Lithium Oxalate as Capacity and Cycle-Life Enhancer in LNMO/Graphite and LNMO/SiG Full Cells

In the present study, we explore the use of lithium oxalate as a “sacriﬁcial salt” in combination with lithium nickel manganese spinel (LNMO) cathodes. By online electrochemical mass spectrometry (OEMS), we demonstrate that the oxidation of lithium oxalate to CO 2 (corresponding to 525 mAh/g) occurs quantitatively around 4.7 V vs. Li/Li + . LNMO/graphite cells containing 2.5 or 5 wt% lithium oxalate show an up to ∼ 11% higher initial discharge capacity and less capacity fade over 300 cycles (12% and 8% vs. 19%) compared to cells without lithium oxalate. In LNMO/SiG full-cells with an FEC-containing electrolyte solution, lithium oxalate leads to a better capacity retention (45% vs 20% after 250 cycles) and a higher coulombic efﬁciency throughout cycling ( ∼ 1%) compared to cells without lithium oxalate. When CO 2 from lithium oxalate oxidation is removed after formation, a similar capacity fading as in LNMO/SiG cells without lithium oxalate is observed. Hence, we attribute the improved cycling performance to the presence of CO 2 in the cells. Further analysis (e.g., FEC consumption by 19 F-NMR) indicate that CO 2 is an effective SEI-forming additive for SiG anodes, and that a combination of FEC and CO 2 has a synergistic effect on the lifetime of full-cells with SiG anodes. © The Author(s) 2018. Published by ECS. This is an open access article distributed under the terms of the Creative Commons Attribution 4.0 License (CC BY, http://creativecommons

Lithium nickel manganese spinel (LiNi 0.5 Mn 1.5 O 4 , LNMO) is a promising cathode material for high energy lithium ion batteries due to its high operating potential around 4.7 V vs. Li/Li + , its high rate capability, structural stability and the absence of cobalt. However, its lower specific capacity (146 mAh/g LNMO ) compared to layered oxide materials (e.g. lithium nickel manganese cobalt oxide (NMC), specific capacity 150-250 mAh/g NMC ) 1 is regarded as a major drawback. In full-cells, the practically achievable capacity of LNMO is even lower, as the formation of the solid-electrolyte interphase (SEI) on the graphite anode consumes active lithium. For many layered oxide cathodes, however, the first cycle irreversible capacity of the cathode is similar to the capacity needed for SEI formation (∼20 mAh/g NMC ), and hence the practical discharge capacity of the cathode and the remaining active lithium are more or less balanced again. 2,3 In contrast, the first cycle irreversible capacity of LNMO (∼6 mAh/g LNMO ) is much lower than the capacity needed for SEI formation. This leads to a mismatch between active lithium and practical cathode capacity, i.e., there is not enough active lithium available to fully discharge the LNMO cathode during subsequent cycling.
In cells with silicon-based anodes, active lithium losses on the anode are even higher compared to graphite, as the expansion of the silicon particles during the first lithiation leads to a continuous exposure of fresh, unpassivated silicon surface. 4 On this new surface, electrolyte reduction occurs instantaneously, which reduces the total lithium reservoir in the cell. Therefore, different ideas to increase the amount of active lithium in lithium ion full-cells have been suggested, for example via prelithiation of silicon anodes with metallic lithium. [5][6][7][8] Recently, Gabrielli et al. 9 successfully used LNMO that had been chemically overlithiated to compensate for the initial lithium loss in LNMO/silicon-carbon full cells. Alternatively, Shanmukaraj et al. 10 proposed "sacrificial salts" as an additional source of lithium ions: A lithium salt is incorporated in the active material/carbon black matrix of the cathode. During the initial charge, the anion of the sacrificial salt is oxidized yielding mostly gaseous products, while the corresponding lithium cation is intercalated into the graphite anode; the gas can then be removed after formation. Lithium oxalate has a high specific charge capacity of 525 mAh/g (based on Li 2 C 2 O 4 → 2 Li + + 2 e − + 2 CO 2 ), but was disregarded as a sacrificial salt for typical 4 V cell chemistries due to its high oxidation potential around 4.6-4.7 V vs. Li/Li + . 10 However, this potential matches well with the charging plateau of LNMO. Additionally, lithium oxalate releases only CO 2 during oxidation, which is considered to improve the interfacial stability of graphite as well as of lithium metal anodes. [11][12][13][14][15] Strehle et al. 16 showed that the presence of CO 2 can suppress the formation of soluble lithium alkoxides and the follow-up electrolyte transesterification reactions. CO 2 can act further as a scavenger for detrimental trace water and protons. 17 Recently, Krause et al. 18 demonstrated that CO 2 is also an effective additive for silicon-based cells, increasing their capacity retention and coulombic efficiency.
In the present study, we use lithium oxalate as a capacity enhancer in LNMO/graphite and LNMO/silicon-graphite (SiG) full cells. First, we investigate the electrochemical oxidation of lithium oxalate and the resulting gas evolution by online electrochemical mass spectrometry (OEMS) in order to quantify its decomposition reaction and potential. As a next step, we test the addition of 2.5 or 5 wt% of lithium oxalate to LNMO composite electrodes in half cells, which increases the theoretical capacity of the initial charge by about 10% or 20%, respectively. To investigate the effect of the increased lithium inventory in full-cells, we cycle LNMO composite electrodes containing 0, 2.5, or 5 wt% lithium oxalate in full-cells against graphite anodes. We further test LNMO composite electrodes with 0 or 5 wt% lithium oxalate with silicon-graphite electrodes (SiG, 35 wt% silicon, 45 wt% graphite) in an electrolyte solution containing fluoroethylene carbonate (FEC). This additive is known to improve the cycling stability of silicon-based electrodes [19][20][21][22][23][24][25] and is commonly employed for cell chemistries containing silicon. As the LNMO/SiG cells with lithium oxalate showed less capacity fade and a higher coulombic efficiency throughout cycling than their counterparts without lithium oxalate, we investigate the effect of CO 2 on cycling performance and FEC consumption in LNMO/SiG cells. Lastly, we discuss the opportunities and challenges associated with the use of lithium oxalate as a sacrificial salt in LNMO/graphite and LNMO/SiG cells.

Table I. Electrode composition and properties of cathodes and anodes used for LNMO/Li half-cells and LNMO/graphite and LNMO/SiG full-cells.
Theoretical capacities for LNMO, lithium oxalate (LO), graphite, and silicon were taken as 146 mAh/g LNMO , 525 mAh/g LO , 372 mAh/g graphite , and 3579 mAh/g Si . Practical 1 st charge/discharge capacities for LNMO and graphite electrodes were determined at a C-rate of 0.1 h −1 in lithium half-cells, while the practical capacities for SiG electrodes were taken from Ref. 25 BET /g, BASF SE, Germany) with 2.5 or 5 wt% lithium oxalate were prepared by first dispersing C65 carbon and lithium oxalate (1:1 by weight) in NMP with an ultrasonication horn as described above. LNMO, PVDF and C65 were added to the dispersion according to the compositional ratios given in Table I. The compositions were chosen in order to keep a fixed weight ratio of LNMO:C65:PVDF of 90:5:5 in all electrodes once all lithium oxalate would be oxidized. NMP was added to yield a solid content of 40%, and the slurry was mixed in a planetary mixer (2000 rpm, 15 min). LNMO electrodes without lithium oxalate were prepared by combining LNMO, C65, and PVDF in ratios according to Table I (using the same mixing procedure). The slurries were then coated onto aluminum foil (15 μm, MTI, United states) using a gap bar (300 μm wet film thickness, ∼11-13 mg LNMO /cm 2 ).
