Revealing electrolyte oxidation via carbonate dehydrogenation on Ni-based oxides in Li-ion batteries by in situ Fourier transform infrared spectroscopy

Understanding (electro-)chemical reactions at the electrode–electrolyte interface (EEI) is crucial to promote the cycle life of lithium-ion batteries. In this study, we developed an in situ Fourier-transform infrared spectroscopy (FT-IR) method, which provided unprecedented information on the oxidation of carbonate solvents via dehydrogenation on LiNi x Mn y Co 1 (cid:2) x (cid:2) y O 2 (NMC). While ethylene carbonate (EC) was stable against oxidation on Pt up to 4.8 V Li , unique evidence for dehydrogenation of EC on LiNi 0.8 Co 0.1 Mn 0.1 O 2 (NMC811) at voltages as low as 3.8 V Li was revealed by in situ FT-IR measurements, which was supported by density functional theory (DFT) results. Unique dehydrogenated species from EC were observed on NMC811 surface, including dehydrogenated EC anchored on oxides, vinylene carbonate (VC) and dehydrogenated oligomers which could diffuse away from the surface. Similar dehydrogenation on NMC811 was noted for EMC-based and LP57 (1 M LiPF 6 in 3:7 EC/EMC) electrolytes. In contrast, no dehydrogenation was found for NMC111 or surface-modified NMC by coatings such as Al 2 O 3 . In addition, while the dehydrogenation of solvents was observed in 1 M electrolytes with different anions, they were not observed on NMC811 electrode form solid electrolyte interphase but lithium-ion conducting enable reversible lithium intercalation, but their possible (electro)chemical reactions on positive electrode materials are poorly understood. Ni-Rich (NMC) materials provide enhanced energy densities but at the expense of shorter cycle life, and recent computational studies have shown that carbonate solvent molecules can be chemically oxidized or dehydrogenated on NMC, especially with increasing Ni content. In this study, we developed an in situ FT-IR method and revealed dehydrogenation pathways of carbonate solvents in the electrolyte (EMC/EC with 1 M LiPF 6 or LiClO 4 ) on composite LiNi 0.8 Mn 0.1 Co 0.1 O 2 (NMC811) electrodes in situ at voltages as low as 3.8 V Li , which can be suppressed by decreasing surface oxygen reactivity of Ni-based oxides through coating or decreasing carbonate activity by using concentrated electrolytes (EMC/EC with 3.1 M LiPF 6 ). These findings highlight the importance of in situ studies to capture reaction intermediates and suggest design strategies for more stable high-energy positive electrode materials.

Multiple reaction mechanisms on carbonate electrolyte oxidation have been proposed, including nucleophilic attack reactions between oxides and carbonate molecules, [31][32][33] electrophilic attack, 34,35 and dehydrogenation reactions, including EC dissociation by breaking C-H bond, 36,37 and dissociation with oxygen vacancy formation. 34 However, recent density functional theory (DFT) results 13 show that EC dissociation on layered Ni-rich oxides is more energetically favorable than the other processes reported. The enhanced oxide-electrolyte reactivity for Ni-rich oxides 10,12,38,39 can be attributed to having more metal-oxygen covalency or more oxygen p states pinned at the Fermi level, 13,40,41 where there is greater driving force for the surface oxygen to dissociate or oxidatively dehydrogenate more carbonate solvents such as EC to form surface protic species (C 3 O 3 H 4 -*C 3 O 3 H 3 + + *H + ) and reduce transition metal ions. 13,42 For example, DFT calculations have shown that the dissociation energetics of EC (BÀ2.6 eV) on NMC811 is thermodynamically favorable. 13 Such proposed dehydrogenation of carbonate solvents is supported by ex situ diffuse reflectance infrared Fourier transform spectroscopy (DRIFTS) and Raman spectroscopy, where EC dissociation on charged NMC is more visible with increasing Ni content in NMC or increasing lithium de-intercalation in NMC811. 13 The protic species produced by dehydrogenation such as surface hydroxyl groups, can further react with LiPF 6 salt, forming lithium nickel oxyfluorides, 14 PF 3 O, 43 Li x PF y O z 14 and HF species, 14 validated by online electrochemical mass spectrometry and X-ray photoelectron spectroscopy (XPS). 14,44,45 Although ex situ techniques have provided information on oxidative products from the electrolyte that remain on oxide surfaces, soluble species and dynamic changes of surface reaction intermediates formed as a function of potential are not captured. 46 In situ FT-IR measurements would be ideal to probe the reactivity between organic electrolytes and the oxide surface because it is highly sensitive to covalent bonds such as C-O and CQO bonds. 47 While there are a few in situ FT-IR techniques previously reported in understanding the reduction [48][49][50] or oxidation 51,52 stability of organic electrolytes in lithium-ion batteries, [48][49][50][51] it is not straightforward to apply these methods to study oxide powders, especially Ni-rich NMCs used in practical Li-ion batteries. For example, Kanamura et al. 52 have developed a method for thin-film electrodes such as LiFePO 4 53 and LiCoO 2 54,55 sputtered on a gold plate and placed on the IR window with a micrometer, where the whole cell is flooded with electrolyte. [52][53][54][55] These authors have observed carboxylate groups from propylene carbonate above 4.0 V Li during cyclic voltammogram of the LiCoO 2 thin-film electrodes, 54 while the exact oxidation products and the mechanisms were not well understood. Unfortunately, this method with thin film setups is not applicable to various composite electrodes of oxide powders due to the difficulty in synthesizing thin film electrodes for Li-containing Ni-rich NMCs. 56  In this study, we have designed a new in situ FT-IR setup to examine the parasitic reactions between carbonate electrolytes and composite NMC powder electrodes used in practical Li-ion batteries as a function of voltage (or time) during galvanostatic (or potentiostatic) measurements. We first study electrolytes containing LiPF 6 in EC using in situ FT-IR spectra, to examine its oxidation stability upon linear voltage sweep without NMC oxides, and their oxidation on NMC811 during charging, respectively. The spectra assignments are validated by DFT simulated spectra. We then study LiPF 6 in EMC, and LP57 electrolyte (1 M LiPF 6 in 3 : 7 EC/EMC), and examine their reactivity with NMC811 and NMC111 as a function of voltage. Next, we correlate the oxidation of different electrolytes by carbonate dehydrogenation on NMC811 and NMC111 with charge transfer impedance measured from electrochemical impedance spectroscopy (EIS). Finally, we propose the pathways and mechanisms for carbonate-based electrolyte decomposition on NMC that account for impedance growth and capacity fading. Through the in situ spectra on NMC surfaces, we track the interfacial reactions and understand the electrolyte decomposition mechanism on composite NMC powder surfaces in practical battery operations during galvanostatic charging, and interpret the strategies for EEI design by tuning electrode reactivity through coating materials, and tuning electrolyte activity by a concentrated electrolyte (EMC/EC with 3.1 M LiPF 6 ).

