Liquid electrolyte development for low-temperature lithium-ion batteries

Lithium-ion batteries (LIBs) power virtually all modern portable devices and electric vehicles, and their ubiquity continues to grow. With increasing applications, however, come increasing challenges, especially when operating conditions deviate from room temperature. While high-temperature performance and degradation has been extensively studied in LIBs, sub-zero Celsius performance has received less attention, despite being critical for batteries in transportation roles. Although many individual processes contribute to the capacity loss commonly observed in LIBs at low temperatures, most of them are governed to some extent by the non-aqueous liquid electrolyte present throughout the cell interior. Therefore, electrolyte engineering presents an unparalleled opportunity to study and address the fundamental causes of low-temperature failure. In this review, we first briefly cover the various processes that determine lithium-ion performance below 0 1 C. Then, we outline recent literature on electrolyte-based strategies to improve said performance, including various additives, solvents and lithium salts. Finally, we summarize these findings and provide some perspectives on the current state of the field, including promising new areas of investigation. reliably as during summer. In the course of our research, we have found that LIB operation at low temperature is often governed most strongly by electrolyte composition. Therefore, we present this review on liquid electrolyte design for LIBs under low-temperature conditions. It is our hope that this article will inspire additional advancements in the field so that clean energy will become available to more people in more environments than ever before.


Introduction
It is difficult to overstate how thoroughly the lithium-ion battery, or LIB, has permeated life in the 21st century thus far. At the time of this writing, nearly all portable electronics and electric vehicles contain some form of LIB, and more recently they have garnered serious consideration for expanded roles in aerospace 1 and power distribution. 2,3 However, as their applications increase in significance, so too do their drawbacks, including their notorious temperature sensitivity.
LIBs function acceptably between 0-40 1C, but they encounter severe problems under harsher conditions. [4][5][6] Performance beyond the high end of this range has been extensively studied, including the mechanisms of high-temperature battery degradation and its consequences, ranging from capacity fade to thermal runaway. 7,8 By comparison, the low-temperature performance of LIBs has received considerably less attention, especially in the preceding 15 years. Nonetheless, this topic continues to gain relevance, as emerging applications in energy storage require increasing tolerance to frigid conditions. For example, widespread adoption of electric vehicles will require them to function during winter just as reliably as during summer.
Successful LIB operation depends on a complex web of physical and chemical processes that must function harmoniously; as a result, it is hardly possible to change one aspect of battery design without affecting many others. This is especially true for the non-aqueous liquid electrolytes used in LIBs, which contact nearly all internal surface area. Indeed, the early success of lithium-ion chemistry was spurred by the discovery that ethylene carbonate (EC) can form a dimensionally-stable solid electrolyte interphase (SEI) on carbon when included as an electrolyte co-solvent. 9 On the other hand, organic electrolyte chemistry may also contribute to capacity loss, 10 poor power density 11 and flammability, 12 among other concerns. Therefore, the prudent battery scientist should view electrolyte design as a ''double-edged sword'' with the potential to address multiple points of failure, but conversely introduce multiple new complications.
In light of its key role in LIB function, it should be unsurprising that early studies identified electrolyte chemistry as one of the main governing factors in low-temperature performance, 13 a consensus that largely remains to this day. In an effort to inspire renewed research in this area, we hereby present this concise review of liquid electrolyte design for low-temperature LIBs. We first summarize the various major processes that determine lithium-ion performance metrics at low temperature (o0 1C) and how the electrolyte may influence them. Then, we review research geared towards improving low-temperature performance using electrolyte engineering (additives, solvents and salts; Fig. 1), focusing primarily on recent reports (2010-now). While the majority of this work covers LIBs with traditional active materials -graphite (Gr) or Li 4 Ti 5 O 12 (LTO) anodes plus lithium transition-metal oxide or phosphate cathodes -we also include a brief summary of reports related to silicon anodes at low temperature, since this material is a likely candidate for next-generation commercial LIBs. Finally, we conclude with some suggestions as to future directions of this field.

Chen Fang
Chen Fang received his PhD degree in Chemistry from Brown University in 2018. His research was in the areas of organic synthesis, selfassembled monolayers and bottomup fabrication. In 2019, he started to work as a postdoctoral researcher at Lawrence Berkeley National Laboratory. His current research activities include electrolyte and binder materials for advanced rechargeable batteries, surface and interface chemistry inside lithiumion batteries, and solid-state batteries.
Most of all, however, EC is prized for its unique ability to form a protective SEI layer on graphite upon its decomposition at low potentials. 13 On the other hand, the melting point of EC (36 1C) 17 is well above the operating temperature of most LIBs. Therefore, it is generally blended with 50-70% of a linear carbonate (e.g., dimethyl carbonate [DMC], diethyl carbonate [DEC] and/or ethyl methyl carbonate [EMC]) to widen the working temperature range and lower the viscosity, which improves conductivity per the Stokes-Einstein and Nernst-Einstein relations. In particular, LiPF 6 /EC/EMC electrolytes have been well-characterized across a large range of compositions and ambient conditions; the conductivity of a typical blend is 5-10 mS cm À1 at room temperature. 18 Within the operating range of LIBs, electrolytic conductivity universally decreases with temperature -a trend primarily attributable to viscosity effects. Naturally, this causes the internal resistance to rise. Because viscosity correlates with EC content, it is generally accepted that this solvent should be minimized in cells designed for low temperature; however, this task has proven challenging due to the vital role of EC in low-resistance and stable SEI formation, causing lean-EC electrolytes to commonly suffer from reduced cycling stability and Coulombic efficiency. Additionally, bulk ohmic resistance is rarely the largest component of total battery impedance, † meaning that conductivity alone does not typically predict low-temperature performance with any degree of accuracy.
Liquidus point. Perhaps the most important reason to minimize EC content in low temperature LIB electrolytes is not viscosity, but crystallization tendency. Because of this solvent's high melting point, even its concentrated electrolyte solutions tend to solidify easily. Blending with lower-melting carbonates can decrease the melting point somewhat, but nonetheless the liquidus point (where solids first appear during cooling) of the common EC/EMC system remains above 0 1C for concentrations 430 mol% EC (Fig. 3a). 19 Freezing point depression through the addition of lithium salt (1 M LiPF 6 ) further lowers the liquidus to À20 to À30 1C for practical electrolytes, [19][20][21] which may partially explain why this Fig. 1 Schematic of electrolyte engineering strategies for improved low-temperature performance. A typical LIB electrolyte contains lithium hexafluorophosphate dissolved in a mixture of ethylene carbonate and linear carbonates. However, incorporation of non-traditional solvents (top row, from left: propylene carbonate, gamma-butyrolactone, methyl [3,3,3-trifluoroethyl]

Gao Liu is a Senior Scientist and Group Leader of the Applied Energy Materials Group in the Energy Storage and Distributed Resources Division at Lawrence Berkeley National
Laboratory. His laboratory combines synthetic chemistry, composite engineering and electrochemistry to solve interdisciplinary problems. Liu Lab's current research in battery and energy storage include silicon materials and electrode binders for advanced lithium-ion battery, lithium-sulfur battery, solid-state battery, and electrode interfaces. His lab also performs research in hydrogen generation, plastic recycling, and advanced manufacturing. More information about the Liu Lab can be found at the Lab's website: https://liulab.lbl.gov. temperature range is generally considered a ''turning point'' for capacity loss in LIBs. As compared to linear carbonates, the structurally-related molecule propylene carbonate (PC) suppresses EC crystallization more effectively when added as a cosolvent (Fig. 3b). PC is quite popular as a low-temperature solvent component for this reason among several others (see the relevant section below). Generally speaking, when crystallization of EC does occur, it lowers the conductivity of the remaining liquid phase and may even block electrode pores, temporarily isolating sections of active material. Both effects cause capacity loss and are therefore critical to avoid.
Transference number. One additional aspect of lowtemperature electrolyte design that has been mostly overlooked is the thermal dependence of Li + transference number (t + ). This parameter describes the fraction of current attributable to movement of Li + , as opposed to its counterion, and is directly linked to polarization resistance in working batteries 22,23 as well as lithium dendrite growth. 24 While simple in concept, this parameter is difficult to accurately measure for non-dilute electrolytes, and is usually determined at room temperature only (t + E 0.3-0.5 for typical formulations). 25,26 Nonetheless, recent work by Landesfeind and Gasteiger, 27 among others, [28][29][30] indicates that t + may decrease with decreasing temperature for typical LIBs electrolytes. The resulting increase in polarization could contribute to premature voltage drop, which is often observed during low temperature discharge. While the  The open dots represent measured data from which the solubility curve and the solidus line have been obtained through data fitting. The dotted curve is an estimated extension of the measured liquidus curve. The closed dots represent measured data for three different solutions of LiPF 6 in an EMC-EC solvent, plotted to demonstrate the change of the phase transition temperatures with the concentration of a lithium salt. (b) Liquid-solid phase diagram of PC-EC. The dotted solubility curve and the solidus line are both estimated due to the tendency of PCrich mixtures to supercool, precluding accurate measurement. Compared to EMC, PC reduces liquidus temperature more effectively when present 420 mol% in the mixture. Reprinted from ref. 19, copyright The Electrochemical Society. Reproduced by permission of IOP Publishing Ltd. All rights reserved. mechanistic underpinnings of this parameter's temperature dependence have yet to be clarified, Landesfeind and Gasteiger suggest that large hydrodynamic volume of (solvated) Li + relative to PF 6 À may play a role, as well as increased formation of ion aggregates like [Li(PF 6 ) 2 ] À . This topic deserves further experimental and computational exploration in order to determine the precise role that t + plays in low-temperature performance characteristics.