To match the higher areal capacities of the SiG electrodes, additional LNMO coatings with 0 and 5 wt% lithium oxalate were prepared with a wet film thickness of 450 μm (∼14-16 mg LNMO /cm 2 ). For OEMS measurements with LNMO or LNMO + 5 wt% lithium oxalate electrodes, the corresponding slurries were coated onto perforated aluminum foil (Microgrid Al 25, Dexmet, United States; ∼25 μm thickness). All LNMO coatings were dried at 50 • C for 6 h in a convection oven. Electrodes with a diameter of 14 mm were punched out and compressed with 150 MPa, which resulted in electrode thicknesses ∼55 μm and ∼75 μm, respectively, and a porosity of 35%.

Graphite and silion-graphite (SiG) electrode preparation.-
Graphite electrodes were prepared by mixing graphite (T311, BET surface area 5 m 2 /g, SGL Carbon, Germany) and PVDF according to the ratio given in Table I with NMP in a planetary mixer (2000 rpm, 10 min). The ink (50% solid content) was coated onto copper foil (MTI, United States) using a 150 μm gap bar and dried at 50 • C for 6 h in a convection oven. Silicon-graphite electrodes were prepared from silicon nanoparticles (∼200 nm, Wacker Chemie AG, Germany), graphite, vapor grown carbon fibers (VCGF-H, Showa Denko, Japan) and lithium poly(acrylate) (LiPAA, Sigma-Aldrich, Germany) by a ballmilling routine as described in Wetjen et al. 25 Graphite and SiG coatings were punched into 15 mm electrodes. The final loadings and compositions for all electrodes are given in Table I. Prior to cell assembly, all electrodes were dried under dynamic vacuum at 120 • C for at least 12 h and then transferred into an argon-filled glove box (MBraun, Germany) without exposure to air.
Cell assembly and cycling.-2032 coin cells were assembled in an Ar-filled glove box (MBraun, H 2 O, O 2 < 0.1 ppm) with LNMO (with or without lithium oxalate) as cathodes (Ø 14 mm) and either graphite or SiG electrodes as anodes (Ø 15 mm). Individual anodes and cathodes were paired in a such a way that the anode/cathode areal capacity balancing for all cells was ∼1.2-1.3 (based on their practical 1 st discharge capacity, see Table I), in order to accommodate any excess A514 Journal of The Electrochemical Society, 165 (3) A512-A524 (2018) lithium from the lithium oxalate oxidation. The electrodes were separated by 2 glass fiber separators (Whatman, Ø 16 mm) and wetted with 80 μL LP57 electrolyte solution (30 wt% ethylene carbonate (EC), 70 wt% ethyl methyl carbonate (EMC), 1 M lithium hexafluorophosphate (LiPF 6 ), Selectilyte, BASF SE) or 80 μL LP57 + 5 wt% fluoroethylene carbonate (FEC, BASF SE). LNMO/graphite cells were cycled galvanostatically between 3.5-4.8 V at 1C (≈1.7 mA/cm 2 ) after 2 formation cycles at a rate of C/10. LNMO/SiG cells were cycled galvanostatically between 4.0-4.8 V at C/2 (≈1.1 mA/cm 2 ) after 3 formation cycles at a rate of C/10. The lower cutoff voltage of 4.0 V was chosen, as Si-based cells have shown better cycling stability when high anode potentials were omitted. 21 Note that due to the lower potential of silicon during its first lithiation, the upper cutoff voltage was set to 4.9 V in the first cycle of the LNMO/SiG cells. For both cell chemistries, a constant voltage (CV) step was performed with a current limit of C/20 after each galvanostatic charging step. All cells were cycled with a battery cycler (Series 4000, Maccor, USA, coulombic efficiency accuracy ∼300 ppm 26 ) in a climate chamber (Binder, Germany) at 25 • C. Specific capacities and coulombic efficiencies are given as the average of two duplicate cells, whereas error bars represent the deviation of these cells from the average. Note that the Maccor coulombic efficiency accuracy is not included in the error bars, because the focus of this study lies on comparing different cell configurations and not on an exact estimate of the cells' lifetime.

On-line electrochemical mass spectrometry (OEMS).-The cell
design of the on-line electrochemical mass spectrometry (OEMS) system has been described in previous publications. 27 To study the oxidation of lithium oxalate, a lithium oxalate / carbon black electrode was charged galvanostatically vs. a lithium metal counter electrode (Ø 17 mm, 450 μm thickness, Rockwood Lithium, United States) with a nominal rate of C/10 (≈0.02 mA/cm 2 , based on the theoretical capacity of 525 mAh/g for lithium oxalate). To avoid reactions of CO 2 with the lithium anode, a sealed 2-compartment setup 28 where the cathode and anode compartment are separated by an aluminumsealed lithium-ion conductive glass ceramics (LICGC, Ohara Corp., Japan) was used, with a glass fiber separator soaked with 250 μL LP57 in the lithium anode compartment, and a Celgard trilayer separator (H2013) soaked with 80 μL LP57 in the lithium oxalate/carbon black cathode compartment, respectively.
OEMS measurements on LNMO/SiG full-cells were performed in 1-compartment OEMS cells (i.e., without barrier between anode and cathode compartment) using LNMO electrodes (∼11.4 mg LNMO /cm 2 ) without or with 5 wt% lithium oxalate coated onto perforated aluminum foil as cathodes in combination with SiG anodes, and using 150 μL of electrolyte solution (LP57 pure or with 5 wt% FEC) and two Celgard separators (H2013, Ø 28 mm). To eliminate the effect of the SiG electrode, we also performed an OEMS measurement using the same cathode, separator and electrolyte solution, but a Ø 15 mm delithiated LiFePO 4 electrode (LFP, 3.5 mAh/cm 2 , Custom Cells, Itzehoe) as anode. Prior to the OEMS measurement, the LFP electrode was cycled once at a rate of C/10 vs. a lithium metal electrode between 2 V and 4 V, and then electrochemically delithiated to a lithium content of 0.15 by a galvanostatic charge at C/10 with a capacity cutoff at 3 mAh/cm 2 . The cell was then disassembled inside an Ar-filled glove box and the LFP electrode directly transferred to the OEMS cell. All OEMS experiments were performed in a climate chamber (Binder, Germany) at 25 • C.