Results and discussion
In situ FT-IR measurements of NMC811-mediated oxidation of EC upon charging While EC with 1.5 M LiPF 6 was stable against (electro)chemical oxidation up to 4.8 V Li on Pt, species derived from dehydrogenation of EC, including dehydrogenated EC (de-H EC, one hydrogen removed), vinylene carbonate (VC, two hydrogens removed), and oligomers with EC-like rings, were detected in situ upon charging NMC811 at voltages as low as 3.8 V Li . FT-IR spectra were first collected in situ at selected voltages with Pt only as the positive electrode, as the voltage of Li/Pt cells was linearly swept from open circuit voltage (OCV) at 3.1 V Li to 4.8 V Li (Fig. S1b, ESI †). Since the CQO stretching region gives the strongest signal and is the most sensitive to molecule structure changes, the focus is to probe this CQO stretching region. Two pronounced peaks in the CQO region (1900 to 1700 cm À1 ) were revealed (Fig. S1c, ESI †), corresponding to CQO stretching in EC (at B1800 cm À1 ) and Li + -coordinated EC (B1773 cm À1 ), 14,49 respectively. This assignment is in agreement with attenuated total reflection (ATR) measurements of 1.5 M LiPF 6 in EC (Fig. S1a, ESI †) and DFT-simulated spectra (Fig. 1a). Voltage-dependent difference spectra obtained by subtracting each spectrum (in Fig. S1c, ESI †) by the spectrum at OCV revealed no new peaks appearing upon charging to 4.8 V Li (Fig. S1d, ESI †), indicating no significant electrolyte oxidation. This result is consistent with previous theoretical [60][61][62] and experimental results [63][64][65] for linear sweep voltammograms showing that carbonate electrolytes are stable against oxidation below 5 V Li on inert metals.
Unfortunately, upon charging NMC811 from OCV to 4.4 V Li (Fig. 1b), significant changes were observed in the difference spectra ( Fig. 1c) obtained from IR spectra collected in situ (Fig. S2, ESI †), indicative of oxidative instability. Broad intensities in the wavenumber range from 1850 to 1750 cm À1 were found to grow significantly from 3.8 V Li to 4.4 V Li , where individual peaks between 1830 and 1800 cm À1 became resolved at 3.8 V Li and grew in intensity. In addition, two small, sharp peaks at B1773 and B1763 cm À1 became visible upon charging from 3.9 V Li to 4.4 V Li . Upon further oxidation from 4.4 V Li to 4.8 V Li (Fig. S3, ESI †), the difference spectra remained unchanged. The peak at B1830 cm À1 that appeared at 3.8 V (Fig. 1c) could be assigned to VC, having two hydrogens removed from EC to form a CQC bond in the ring, which assignment was supported by experimental ATR spectrum of VC-containing 1.5 M LiPF 6 in EC (Fig. 1c), and calculated spectrum of VC by DFT (Fig. 1a). The peaks at B1800 cm À1 and B1773 cm À1 came from the CQO stretching in EC and Li + -EC, respectively, and the peak intensity of EC grew with increasing voltage, possibly because bulk EC from electrolyte were attracted to surface during the adsorption and oxidation of EC on surface which could create a concentration gradient of EC in the electrolyte. The increased intensity ratio of VC to EC from 3.8 V Li to 4.8 V Li (Fig. S3c, ESI †) indicate more VC formed on the surface with increasing voltage. In addition, the peaks in the broad feature between 1820 to 1810 cm À1 and the peak at B1763 cm À1 (black arrow in Fig. 1c) can be attributed to small oligomers with EC-like rings, which could come from EC ring opening and polymerization. [66][67][68] This assignment is supported by the DFT spectra (Fig. S3d, ESI † and Fig. 1a) of oligomers including C 6 H 8 O 6 (computed 1813 and 1755 cm À1 ), C 9 H 14 O 8 (computed 1819 and 1757 cm À1 ), and C 7 H 10 O 6 (computed 1813 and 1750 cm À1 ). Further support came from timedependent intensity measurements at OCV in Fig. 1e, which will be discussed below. Moreover, the feature around 1813 cm À1 (convoluted between 1820 to 1810 cm À1 , black arrow in Fig. 1c) that appeared the earliest at B3.8 V Li could also contain CQO stretching of dehydrogenated EC (de-H EC), having one hydrogen removed from EC and whose formation was energetically favorable with driving force around 2.6 eV (Fig. 1f). 13,14 This peak position of de-H EC fell between EC and VC, which is in agreement with DFT ( Fig. 1a), and the assignment was further supported by timedependent intensity measurements at OCV in Fig. 1e to be discussed later. The deconvoluted peak areas of VC and oligomers were found to grow comparably with increasing voltage, indicating that they were gradually generated from EC and/or de-H EC, while the peak area of de-H EC grew less with increasing voltage (Fig. 1b), which might due to its consumption to further generate VC and oligomers. In summary, the appearance and growth of new peak features at B1830, B1820 to 1810 cm À1 , and B1763 cm À1 provided direct evidence of electrolyte oxidation via EC dehydrogenation on charged NMC811 starting at voltages as low as B3.8 V Li , where EC could dissociate on electrode surface to remove one hydrogen and generate de-H EC, further remove another hydrogen to form VC, or combine with another EC to form oligomers.