Interfacial considerations
Charge transfer resistance. At this time, it is widely accepted that electrolyte/active material interfacial processes make up the bulk of LIB impedance below freezing, although the exact identity of these processes remains an active subject of research. Early impedance studies hinted towards a kinetic origin for the limiting resistance, as work by Zhang, Xu and Jow 31 revealed a strong dependence of LIB overpotential on state-of-charge, especially below À20 1C (Fig. 4a). Furthermore, impedance spectra showed increasing dominance of a midfrequency component as temperature decreased, a region generally assigned to charge transfer resistance (R ct ). However, the exact physical meaning of this term is often ambiguous, as ''charge transfer'' may itself consist of multiple, reactiondependent steps. Furthermore, these processes are likely to generate different resistances at either electrode, which this experiment did not distinguish between. These authors' follow-up work noted an Arrhenius-type relationship between R ct and temperature (Fig. 4b), and that the activation energy of the charge transfer process (E a,ct ) was strongly dependent on electrolyte composition (both salt and solvent). 32 This E a , ct term varied independently from bulk ion transport resistance. Furthermore, Abraham et al. identified near-identical E a , ct trends for binderand-carbon-free LTO|LiMn 2 O 4 cells, despite the absence of any significant interphase on these materials. 33 Taken together, these results led the authors to strongly suggest Li + desolvation as the rate-limiting process associated with charge transfer, as opposed to transport through an SEI or across the SEI-active material boundary.
Recent works by various authors have strengthened the conclusion that Li + desolvation is the rate-limiting process at low temperature (Fig. 5a). Xu and coworkers corroborated previous results by demonstrating that E a,ct on LTO varies with the electron donicity of the solvent, which is a major factor in solvation strength. These authors obtained values of 40 kJ mol À1 for 1 M LiTFSI in tetrahydrofuran vs. 52.5 kJ mol À1 for 1 M LiPF 6 in EC/EMC 3 : 7. 34 Their report also noted similar resistance trends on graphite anodes, but with values consistently offset upwards by B20 kJ mol À1 ; the authors attributed this to the additional effect of ion transport through a resistive SEI, which appeared at a similar resonant frequency and was difficult to separate out. Similarly, Li et al. studied Li8Gr and Li8LiFePO 4 half cells with identical electrolytes, finding graphite to be limiting in both charge transfer and interfacial transport (although electrolyte design had a strong effect). 35 In 2017, a team from Pacific Northwest National Laboratory systematically compared carbonate electrolytes of varying composition in Gr8Gr, NCA8NCA and LTO8LTO symmetric cells, thereby eliminating any competing effects from Li metal or multiple electrodes. 15 Despite the disparate chemical structures of these materials and their interfaces, their mid-frequency EIS responses were nearly identical at À40 1C, implying that ion desolvation -their only common factor -is the rate-limiting process in each case. Furthermore, when graphite anodes were subjected to an SEI-formation process in one electrolyte, then re-assembled into Gr8NCA full cells with another electrolyte, the discharge capacity of the cells at Fig. 4 (a) Variation of bulk electrolyte resistance (R b ), SEI resistance (R SEI ), charge-transfer resistance (R ct ) and the percentage of total resistance made up of R ct (R ct %) with temperature in a commercial Li-ion cell discharged to 3.87 V. Reprinted from ref. 31  À20 1C had virtually no correlation with the former electrolyte composition, but correlated strongly with the latter (Fig. 5b and c).
The overall lesson appears to be that electrolyte design affects low temperature performance primarily by determining the makeup of Li-ion solvation shells -the breakup of which requires the greatest energy input out of all simultaneous processes during operation.
Solid-electrolyte interphase resistance. The additional major impact of electrolyte (on graphite especially, but also on 44 V cathode materials) is that of interphase composition. Interphase layers may form on either anode (SEI) or cathode (CEI) surfaces, and they contribute a bulk transport resistance (R SEI ) to cell impedance, which can vary in magnitude from inconsequential to nearly as large as R ct . 36 In fact, some early research pointed to SEI conductivity as the main limiting factor in low-temperature LIB capacity, 37 a common misconception that remains to this day. Of course, as discussed above, we now know that R SEI is unlikely to be the largest component of internal resistance below 0 1C, at least for ''typical'' cell designs, i.e. a graphite anode, oxide cathode and EC-based electrolyte.
Why, then, does this misconception persist? One likely reason is that R SEI and R ct can be difficult to distinguish. identical pre-cycled electrodes, but different electrolytes. E9A refers to an anode that has undergone formation cycles in E9 electrolyte (an EC-lean formulation), while E2C refers to a cathode that has undergone formation in E2 (an EC-rich formulation). Despite identical pre-formed interphase layers, major differences in discharge capacity below À20 1C are observed, consistent with lower R ct in the EC-lean electrolyte E9. Reprinted/adapted with permission from ref. 15. Copyright 2017 American Chemical Society.
The impedances associated with interphase ion transport and charge transfer often overlap quite substantially in characteristic frequency, which can cause them to appear as a single semicircle in Nyquist plots generated by EIS ( Fig. 6a and b). Attempting to extract separate, accurate resistance values from such a feature becomes an exercise in futility, and it is easier to refer to a single, combined value for ''interfacial resistance.'' Another possible reason is that R SEI , while not the largest component of total resistance, may still compose a sufficient fraction of it to notably affect performance. An excellent example of this can be found in the recent work of Liu et al., who systematically studied interphase-forming additives, while keeping the base electrolyte constant. 38 Under such conditions, R ct should hardly change, but capacity improvements of up to 7% were observed at À40 1C and C/5 rate, which correlated well to changes in interphase chemistry/thickness with different additives.
The third possible answer is that interfacial chemistry may not be totally separable from charge-transfer kinetics. After all, if Li + desolvation at the electrolyte/interphase boundary is responsible for R ct , it follows that surface chemistry may play a role in that process -partially coordinating the ion as its solvation shell strips away, for example. However, such an effect would be very difficult to study, especially given how little we still know about SEI/CEI chemistry on the molecular scale, which is notoriously difficult to characterize. [39][40][41] Nonetheless, the fact remains that changes to SEI composition are often reported to produce outsized effects on lowtemperature capacity (vide infra). Therefore, when posed the question ''Exactly what role does SEI/CEI play in lowtemperature performance?'', we can unfortunately provide no definitive answer other than ''It depends.'' That being said, even if R SEI does not ordinarily limit low-temperature capacity, the stabilizing role of the interphase cannot be ignored. In order to successfully cycle a graphitebased LIB at any temperature, the electrolyte must form a dimensionally-stable layer to protect against continued decomposition, as well as exfoliation. Therefore, the issue is often less about optimizing interphase composition for subfreezing performance directly, and more about adjusting other factors (freezing point, charge transfer resistance, etc.) without compromising the robustness of anode/cathode interphases. This is of special concern for low-liquidus-point electrolytes with little-to-no EC content, which must rely on additives for stable cycling.