FEC quantification by NMR.-The consumption of fluoroethylene carbonate (FEC) during cycling of LNMO/SiG cells with and without lithium oxalate was investigated by 19 F-NMR of the recovered electrolyte solutions from full-cells. To this purpose, LNMO/SiG full-cells with or without lithium oxalate and LP57 + 5 wt% FEC were cycled as described above, and carefully disassembled after 50 cycles or 250 cycles. One of the glass fiber separators was removed and immersed into 700 μL deuterated dimethyl sulfoxide (DMSO-d6, anhydrous, Sigma-Aldrich, USA). The solutions were then filled into air-tight NMR tubes and 19 F-NMR spectra were measured on a Bruker Ascend 400 (400 MHz). As the obtained 19 F-NMR spectra show only peaks that can either be ascribed to PF 6 − or FEC, and assuming that changes in the PF 6 − concentration in the electrolyte solution during cycling are negligible, the PF 6 − anion can be used as an internal standard. 22 The amount of FEC remaining in the electrolyte solution can thus be determined by the ratio of PF 6 − and FEC peak integrals. A more detailed description of this method can be found in a recent publication by Jung et al. 22

Results
Electrochemical characterization of lithium oxalate/carbon and lithium oxalate/LNMO electrodes.-To investigate the electrochemical oxidation of lithium oxalate within the potential window of an LNMO cathode, we first tried to understand the electrochemical oxidation of lithium oxalate by itself, i.e., in the absence of any active material. Therefore, we fabricated model electrodes that contain only lithium oxalate in a conductive carbon matrix, similar to the ones used by Meini et al., 29,30 who studied the anodic decomposition of lithium salts (Li 2 O 2 , Li 2 CO 3 , LiOH, and Li 2 O). Figure 1a shows the potential during the galvanostatic charge of an electrode consisting of 45 wt% sub-micron lithium oxalate, 45 wt% carbon black and 10 wt% binder, with a nominal rate of C/10 (based on its theoretical decomposition capacity of 525 mAh/g LO ; see reaction 1 below). The potential profile shows a flat plateau around 4.7 V vs. Li/Li + until ∼90% of the theoretical capacity (∼470 mAh/g LO ) is reached, from which point on the potential rises continuously. The concomitant gas evolution, as measured by on-line electrochemical mass spectrometry (OEMS) during this first charge, is shown Figure 1b. Here it is important to note that this experiment was conducted in a sealed 2-compartment cell, 28 so that there are no contributions to the measured gas evolution from the lithium metal counter electrode nor any consumption of the gas evolved at the working electrode by the lithium metal electrode. During charge, a linear increase of the CO 2 (m/z = 44) concentration in the cell headspace is observed (pink line in Figure 1b), indicating a constant gas evolution of CO 2 . Carbon monoxide (m/z = 28, green line) and particularly oxygen (m/z = 32, blue line) are only detected in negligible trace amounts throughout the entire measurement. The slope of the CO 2 evolution vs. charge during the bulk oxidation of lithium oxalate is close to the 1 e − /CO 2 line (gray dashed line in Figure 1b), suggesting a 1-electron process. As the CO 2 evolution from electrolyte oxidation is negligible at potentials ≤ 5 V vs. Li/Li + on a carbon black electrode at room temperature, 31 we can attribute the CO 2 evolution entirely to the oxidation of lithium oxalate. Once the theoretical charging capacity of Li 2 C 2 O 4 has been reached (black dashed line in Figure 1), the CO 2 evolution slows down. This indicates that lithium oxalate is quantitatively oxidized around 4.7 V vs. Li/Li + according to reaction 1: 32 The slight deviation downwards from the 1 e − /CO 2 line can be explained by two effects: i) the lithium oxalate used here contains about 1% Li 2 CO 3 as impurity, whose oxidation also gives CO 2 , but in a ≥ 2 e − /CO 2 process; 29 and ii) about 2.5% of the total CO 2 remain dissolved in the electrolyte solution and are thus not detectable by OEMS (for details see Discussion section of this paper).
The oxidation potential of 4.7 V vs. Li/Li + for lithium oxalate lies well within the plateaus of the Ni 2+ /Ni 3+ and Ni 3+ /Ni 4+ redox couple of high-voltage lithium nickel manganese spinel (LiNi 0.5 Mn 1.5 O 4 , LNMO) at 4.7 V and 4.75 V vs. Li/Li + , respectively. 33 In the following, we explore the use of lithium oxalate as a sacrificial salt in combination with LNMO cathodes. Figure 2a shows the first charge/discharge profile of a LNMO cathode vs. Li at C/10 with 0 wt% (black line), 2.5 wt% (purple line) or 5 wt% lithium oxalate (pink line) added to the cathode slurry during electrode fabrication. The ratios of LNMO to binder and conductive carbon were kept constant in all electrode compositions (see Table I). During charge, the LNMO cathode without lithium oxalate (black lines in Figure 2a) can be delithiated to a capacity of 145 mAh/g LNMO , which is essentially identical with the theoretical capacity of 146 mAh/g LNMO (black dashed line in Figure 2). LNMO cathodes with 2.5 wt% lithium oxalate (purple lines in Figure 2a) deliver ∼160 mAh/g LNMO charge capacity, which also comes close to the theoretically expected 161 mAh/g LNMO for these electrodes based on the combined capacity of LNMO and lithium oxalate (see Table I and purple dashed line in Figure 2). Accordingly, the charge capacity of the cathode with 5% lithium oxalate (pink lines in Figure 2) is about 173 mAh/g LNMO (theoretical capacity: 175 mAh/g LNMO , see pink dashed line). Note that all capacities are given per gram of LNMO, neglecting the weight of the lithium oxalate, as the latter is virtually completely oxidized to CO 2 and Li + -ions during the first charge. Considering the potential profiles, one can see that the transition between the Ni 2+ /Ni 3+ and the Ni 3+ /Ni 4+ plateaus is gradually shifted toward higher specific capacities with increasing lithium oxalate content. This indicates that a large fraction (but not all) of the lithium oxalate is already oxidized during the Ni 2+ /Ni 3+ plateau. The subsequent discharge is equally long for all three electrode compositions, i.e., amounting to the same first discharge capacity of 140 mAh/g LNMO (see Table I). Furthermore, the charge/discharge capacities as well as the potential profiles of the second cycle at C/10 are identical for the three different compositions (see Figure 2b). Theoretically, the oxidation of all lithium oxalate will increase the LNMO electrode porosity from ∼35% to ∼38% or ∼40% for electrodes containing 2.5 wt% or 5 wt% lithium oxalate, respectively. However, based on the capacities and potential profiles  Table I. For LNMO loadings, see Table I. obtained during the second cycle (see Figure 2b), we can conclude that the effect of this porosity change is only minor, and the oxidation of lithium oxalate within the cathode has not altered the subsequent electrochemical behavior.

The effect of lithium oxalate in LNMO/graphite full cells.-
In the above shown LNMO/Li half-cells, the anode consisted of a massive lithium reservoir, meaning that at least with regards to the availability of active lithium, the LNMO cathode could be fully relithiated in all cases. However, in full-cells with a graphite anode, the amount of active lithium in the cell is limited, and the relithiation capacity of the cathode depends on the amount of lithium still available, i.e., lithium that has not been consumed for SEI formation. Therefore, we expect higher discharge capacities for cathode compositions that have shown higher capacities in their first charge in LNMO/graphite cells, i.e., for those cells which have a higher lithium oxalate content in the cathode. Figure 3a shows the discharge capacities (closed symbols) and 1 st cycle charge capacities (open symbols) for LNMO/graphite fullcells with different lithium oxalate contents (0, 2.5, and 5 wt%) in the LNMO cathode for two formation cycles at C/10 and further cycling at 1C (all at 25 • C). The areal capacity of the graphite anodes was kept  Table I. constant in order to have a comparable first cycle irreversible capacity for all cells. Similar to the LNMO/Li half cells, the first charge capacity is 146 mAh/g LNMO , 159 mAh/g LNMO , and 173 mAh/g LNMO for cells with 0, 2.5, and 5 wt% lithium oxalate in the cathode (closed symbols in Figure 3a). The discharge capacity is now however different for the three compositions: while the cells without lithium oxalate reach a first cycle discharge capacity of 125 mAh/g LNMO , the cells with 2.5 wt% lithium oxalate have a first cycle discharge capacity of 138 mAh/g LNMO , which means that the first irreversible capacity of these different cells is in both cases ∼21 mAh/g LNMO . The cells with 5 wt% lithium oxalate have a first cycle discharge capacity of around 139 mAh/g LNMO . Assuming a similar irreversible capacity as for the cells with 0 wt% and 2.5 wt% lithium oxalate, one would expect a discharge capacity of around 152 mAh/ g LNMO ; however, this is above the practical reversible capacity of LNMO of 140 mAh/g (as shown in Figure 2). The additional lithium (corresponding to ∼12 mAh/g LNMO ) thus remains as a reservoir in the graphite anode.