While no significant changes were noted for the CQO stretching region (1900 to 1700 cm À1 ) in the difference spectra during voltage holding at 4.4 V Li (Fig. 1d), most peaks were reduced in intensity with increasing time upon OCV following the 4.4 V Li voltage hold ( Fig. 1b and e). These potentiostatic and open circuit voltage measurements were conducted followed by charging to 4.4 V Li to examine whether the species could stick on surface or could diffuse or dissolve away from surface. The spectrum at OCV (after 60 min) could be splitted into four features, VC at B1830 cm À1 , de-H EC at B1813 cm À1 , and oligomers at B1820 cm À1 (and 1763 cm À1 ) and EC at 1800 cm À1 . The peaks of VC, oligomers and EC were reduced rapidly with increasing time at OCV, indicating these species diffused away from charged NMC811 into the electrolyte. Of significance to note that VC intensity was reduced faster than oligomers and EC, indicative of greater diffusivity or solubility, which is in agreement with the fact that VC was not detected in the ex situ DRIFT spectra. 14 In contrast, the peak intensity of de-H EC (B1813 cm À1 ) did not change significantly with increasing time (Fig. 1b and e), and it became the most dominant feature after resting for 60 minutes, suggesting de-H EC was anchored on the oxide surface, presumably by a C-O surface bond formed during EC dehydrogenation on surface oxygen of NMC811 13 (Fig. 1f), which is in agreement with the observation of this species in ex situ DRIFT spectra of charged NMC811 electrodes dried and removed from electrochemical cells. 14

View Article Online
Next we discuss NMC811-mediated oxidation of EC in LP57 (1 M LiPF 6 in EC/EMC) -a commonly used electrolyte for Li-ion batteries. While LP57 (1 M LiPF 6 in EC/EMC) was stable against oxidation on Pt, in situ FT-IR spectra revealed that EC was dehydrogenated to form de-H EC, vinylene carbonate (VC), and oligomers during galvanostatic charging similar to those discussed earlier for 1.5 M LiPF 6 in EC. ATR measurements of bulk LP57 had four features (Fig. S5a, ESI †), including EC

View Article Online
(B1800 cm À1 ), Li + -coordinated EC (B1773 cm À1 ), EMC (B1750 cm À1 ) and Li + -coordinated EMC (B1718 cm À1 ). On Pt, no new species appeared in the difference spectra during linearly voltage sweeping from OCV (B3 V Li ) to 4.8 V Li (Fig. S4, ESI †), indicative that the electrolyte was stable up to 4.8 V Li , which is in consistent with having both 1.5 M LiPF 6 in EC ( Fig. S1, ESI †) and 1 M LiPF 6 in EMC (Fig. S5, ESI †) being stable on Pt. In contrast, spectra collected on NMC811 from OCV (3 V Li ) to 4.8 V Li (Fig. S6, ESI †) was subtracted from the OCV spectrum to yield difference spectra (Fig. 2b), which revealed new peaks of EC-derived dehydrogenation species at voltages greater than 3.8 V Li , including de-H EC (B1813 cm À1 , greater In situ FT-IR cell was galvanostatically charged at 27.5 mA g À1 . In situ FT-IR difference spectra in CQO stretching region (in red) on (b) NMC811 and (d) NMC111 surface during charging in LP57 electrolyte, and ATR spectra (in black) for LP57 and LP57/VC solutions (1 : 1). On NMC811 surface, de-H EC (1813 cm À1 , marked in arrow), VC (B1830 cm À1 ), and oligomers with EC-like rings (B1820 and B1763 cm À1 , marked in arrows) arose in the spectra upon charging, while no obvious peaks appeared for NMC111 during first cycle charging. View Article Online than 3.9 V Li ), VC (B1830 cm À1 , greater than 4.1 V Li ), and oligomers with EC-like rings (B1820 and B1763 cm À1 , greater than 4.1 V Li ) similar to Fig. 1. Changes in the difference spectra unique to the dehydrogenation of EMC during galvanostatic charging of NMC811 was not observed, which can be interpreted by the convolution of de-H EMC (discussed below in Fig. 3) with bulk EMC and with EC-derived oligomers, and/or having EC preferentially adsorbed on the NMC811 surface than linear alkyl carbonate EMC 69,70 due to its higher dielectric constant. 61 Experimental sum frequency generation results show that EC occupies over 90 mole% on lithium metal oxide (LiCoO 2 ) surfaces in mixed carbonate solutions such as EC + DEC or EC + DMC, much higher than that in the bulk solution (43 mole%). 70 After charging, we also discharged the cell and the difference spectra (Fig. S7, ESI †) showed that upon discharging from 4.8 V Li to 2 V Li , VC peak (1830 cm À1 ) and oligomer peaks (1820 and 1763 cm À1 ) gradually decreased in intensity, which agrees with the results during OCV (Fig. 1e) that VC and oligomers could diffuse or dissolve away. De-H EC still anchored on the surface, which is also in agreement with the results at OCV. Another interesting finding is that Li + -EC (1773 cm À1 ) peaks gradually grew larger, which was not observed during OCV, and this was probably because during discharge, Li + migrated from bulk electrolyte to oxide surface to be intercalated, making more EC to become coordinated. As EC was found to oxidize by dehydrogenation upon charging at  voltages as low as 3.8 V Li , we further examine and discuss NMC811-mediated oxidation of 1 M LiPF 6 in EMC electrolyte to better understand the different reactivity of EC and EMC.