Lithium plating
Our discussion thus far has centered on discharge capacity as the primary metric by which low-temperature performance is measured, with the implicit assumption that charging is done at 25 1C or higher. This is because charging a graphite-anode LIB at lower than 25 1C and/or at a high rate introduces significant risk of lithium metal plating, a major safety hazard and cause of premature capacity fade. Much of the work to detect the onset of -and to mitigate -lithium metal deposition on graphite anodes has focused on achieving fifteen-minute charge times at room temperature. This has been well summarized and described in other reviews, including Waldman et al., 42 Liu et al. 43 and Tomaszewska et al.; 11 the latter review also includes some analysis of work to enable low-temperature LIB operation. While ''extreme fast charging'' at around 25 1C has numerous issues, low-temperature charging brings many similar challenges associated with undesired lithium metal nucleation and growth, even at relatively slow charging rates.
During ''typical,'' facile charging of a graphite-anode LIB, Li + ions are reduced at the graphite anode and intercalate between individual graphene layers. However, if the electrochemical potential of the graphite particle dips below 0 V vs. Li/Li + , lithium metal formation becomes thermodynamically possible. Graphite is particularly susceptible to this phenomenon because its operating potential (under open-circuit conditions) is only B100 mV vs. Li/Li + . 44 When current is applied, anode potential may further drop into the lithium plating regime, given sufficiently-large overpotential at the graphite/electrolyte interface (Fig. 7a). As discussed above, detrimental overpotentials can arise from a number of factors, including ohmic losses-which scale with applied charging current-as well as where only one semicircle is distinguishable. In this case R SEI cannot be separated from R ct , either due to similar time constants for both processes or because one is much larger than the other. Additional context is required to interpret the combined ''interfacial resistance.'' Reprinted from ref. 66 23,45 Kinetic overpotentials are additionally impacted by such concentration gradients, as charge transfer resistance is dependent on the surface concentration of Li + . Mass transport and kinetic overpotentials will also both increase with decreasing temperature, generally following Arrhenius-type dependencies. Waldmann et al. used a 3-electrode cell to demonstrate this temperature-dependent decrease in the anode potential; 46 during low-temperature charging, these overpotentials led to large amounts of Li plating at relatively low charging rates. A fraction of such Li metal deposits cannot be reversibly stripped, instead forming ''dead'' lithium which includes electronically-isolated metallic Li and graphite, as well as degradation products from the reaction of electrolyte with metallic Li. The formation of dead Li necessarily means a reduction in cyclable Li capacity; this has been quantified as a function of rate and cycle life at room temperature using destructive mass spectrometry titration. 47 While the majority of literature has focused on fast charging at ambient temperatures (420 1C), there have still been numerous studies characterizing Li plating on graphite during low-temperature operation. Electrochemical methods have been used to detect when Li plating has occurred at low temperature, including observation of a stripping plateau during discharge of spiral-rolled, 3-electrode LIBs (300-400 mA h, Gr8NCO8Li Ref.) operated at À40 1C (Fig. 7b), 48 along with dV/ dQ analysis of discharge curves for LIBs (2.5 A h, 26 650 cylindrical Gr8LFP) cycled at temperatures down to À30 1C. 14,49 The latter technique relies on the presence of the stripping plateau seen in the former. These dV/dQ studies also reveal an impedance rise after low-temperature charging, which the authors attributed to SEI growth from Li metal reacting with the electrolyte. This reaction of plated Li with the electrolyte is also noted by Ng et al. as the primary culprit for significant gas formation in their LIBs (50 A h, Gr8NMC 532 prismatic) cycled at À29 1C. This gas formation was shown to cause detrimental additional stresses on the electrodes, leading to eventual cell failure. 50 Still, simple optical inspection of the graphite electrode remains the most common technique for Li plating detection upon post-mortem analysis. This necessarily requires enough Li to have plated for visual inspection to be possible. However, certain groups have expanded upon this relatively simple analysis to allow for more robust diagnoses; one of the previously-mentioned studies using dV/dQ and impedance spectroscopy also analyzed the deposited Li layer thickness to quantify the amount of Li plated. 14 Unfortunately, this still requires significant Li plating to have already occurred. Waldmann et al. have utilized optical inspection, combined with capacity fade monitoring, to label Li plating as the primary capacity fade mechanism for LIBs (1.5 A h, 18 650 cylindrical Gr8NMC 111 + LMO) cycled at temperatures down to À20 1C, relative to cells cycled up to 70 1C. 51 Ghanbari et al. used a more specialized technique-glow discharge optical emission spectroscopy-to characterize Li plating in cells, which showed homogeneous Li plating in cells cycled at À20 1C (2.5 A h, 26 650-format Gr8LFP) relative to the ''island'' deposits formed in cells cycled at 45 1C (16 A h, Gr8NMC). 52 It should be pointed out, though, that some of the differences in plating morphology could be attributed to different cell geometries. This optical emission technique can also discriminate between aging related to SEI growth vs. that related to Li plating. Each of these studies seems to suggest that Li plating is especially prevalent at low temperatures, and that it is the primary culprit for capacity fade of LIBs repeatedly charged under such conditions.
As others have noted, though, no technique to-date has been able to reliably detect the onset of Li plating in LIBs with high general sensitivity. dV/dQ analysis relies on the presence of a stripping plateau which is not always present, even when   53 and requires a slow discharge immediately after charging. Incidentally, a concise summary of all differential voltage analysis work performed for low temperature LIB applications thus far may be found in ref. 47 above. Optical inspection -perhaps the most universally-used technique -necessarily requires a large amount of Li deposition to have occurred; therefore, the damage of substantial, irreversible capacity loss must already be done. Differential open-circuit voltage analysis has recently shown some promise as a more general method for plating detection, 54 but requires further validation. Li plating detection techniques and their limitations, as well as ideas for Li plating mitigation, are more-thoroughly discussed in the previously mentioned fast charging and Li plating review articles. However, it should be emphasized that this current limitation in the field will need to be addressed in order to make low-temperature LIBs operation-where Li plating more readily occurs-a reality. As a final comment, it should be noted that, under realistic operating conditions, i.e. large-format cells assembled in packs, the situation may be less bleak than it seems. Several authors have proposed tailoring battery pack design and/or charging protocols to take advantage of self-heating, so that overpotential is minimized by the end of charge and lithium plating is mitigated. Details are outside the scope of this review but may be readily found elsewhere. 55,56 Non-electrolyte factors There are, of course, physical processes relevant to lowtemperature performance which are altogether divorced from electrolyte choice. While a detailed discussion falls outside the scope of this review, these factors may sometimes play a role in observed performance and, therefore, should be kept in mind. One such process is electron transport, which depends not only on bulk electrical conductivities (often quite low for cathode materials), but also on particle-particle interfacial contact. Luckily, modern electrode designs minimize these problems quite effectively, to the point where internal electrical resistance is generally negligible. 57 Solid-state properties of the cathode and anode active materials also influence sub-zero behavior, most obviously through the dependence of surface area (and, thus, effective current density) on particle size. Indeed, an early study found that coke anode capacity at rÀ20 1C was improved when average particle size was reduced from 25 mm to 6 mm. 58 More subtly, limited solid-state diffusion of lithium naturally causes the surface concentration of lithium to differ from the bulk when current is non-zero, which produces concentration polarization. The associated polarization resistance, although usually mild at room temperature and low rate, may become more significant with dropping temperature due to a decrease in lithium-ion solid-state diffusion coefficient. 59 A coupled electrochemical-thermal modeling study performed by Ji, Zhang and Wang has demonstrated that the limiting factors in large-format cells (2.2 A h 18 650 cylindrical) may be significantly different than in smaller cells due to non-isothermal conditions. 60 At À20 1C and low-rate discharge (0.001C), their model indeed predicted that interfacial kinetics dominate internal resistance; however, at higher rate (1C), self-heating effects reduced the influence of kinetic factors very early in the discharge process, whereas solid particle resistance at the anode grew continuously, becoming dominant towards the end of discharge. In all cases, accounting for self-heating resulted in higher capacity than predicted by an isothermal model. This result should caution the reader against generalizing any particular low-temperature result to batteries of dissimilar size or geometry.