During the subsequent cycles at 1C, the difference in discharge capacity between the three different cathode compositions is retained up to 300 cycles. Interestingly, the cells with lithium oxalate also show better capacity retention: While cells without lithium oxalate drop about 19% in capacity between cycle 3 (i.e., the first cycle at 1C) and cycle 300, the cells with 2.5 and 5 wt% lithium oxalate lose only about 12% and 8% capacity, respectively. These cells also show an improved average coulombic efficiency between cycles 3-300 of 99.88% (2.5 wt% lithium oxalate) and 99.92% (5 wt% lithium oxalate), compared to 99.81% for cells without lithium oxalate (see Figure 3b). This phenomenon could be explained by i) the additional lithium reservoir and its effect on the graphite potential at the end of discharge, and/or, ii) by the effect of CO 2 as an SEI-forming additive. In the cells with 5 wt% lithium oxalate, some lithium remains in the graphite after formation. Hu et al. 34 have shown that LNMO/graphite cells show an improved cycling behavior as long as the cells contain an excess of active lithium (added in their case as metallic lithium or by exsitu pre-lithiation of the graphite anode). However, the amount of additional lithium in our case is much lower compared to Hu et al. 34 (∼10% vs. ∼100% of the initial LNMO capacity) and should therefore only affect the very first cycles (i.e., until the additional capacity of ∼12 mAh/g has been consumed). Still, if lithium is remaining in the graphite anode at the end of a discharge, the maximum graphite potential is lower compared to cells with less or without lithium oxalate. It has been shown that enhanced gas evolution related to SEI damage can occur when graphite is polarized to high potentials (>1.2 V vs. Li/Li + ). 35 Although a precise determination of the anode potential without a reference electrode is hardly possible, we can use following approximation: If we assume a maximum potential of 4.7 V vs. Li/Li + for LNMO at the end of discharge and consider that the difference between cathode and anode potential has to be at least 3.5 V (which is our lower cutoff penalty), the graphite potential is limited to a maximum of 1.2 V vs. Li/Li + ; hence, SEI damage should be avoided. Furthermore, the observed improvements in cycling stability of cells with 2.5 wt% lithium oxalate cannot be explained by different maximum potentials for graphite, as in this case, no lithium is remaining in the graphite and the upper cutoff potential during graphite delithiation should be very similar to cells without lithium oxalate.
Consequently, this brings us back to the effect of CO 2 as an SEI additive. In general, SEI instability and the consequent active lithium loss is regarded as a major fading mechanism in LNMO/graphite cells. [34][35][36] Pritzl et al. 37 recently showed that the cycling stability of LNMO/graphite cells can be improved by very small amounts of VC, an effective SEI former; however, if the amount of VC gets too large, the competitive oxidation of VC on LNMO counteracts its beneficial effect on the anode. 37,38 CO 2 has long been known to improve SEI properties on both lithium metal 14,15 and graphite. 12,13,39 Xiong et al. found that almost all CO 2 generated on the cathode can be consumed on the graphite anode in commercial-scale NMC422/graphite fullcells, given that no other strong SEI-forming additives are present. 40 Krause et al. showed that CO 2 can have a similar effect on the cycling stability of graphite electrodes in EC-free electrolyte solutions as VC. 18 Recently, our group demonstrated that CO 2 can stop the transesterification reactions occurring from alkoxides generated through the reduction of linear carbonates. 16 Furthermore, the CO 2 − radical from the reduction of CO 2 on graphite can act as an H 2 O/H + scavenger, yielding lithium formate and lithium carbonate as SEI products. 17 As protons are a possible product of electrolyte oxidation, 28 this effect would be especially relevant for high-voltage lithium ion cell chemistries like the LNMO/graphite system.
Sloop et al. 41 suggested that the reduction of CO 2 could lead to lithium oxalate formation at the anode, which could dissolve and be re-oxidized at the cathode, generating a shuttling current followed by self-discharge. To assess whether the presence of CO 2 indeed leads to enhanced side reactions, we calculated the charge end point slippage (the cumulative irreversible charge capacity, i.e., the charge capacity of each cycle subtracted by the previous discharge, summed up over all cycles), which is an indicator for oxidative or shuttling side reactions. 42 The charge end point slippage for LNMO/graphite cells with 0, 2.5, or 5 wt% lithium oxalate is shown in Figure 3c. As a CO 2 /oxalate shuttle mechanism would contribute to the charge capacity but not the discharge capacity, cells with lithium oxalate should show a higher charge end point slippage compared to cells which do not contain lithium oxalate or CO 2 . However, it becomes clear from Figure 3c that the charge end point slippage for cells with lithium oxalate is lower compared to cells without lithium oxalate. Hence, under the present conditions, the CO 2 /oxalate shuttle effect is either negligible or nonexistent and does not contribute to the side reactions in the cell. This is in agreement with Xiong et al., 40,43 who showed that there is no re-generation of CO 2 from a lithium oxalate/CO 2 shuttle detectable in NMC422/graphite cells.

The effect of lithium oxalate in LNMO/SiG full cells.-
The use of a sacrificial salt to compensate for SEI losses is even more relevant if LNMO cathodes are combined with silicon or silicon/graphite anodes, which typically have much higher SEI losses due to i) the high specific surface area of the nanometer-sized silicon particles, and, ii) the expansion of the silicon particles during their lithiation, creating fresh surface area in every cycle that triggers further SEI growth. Hence, we investigate the use of lithium oxalate as capacity enhancer in combination with silicon/graphite (SiG) electrodes containing 35 wt% nano-Si and 45 wt% graphite. These electrodes, which have been investigated in more detail in a previous study by our group, 25 show a typical first cycle coulombic efficiency of ∼85%. Therefore, we combine them with LNMO cathodes containing 5 wt% lithium oxalate, as the amount of lithium oxalate in these electrodes should largely compensate the irreversible loss during the first cycle. As electrolyte solution, we use LP57 + 5 wt% fluoroethylene carbonate (FEC), as this additive is known to improve the capacity retention of silicon-based electrodes. [19][20][21][22][23][24][25] The capacity retention and the coulombic efficiency of the LNMO/SiG cells with 5 wt% (green symbols) and without lithium oxalate (black symbols) in the cathode are shown in Figures 4a and 4b, respectively. The first charge capacities are 145 mAh/g LNMO for cells without lithium oxalate and 173 mAh/g LNMO for cells with 5 wt% lithium oxalate (open symbols), similar to the corresponding cells with graphite anodes. The first discharge capacities of 128 mAh/g LNMO for cells with 5 wt% lithium oxalate and 109 mAh/g LNMO without lithium oxalate are however lower compared to the corresponding LNMO/graphite cells. This was expected due to the higher irreversible capacity of the SiG anodes. It is to note that the first cycle irreversible capacity of cells with 5% lithium oxalate is slightly higher compared to the cells without lithium oxalate (45 mAh/g LNMO vs. 36 mAh/g LNMO ); this effect can be explained as the 20% larger charge capacity results in a ∼17% higher degree of lithiation (considering the balancing factor of ∼1. 3) The stronger expansion of the silicon particles creates more fresh surface and requires a higher irreversible capacity to passivate the selfsame.