In situ FT-IR measurements of NMC811-mediated oxidation of EMC upon charging While EMC with 1 M LiPF 6 was stable against oxidation to 4.8 V Li on Pt, in situ FT-IR spectra revealed that EMC became dehydrogenated (de-H EMC) on NMC811 with increasing voltage during galvanostatic charging (Fig. 3a). Ex situ ATR spectrum of pristine 1 M LiPF 6 in EMC revealed two features of CQO stretching at B1750 and B1718 cm À1 (Fig. 3b and Fig. S5a, ESI †), which can be assigned to EMC and Li + -coordinated EMC, respectively. 14 (Fig. 3a) revealed that the peak of EMC at B1750 cm À1 shifted towards higher wavenumbers with increasing voltage (Fig. 3b), which could be assigned to de-H b EMC, indicative of EMC oxidation by EMC dehydrogenation. This assignment, having one b-site hydrogen from carbon removed and a C-O surface bond formed on the oxide in Fig. 3c, is supported by DFT computed spectra (Fig. 3d). As both EC and EMC (in 1.5 M LiPF 6 in EC, 1 M in EMC and 1 M in EC/EMC mixture -LP57) were found to oxidize by dehydrogenation on NMC811 upon charging, we further examine the oxidation of LP57 by replacing NMC811 with NMC111, NMC622, and Al 2 O 3 -coated NMC622 upon charging to 4.8 V Li .
In situ FT-IR measurements of NMC111-and NMC622mediated oxidation of LP57 upon charging The dehydrogenation tendency of EC in LP57 on Ni-based oxide electrode decreased as decreasing metal-oxygen covalency 13,72 or moving the Fermi level away from the oxygen p band center as predicted by DFT. 13,73 Unlike NMC811, where EC was dehydrogenated upon charging to voltages as low as 3.8 V Li in the first cycle, no change was observed upon charging NMC111 to 4.8 V Li in the first cycle and subsequent two cycles. No new peaks were found for difference spectra collected for NMC111 upon the first galvanostatic charging from OCV to 4.8 V Li ( Fig. 2c and d), indicative of no obvious solvent oxidation upon first cycle charging for NMC111, which is in agreement with recent ex situ FT-IR studies. 14 In the subsequent third cycle, still no new peaks were found for difference spectra collected for NMC111 upon galvanostatic charging from OCV to 4.8 V Li (Fig. S9, ESI †). The difference in the dehydrogenation of EC or oxidation of LP57 between NMC811 and NMC111 observed in the in situ FT-IR measurements provides further experimental evidence to support previous DFT results, where the driving force for EC dehydrogenation on NMC811 with greater metal-oxygen covalency and more oxygen 2p states pinned at the Fermi level is much greater than that on NMC111. 14,15,74,75 While NMC622 did not show any evidence for carbonate dehydrogenation during the 1st cycle (Fig. S10, ESI †), dehydrogenated EC products including de-H EC (B1813 cm À1 , marked in black arrow), VC (B1830 cm À1 ), and oligomers with EC-like rings (B1820 and B1763 cm À1 , marked in black arrows) were detected in the FT-IR difference spectra obtained upon the third charge ( Fig. 4a and b). In contrast, the dehydrogenation of LP57 was not detected for NMC622 coated with Al 2 O 3 in the 3rd charging ( Fig. 4c and d). This difference in the dehydrogenation of LP57 between coated and uncoated NMC622 demonstrates the effect of surface ''inert'' coating layers 76 [77][78][79]83 can be attributed to enhanced stability against dehydrogenation compared to electrodes without coatings. Next, we further examine the influence of carbonate solvent activity on the oxide-mediated electrolyte oxidation by changing electrolyte salt concentrations and salt anions.