Progress in low-temperature electrolyte design
When formulating a liquid electrolyte, there are three general components to adjust: additives, solvents and salts. The lines between these descriptors sometimes blur, as a particular chemical species may fall into multiple categories. For the sake of this review, we will consider a ''solvent'' to be a component with a melting point r40 1C that is present at 45 wt% concentration in the final mixture. Likewise, we will consider a ''primary salt'' to be a dissociable ionic substance containing Li + , which is present at either 40.5 M or 45 wt% in the final mixture. Any reagent that falls outside these guidelines -for example, an ionic substance that is solid at room temperature and does not contain lithium, or a molecular liquid added to merely 2 wt% -would be considered an additive. These guidelines are largely arbitrary and chosen purely for convenience. Note that a single substance could be considered an additive in one study and a primary salt or solvent in another; the difference depends solely on the concentrations used. For the sake of relevance, we exclude LiPF 6 and EC/DEC/DMC/EMC from receiving focused discussion, since these materials are all well-studied and used commercially. More background on these standard electrolyte materials can be found in existing reviews. 13,61,62 Additives Electrolyte additives, by nature of their broad definition, take many forms and produce a variety of effects, such as reduced gas evolution or overcharge protection. 63 Most commonly, however, additives are intended to improve the cycling performance of LIBs, usually by stabilizing electrode interfaces to chemical or morphological changes over time. Given the strong correlation between electrode interfacial chemistry and low temperature performance, it is unsurprising that many additives have been investigated specifically for their effects below 0 1C. Additive formulations used by the references cited here may be found summarized in Table 1 along with their structures.
Fluoroethylene carbonate (FEC). Fluoroethylene carbonate (FEC) is one of the most popularly-studied additives for lithium-based batteries due to its favorable SEI-forming properties. 61   and Coulombic efficiency on mesocarbon microbead (MCMB) graphite anodes at temperatures from À40 to 20 1C. 66 Combined electrochemical impedance spectroscopy (EIS) and X-ray photoelectron spectroscopy (XPS) results led the authors to conclude that the improvement resulted from a smoother, more conductive interphase formed by the additive, with the difference becoming more significant as temperature decreased. However, the authors' own data seems to indicate reduced charge transfer resistance in the presence of FEC as well, although they were unable to obtain accurate numbers for this component. Interestingly, the same group also showed a beneficial effect of FEC on a LiFePO 4 cathode, 67 although it should be mentioned that both studies presented data for only half cells with Li metal, for which FEC also produces notable effects. 68 More commonly, FEC and other carbonates are used as individual components of a multi-additive mixture. For instance, Liu and coworkers demonstrated a complex, multiadditive system in which 0.5 wt% FEC played a crucial role to stabilize Gr8NMC 111 pouch cells at both À40 1C and 60 1C. 38 Another recent report by Wotango et al. demonstrated additive mixtures that included up to 5 wt% FEC. 69 Often, this additive is used in large enough concentration to qualify as a ''solvent'' component; such instances are discussed in the relevant section below.
Tris(trimethylsilyl)phosphite. Organophosphites have been proposed as LIB electrolyte additives for a variety of purposes, including flammability reduction 70 and high-voltage stability improvement. 71 In particular, tris(trimethylsilyl)phosphite a.k.a. TMSP or TMSPi has enjoyed recent popularity due to its passivating effect on metal oxides, resulting from the formation of a stable cathode electrolyte interphase (CEI) layer. 72 TMSPi has also been shown to participate in SEI formation on graphite  due to its reactivity with lithium alkoxides formed during EC degradation. 73 Given its influence at both sides of the cell, this molecule has received some recent study in low-temperature systems. A 2019 report by Liu et al. found TMSPi to be one of the most effective additives out of several at reducing Gr8NCA full cell overpotential at À40 1C, thereby increasing capacity. 38 Interestingly, only 0.5 wt% was required for such improvement, while 1 wt% had a detrimental effect. Another paper from the same year demonstrated an electrolyte design including TMSPi which allowed a 5 V-class MCMB8LiNi 0.5 Mn 1.5 O 4 (a.k.a. LNMO) cell to operate at temperatures down to À60 1C, including repeated charge/discharge at À5 1C at 0.3C rate without capacity fade ( Fig. 8a and b). 74 Besides the expected cathode passivation, this additive was shown to significantly stabilize the anode SEI as well, preventing lithium metal deposition and reducing microscopic crack formation (Fig. 8c). These promising early results merit additional investigation, especially to untangle the relative contributions of TMSPi-formed CEI and SEI to the observed improvement at low temperatures. Sulfur-containing molecules. Organosulfur compounds of varying structure and oxidation state have been heavily investigated as interphase-altering LIB electrolyte additives, including for low temperature cells. Butyl sultone (BuS) was reported for this purpose as early as 2006, 75 with the authors observing significant improvement in Gr8LiCoO 2 full cell capacity and overpotential at temperatures from À20-70 1C. Detailed EIS study of a graphite anode revealed a notable drop in R ct when 1 wt% BuS was present. A more current study found that BuS also has a positive effect on LiFePO 4 cathode capacity at low temperature, with up to 35% higher capacity (64.8 mA h g À1 ) being available at 1C rate and À20 1C. 76 Additive amounts greater than 1%, however, proved detrimental. Dimethyl sulfite (DMS), a close chemical relative of DMC, has also been investigated by multiple groups. 77,78 Guo et al. very recently found that this additive is more effective than the commercially-popular 1,3,2-dioxathiolane-2,2-dioxide (DTD) at reducing capacity fade in Gr8NMC 523 pouch cells at À10 1C. 78 Li8Gr half-cell studies indicated that both additives produced similar benefits in R SEI and R ct at room temperature, but that R SEI was significantly lower for the DMS additive as temperature decreased. TEM imaging confirmed that the interphase formed by DMS was thinner and more uniform than that formed by DTD or in baseline electrolyte. Other compounds which have received focused study at subzero temperatures include View Article Online 1,3-propane sultone, 38 prop-1-ene-1,3-sultone 79 and 1,3-propanediol cyclic sulfate (see Fig. 7). 74 More unique functionalities have also proven useful. Shi and coworkers recently demonstrated fluorosulfonyl isocyanate as a novel sacrificial additive which preferentially reduces to form a low-impedance SEI on graphite. 80 The resulting electrodes showed dramatically-improved rate performance in half cells with lithium, especially at low temperatures. Wotango et al. devised a novel chemical derivative of ethylene sulfite: 4-chloromethyl-1,3,2-dioxathiolane-2-oxide, which was used as a one of several additives to significantly improve Li8MCMB half-cell performance at À10 1C. 69 Lithium salts. Although many diverse lithium salts have been proposed as LIB additives, surprisingly few of these have been studied for their effect at low temperature specifically. One such salt is lithium difluorophosphate (LiPO 2 F 2 ), which has been reported elsewhere to modify both cathode 81 and anode 82 interfacial chemistry. Yang et al. found that merely 1 wt% LiPO 2 F 2 added to an electrolyte of 1 M LiPF 6 in EC/EMC/PC 4 : 7 : 1 w/w improved the relative capacity retention of Gr8NMC 523 cells at À30 1C from 9.6% to 57.9% (Fig. 9). 83 Galvanostatic cycling performance was also impacted: cells with control electrolyte experienced rapid capacity fading during repeated 0.5C charge/ discharge at À20 1C, while additive-containing cells retained 91% of their initial capacity over 100 such cycles. Most recently, some of the same authors reported 2 wt% LiPO 2 F 2 to improve the temperature range of 4.4 V pouch cells containing an electrolyte of 1 M LiPF 6 in EC/EMC 1 : 2 w/w. 84 Combined analytical results from both studies indicate that this additive acts as an interface-former on both graphite and NMC surfaces, with the resulting SEI and CEI exhibiting vastly-reduced resistance. Further study may be required to probe the mechanistic underpinnings of this effect. A structurally-related salt, lithium difluorobis(oxolato)phosphate (LiDFBOP), has also been shown to dictate anode and cathode interfacial makeup when added at 1%, with associated full-cell performance improvement at 0 1C. 79 A particularly unique design comes from Ko et al., who reported on a lithium-modified silica nanosalt (''Li202'') synthesized by treating hydrophobic silica with LiH, followed by 1,3-propane sultone. 85 The resulting sulfonate-rich surface allowed the silica to form a stable dispersion (2.5 wt%) with an electrolyte solution of 1 M LiPF 6 in EC/PC/EMC/DEC 20 : 5 : 55 : 20 v/v + 2 wt% vinylene carbonate (VC). The presence of Li202 additive slightly improved the cycling performance of Gr8LiCoO 2 coin cells at À20 1C when compared to the electrolyte alone and a non-functionalized silica dispersion. EIS analysis determined this effect to come from reduced R SEI in the presence of the additive, a hypothesis which was corroborated by SEM images showing thinner interfacial layers. Several of the same authors also published a follow-up report showing a synergistic effect of Li202 additive with an acrylate-grafted PDMS additive. 86 Altogether, graphite8LiCoO 2 cells with this dual-additive formulation showed improved capacity of up to 110 mA h g À1 at À20 1C and 0.1C discharge rate, as compared to 95 mA h g À1 without the additives. The combination appeared to significantly reduce both interfacial resistance and bulk electrolyte resistance as measured by EIS, although the exact reason was not explored.
Non-lithium salts. One of the most intriguing additives recently reported to improve low-temperature performance is CsPF 6 , which has been systematically studied by Xu et al. (Fig. 10). 15,38,[87][88][89] The effect of Cs + is reportedly due to its role in directing SEI formation: the cesium ion can accommodate only 1-2 EC molecules in its solvation shell and the reduction potential of [Cs(EC) 1-2 ] + solvates is higher than that of the [Li(EC) [3][4] ] + solvates also present. This results in an SEI dominated by the (desirable) decomposition products of EC, 87 even in electrolytes with competing solvents 88 and/or low EC content. 89 After optimization of both solvent and additive composition, 0.05 M CsPF 6 allowed Liu et al. to demonstrate 1 A h pouch cells (Gr8NMC 111 ), which retained 0.69 A h discharge capacity at À18 1C and an impressive 0.37 A h discharge capacity at À40 1C, both at 1C rate. This was achieved without sacrificing room temperature or high temperature characteristics, either, as the same cells were able to cycle 1000 times with 485% capacity retention at 25 1C, and Gr8NCA coin cells with the same electrolyte showed 460% retention after 300 cycles at 60 1C.

