Investigating and inhibiting cathode-side gas generation for practical low-temperature and fast Li-ion batteries

Abstract

Weakly solvating electrolytes attract increased interest in low-temperature lithium-ion batteries since they can overcome the kinetic limitations of lithium-ion desolvation. However, the high content of weakly-coordinating/free solvents poses a threat to the oxidative stability of weakly solvating electrolytes. Herein, we investigate the oxidation and gas-producing behaviour of weakly solvating electrolytes with typical low-melting-point solvents inside low-temperature pouch cells. Excluding the influence of the anode, the results show that the continuous gassing from the oxidation of solvents in weakly solvating electrolytes on the cathode significantly affects the lithium-ion chemistry, leading to low-temperature cells with a fast capacity decay. Constructing a stable cathode electrolyte interphase, such as adding lithium difluoro(oxalato)borate into weakly solvating electrolytes, could suppress oxidation-induced gas production and therefore enable the 3.5 Ah cylindrical cell with nearly two years of stable operation (2100 cycles with 81.1% retention) at −20 °C. These findings highlight the importance of a trade-off that balances the lithium-ion kinetics and oxidative stability in weakly solvating electrolytes when the low-temperature issue of lithium-ion batteries is overcome.

Weakly solvating electrolytes can overcome the kinetic limitations of lithium-ion desolvation at low temperatures, but their high content of weakly coordinating/free solvents threatens oxidative stability. Here, the authors introduce lithium difluoro(oxalato)borate to construct a stable cathode electrolyte interphase, suppressing oxidation-induced gas production and enabling a 3.5 Ah cylindrical cell to achieve stable operation for nearly two years at −20 °C.

Introduction

Lithium-ion batteries (LIBs) are now ubiquitous in terrestrial applications ranging from smartphones to electric vehicles (EVs). However, their performance is only satisfactory at moderate operating temperatures (~25 °C)^1–3. Due to the limitations of battery chemistry, including electrolyte conductivity, Li-ion reaction kinetics, and solid-state diffusion, the drastically increased overpotential prematurely drives the voltage to the end-of-discharge voltage under low temperatures; this diminishes the reversible capacity by almost 60% compared to that at room temperature^4–6. Therefore, the low-temperature application of current LIBs, such as EVs in colder climates, is insufficient and suffers from many concerns. The U.S. Advanced Battery Consortium requires EVs to have the ability to stably cycle above −30 °C^7–12. This requirement poses formidable challenges to the LIB community, especially for researchers focusing on electrolyte development^13,14.

Chemically stable solvents with low freezing points and adequate dielectric constants, such as methyl propionate (MP), methyl acetate (MA), ethyl methyl carbonate (EMC), and diethyl ether (DEE), are useful for accessing electrolytes with high ionic conductivity and sufficient liquid ranges at low temperatures¹⁵. As reported by Smart et al.¹⁶, the reversible capacity of the cell could improve by more than six times at −50 °C if MP is substituted for 20 vol.% ethylene carbonate (EC) in the baseline electrolyte. Reducing the energy barrier of the rate-limiting process could further modify the low-temperature performance¹⁷. Since the Li-ion desolvation on anodes is the energy-consuming step^18,19, the vast majority of research has been dedicated to exploring the weakly solvating electrolyte (WSE) formulations that support weak Li⁺–solvent interactions^20,21. A 2021 report by Liu et al. reveals DEE to be one of the most effective solvents for reducing the desolvation energy, allowing the cell to retain 84% of its room temperature capacity (RTC) when cycled at −40 °C¹⁷. Wang and co-workers further noted that the cell with methyl difluoroacetate-based WSE delivers 83% RTC at −30 °C⁴.

Despite the excellent efficacy of WSEs in low-temperature LIBs, their development should be cautiously viewed as a “double-edged sword” with the potential to address the sluggish kinetics but, conversely, introduce some new complications. In this regard, the oxidation behavior of WSEs warrants further examination. In regular carbonate-based electrolytes (RCEs), the solvents react with Li ions to form universal tetra-coordinated solvation structures via strong ion−dipole interactions; this results in a decrease of solvent molecules in the lowest unoccupied molecular orbital (LUMO) and highest occupied molecular orbital (HOMO) energy levels. Therefore, in comparison to free solvents, Li-coordinated solvents have a greater tendency to undergo decomposition on the anode surfaces while showing high endurance to oxidization on the cathode²². By contrast, from the perspective of electronic energy levels, WSE with high free solvent content is supposed to present a high tendency to be oxidized on the cathode during cell cycling, even at a low temperature^23–25. However, the understanding on WSE oxidization behavior and potential gas production at low temperatures is still limited, which hinders the application of WSE in high-voltage cathodes and increases the difficulty of improving the lifespan of WSE batteries.