In situ FT-IR measurements of NMC811-mediated oxidation of concentrated and non-PF 6 À containing electrolytes on NMC811 upon charging Increasing the salt concentration in the electrolyte was found to reduce NMC811-mediated dehydrogenation of EC upon charging. Ex situ ATR spectra (Fig. S13, ESI †) of EC/EMC with 1 M, 2 M, and 3.1 M LiPF 6 revealed that increasing the salt concentration resulted in more Li + -coordinated solvent molecules (Li + -EC/EMC) but fewer free EC/EMC solvent molecules (low carbonate activity), in agreement with previous work. [84][85][86][87] NMC811 charged in the concentrated electrolyte (3.1 M LiPF 6 in EC/EMC), which contained fewer free EC or free EMC (molar fraction of free solvent in general concentrated electrolyte with dissociative salt is reported to be less than 10%), 88-92 did not show dehydrogenation or oligomerization products from EC or EMC in the FT-IR difference spectra (Fig. 5a) upon charging to 4.8 V Li , unlike in LP57 (with 1 M LiPF 6 ) where EC became dehydrogenated as discussed before (Fig. 2b). The overpotential in Fig. 5a  . 101 With the different salt (1 M LiClO 4 in 3 : 7 EC : EMC) but the same concentration (1 M) as LP57, NMC811 also showed dehydrogenation products from EC during the first charging in EMC/EC/1 M LiClO 4 . VC (B1830 cm À1 ) appeared from 3.8 V Li , and the intensity increased as it was charged to 4.8 V Li (Fig. 5d), similar to LP57 (with 1 M LiPF 6 ) case. De-H EC Fig. 4 In situ FT-IR measurements on NMC622 surfaces with and without surface coating upon the 3rd cycle charging to 4.8 V Li in LP57. Voltage profile during charging of (a) NMC622, (c) NMC622 with Al 2 O 3 coating. In situ FT-IR cell was galvanostatically charged at 27.5 mA g À1 . In situ FT-IR difference spectra in CQO stretching region (in red) on (b) NMC622, (d) NMC622 with Al 2 O 3 coating electrodes during the 3rd cycle charging in LP57 electrolyte, and ATR spectra (in black) for LP57 solutions and LP57/VC solutions (1 : 1). On uncoated NMC622 surface, de-H EC (1813 cm À1 , marked in arrow), VC (B1830 cm À1 ), and oligomers with EC-like rings (1820 and 1763 cm À1 , marked in arrows) arose in the spectra upon charging, while no obvious peaks appeared for coated NMC622 even during 3rd cycle charging. (B1813 cm À1 ) and oligomer (B1820 cm À1 ) features were convoluted between the VC (B1830 cm À1 ) and EC (B1800 cm À1 ) features, and they were less observable than in LP57 (Fig. 2b)

View Article Online
Li + coordinated carbonate, giving lower energetic barrier to free carbonate to be decomposed on LiMO 2 surface. 75,87 The observation of VC in both EMC/EC with 1 M LiClO 4 and LP57 confirmed the universal dehydrogenation of EC in 1 M electrolytes, which had higher carbonate activity or lower oxidative stability 103 than concentrated electrolytes (3.1 M).

Correlating impedance growth with carbonate solvent dehydrogenation
While in situ FT-IR spectra revealed the dehydrogenation of solvents on NMC in situ as a function of voltage/time, we employed EIS (Fig. S16-S22, ESI †) to further reveal that the charge transfer and EEI impedance on NMC811 grew much more significantly than on NMC111 with increasing voltage, which could be attributable to dehydrogenated products on NMC811 surfaces (Fig. 2a and 6). NMC111, NMC622 with surface coating, and NMC811 in the concentrated electrolyte (3.1 M LiPF 6 in EC/EMC) exhibit no significant impedance growth in contrast to NMC811 in LP57 (Fig. 6). EIS on NMC811 positive electrode collected at 3.9 V Li , 4 V Li and up to 4.6 V Li during the 1st cycle charging in LP57 showed huge growth in the low-frequency semicircle (right-hand side semicircle in Fig. 6a), which can be attributed to the charge transfer and EEI resistance (R CT+EEI , low-frequency resistance) 104 (Fig. 6c), growing from B10 to over 300 Ohm during charging. The R CT+EEI impedance at 4.6 V Li was beyond the detected amplitude in this frequency range (down to 100 mHz) because it was almost at the end of charge. Similarly, for NMC811 in other nonconcentrated electrolytes including EC with 1.5 M LiPF 6 , EMC with 1 M LiPF 6 , and EMC/EC with 1 M LiClO 4 , R CT+EEI all grew dramatically from B10 Ohm to B400 Ohm (Fig. 6c). The difference in R CT+EEI between EC with LiPF 6 , EMC with LiPF 6 , and LP57 at each voltage could come from the different dielectric constants 105 and physical properties such as ionic conductivity and viscosity (Table 1) which could affect the solvation and de-solvation processes during charge transfer. This great increase of R CT+EEI upon NMC811 charging in all four 1 M or 1.5 M electrolytes can be correlated with the dehydrogenation and/or oligomerization products detected by in situ FT-IR reported above for 1 M or 1.5 M electrolyte cases (Fig. 1c, 2b, 3b and 5d), indicating that these organic products likely formed a resistive layer on NMC811 which passivated the surface and impeded charge transfer. The comparable impedance growth and similar dehydrogenation reactions between the PF 6 À -based electrolyte and ClO 4 À -based electrolyte noted for NMC811 further supports the hypothesis that solvent (carbonate) decomposition contributed most to the impedance growth, rather than commonly-perceived salt decomposition. Although LiPF 6 salt could be attacked by surface protic species to form metal fluorides (MF) 14,55,106 and resistive LiF, 32 our result of a non-PF 6 À electrolyte revealed that even without severe salt decomposition, solvent dehydrogenation itself could account for great impedance growth on NMC811 in the ClO 4 À -based electrolyte.