Solvents
With few exceptions, solvents make up the largest portion of any electrolyte, whether by weight, volume or mole fraction.

View Article Online
Their principal job is to dissolve and dissociate the primary salt into free ions, which carry the internal current of the battery. This process is facilitated by solvents with high dielectric permittivity and electron donor number, such as EC, which forms strong solvation complexes with Li + . Unfortunately, this presents a dilemma at low temperatures, where desolvation becomes the rate-limiting step in charge transfer (vide supra). This has driven a large body of research into alternative solvents which can replace or augment EC without sacrificing the favorable ion transport properties it endows. Such efforts are additionally complicated by the secondary role of LIB electrolyte solvents: interfacial compatibility. Solvent blends must be stable to oxidation at high cathode potentials (44 V) and form sufficiently inert SEI layers at graphite anodes, as discussed previously. Therefore, alternative solvents are often used in combination with passivating additives in order to realize the benefits of low-EC-content electrolytes at sub-zero temperature without sacrificing cycle life. Solvent formulations investigated by the references cited here are summarized in Table 2 along with their structures. Propylene carbonate. Propylene carbonate (PC) is a wellknown polar aprotic solvent (e = 65) 16 with a wide liquid range (À49 to 242 1C), 90 properties which have garnered attention since the early days of LIB development. While the idea of a PC-based electrolyte was initially shelved due to its tendency to co-intercalate into graphite, modern developments in additive and solvent formulation have enabled its reintroduction with the suppression of harmful side effects. Because the melting point and viscosity of PC compare favorably to EC, 17 replacing some or all of the latter can naturally improve conductivity at low temperature and dramatically reduce crystallization tendency, a factor that has been exploited by many researchers in this area. In fact, many of the studies summarized in this review use baseline electrolytes containing some quantity of PC.
For instance, as mentioned previously, Xu and coworkers have published a series of reports on CsPF 6 additive, the original purpose of which was to stabilize graphite anodes in a majority-PC electrolyte. 15,38,[87][88][89] In one study from 2017, Li et al. thoroughly compared the thermal properties and ionic conductivity of electrolytes containing varying ratios of EC, PC and EMC with constant salt concentrations (1.0 M LiPF 6 and 0.05 M CsPF 6 ). 89 Addition of PC was found to significantly reduce both thermodynamic liquidus temperature and the kinetic precipitation point, with a 1 : 1 : 8 blend exhibiting values of À58.4 1C and À67.2 1C, respectively (Fig. 11a). Moreover, conductivity uniformly increased with decreasing EC content at temperatures below À20 1C, with     values as high as 1 mS cm À1 at À40 1C being obtainable for the 1 : 1 : 8 blend (Fig. 11b). This relatively simple ternary mixture enabled both Gr8NCA coin cells and Gr8NMC 111 pouch cells to discharge 465% of their normal capacity at À40 1C and C/5 discharge rate (Fig. 11c). FEC, VC and other carbonates. Other carbonates have also been tested for their ability to fully replace EC, including VC 91 and FEC. [92][93][94][95] Dahn and coworkers have recently reported on electrolytes containing VC as the sole carbonate component, which allowed cells to retain 40% of their normal discharge energy at À14 1C and a blistering 4C rate, whereas a traditional EC/EMC electrolyte with 2 wt% VC could not discharge at all at rates above 2C at this temperature. 91 A later study from the same group found that a 5 : 95 FEC/EMC electrolyte produced   The concentrated electrolytes were then further diluted into inert fluorinated solvents, which cannot ordinarily dissolve lithium salts, but are miscible with the pre-solvated blends. The resulting localized high-concentration electrolytes remained liquid below À120 1C (Fig. 12a), an astonishing low-temperature limit for any organic fluid. Their conductivities also showed unexpectedly weak thermal dependence, varying merely two orders of magnitude (10 À5 -10 À3 S cm À1 ) from À80 1C to 25 1C (Fig. 12b). Moreover, due to the high ratio of salt to (coordinating) solvent molecules, the average solvation energy of Li + -and thus, indirectly, the charge transfer resistance -was predicted to be significantly reduced (although it was not directly measured). Altogether, this combination of favorable electrolyte properties allowed the authors to successfully operate Li8NCA cells from À95 to 70 1C (Fig. 12c and  d). To our knowledge, this is the widest temperature range ever reported for a lithium-based battery with a liquid electrolyte. It is also notable that the authors reported Coulombic efficiencies 499% for a large number of anode materials (lithium metal, Gr, LTO) and cathode materials (NMC 811 , NCA, LiNMO, LiCoMnO 4 ) with this electrolyte, demonstrating the universality of the design approach. Esters/lactones. Besides carbonates, esters have received the most consideration as electrolyte solvents for low temperature LIBs, along with their cyclic variants known as lactones. This is largely due to the pioneering early work of M. C. Smart and coworkers, 96,97 who investigated many of these molecules as solvent components to supplement the traditional EC, DEC, etc. In particular, linear aliphatic esters (or simply ''esters'') possess a winning combination of low viscosity, low melting point and moderate polarity -all characteristics that encourage fast Li + transport over a wide temperature range. Nonetheless, they have several critical drawbacks, including high volatility (i.e. flammability) and reduced electrochemical window compared to carbonates. The simplest esters, such as methyl formate 98 and ethyl acetate, 96 are cathodically unstable to lithiated graphite and do not produce an effective passivating layer upon their breakdown. Although this may be remedied somewhat by inclusion of EC, R SEI and R ct become quite high in such electrolytes after extended cycling, rendering them useless in the long-term. However, slightly-larger esters like ethyl butyrate appear to be less reactive in this regard and can be successfully added to carbonate mixtures to improve low-temperature capacity without incurring large penalties elsewhere. 97 It should be noted that, according to the preceding reference, ester cosolvent addition reduces high-voltage stability as well, necessitating careful control over charge conditions. Several design strategies have emerged to balance the advantages and drawbacks of ester (co-)solvents. Structure/ blend optimization is perhaps the simplest solution, as demonstrated by the comprehensive screening published by Smart et al. in 2010. 99 Amongst the many formulations studied by these authors, methyl propionate (MP) was found to be most effective (Fig. 13a), as an electrolyte of 1 M LiPF 6 in EC/EMC/MP 2 : 6 : 2 v/v enabled nominal 7 A h Gr|LiNi 0.8 Co 0.2 O 2 cells to retain over 5 A h down to À60 1C. Another recurring concept involves partial fluorination, as in the case of 2,2,2trifluoroethyl butyrate 100 and 2,2,2-trifluoroethyl n-caproate 101 among others, which generally tends to improve high-voltage stability and modify SEI composition. The most common recent development, however, has seen ester-based formulations paired with interface-modifying additives, including VC, 91,102-104 FEC 103,105 and others. 74,104 For instance, Jones et al. tested a series of additives for their ability to inhibit lithium plating during low-temperature charge in a MP-rich electrolyte, finding 0.1 M LiFSI to be most effective. 104 In at least one case, additives have made it possible to eliminate EC from the electrolyte entirely, with a MP : VC 95 : 5 w/w electrolyte allowing acceptable capacity retention during 40 1C cycling of Gr8NMC 111 pouch cells and enabling significantly-improved rate performance at À14 1C compared to an EC/EMC/VC mixture. 91 Inclusion of such additives also frees up the design space to include more-reactive, but less-viscous and higher-polarity esters like methyl acetate. 103 Shorter esters have also proven valuable to the performance of

