35]. Here, we observe that the performance of LiPF6 is lower than that of LiDFOB (Figure 4g). In fact, most of the electrolytes perform worse than LP47 (Figure 4h), a moderately conductive commercial electrolyte, with the exception of the LiDFOB electrolyte which allows both higher rate capability and stable cycling at C/2. cycles, the two cells with the LiPF6 electrolytes start decaying at a faster rate, probably due to electrolyte consumption. The slopped voltage profiles (shown in Figure 5c) show that the diffusion of Li + is becoming significantly slower from cycle 10 to cycle 40 and even more obviously at cycle 55. Even in the first charge at C/2, the capacity inserted galvanostatically is rather low. On the other hand, as seen Figure 5d, LP47 allows inserting galvanostatically ca. 180 mAh g -1 in the first cycle and still above 100 mAh g -1 in cycle 55.
The 1 M LiDFOB electrolyte exhibits a 'recovery' behavior after ca. 55 cycles. It suggests that the issue originates from the Li metal counter electrode, which suddenly starts performing better (i.e. the Li metal counter is 'depassivated' by surface rearrangement/dendrites). After around 55 cycles, the capacity becomes very stable and higher than those of the other electrolytes, with also higher efficiencies (above 99.6%). The voltage profiles (shown in Figure 5b) show that the discharge voltage plateau of cycle 40 is much shorter compared with that of cycle 55. It probably means that full areas of the graphite electrode were not accessed (as a result of partial blocking of the Li counter electrode facing them) and are then depassivated during the cycling. Thus, LiFePO4 electrodes were used to substitute the Li metal counter electrode, as a common feature of SL:DMC electrolytes seems to be a poor Li metal counter electrode performance at room temperature. As shown in Figure 6a, the 1 M LiDFOB SL:DMC (1:1, wt.) electrolyte, when lithium is not used as counter electrode, allows efficient and steady cycling of graphite, with efficiencies close to 100% (99.83% average efficiency after the first cycle). The rate performance is significantly improved as most of the insertion occurs during the galvanostatic step at C/2, as shown in Figure 6c, resulting in a full charge in less than 2 h 30 min. On the other hand, the 1 M LiFSI SL:DMC (1:1, wt.) electrolyte, leads to much lower efficiencies ( Figure 6b) and, despites its higher conductivity, the graphite electrode is mostly charged during the constant voltage step (Figure 6d), which indicates that the SEI formed on graphite limits the charge. Figure   6c also shows that, for the LiDFOB electrolyte, the insertion voltage profiles are well superimposed from cycle 3 to cycle 100, showing no sign of Li + diffusion hindrance by SEI build-up. If we compare these results with the half-cells results, it is clear that Li metal rather than graphite is problematic with the LiDFOB electrolyte. On the other hand, even with a LiFePO4 counter electrode, Figure 6d shows a clear evolution of the potential profiles of the graphite electrode when the LiFSI electrolyte is used, which indicates that the SEI evolves with cycling in accordance with the lower efficiencies and fast capacity decay.

Conclusion
Sulfolane can be used as an EC alternative for cycling graphite-based Li-ion batteries. Without additive, efficiencies around 90% are obtained in the first cycle. Out of LiBF4, LiPF6, LiFSI, LiTFSI, LiDFOB and LiTDI, only LiTDI and, more surprisingly, LiFSI, did not allow graphite electrode cycling in half-cells.
LiDFOB even allowed better performance than the EC-based LP47 electrolyte. However, the performances in half-cells are strongly influenced by the Li metal counter electrode. By substituting Li metal with a LiFePO4 counter electrode, it is possible to show improved graphite performance, with efficiency above 90% in the first cycle and above 99.8% in the following cycles with LiDFOB. LiFSI, however, even without Li metal counter electrode, does not form a stable and protective SEI on graphite in SL:DMC. It is likely that the use of well-known additives, such as VC, FEC or LiBOB could help the SEI formation on graphite and lead to further improvements.