In contrast, NMC811 in the concentrated electrolyte did not exhibit as large impedance growth and R CT+EEI remained around 30-100 Ohm (Fig. 6c), which could be also be correlated with the more stable interface and lack of dehydrogenation from FT-IR result (Fig. 5a). This observation further supports the role of undesirable carbonate dehydrogenation in passivating the surface and increasing interfacial impedance, suggesting that lowering free solvent activity could facilitate stable cycling of NMC811. For NMC111 and NMC622, R CT+EEI stayed less than 50 Ohm through all the voltages (Fig. 6b and c) during the first cycle, which matched well with in situ FT-IR spectra that there was no profound solvent dehydrogenation in the first charging. The more stable interface and lower impedance growth is also in agreement with the better cycling stability of NMC111 and NMC622 than NMC811. 11,107 Mechanistic discussion on electrolyte oxidation on Ni-rich NMC We propose detailed mechanisms of electrolyte oxidative decomposition on layered Ni-rich metal oxide positive electrodes (Fig. 7) based on the species detected by in situ FT-IR experiments and their correlated contribution to EEI impedance growth and capacity loss. EMC molecules can dissociate b-site hydrogen (Fig. 3d) and adsorb on oxide surface through a C-O surface bond, revealed by a blueshifted shoulder at B1757 cm À1 in Fig. 3b. The reaction is energetically favorable, which is closer to the dissociation on LiNiO 2 surface with DE EMC dissociation = À2.7 eV calculated from DFT (Fig. 3c). EC molecules also dehydrogenate and form de-H EC that bonds with surface oxygen, revealed by the feature at B1813 cm À1 (Fig. 1). EC dissociation is also energetically favorable, with DE EC dissociation close to À2.6 eV calculated from DFT (Fig. 1f). The driving force becomes even greater during delithiation and decreasing of x in Li x MO 2 . 14,15 EC can also further dehydrogenate a second hydrogen on the other carbon and form VC, which corresponds to the peak at B1830 cm À1 (Fig. 1c  and 2b). Another pathway is that de-H EC could be further oxidized to open the ring, or form oligomers with EC-like rings such as C 6 H 8 O 6 , C 9 H 14 O 8 and C 7 H 10 O 6 , revealed by the features at B1820 and B1763 cm À1 (Fig. 1c). The solvent dehydrogenation products may also be eventually oxidized to CO 2 or CO. 107 Organic products from carbonate decomposition can form a passivating layer on the oxide surface, through which Li + needs to migrate to reach the oxides and undergo charge transfer, 32 and therefore R CT+EEI shows great impedance growth during charging of NMC811 measured by EIS (Fig. 6c). The dehydrogenation of EC and EMC also generates protic species on the surface, 13,107,108 which can further trigger reactions 106 with the widely used LiPF 6 salt to form HF 14,108 and less-fluorinecoordinated species such as Li x PF y O z , 14,78 transition metal fluorides (MF) 32 and PF 3 O. 109,110 Salt decomposition might also contribute to EEI impedance, but our comparable results between LP57 (EMC/EC with 1 M LiPF 6 ) and EMC/EC with 1 M LiClO 4 show that solvent decomposition plays a much larger role. Combining in situ FT-IR spectra and EIS reveals that when there is undesirable dehydrogenation of solvents, there is great interfacial impedance growth at higher voltage, eventually leading to the loss of capacity when NMC811 is cycled to high voltages. 107 On the NMC111 surface, which is less reactive, solvent is more stable against dehydrogenation, and the undesirable reactions to form passivating layers barely happen, leading to lower impedance and better capacity retention. The reactivity of NMC622 is between that of NMC811 and NMC111, so it did not yield dehydrogenated EC visibly during the first cycle, but gradually caused dehydrogenation in following cycles (such as the third cycle analyzed above). The understanding of the electrolyte decomposition pathway suggests strategies to design a more stable interface, based on the principle of eliminating undesirable solvent dehydrogenation reactions that can be triggered followed by oligomerization and salt decomposition. This can be achieved by tuning the surface chemistry of electrode materials, e.g. using inert surface coating, such as ceramic thin films, which reduce the driving force for solvent dissociation. Although NMC622 yielded EC dehydrogenation after 3 cycles, NMC622 with Al 2 O 3 ) decomposition. Solvent decomposition happened by dehydrogenation first, and de-H EC could be further decomposed by removing another hydrogen or by oligomerization. The protic species on surface coming from dehydrogenation could further attack PF 6 À and lead to coupled salt decomposition. 14 The decomposed species (mainly solvent decomposed species) would form a resistive layer at the EEI and lead to great impedance growth, eventually resulting in capacity loss for NMC811. increase cycling stability and capacity retention. Another strategy to design a stable interface is to tune the solvent activity, e.g. by decreasing the number of free solvent molecules, which are less oxidatively stable than the molecules in the solvation shells. 84,85,97 While NMC811 charged in non-concentrated electrolytes (1 M or 1.5 M), all showed undesirable solvent dehydrogenation and/or oligomerization, no obvious decomposition products were observed in concentrated electrolyte containing fewer free solvent molecules and the interface impedance didn't grow as much as 1 M or 1.5 M cases, which could promote the cycling stability of NMC811.
Other approaches towards suppressing the undesirable solvent decomposition include adding additives in the electrolytes, 111,112 with the idea of sacrificing additives to be oxidized rather than solvents. Interfacial stability can also be promoted by new types of solvents that are less likely to dissociate on oxides and new types of salts that are more stable and less likely to be attacked by protons. Exploring such strategies, including various electrolyte additives, will be the subject of our future work. The method of in situ FT-IR is also applicable to study the SEI formation on anodes, which is also within our plan to extend this method.