During cycling, all SiG-based cells show a much stronger capacity fade compared to the LNMO/graphite cells (Figure 3). Yet, also for the LNMO/SiG system, the cells with lithium oxalate show a more stable cycling behavior and a higher coulombic efficiency compared to their wt% (black and gray squares) or 5 wt% (green and orange circles) lithium oxalate in the cathode matrix during cycling in LP57 + 5 wt% FEC at C/2 and 25 • C between 4.0-4.8 V. The first three cycles were performed at C/10. Cells represented by black and green symbols ("not opened") were then cycled at C/2. Cells represented by gray and orange symbols ("opened after formation") were reopened after the third cycle inside an Ar-filled glove box and both electrodes of each cell were transferred to a new cell with fresh separators and electrolyte solution. Cycling was then continued at C/2. All symbols represent the average of two replicate cells, whereas error bars represent the deviation between the replicates. The LNMO and SiG loadings are given in Table I. counterparts without lithium oxalate. While this difference was rather small in the LNMO/graphite cells, in the case of the LNMO/SiG cells, the average coulombic efficiency is almost 1% point higher during the first 50 cycles for cells containing 5 wt% lithium oxalate (black vs. green symbols in Figure 4b). At the same time, the capacity retention after 250 cycles is ∼45% for cells with 5 wt% lithium oxalate and only ∼20% for cells without lithium oxalate (referenced to the discharge capacity of cycle 4, i.e., the first cycle at C/2). Again, this effect could be attributed to either the additional lithium, or the CO 2 present in the cell. Markevich et al. 21 reported that the cycling stability of silicon is improved if complete delithiation is omitted; and similar to LNMO/graphite cells, LNMO/Si cells show a stable cycling performance if a sufficiently large lithium reservoir is available. 44 However, in both cell types used here (i.e., LNMO/SiG with and without lithium oxalate), the discharge capacity is always at least 20 mAh/g LNMO lower than the maximum relithiation capacity of the LNMO cathode, which means that the potential of the LNMO cathode is still around ∼4.7 V when the cells reaches the lower cutoff voltage (compare the half-cell potentials from Figure 2). As the lower cell cutoff is limited to 4.0 V, this corresponds to a maximum voltage of ∼0.7 V at the anode for both cells, where structural damage due to complete delithiation is unlikely.
In the LNMO/graphite cells, we have attributed the clearly improved cycling performance of cells with lithium oxalate to the effect of CO 2 as an SEI-forming additive. In contrast, in the LNMO/SiG cells there is already an SEI-forming additive (namely FEC), yet the differences between cells with and without lithium oxalate are much more pronounced for SiG anodes. In order to understand whether the additional lithium or CO 2 leads to the improved capacity retention, we repeated the cycling experiments with SiG anodes and LNMO cathodes containing either 0 or 5 wt% of lithium oxalate. This time, however, we stopped the cells after formation (i.e., after the initial 3 cycles at C/10) in the discharged state and disassembled them inside an Ar-filled glove box. The electrodes were then reassembled in new coin cells with fresh electrolyte solution and separators. In this way, all CO 2 was removed from the void space in the cell body and from the electrolyte solution, while the amount of active lithium was not altered. The gray and orange symbols in Figure 4 show the capacity retention and coulombic efficiency of these reassembled cells. The capacities of the first three cycles of both cell types look identical to Figure 4, as expected. After reassembly, the performance of the cells without lithium oxalate (gray symbols in Figure 4a) is slightly worse than the same cells which had not been opened (black symbols in Figure 4a), which can be attributed to a partial re-dissolution of the SEI in the fresh electrolyte solution. The minor differences however indicate that the reassembly procedure did not alter the cell performance substantially. The cells with lithium oxalate (orange symbols in Figure 4a) show the same discharge capacity directly after reassembly as before the reopening procedure, which also indicates that they were not damaged during the reassembly. However, from cycle 20 onwards, these cells show a much stronger capacity decrease compared to the cells with lithium oxalate that had not been opened (green symbols in Figure 4a). From cycle 10 on (i.e., almost directly after the reassembly), the coulombic efficiency of the reassembled cells with lithium oxalate (orange symbols in Figure 4b) starts to decline, while it is continuously rising in the cells with lithium oxalate that have not been opened (green symbols in Figure 4b). Around cycle 50, the coulombic efficiency of the reassembled cells with lithium oxalate (orange symbols) has reached the level of the cells without lithium oxalate (black and gray symbols). Apparently, the removal of CO 2 leads to an increase in irreversible reactions, which lowers the coulombic efficiency and depletes the active lithium, ultimately causing a drop in capacity.

FEC consumption in LNMO/SiG full cells with and without lithium oxalate.-To
understand to which extent CO 2 participates in SEI formation when FEC is present, we conducted a post-mortem analysis of the electrolyte solution of aged LNMO/SiG cells without lithium oxalate and with 5 wt% lithium oxalate after 50 and 250 cycles, quantifying the amount of residual FEC by 19 F-NMR (for more details see Jung et al. 22 ). The upper x-axis of Figure 5 shows the amount of residual FEC found in the electrolyte solution. As more remaining FEC is found in the cells with lithium oxalate (green symbols) compared to their counterparts without lithium oxalate that underwent the same number of cycles (black symbols), it is apparent that the FEC consumption per cycle for cells containing both FEC and lithium oxalate is lower. Previous studies have shown that the amount of consumed FEC correlates linearly with the cumulative irreversible discharge capacity (i.e., the sum of the differences between discharge and charge capacity over a certain amount of cycles), 22 which has recently also been demonstrated for SiG anodes with identical composition. 25 Therefore, the cumulative irreversible discharge capacity for each of the analyzed cells is shown on the y-axis of Figure 5  observed by Jung et al. 22 The offset of the e − /FEC ratios on the y-axis in Figure 5 can be explained by considering the following: As the lower cell cutoff potential is restricted to 4.0 V, the potential of the SiG anode is limited to a maximum of ∼0.7 V vs. Li/Li + (assuming a maximum cathode potential of 4.7 V vs Li/Li + ). At this potential, only about 85% of all lithium is extracted from the SiG electrode during discharge, 25 leading to an apparently higher irreversible capacity for the first cycle, which is however not related to FEC consumption.
It is to note that in contrast to References 22 and 25, in the present study the silicon-based anode is not the capacity-limiting electrode, and side reactions occurring at the LNMO cathode may not be negligible, which makes an analysis of the cumulative irreversible capacity less obvious. Still, the cells without lithium oxalate (black symbols) lie reasonably close to the previously found 4 e − /FEC linear correlation, which holds also true for the cells with 5 wt% lithium oxalate after 50 cycles (open green circles). After 250 cycles, however, the irreversible capacity is about 1 mAh higher (or the FEC consumption is 10 μmol lower) than the 4 e − /FEC correlation for cells with 5 wt% lithium oxalate (closed green circles in Figure 5). This indicates that there is an additional process related to the irreversible discharge capacity, which does not consume FEC. It is likely that this additional process is the electrochemical reduction of CO 2 on the SiG anode, as this reaction would contribute to the cumulative irreversible capacity, but not the FEC consumption.

Consumption of carbon dioxide in LNMO/SiG full cells.-To
investigate the consumption of CO 2 on the SiG anode, we performed 1-compartment OEMS measurements on the first cycle of LNMO/SiG full-cells. If a significant amount of the CO 2 from lithium oxalate oxidation were to be consumed on the SiG anode during this first formation cycle, the overall CO 2 evolution should be lower than expected for the essentially complete lithium oxalate oxidation in the first cycle. As a benchmark for the maximum CO 2 evolution that can be practically achieved from our LNMO/lithium oxalate cathodes, we also measured a LNMO electrode containing 5 wt% lithium oxalate vs. an oversized delithiated LFP electrode. Due to the high potential of the LFP (∼3.4 V vs. Li/Li + ), we do not expect any reductive consumption of CO 2 in this system. The yellow line in Figure 6a shows the potential profile of an LNMO/LFP cell with 5 wt% lithium oxalate (the cell potential was converted to the Li/Li + scale by adding 3.42 V, the equilibrium potential of LFP 45 ), whereas the yellow line in Figure 6b shows the corresponding CO 2 evolution. The first charge capacity (174 mAh/g LNMO ) in the LNMO/LFP + 5 wt% lithium oxalate OEMS cell is similar to the capacity achieved in LNMO half cells containing 5 wt% lithium oxalate vs. lithium (compare Figure  2). The CO 2 evolution in the LNMO/LFP + 5 wt% lithium oxalate OEMS cell rises to 1.02 mmol/g LNMO , which corresponds to ∼95% of the theoretical amount of CO 2 based on conversion of all lithium oxalate (see right y-axis in Figure 6b). The deviation from 100% is likely due to Li 2 CO 3 impurities as well as partial dissolution of CO 2 in the electrolyte solution, as discussed previously.