In this study, we investigated the low-temperature oxidization and gas evolution of WSEs inside 1.2 A h pouch cells with a LiNi_0.5Co_0.2Mn_0.3O₂ cathode and graphite-SiO_x composite (GSiO) anode by Operando differential electrochemical mass spectrometry (DEMS). Nine solvents with low freezing points are shown to be separately involved in the WSE formulations, and their HOMO/LUMO energy levels and solvation structures/energies are identified by Raman and nuclear magnetic resonance (NMR) combined with theoretical calculations. In addition, the relationships between the solvent properties and solvation structure/solvation energy/gas production are explored. As a result, all WSEs experience severe oxidization and gas production at −20 °C. The expanded cells are noninvasively detected by nano-computed tomography (nano-CT). A simple solution for suppressing oxidation-involved gas production inside the WSE cell is further clarified by introducing the lithium difluoro(oxalato)borate additive. The optimized WSE with high oxidation resistance and favorable desolvation enables the pouch cell to retain an 80% RTC at −20 °C. The cycling of 3.5 A h GSiO||LiNi_0.8Co_0.1Mn_0.1O₂ cylindrical cells is also evaluated at −20 °C, exhibiting a compelling 81.1% capacity retention after 2100 cycles.

Results

Li-ion solvation and transport in WSEs

Ten low-freezing-point solvents were employed in the electrolyte preparation: propylene carbonate (PC, 1), methyl acetate (MA, 2), ethyl acetate (EA, 3), ethyl propionate (EP, 4), ethyl methyl carbonate (EMC, 5), diethyl carbonate (DEC, 6), diethyl ether (DEE, 7), dipropyl ether (DPE, 8), dibutyl ether (DBE, 9), and isopropyl ether (IPE, 10). The detailed information of the solvents is listed in Supplementary Table 1. Lithium bisfluorosulfonimide (LiFSI, a Li salt) was used for electrolyte preparation to maintain the high liquid ranges of electrolytes at low temperatures. After stirring and resting, the electrolyte (S-0, where S is the solvent) consisting of 1.0 M LiFSI and the mono-solvents all exhibited good liquidity at −20 °C (Supplementary Fig. 1). Notably, 10 wt% ethylene carbonate (EC, 11) was added to PC-based electrolytes to inhibit the co-intercalation of the PC molecules into the graphite interlayer of the graphite-SiO_x (GSiO) anode during the lithiation process. It is worth noting that EC was only used in the PC-0 to form SEI on the anode, and was not employed in subsequent theoretical calculations nor included in the electrolyte thereafter.

The Li⁺–solvent interactions were determined by density functional theory (DFT). As shown in Fig. 1a and Supplementary Fig. 2, the binding energy for Li⁺–PC was highest (below −1.78 eV) among the studied solvents. NMR results also exhibited that the PC molecules tended to coordinate closely with the Li ions (Supplementary Fig. 3). By contrast, other solvents, to a certain extent, all weakly interacted with the Li ions, showing a binding energy of above −1.55 eV. This result indicates that the PC-involved solvation may have a more robust structure relative to the solvation with other solvents²⁶. Since the electrostatic interaction between the Li ion and solvent affects their binding energy, the electrostatic potential (ESP) distribution of the solvents was further calculated to determine the strength of the solvation. As shown in Fig. 1b, the lowest negative electrostatic potential (ESP_min) and highest positive electrostatic potential (ESP_max) for the ten solvents were noticeably different (Supplementary Table 2). Except for the PC, the other nine solvents all exhibited |ESP_min| » ESP_max, corresponding to the feature of the weakly solvating solvent (WSS)²⁷. By contrast, the positive and negative surface electrostatic potentials of PC were both relatively strong (|ESP_min| ≈ ESP_max), which, to some extent, indicated the high polarity of the PC, thereby resulting in a high Li⁺–PC binding and stronger solvation. Combining the ESP and binding energy results, the EA and DEC with the medium Li binding energy (−1.55 eV) among the studied solvents were used as a navigation mark to briefly distinguish the strength of solvation structures. Solvents with a binding energy lower than −1.55 eV are considered strongly solvating solvents (like PC, −1.78 eV), and solvents with a binding energy greater than or equal to −1.55 eV are considered WSS in this work. It is important to emphasize that the single, the binding energy threshold of −1.55 eV does not possess universal physicochemical significance. The Li⁺-solvent binding energy is influenced by multiple factors, such as the type of lithium salt, salt concentration, and solvent mixtures. The value of the −1.55 eV threshold proposed in this manuscript lies primarily in its role as an effective classification tool for the specific system studied herein.