Conclusions
In this study, we developed an FT-IR method, which allowed the in situ studies on the reactivity of the electrolyte on NMC surfaces as a function of voltage. While ethylene carbonate (EC) remained stable against (electro)chemical oxidation on Pt up to 4.8 V Li , we found unique evidence for dehydrogenation of EC on LiNi 0.8 Co 0.1 Mn 0.1 O 2 (NMC811) surface at voltages as low as 3.8 V Li . Three unique dehydrogenated species from EC were observed on the NMC811 surface, which included dehydrogenated EC (de-H EC, anchored on oxides), vinylene carbonate (VC), and dehydrogenated oligomers with EC-like rings, while the latter two can diffuse away from the NMC811 surface into the electrolyte. These observations indicate that electrolyte oxidation on NMC811 might not generate a protective film unlike the solid electrolyte interface (SEI) formed by EC reduction on graphite. Similar dehydrogenation was observed for EMC-based and LP57 electrolytes on NMC811. In contrast, no dehydrogenation was found for NMC111 or modified NMC surface by coatings such as Al 2 O 3 . Such oxidative dehydrogenation tendencies on the oxide chemistry is in good agreement with the driving force for EC dehydrogenation on surface oxygen of NMC predicted by recent density functional theory (DFT) calculations, which increased with lower Fermi level into the oxygen p band of oxides associated with greater Ni and less lithium in NMC. Interestingly, the dehydrogenation of EC and EMC was observed in different anions in the 1 M electrolytes, but not observed in the concentrated electrolyte (EC/EMC with 3.1 M LiPF 6 ) on NMC811, indicating that the dehydrogenation of carbonates can be suppressed by increasing lithium coordination with solvents. Dehydrogenation of carbonates (EC and EMC) on the NMC811 surface was found to accompany with rapid growth of interfacial impedance with increasing charging voltage measured from EIS. In contrast, those electrodeelectrolyte combinations without dehydrogenation were found to have no significant impedance growth. Therefore, it is crucial to minimize carbonate dehydrogenation on the NMC surface by tuning electrode reactivity and electrolyte reactivity for improving the cycle life and high energy of lithium-ion batteries.

Experimental methods
Materials and electrode preparation. The composite electrode for in situ FT-IR was composed of NMC (85 wt%, from Ecopro and Umicore) as the active material, carbon black (5 wt% KS6 and 2 wt% Super P, both from Timcal) as an electrically conductive carbon, and poly(vinylidene fluoride) (8 wt%, PVDF, Kynar) as the binder. These materials were mixed thoroughly with N-methylpyrrolidone (NMP, Aldrich) in a 1 : 15 mass ratio, using a planetary centrifugal mixer (THINKY AR-100). The obtained slurry was drop-casted onto glassy fiber substrates (Whatman 934-AH, 10 mm in diameter) and dried at 100 1C. Then the composite composites were compressed at 0.5 T cm À2 using a hydraulic press to improve electrical conductivity. Finally the electrodes were completely dried at 120 1C under vacuum for 24 h. The active material loading was B6.5 mg cm À2 , but it is noted here that the effective loading is smaller because some particles could permeate into the glassy fiber and not all of the loading is on the top surface.
The positive electrode for EIS experiments was composed of 85 wt% of NMC, carbon black (5 wt% KS6 and 2 wt% Super P, both from Timcal), and poly(vinylidene fluoride) (8 wt%, PVDF, Kynar). The mesh reference electrode was also composed of 80 wt% of Li 4 Ti 5 O 12 (Itasco, 499.5%), 10 wt% of acetylene black (C-55, Chevron) and 10 wt% of PVDF. These materials were mixed together and thoroughly agitated in NMP. The obtained slurry was applied with a blade applicator onto aluminum foil (for NMC, 16 mm thickness) or 316 stainlesssteel mesh (for Li 4 Ti 5 O 12 , 325 Â 325 mesh, opening size 0.0017 00 ), and resulting sheet/mesh were dried at 70 1C. Next, each electrode was punched with a 1/2 inch diameter (1.27 cm +) for NMC and 18 mm diameter for Li 4 Ti 5 O 12 mesh reference. The NMC composite were compressed at 6.3 T cm À2 using a hydraulic press. All of the electrodes were further dried in vacuum at 120 1C prior to cell assembling. The active material loading was B2.7 mg cm À2 for NMC and B1 mg cm À2 for Li 4 Ti 5 O 12 respectively. Particle size of NMC111, NMC622 and NMC811 was examined with a scanning electron microscope (JEOL 5910, with secondary electron detector at accelerating voltage of 15 kV) and shown in Fig. S23 (ESI †) (d = 5-10 mm).
In this study, the electrolytes include 1 M LiPF 6 in a 3 : 7 wt:wt ethylene carbonate (EC) : ethyl methyl carbonate (EMC) (LP57, battery-grade, BASF), 1.5 M LiPF 6 (499.99%, batterygrade, Aldrich) in EC (battery-grade, BASF), 1 M LiPF 6 in EMC Comprised composite NMC/glassy fiber is the positive electrode, and a lithium metal foil is the negative electrode, separated by two pieces of polypropylene separator (2325 Celgard) or Whatman GF/A separator (for 1.5 M in LiPF 6 /EC and 3.1 M LiPF 6 in EMC/EC), with 100 mL of electrolytes. The lithium foil was covered by a stainless steel plate as a spacer, and a spring was compressed by a cap on top to reduce composite electrode electronic resistance. 104 In this setup, NMC on the glassy fiber was placed downward and faced towards the prism, so that the IR signals could collect the surface information on NMC particles. Electrolytes could permeate through the glassy fiber, so that the diffusion is not restricted.
After assembly, the cell was first rested for 6 hours. Next, galvanostatic charge were performed using BCS-COM (Biologic) and VMP3 (Biologic). Electrochemical behavior of the electrodes in the in situ FT-IR spectro-electrochemical cell was confirmed by galvanostatic measurements above (27.5 mA g À1 ). The actual or effective loading which could be charged might be smaller than measured: during electrode preparation (dropcasting) some particles could permeate through the glassy fiber substrate and stay on the other side of the substrate, and they may not be charged actually. Fig. S24 (ESI †) summarizes the charging time to reach 4.8 V Li during the first cycle for different electrode-electrolyte combinations in this paper, which shows comparable SOCs (8.2 hours on average with same charging rate at 27.5 mA g À1 , variance within 20%, so the SOC reaches around 82%). For the measurements without oxides, sputtered Pt was the positive electrode and linear sweep voltammetry was performed.