As a next step, we repeated this experiment but replaced the LFP counter electrode with a SiG anode, while the cathode (LNMO + 5 wt% lithium oxalate) and the electrolyte solution (LP57 + 5 wt% FEC) remained identical (green lines in Figures 6a and 6b). With the SiG anode, the total CO 2 evolution was significantly lowered to 0.83 mmol/g LNMO (∼77% of the theoretical CO 2 ). In the absence of FEC (i.e., in pure LP57), the total CO 2 evolution of the LNMO/SiG cells with 5 wt% lithium oxalate (red lines in Figures 6a and 6b) is further decreased to 0.73 mmol/g LNMO , i.e., to ∼68% of the theoretical CO 2 . These results indicate that a significant amount of the practically available CO 2 (∼19% for cell with LP57 + 5 wt% FEC or ∼28% for cells with LP57) is already reduced in the first cycle of LNMO/SiG cells. To evaluate whether the apparently lower CO 2 consumption in FEC-containing vs. FEC-free electrolyte solutions may not simply be due to the additional release of CO 2 by the reduction of FEC, 22,46 we also investigated an LNMO/SiG cell without lithium oxalate in LP57 + 5 wt% FEC by OEMS (black lines in Figures 6a and 6b). While a minor extent of CO 2 is formed in this case, it amounts to only 0.03 mmol/g LNMO , which is small compared to the ∼0.10 mmol/g LNMO difference between the lithium oxalate containing LNMO/SiG cells with and without FEC. These observations are consistent with the findings by Krause et al., 18 who showed that CO 2 dosed to cells with silicon anodes gets consumed at the silicon anode, and that its consumption rate is reduced in the presence of FEC.
In contrast to FEC, where the consumed amount can be easily determined by 19 F-NMR, a quantification of the remaining CO 2 after extended cycling is not easily possible from the coin cells used in this study. However, Krause et al. 18 showed that once all added CO 2 is consumed, Si alloy-based cells suffer from a severe drop in coulombic efficiency and capacity retention, analogously to what both Jung et al. 22 and Petibon et al. 24 demonstrated for the complete consumption of FEC from Si-based cells. Additionally, also the significantly different coulombic efficiencies from LNMO/SiG cells where CO 2 was either left in the cells (green symbols in Figure 4) or purposely removed (orange symbols in Figure 4) indicate that a drop in coulombic efficiency can be expected at the point where all CO 2 is depleted. Therefore, we repeated the cycling experiment from Figure 4 with LNMO/SiG cells containing different amounts of lithium oxalate (namely 0, 2.5, and 5 wt%) in pure LP57, i.e., without FEC in the electrolyte solution. Furthermore, with this experiment we investigate the effectiveness of CO 2 by itself as an SEI-forming additive for silicon-based anodes. Figure 7 shows the capacity retention for LNMO/SiG cells with LP57 and 0, 2.5, and 5 wt% lithium oxalate in the LNMO cathode. The initial charge/discharge capacities for cells with 0 wt% and 5 wt% lithium oxalate are similar to the cells with the same electrodes in FEC-containing electrolyte solution (see Figure 4), namely 145/110 mAh/g LNMO and 173/123 mAh/g LNMO , respectively, whereas the first cycle charge/discharge capacity of cells with 2.5 wt% lithium oxalate lies in between these two (160/112 mAh/g LNMO ). During cycling, the cells with 0 wt% lithium oxalate decline dramatically in capacity, while their coulombic efficiency drops to <90% (black symbols). After 100 cycles, there is essentially zero capacity left in these cells. Comparing the cells without lithium oxalate (i.e., without CO 2 ) from Figure 4 and Figure 7 (both black symbols), once again illustrates how important SEI-forming additives like FEC are for silicon-based anodes to achieve a minimum of stable cycling.
The LNMO/SiG cells with pure LP57 and 5 wt% lithium oxalate (red spheres in Figure 7), however, show initially a much improved cycling stability and a coulombic efficiency of around 99.5%, which is much higher than the ∼98.5% of the cells with FEC but without lithium oxalate (i.e., without CO 2 ; black symbols in Figure 4) and essentially identical with that of the cells with FEC and 5 wt% lithium oxalate (i.e., with CO 2 ; green symbols in Figure 4). This comparison clearly demonstrates that CO 2 forms an even more effective SEI than FEC. The cells with 2.5 wt% lithium oxalate (brown triangles in Figure 7) show a ∼9 mAh/g lower capacity compared to their counterparts with 5 wt% lithium oxalate, which we ascribe to the lower amount of additionally available active lithium in the former. However, their coulombic efficiencies in the first tens of cycles are essentially identical due to the presence of CO 2 . Around cycle 36 and 89, respectively, the cells with 2.5 wt% and 5 wt% lithium oxalate show a distinct decline in coulombic efficiency from ∼99.5% to ∼94% (for cells with 2.5 wt% lithium oxalate) or 97% (for cells with 5 wt% lithium oxalate), which in both cases is followed by a rapid decay in capacity.
In analogy to the rapid capacity fade and coulombic efficiency loss which was observed for silicon electrodes in FEC containing electrolye (without CO 2 ) once all FEC was being consumed, 22,24 and the observations made by Krause et al., 18 it is very likely that it is the complete consumption of CO 2 which leads to the onset of the decline in coulombic efficiency around cycle 36 and 89 in cells with lithium oxalate (i.e., with CO 2 ) in the FEC-free electrolyte solution  Table I. (see Figure 7). Under this assumption, we can examine whether there is a similar correlation between consumed CO 2 and cumulative irreversible capacity as we had done for FEC-containing electrolyte solution (see Figure 5), by taking the cumulative irreversible discharge capacity after cycle 36 (for cells with 2.5 wt% lithium oxalate) and cycle 89 (for cells with 5 wt% lithium oxalate) and the total theoretical amount of CO 2 that was available from lithium oxalate oxidation (see reaction 1) in these cells. Note that due to the ∼30% lower loading of the LNMO electrodes with 2.5 wt% lithium oxalate (see Table I), the total theoretical amount of CO 2 in these cells is not 50% of the amount in the cells with 5 wt% lithium oxalate, but somewhat lower. The resulting correlation is shown in Figure 8; interestingly, the points lie close to a 2 e − /CO 2 linear slope, whereas the often assumed reduction of CO 2 to carbonate and CO (acc. to: 2 CO 2 + 2 e -→ CO 3 2− + CO) [47][48][49] as well as the reduction of CO 2 to oxalate (acc. to: 2 CO 2 + 2 e -→ C 2 O 4 −2 ) 49-51 both would correspond to 1 e − /CO 2 . Apparently, a more complex reduction mechanism is taking place. One possible pathway is the formation of formate anions from CO 2 and protic species (i.e., protons from trace HF or electrolyte oxidation products), which would correspond to a total of 2 e − /CO 2 (acc. to: CO 2 + 2 e -+ H + → HCOO -). 47,51 Furthermore, it may be possible that a fraction of the cumulative irreversible capacity is related to decomposition reactions of the electrolyte solvent. A more detailed analysis of the possible reduction reactions of CO 2 on lithium ion battery anodes is currently under investigation. 17 In the LNMO/SiG cells with pure LP57 and 5 wt% lithium oxalate (red spheres in Figure 7) as well as in the reassembled LNMO/SiG cells with 5 wt% lithium oxalate, i.e., after the removal of CO 2 (orange spheres in Figure 4), we have observed a rapid drop in coulombic efficiency (to a less pronounced degree also in capacity) at the point where no CO 2 was left. However, such a coulombic efficiency drop is not observable for the cells with 5 wt% lithium oxalate with LP57 + 5 wt% FEC (green spheres in Figure 4). To our current understanding, this would indicate that CO 2 has not been completely consumed in these cells even after 250 cycles. To estimate the amount of remaining CO 2 in these cells, we again consider the irreversible capacity vs. FEC consumption relationship shown in Figure 5. As previously discussed, the irreversible capacity for the cells with 5 wt% lithium oxalate is higher compared to the experimentally found 4 e − /FEC (= 0.107 mAh/μmol FEC ) line (green circles in Figure 5). If we assume that this additional irreversible capacity (∼1 mAh) is associated with the reduction of CO 2 , we can use the empirically found correlation of 2 e − /CO 2 (= 0.0536 mAh/μmol CO2 ) from Figure 8 to estimate that ∼18.6 μmol of CO 2 have been consumed after 250 cycles. The fact that this amount is still lower than the available amount of CO 2 (28.2 μmol) suggests that CO 2 is still remaining in these cells after 250 cycles, which would explain why no rapid coulombic efficiency drop has occurred until this point for the cells with 5 wt% lithium oxalate in FEC-containing electrolyte solution (green circles in Figure 4). For the same cells after 50 cycles, the irreversible capacity lies close to the 4 e − /FEC line, which means that an assessment of the CO 2 consumption through the irreversible capacity cannot be undertaken. We believe that due to both the low consumption of FEC after 50 cycles as well as the low irreversible capacity, a definitive correlation cannot be made. Nevertheless, it is possible that a fraction of the irreversible capacity is used for the reduction of CO 2 ; the large differences in coulombic efficiency between cells with and without lithium oxalate (see Figure 4) as well as the results from OEMS measurements (see Figure 6) suggest that CO 2 is modifying the SEI already during early cycles.