View Article Online
LIBs without graphite: Chen and coworkers tested LTO8lithium manganese oxide cells with a variety of electrolytes and found an ethyl acetate-containing commercial blend to produce the lowest total impedance, with resulting cells able to pass the USABC cold cranking test at À30 1C. 106 Similarly, an all-organic battery design has been demonstrated at À70 1C using an electrolyte with ethyl acetate as the sole solvent component. 107 We would be remiss to conclude our discussion of ester electrolytes without mentioning lactones. Unlike linear esters, which generally play a similar role to linear carbonates in electrolyte design, cyclic lactones are polar enough to replace EC/PC partially or entirely. The best-studied member of this class, gamma-butyrolactone (GBL), has a relative dielectric constant of 42 at room temperature 108 -around half that of EC, but still large enough to permit effective ionization of Li + salts. It also remains liquid over a wider range (À44 to 204 1C) and possesses lower viscosity (1.7 cP at 25 1C) than EC, both valuable qualities for a low-temperature electrolyte solvent. Like other esters, however, it is unable to form a passivating SEI on graphite, 13 a discovery that all-but-killed research into GBL electrolytes early in the development of LIBs. Yet recent progress may provide a way forward, as the responsibility for interfacial stability shifts away from solvents and towards salts/ additives. For instance, Lazar and Lucht reported successful cycling of Gr8NCA coin cells in an all-ester electrolyte of GBL and methyl butyrate (1 : 1 vol) containing 1 M lithium difluoro(oxalato)borate (LiDFOB). 109 Shi and coworkers also developed a similar electrolyte of 1 M lithium bis(oxalato) borate (LiBOB) in a novel blend of GBL with inert hydrofluoroether. 110 This unique combination proved effective across a wide temperature spectrum (À40 to 60 1C) in Gr8NMC 111 cells, which could still deliver 74 mA h g À1 at the low end of this range, but did not function at all with a conventional electrolyte (Fig. 13b and c). A follow-up study by the same group also demonstrated excellent safety and processability characteristics for a similar electrolyte containing LiDFOB. 111 Given these promising early results and the wealth of modern additive choices, it may be time to revisit GBL-based electrolytes as a serious option for next-generation LIBs.
Room-temperature ionic liquids. Room-temperature ionic liquids (RTILs or simply ILs) are a diverse group of materials widely studied for their unique physical, electrochemical and solvency properties. Many ILs have been proposed as components of LIB electrolytes, especially for improved safety and cyclability at high temperatures. However, in general, their high viscosities/ melting points and poor wettability with commercial separators require that they be blended with organic solvents to achieve reasonable performance below 0 1C. For instance, Xiang and coworkers investigated electrolytes containing 0.1-0.4 mol kg À1 lithium bis(trifluoromethanesulfonyl)imide (LiTFSI) in N-methyl-N-propylpiperidinium bis(trifluoromethanesulfonyl)imide (PP 13 TFSI) in Li8LiCoO 2 half cells. 112 While this system performed quite well at room temperature and C/10 rate (137 mA h g À1 by cathode active weight), capacity dropped to 119 mA h g À1 at only 10 1C, and the cell became inoperable at lower temperatures due to freezing of the electrolyte. Addition of only 20 wt% DEC to the ionic liquid, however, reduced the liquidus temperature to À19 1C, while 40 wt% DEC suppressed crystallization altogether. All electrolytes were confirmed to be non-flammable. The 4 : 1 PP 13 TFSI : DEC blend with 0.4 mol kg À1 LiTFSI allowed Li8LiCoO 2 half cells to operate at À10 1C with 102 mA h g À1 capacity at C/10 rate -72% of its room-temperature value.
An additional challenge facing RTILs is their general instability at low voltages. In order to intercalate and deintercalate lithium reversibly, low-potential anodes, i.e. graphite or lithium metal must form a dimensionally-stable SEI that prevents continuous decomposition. For graphite, ionic liquids based on the TFSI anion cannot accomplish this. However, many researchers have partially addressed the issue by adding significant amounts of the structurally-similar bis(fluorosulfonyl) imide (FSI) anion, which degrades rapidly to form inorganic protecting layers. 113 This strategy also tends to suppress crystallization and lower viscosity. Kunze et al. demonstrated that mixtures of N-butyl-N-methylpyrrolidinium (Pyr 14 ) TFSI and N-methyl-N-propylpyrrolidinium (Pyr 13 ) FSI have liquidus temperatures far below that of either parent compound (À18 1C and À8 1C, respectively). 114 The (Pyr 13 FSI) 0.836 (Pyr 14 TFSI) 0.164 intermediate blend maintained a conductivity 410 À3 S cm À1 down to À20 1C, while (Pyr 13 FSI) 0.568 (Pyr 14 TFSI) 0.432 remained liquid with conductivity 10 À4 S cm À1 at À40 1C. These values are comparable to state-of-the-art organic electrolytes, although the relevance of this data may be limited, since none of the reported blends contained lithium and the lithium transference numbers of RTIL blends are generally quite low. 115 Yamagata et al. tested a mixed electrolyte of 0.45 M LiTFSI in 1-ethyl-3-methylimidazolium (EMIM) FSI in Li8Gr half-cells, comparing the performance to a standard electrolyte (1 M LiPF 6 in EC/DMC 1 : 1 v/v). 116 The IL electrolyte outperformed the standard at rates up to 5C and temperatures from 25-60 1C, but was initially inoperable at 0 1C. However, a small amount of LiBOB additive dramatically improved the performance, allowing these cells to pass 67% of their room-temperature capacity after 25 cycles at 0 1C (C/10 rate).
Overall, the concept of low-temperature electrolytes based purely on ILs appears to have some potential, but more focused study would be required to realize competitive performance. When the additional drawbacks of IL electrolytes are factored in, such as high cost and low t + , the outlook becomes discouraging, at least for this application space. We suggest that additional research might focus on IL/organic solvent mixtures rather than pure ILs. At least one existing study has validated this approach: Wang and coworkers found that 1 wt% EMIMBF 4 added to a standard EC/EMC electrolyte could nearly double the discharge capacity of Gr8NMC 523 pouch cells at À30 1C. 117 Furthermore, it remains to be seen whether the general trends discussed in this review still apply to a majority-ionic system, e.g. is charge transfer still the limiting step at low temperature, and is solvation structure still predictive of R ct when the coordinating species are themselves charged? Further research into these fundamental questions is encouraged.
Other solvent types. The strong electron-donating properties of organosulfur compound tetrahydrothiophene-1-oxide (THTO) were taken advantage of by Oldiges and coworkers,

Review
Energy & Environmental Science who blended this liquid in varying ratios with PC to form an electrolyte solvent base. 118 Combined experimental and simulation data indicated that THTO preferentially replaced PC in the solvation shell of Li + , which was reduced to form a passivating SEI which prevented PC co-intercalation. When compared to a standard electrolyte (1 M LiPF 6 in EC/ DEC 1 : 1 w/w), the mixed electrolyte 1 M LiPF 6 in THTO/PC 15 : 85 mol/mol roughly doubled the capacity of Gr8NMC 111 cells at 0 1C and 1C rate. One extraordinarily unique low-temperature system comes from Rustomji et al., who reported electrolytes based on liquefied gases. 119 Using a pressurized cell, these authors were able to dissolve 0.1 M LiTFSI in fluoromethane, a gas with a boiling point of À78 1C at 1 atm. Due to the extreme low viscosity of this material (owing to the weak interactions between molecules), the resulting electrolyte displayed high conductivity B10 À3 S cm À1 across all temperatures from À60 1C to 25 1C, despite the comparatively low salt concentration. Addition of 5% CO 2 enabled formation of a stable SEI on lithium metal and allowed more salt to be dissolved. Li8LiCoO 2 half cells cycled perfectly well in a liquefied gas electrolyte (0.2 M LiTFSI in fluoromethane/CO 2 19 : 1) down to À60 1C (60.6% capacity compared to room temperature) and performed similarly at 25 1C to 1 M LiPF 6 in EC/DEC 1 : 1 w/w. The same group recently followed up on this work using either tetrahydrofuran 120 or acetonitrile 121 as solvating additives to achieve higher salt concentrations and improved performance with lithium metal. It should be noted, of course, that the vapor pressures of these liquified gas solvents are multiple MPa under ordinary conditions, creating an inherent challenge to any practical application of this system.