At the same time during charging or linear voltage sweeping, in situ FT-IR measurements were performed on a Tensor II (Bruker) FT-IR equipped with deuterated triglycine sulfate (DTGS) detector inside an argon-filled glovebox. The FT-IR spectra were acquired in the single-reflection mode using an attenuated total reflection (ATR) accessory (Pike Vee-Max II, Pike Technologies) at an incident angle of 50 degrees. The spectral resolution was 4 cm À1 and the scan velocity was 1.6 kHz. Each spectrum was measured by superimposing 32 interferograms. All spectra were presented in the form of absorbance according to log(I 0 /I 1 ), where I 0 and I 1 are the spectrum of background (blank Pt surface without electrolytes) and in situ spectrum of the sample, respectively.
For ex situ measurements on pristine electrolytes, the spectra were acquired in the ATR mode using a germanium (Ge) prism (Pier optics) at an incident angle of 50 degrees. Spectral settings were the same as in situ measurements. Although we assign the observed species based on DFT and solution ATR spectra, other species such as semicarbonates cannot be excluded, because there could be other features in the convoluted bands.
EIS measurements. EIS was measured using a threeelectrode cell. The three-electrode cell was assembled in an argon-filled glove box with a Li metal foil (15 mm +), 2 pieces of Celgard 2325 (19 mm +) as the separators, Li 4 Ti 5 O 12 mesh reference electrode (18 mm +), 2 pieces of Celgard 2325 (19 mm +) again, and NMC composite electrode (1/2 inch +) from bottom to top, where a mesh Li 4 Ti 5 O 12 reference electrode was placed between positive and negative electrode with two separators. Detail cell configuration can be found in previous paper. 104 200 mL of electrolyte was added to the cell. For the test with 1.5 M LiPF 6 /EC solution, 1 piece of Whatman GF/A (19 mm +, dried at 150 1C in vacuum overnight prior to use) was used as separator instead of 2 pieces of Celgard 2325 due to wettability problem. Mesh reference electrode was used to avoid inhomogeneous electric field during EIS measurement, which is known to cause artifactual EIS response (e.g. ''spiral'' behavior on Nyquist plots with Li rod reference electrode). 104,[113][114][115][116] Galvanostatic and potentiostatic charge and EIS tests were performed using VMP3 (potentiostat with frequency response analyzer, Biologic) with thermally equilibrated by thermostat chamber (SU-241, Espec) at 25 1C. After cell assembly, Li 4 Ti 5 O 12 mesh reference electrode was electrochemically lithiated (negatively polarized at constant current of 500 mA against Li metal counter electrode with cut-off voltage of 1.3 V Li ) and got stable reference electrode potential at 1.56 V Li . 114 Then NMC working electrode was charged with different end-of-charge potential (3.9-4.6 V Li ) at 27.5 mA g À1 (0.1C rate based on theoretical capacity of 275 mA h g À1 , which corresponds to full lithium deintercalation: LiNi x Mn y Co 1ÀxÀy O 2 -Li + + e À + Ni x Mn y Co 1ÀxÀy O 2 , 27.5 mA g À1 corresponds to B70 mA cm À2 for average loading density of B2.7 mg cm À2 ), hold end-of-charge potential for 1 hour, and relax for 1 hour. After relax, EIS measurements were carried out at open circuit potential with 10 mV amplitude and frequency range from B10 À1 to 10 6 Hz. Obtained EIS data (excluded very high frequency region 4100 kHz if needed) were analyzed using ZView2 (Scribner).

View Article Online
Viscosity, density and conductivity measurements. The viscosities and densities of the electrolytes were measured by using Stabinger viscometer (SVM3001, Anton Paar). Ionic conductivity was measured by using the complex impedance method in the frequency range of 10 0 to 10 6 Hz with 100 mV amplitude with VMP3 potentiostat/impedance analyzer (Bio-Logic). Two platinum-black electrodes (Conductivity cell; CONPT-BTA, Vernier) were dipped in the electrolyte solution, and the cell was thermally equilibrated at 25 1C for 1 hour before conductivity measurement using an environmental chamber (SU-241, Espec). Cell constant of conductivity probe was calculated by measuring conductivity standard solution (0.01 M KCl aqueous solution, 1.413 mS cm À1 at 25 1C, VWR). Measured values are shown below in Table 1.

Computational methods
FT-IR simulation. The FT-IR spectra were simulated by computing the vibrational frequencies of solvent molecules, salt ions and solvent-ions complexes in an implicit solvation model (PCM 107 ), and dimethyl sulfoxide (DMSO, dielectric constant e = 47) was used as solvent. We used the B3LYP functional and 6-311++G** basis set, as implemented in the Gaussian (g09) suite. 117 Surface calculations. EC and EMC adsorption on the surface of LiNiO 2 were studied with a periodic approach, where we used the Perdew-Burke-Enzerhof (PBE) functional 118 and projected augmented wave (PAW) potentials, as implemented in Vienna ab initio simulation package (VASP). 119,120 As in our previous studies, [13][14][15] we employed a Dudarev's rotationally invariant DFT+U approach 121,122 with effective Hubbard-type U values U eff = 6.4 eV for Ni 123,124 and a ferromagnetic ordering. We used the (10% 14) surface, modeled by a five-layer thick slab, as it has the lowest surface energy non-polar surface, 125 and exposes the Li intercalation channels. The slabs were separated by at least 13 Å of vacuum and a dipole correction was applied to eliminate spurious interactions across the periodic boundary in the direction perpendicular to the surface. The coordinates of the adsorbate and the two upmost surface layers were allowed to fully relax, while the three bottom layers were fixed at the bulk positions.

Conflicts of interest
There are no conflicts to declare.