Implications for the deliverable capacity of LNMO / SiG cells.-
Besides the continuous loss of active lithium, the lifetime of cells with Si-based anodes is largely dependent on the amount of SEIforming additives available. 18,22,24,25 The experiments in the present study have shown that LNMO/SiG cells with lithium oxalate show a better capacity retention and coulombic efficiency compared to their counterparts without lithium oxalate even in the presence of FEC, and that the co-reduction of CO 2 and FEC occurs simultaneously. However, the question remains how "efficient" the combination of FEC and CO 2 is in terms of additive consumption and active lithium loss. To elucidate this, Figure 9a shows the cumulative delivered discharge capacity, i.e., the sum of discharge capacities Figure 9. Cumulative delivered discharge capacity a) per cumulative irreversible discharge capacity, and, b) per μmol of additives (FEC+CO 2 ) for LNMO/SiG cells without lithium oxalate and LP57 + 5 wt% FEC (gray bars), cells with 5 wt% lithium oxalate and LP57 + 5 wt% FEC (green bars), and cells with 2.5 or 5 wt% lithium oxalate and LP57 (red bars). The solid bars represent the delivered capacity after 50 (36 for cells without FEC) cycles, while the dashed bars represent the delivered capacity after 250 (89 for cells without FEC) cycles. Bars represent the average result from two replicate cells, whereas error bars represent the deviation between the replicates. over all cycles up to a certain point, per cumulative irreversible discharge capacity for cells with only FEC (gray bars), only CO 2 (red bars) or both additives (green bars). While for cells containing FEC, the data after 50 and 250 cycles (gray solid and dashed bars) are shown, we plotted the data from cycle 36 and 89 for cells without FEC and only CO 2 , as these are the cycles for which there is strong evidence that all CO 2 has been consumed. Comparing only FEC-or only CO 2 -containing cells (gray and red bars in Figure 9a) , it becomes clear that CO 2 improves the delivered capacity per irreversible capacity. The largest impact, however, has the combination of FEC and CO 2 (green bars in Figure 9a), which leads to a doubling of the delivered capacity per irreversible capacity after 50 and 250 cycles compared to FEC by itself. Figure 9b shows the cumulative delivered discharge capacity per μmol of consumed additive (i.e., the sum of CO 2 and FEC), for the same points as in Figure 9a. To estimate the amount of CO 2 consumed in the LNMO/SiG cells with both lithium oxalate and FEC, we used the above described approximation based on the additional irreversible capacity. From Figure 9b, it is apparent that the combination of CO 2 and FEC (green bars) in LNMO/SiG cells leads to an improved additive efficiency (i.e., delivered capacity per mol of additive) compared to the use of single additives (gray and red bars). This effect is especially pronounced during the early stage of cycling (≤50 cycles), where the additive efficiency for cells with lithium oxalate and FEC is more than two times higher compared to the cells with only FEC (13.1 mAh/μmol vs 5.9 mAh/μmol, respectively). Interestingly, the additive efficiency of only FEC cells (gray bars) is similar to only CO 2 cells (red bars), whereas the delivered capacity per irreversible capacity is clearly lower for cells containing only FEC than for only CO 2 cells (gray and red bars in Figure 9a). However, this agrees with the lower number of electrons required for the reduction of CO 2 in contrast to FEC (compare Figure 5 and Figure 8). For all cells, the additive efficiency grows for a higher number of cycles, which fits well to the observation that coulombic efficiencies also tend to increase during cycling as long as FEC and/or CO 2 have not been consumed (see Figure 4 and Figure 7 as well as Reference 25). In the case of graphite anodes (see Figure 3), this is related to the formation of a gradually more passivating and thicker SEI; in the case of silicon anodes, the additional effect leading to an improvement of the coulombic efficiency is the fact that as the capacity fades, the state-of-charge change per cycle becomes lower, which causes less volume expansion/contraction and thus less and less SEI rupture.

Implications for the energy density of LNMO/graphite cells.-
To consider the effect of lithium oxalate on commercial-scale cells, we take a step back to the LNMO/graphite cells that were shown in Figure 3. Figure 10 shows the specific energy density (using the charge-averaged discharge voltage) during cycle 5 and cycle 300 of the LNMO/graphite cells with different amounts of lithium oxalate from Figure 3 at 1C discharge. The addition of 2.5 wt% lithium oxalate increases the initial (5 th cycle) cathode specific energy density from 555 Wh/kg LNMO to 600 Wh/kg LNMO ; the addition of 5 wt% lithium oxalate does not improve the energy density much further in the 5 th cycle (to 615 Wh/kg LNMO ). These values are ∼5-10% lower than the energy density of NMC622/graphite cells cycled at the same conditions to 4.4 V, the highest possible cutoff potential which still shows stable performance. 52 After 300 cycles, the cathode specific energy density of the LNMO/graphite cells without lithium oxalate drops to 453 Wh/kg LNMO (≡82% energy density retention), while ∼16% higher specific energies of 527 Wh/kg LNMO (≡88% energy density retention) are observed for cells with 2.5 wt% lithium oxalate; at the higher level of 5 wt% lithium oxalate, the specific energies of 555 Wh/kg LNMO are ∼23% higher than without oxalate and the energy density retention is 90%. These energy retentions are very comparable to that of the above mentioned NMC622/graphite cells. 52 While the lithium oxalate containing LNMO/graphite cells do have a ∼5-10% lower energy density, they are an interesting option for cobalt-free lithium ion battery cells, which may become critical in the future due to the rising cost and geographic concentration of cobalt. 53,54 Assuming all electrodes used here have the same initial porosity of 35%, the specific volume of electrode (including voids) per gram LNMO increases from 0.44 cm 3 /g LNMO to 0.46 cm 3 /g LNMO or 0.48 cm 3 /g LNMO by adding 2.5 wt% or 5 wt% lithium oxalate, respectively (calculated from electrode compositions given in Table I and bulk densities of 4.4 g/cm 3 for LNMO, 1.8 g/cm 3 for PVDF, 2.2 g/cm 3 for C65 and 2.1 g/cm 3 for lithium oxalate). 55 Accordingly, the oxidation of lithium oxalate leads to a porosity increase from 35% to 38% in the electrodes with 2.5 wt% and to 40% in the electrodes with 5 wt% lithium oxalate. The resulting volumetric energy density, here defined as energy per entire electrode volume including voids, is around 1272 Wh/L electrode at cycle 5 for cells without lithium oxalate and rises about 3% to 1315 Wh/L electrode for cells containing 2.5 wt% lithium oxalate. Cells with 5 wt% lithium oxalate deliver only 1291 Wh/L electrode at cycle 5, as the higher porosity now counteracts the slight increase in gravimetric energy density. This effect can be avoided if electrodes with lithium oxalate are calendered to initial porosities of 32% (2.5 wt% lithium oxalate) or 30% (5 wt% lithium oxalate). In this way, the porosity reaches 35% after lithium oxalate oxidation for all electrodes, and the resulting volumetric energy densities at cycle 5 for electrodes containing 0, 2.5 or 5 wt% lithium oxalate are 1272 Wh/L electrode , 1376 Wh/L electrode or 1412 Wh/L electrode .