Primary salts
No LIB can function without a lithium-containing electrolyte to transport Li + across the cell interior and, when it comes to sources of dissociable lithium, LiPF 6 is the undisputed first choice. It has been said that LiPF 6 earned this position not due to any one outstanding property, but rather due to its lack of any crippling disadvantages. 13 That being said, non-PF 6 À salts can -and have -been utilized in application-specific LIBs, where they may provide certain upside. For example, different salts may have better solubility in non-typical solvents or participate in the passivation of electrode interfaces. These characteristics can be leveraged to the benefit of lowtemperature performance. Salt formulations investigated by the references cited here may be found summarized in Table 3 along with their structures. Borates (LiBF 4 , LiBOB, LiDFOB). Zhang, Xu and Jow reported as early as 2002 that LiBF 4 could enable better cell performance than LiPF 6 at sub-zero temperatures, despite poorer ionic conductivity and SEI characteristics in its electrolytes. 122 Subsequent analysis by these authors found a dramatic reduction in full-cell R ct when LiBF 4 was used (Fig. 14a), 123 although to our knowledge the chemical mechanism behind this effect has yet to be clarified.
Unfortunately, LiBF 4 has not gained widespread popularity as a primary salt due to the inadequate SEI formed in its electrolyte solutions, which precludes extended cycling. On the other hand, lithium bis(oxolato)borate (LiBOB) has received steady investigation precisely because of its SEI-forming abilities, which may reduce or eliminate the need for EC solvent. 124 Naturally, this has attracted interest for low-temperature electrolyte designs, 110,125,126 where it is desirable to minimize EC content. However, LiBOB comes with its own set of drawbacks, including poorer solubility and conductivity than LiPF 6 , which limit its applications as well. Kang Xu has explored the topic of LiBOB electrolyte design in more detail elsewhere. 127 An appropriate compromise can be found in lithium difluoro(oxolato)borate (LiDFOB), which shares the SEIforming characteristics of LiBOB, but with improved solubility and ion dissociation characteristics. This salt has proven to be suitable for a wide range of electrolyte designs, 128 including lactone-based formulations. 109,111 Several groups have reported that a mixed-salt approach -using LiDFOB together with LiBF 4can maximize the advantages of both compounds at low temperature. For instance, Li et al. investigated an EC/DMS/ EMC 1 : 1 : 3 v/v electrolyte containing 0.9 M total of LiDFOB and LiBF 4 in a 5.365 : 1 mass ratio. 77 These authors found the ionic conductivity of this mixture to be similar to that a traditional LiPF 6 -based blend across the temperature range À40 to 20 1C. Importantly, the electrolyte was found to be compatible with both graphite and LiFePO 4 in half-cell tests, enabling drastically-higher capacities than the control electrolyte at À20 1C (Fig. 14b). Zhou and coworkers conducted a detailed study of similar electrolytes (1 M salt in EC/DMC/EMC 1 : 1 : 1 w/w) with varying DFOB/BF 4 ratio across the range À20 to 60 1C. 129 Pure LiDFOB performed universally best at room temperature and higher, producing the best conductivity and discharge capacity in Gr8LMNO cells, but this trend was exactly reversed at À20 1C, with blends showing intermediate properties (Fig. 14c). An ideal balance was struck at 0.8 M LiDFOB and 0.2 M LiBF 4 , since even a relatively small amount of LiBF 4 was found to reduce cell R ct from 482.6 O to 346.3 O at À20 1C, without compromising the effective passivation of LiDFOB. As a result, cells containing this blend showed excellent capacity retention across all temperatures.
Sulfonylimides (LiTFSI, LiFSI, LiBETI). Most of the sparse literature on LiPF 6 -free electrolytes below 0 1C has focused on borate salts, with other structures proving to be quite rare. Nonetheless, exceptions do exist, mostly when unique solvents are involved, e.g. the IL-based, 112,114,116 ester-based 107 or highlyfluorinated electrolytes 94 described in sections above. These designs exploit the high solubility, dissociation ability and chemical stability of lithium sulfonylimide salts (LiTFSI, LiFSI, LiBETI, etc.) -as well as the interface-stabilizing characteristics of LiFSI 113 -in situations where LiPF 6 is unsuitable. Mandal and coworkers did publish an article in 2006 which directly compares LiTFSI to LiPF 6 (0.9 M salt in EC/DMC/EMC 15 : 37 : 48 w/w) across various temperatures. 130 Most interestingly, the LiTFSI electrolyte showed nearly an order-of-magnitude reduction in charge-transfer resistance on lithium metal at 30 1C, although no measurements were done at lower temperatures.  [131][132][133] Overall, LiTFSI and its related compounds appear to be underutilized in low-temperature LIB literature, and it is our opinion that the field could benefit by adding this class of materials to its toolbox. Potential investigators should be warned, however, that these salts may cause severe corrosion to Al current collectors unless additional passivating strategies are employed. 134

A note on silicon anodes
Although graphite has been the anode of choice for virtually the entire commercial history of LIBs, future designs are likely to rely on some combination of silicon and carbon due to the former's high specific capacity, earth abundance and mature commercial production. The various characteristics and challenges of silicon and Si/C composite anodes under ''ordinary'' operating conditions have been covered elsewhere extensively. 64,[135][136][137] Curiously, however, we were only able to locate a handful of reports that directly address silicon performance at low temperatures. One of the earliest reports came in 2005 from Kasavajjula and Wang, who developed a composite anode containing graphite and nano-Si in a 5 : 1 ratio, along with a solid electrolyte filler and PEO-LiClO 4 binder. 138 In half cells with lithium metal and an electrolyte of 1 M LiPF 6 in 1C. Careful impedance analysis led the authors to conclude that increasing charge-transfer resistance was responsible for the capacity loss below À20 1C, although the inclusion of solid electrolyte helped this situation significantly compared to a control electrode without it. Several more articles have been published on this topic within the past five years as silicon has inched closer to widespread adoption. Markevitch, Salitra and Aurbach compared the performance of monolithic amorphous silicon (a-Si) anodes (capacity-limited to 600 mA h g À1 lithiation) to graphite counterparts of identical areal capacity, using half cells with   15 (a-c) Cycling performance of Si nanoflake powder half-cells containing various electrolytes at varying temperatures. The base electrolyte was 1 M LiPF 6 in EC/DEC 1 : 1 v/v, with 10 wt% ''additive'' compounds as indicated. FEC appears to offer the best trade-off between high-temperature stability and low-temperature discharge capacity. (d and e) 10th cycle charge and discharge curves for the above cells at different temperatures. While VC-added electrolyte produces the best capacity (by a small amount) at 25 1C and 60 1C, it creates extreme polarization in the cell at À5 1C. Both FEC and VC additives appear to increase cell resistance at sub-zero conditions relative to the base electrolyte. Reprinted from ref. 93 139 Following lithiation at 30 1C, the delithiation capacity of a-Si exhibited virtually no temperature dependence down to À30 1C; furthermore, the silicon anodes could be charged and discharged under such conditions with no observable capacity drop, while graphite lost capacity rapidly when charged at 0 1C and below. Haruta et al. compared the effects of several common electrolyte additives on Si nanoflake anodes over a wide temperature range (Fig. 15), 93 finding 10 wt% FEC to be the most effective at À5 1C and moderately effective at 60 1C. In contrast, 10 wt% VC produced slightly better capacity at high temperature, but drastically reduced capacity at sub-zero temperature. Notably, all of the above studies derived their conclusions from half cells with lithium metal. In contrast, two articles published just months prior to this writing examined commercial large-format cells containing Si or Si/Gr anodes paired with NCA cathodes. Subburaj and coworkers studied 1 A h pouch cells made from sputter-deposited Si anodes and a conventional electrolyte blend (1 M LiPF 6 in EC : DMC : DEC 1 : 3 : 1 v/v). 140 These cells retained an impressive 65.3% of their 20 1C discharge capacity at À40 1C; however, capacity faded rapidly after 40 low-temperature charge/discharge cycles. Notably, the authors identified cathode structural transformation as the cause of this degradation, rather than lithium plating or other anode processes. Richter and coworkers took a different approach, examining minor-Sicontent (3.5 wt%) anodes in 18 650 format cylindrical cells with operando neutron diffraction. 141 This experiment revealed that, under low-temperature charge conditions, graphite was lithiated more rapidly than silicon despite its lower potential, indicating a kinetic limitation that may be a target for future optimization.
Doubtless, many questions remain about the characteristics of silicon anodes below 0 1C, indicating that this area is ripe for further investigation. We observe that -generally speakingsilicon has been reported to resist low-temperature capacity loss better than graphite, an advantage also shared by the related alloy-type material tin. 142 However, it remains to be seen whether this is an intrinsic benefit, or merely a result of different particle shapes/sizes, electrode fabrication methods and testing protocols. We strongly encourage additional research in this area. Results from the various references cited here are summarized in Table 4.

Summary and outlook
The future of lithium-ion batteries (LIBs) will require them to fill more roles than ever before and do so across a wider range of operating temperatures. It is well-known that the available discharge capacity of lithium-ion cells (even when charged at room temperature) tends to drop off considerably below À20 1C, which poses a significant challenge for applications in aerospace and transportation where sub-zero ambient conditions are common. This is largely due to a massive increase to internal resistance as the temperature is lowered. Solving the issue of low-temperature performance requires careful consideration of the many simultaneous chemical processes occurring within a lithium-ion cell, each of which contributes an associated impedance. Often, the best way to address this complex problem is by way of the one component which affects nearly every internal process: the non-aqueous liquid electrolyte.
In this review, we first summarized the many individual sources of impedance that are associated with liquid electrolytes and the role that they play in low-temperature performance. A typical lithium-ion electrolyte consists of 1 M LiPF 6 dissolved in a liquid mixture of r50% ethylene carbonate (EC) along with linear carbonates (DMC, DEC, EMC) and a few wt% of additives (e.g. FEC, VC). While the bulk physical properties of these formulations, e.g. ionic conductivity, have been well-studied due to their importance at moderate-to-high temperatures and/ or high currents, low-temperature performance does not generally correlate with bulk conductivity. If anything, the most relevant physical property is instead liquidus point, where solids begin to block electrode pores and reduce ion access to active surfaces. However, by far the biggest limiting factor in sub-zero LIB operation is charge-transfer resistance, a process that has been linked by several studies to Li + desolvation at the electrolyte/active material interface. These large desolvation energies and high freezing points are both largely due to the use of EC, which is a high-melting compound (36 1C) that possesses the large dielectric permittivity (e = 90 at 40 1C) necessary to dissociate lithium salts from their counterions. EC is also indispensable because of its ability to passivate graphite anode surfaces upon reductive breakdown, forming a solid-electrolyte interphase (SEI) that stops further electrolyte decomposition and stabilizes the carbon structure to exfoliation upon repeated cycling. While the SEI (and its thinner cathode counterpart, the CEI) do indeed contribute some impedance to the cell, evidence suggests that interphasial ion transport is not the primary limit on low-temperature performance. Rather, the importance of SEI comes from its ability to prevent capacity loss via parasitic corrosion current, which is critical at all temperatures and becomes especially challenging when attempting to reduce EC content in the electrolyte. Attempting to charge a LIB under sub-zero conditions also introduces new capacity loss mechanisms associated with lithium deposition at low anode potentials.
Just as the causes of low-temperature capacity loss are multifaceted, so too are the potential strategies to address them via electrolyte engineering. We have concisely summarized recent developments in this area, which can be broken down into three primary research thrusts: additives, solvents and salts. Successful electrolyte additives generally assist in the formation of robust SEI/CEI layers with low resistance to Li + transport, which reduces reliance on EC as a main solvent component. Studied additives include molecular compounds like fluoroethylene carbonate (FEC), tris(trimethylsilyl)phosphite (TMSPi) and sulfur compounds in a variety of oxidation states, e.g. dimethyl sulfite (DMS) and 1,3,2-dioxathiolane-2,2-dioxide (DTD). Ionic compounds may also fill this role, including novel lithium salts like LiPO 2 F 2 or even non-lithium-containing salts such as CsPF 6 . Alternately or in combination with additives, many authors have sought to augment or replace traditional solvents as well.

Review
Energy & Environmental Science Open Access Article. Published on 14 January 2022. Downloaded on 10/4/2022 7:29:02 PM. This article is licensed under a Creative Commons Attribution 3.0 Unported Licence.

View Article Online
One of the simplest, but most effective, solvent components at low temperature is propylene carbonate (PC), which shares many characteristics of EC, but with a lower melting point (À49 1C). Modern additive developments have made it possible to partially or entirely replace EC with PC without causing undesirable breakdown of graphite, as known from the early days of LIB research. FEC and vinylene carbonate (VC) have also been explored as major solvent components. However, many researchers have turned to another class of solvents entirely: esters, which possess low melting points and low viscosities, while being of moderately polar character. On the other hand, low-molecular-weight esters are often detrimentally unstable at low anode potentials; an exhaustive study has found methyl propionate (MP) to possess the best trade-off in properties, although protective additives or higher-potential anodes like Li 4 Ti 5 O 12 (LTO) can eliminate stability concerns. The cyclic ester gamma-butyrolactone (GBL) has also shown great potential as a replacement for chemically-similar EC. Other creative solvents include room-temperature ionic liquids (RTILs) and liquified gases, i.e. fluoromethane contained under pressure. Finally, other investigators have turned away from the ubiquitous LiPF 6 in favor of other lithium salts. In particular, borates have received significant attention due to the discovery of vastly-reduced charge-transfer resistance in LiBF 4 -based electrolytes at low temperatures. Recently, lithium bis(oxalate) borate (LiBOB) and lithium difluoro(oxalato)borate (LiDFOB) have been characterized for sub-zero applications and been found to possess intriguing advantages. Other salts like lithium bis(trifluoromethane)sulfonylimide (LiTFSI), while wellcharacterized in other contexts, have received comparatively little investigation at low temperature. Finally, we would like to conclude by pointing out several emerging areas of research for low-temperature LIB electrolytes which deserve greater attention. The most obvious of these is in silicon-based anode materials, which have seen increasing commercial attention in combination with graphite and are likely to entirely replace graphite at some point in the future. Despite the explosion of silicon research over the past decade, there have been remarkably few published reports on Si anodes at low temperatures. It remains to be seen how the ''conventional wisdom'' enumerated in this review will apply to Si materials systems, especially since the anode is generally considered to be the dominating factor when it comes to sub-zero battery operation. Additionally, there are many electrolyte components which, for one reason or another, were deemed unsuitable during the early days of LIB research, despite marked advantages at low temperature. A perfect example is PC, which cannot form a protective SEI on its own, but is now enjoying a revival due to the advent of sacrificial additives. GBL, on the other hand, has yet to experience such renewed interest, despite several promising recent results, especially in combination with borate salts like LiDFOB for which GBL seems to exhibit a particular synergy. It should also be mentioned that fluorine-containing solvents, especially hydrofluoroethers, have (deservedly) enjoyed recent popularity as electrolyte components for a variety of applications. 94,110,[143][144][145] While a handful of these reports have addressed sub-zero performance, this angle remains underexplored given their tendencies towards wide liquid range and low Li + solvation energy. In general, while incremental progress remains possible, the most exciting developments in low-temperature LIB performance have often come from the most novel design strategies, such as highly-fluorinated electrolytes, liquefied gases or non-lithium metallic salt additives. With a greater commitment to push the envelope on electrolyte development, we can expect a bright future for batteries that will carry us down the road, across the sky, through space and to the limits of human imagination.

Conflicts of interest
There are no conflicts to declare.