Gas evolution in large-format cells.-As for LNMO/graphite cells, the use of more than 2.5 wt% lithium oxalate shows the biggest improvement factor and perhaps is the best compromise between the amount of electrode additive and specific energy retention. Therefore, the following considerations are all based on electrodes containing 2.5 wt% lithium oxalate. The CO 2 evolution from the oxidation of lithium oxalate during formation could be an issue in commercialscale cells due to swelling (in pouch cells) or pressure buildup (in hard-case cells). This is largely related to the fact that in commercialscale cells, the ratio of active materials to electrolyte solution and void volume is typically ∼10 times higher compared to the lab-scale cells used here. 37,56 Strehle et al. 16 recently showed that under these conditions, the majority of CO 2 released from VC reduction would remain dissolved in the electrolyte solution instead of being released into the gas headspace of the cell. This is illustrated by first estimating the amount of dissolved CO 2 by Henry's law 2: p CO 2 (gas) K H = n CO 2 (el) V el c el +n CO 2 (el) [2]  where n CO2(el) is the amount of CO 2 dissolved in the liquid electrolyte solution, V el is the volume of the electrolyte solution, c el is the total molar concentration of the electrolyte solution (i.e., solvent and salt) and K H is the Henry constant of CO 2 in the electrolyte solution in units of pressure. Combining this with the ideal gas law and the assumption that V el c el + n CO2(el) ≈ V el c el , the fraction of CO 2 dissolved in the electrolyte solution can now be given as 3: n CO 2 (el) n CO 2 (total) = V el RTc el V el RTc el + V gas K H [3] where n CO2(total) is the total amount of CO 2 present in the system and V gas is the volume of the cell's gas headspace. Assuming a constant gas volume, as would be the case for a hard-case cell, the pressure buildup can be expressed as 4: p = n CO 2 (total) K H RT V el RTc el + V gas K H [4] On the other hand, in soft pouch cells, gas evolution would typically lead to expansion (or bulging) of the cell. This volume expansion at a given pressure can be calculated by 5: V = RT n CO 2 (total) p − V el c el K H [5] To assess how much pressure buildup or volume expansion would actually occur in a commercial-scale cell containing 2.5 wt% lithium oxalate in the cathode electrode, we use a similar approximation for a commercial-scale 3 Ah cell as shown in ref. 16, where the weight for cathode active material and electrolyte solution were taken from Wagner et al. 57 Furthermore, we also calculate the expected volume expansion for a 180 mAh pouch cell containing ∼ 0.75 mL electrolyte solution as used by Xia et al., 58 assuming a constant pressure of 1 bar in the cell. In both cases, the composite cathode is approximated to consist of 96% active material and 2.5 wt% lithium oxalate. For comparison, the 2032-type coin cells used in this study are also included in this assessment. Table II summarizes the expected pressure buildup and volume expansion for the respective cells. The pressure increase in coin cells is low (∼0.03 MPa), due to the relatively large void space compared to electrolyte solution and cathode active material volume. In a 180 mAh pouch cell, the estimated gas evolution would be ∼9.3 mL at 1 bar, which is about 5 times larger than the gas evolution normally expected for these cells during formation. 58 The pressure buildup in the hard case 18650 cell is ∼1.2 MPa; this causes that 95% of the CO 2 remains dissolved in the electrolyte solution. However, the oxidation of lithium oxalate is completed after the first charge, which means that the gas evolution will stop thereafter. As many commercial-scale cells are vented during or after formation, the high pressure/volume increase is only a matter of the very first cycles. We have further not considered the consumption of CO 2 on the graphite or silicon/graphite A523 anode: Strehle et al. 16 showed that up to 40 μmol (≈1 mL) CO 2 can be consumed per square meter graphite surface area during the first formation cycle, which agrees well with previous reports by Xiong et al. 40 The graphite electrodes used in the present study have a specific surface area of 0.034 m BET 2 /cm geom 2 , hence 40 μmol/m BET 2 would correspond to a CO 2 consumption of 1.4 μmol/cm geom 2 , which is about ∼20% of the evolved CO 2 (6 μmol/cm geom 2 ). As long as the ratio of lithium oxalate to graphite or the specific surface area of the graphite do not change drastically, the same fraction of CO 2 would also be consumed during formation in other cell formats. For siliconbased anodes, the previous OEMS measurements have shown that the CO 2 consumption of SiG anodes during the first charge can be about 19-28% of the theoretically available CO 2 (see Figure 6). Assuming a consumption of 25% CO 2 in the first cycle, the volume expansion in pouch cells would be decreased to ∼6.5 mL, while the pressure rise in 18650 cells would be limited to ∼0.89 MPa. Although a venting of some of the excess gas is probably still required in this case, a complete removal of CO 2 after formation is not desirable, as the amount of CO 2 within the cell should remain high during cycling to benefit from its properties as an SEI-forming additive, as was shown here and in Reference 18.

Conclusions
In this paper, we assessed the use of lithium oxalate as a "sacrificial salt", i.e., a lithium ion donor, in combination with LNMO cathodes. We have shown that the incorporation of 2.5 wt% or 5 wt% lithium oxalate into the cathode electrode increases the first cycle charge capacity by about 10% and 20%, respectively, without affecting the electrochemical performance of the cathode during subsequent cycles. The effect of lithium oxalate and CO 2 released from its oxidation was investigated in LNMO/graphite and LNMO/silicon-graphite (SiG) cells. The former showed increased initial capacity according to the increased pool of active lithium, as well as a higher coulombic efficieny and capacity retention, when lithium oxalate was added to the cathode matrix. For LNMO/SiG cells, a significantly improved cycling stability and coulombic efficiency was found for cells containing lithium oxalate and FEC compared to cells with only FEC but no lithium oxalate, which we ascribe to the beneficial effect of CO 2 on the cycling stability of silicon-based anodes.
By OEMS measurements and analysis of the cumulative irreversible discharge capacity, we can conclude that CO 2 and FEC are simultaneously reduced, following and overall ∼4 e − /FEC and ∼2 e − /CO 2 process. Furthermore, the combination of these two additives is more efficient in terms of deliverable capacity per irreversible capacity and per mol of consumed additive than either of them alone. In this context, the use of lithium oxalate in the cathode matrix is not limited to its use as a "sacrificial salt" in the original sense, but also displays an easy and controllable way to introduce defined amounts of CO 2 into lithium ion cells with graphite or silicon-based anodes.