Fig. 1. Fast Li-ion desolvation kinetics in WSEs.

a Interaction between the Li ions and solvents. All solvents include propylene carbonate (PC), methyl acetate (MA), ethyl acetate (EA), ethyl propionate (EP), ethyl methyl carbonate (EMC), diethyl carbonate (DEC), diethyl ether (DEE), dipropyl ether (DPE), dibutyl ether (DBE), and isopropyl ether (IPE). b Calculated ESP distribution of different solvents. c Percentages of SSIP, CIP, and AGG in the electrolytes. The S-0 series consists of 1.0 M LiFSI dissolved in various mono-solvents, including PC, MA, EA, EP, EMC, DEC, DEE, DPE, DBE, and IPE. d E_a values of Li-ion desolvation (R_ct) and Li ions crossing the SEI (R_f). e XPS spectra of the GSiO anode after 2 cycles in the MA electrolyte with and without FEC additive. f Cryo-TEM images of the GSiO anode after cycling in the MA electrolyte with and without FEC additive. DRT analyses of half-cells using g PC-1, h MA-0, and i MA-1 at −40 ~ 25 °C.

From the Raman results, the peak at 600–1000 cm⁻¹ significantly changed after adding LiFSI into the solvent (Supplementary Figs. 4 and 5), which was correlated with the Li coordination behavior. Due to the high ion–dipole interaction, numerous PC molecules coordinated with Li ions, resulting in the solvent-separated ion pairs (SSIPs) proportion as high as 51% in PC-0 (Fig. 1c), while the proportion of contact ion pairs (CIPs) and ion aggregates (AGGs) merely accounted for 39% and 10%, respectively. For WSS, the SSIP proportion significantly decreased in MA-0, EA-0, EP-0, EMC-0, and DEC-0 and was very rare or even undetectable in DEE-0, DPE-0, DBE-0, and IPE-0. As shown in Supplementary Fig. 6, MA exhibits a higher Li⁺ transference number (t_Li+). This demonstrates that in WSE, the greater participation of anions in coordination results in t_Li+ that are distinctly different from those in strongly solvating systems. The coordination environment of the electrolytes at different temperatures (25, 0, −20, and −40 °C) was further investigated through molecular dynamics (MD) simulations. The results indicate that at room temperature, due to the weaker solvating power of MA compared to PC, Li⁺ tends to coordinate more with anions, leading to a higher proportion of AGG in the solvation structure. Meanwhile, as the temperature decreases, the proportion of AGG in both electrolytes shows an increasing trend. However, even at −40 °C, the AGG content in the MA-containing electrolyte remains significantly higher than that in the PC-based electrolyte (Supplementary Figs. 7–9 and Supplementary data).

The activation energies (E_a) for Li-ion desolvation and Li-ion crossing of the solid electrolyte interphase (SEI) were investigated by separately fitting the charge-transfer resistance (R_ct) and interfacial resistance (R_f) to the Arrhenius equation. The electrochemical impedance spectroscopy (EIS) profiles shown that the R_ct and R_f values in the Li metal||GSiO half-cells increased as the temperature decreased from 25 °C to −40 °C (Supplementary Fig. 10 and Table 3), this was attributed to the decrease in the Li-ion conductivity and reaction kinetics at low temperatures (Supplementary Fig. 11). With the exception of PC-0, the electrolytes featured with WSS all exhibited an E_a,Rct less than 50 kJ mol⁻¹ for the Li-ion desolvation process (Fig. 1d and Supplementary Fig. 12), corroborating well with the above theoretical analysis. On the other hand, the studied WSEs delivered lower E_a,Rf than the PC electrolyte, indicating the fast Li-ion transport inside the SEI. This is likely due to the high Li-ion flux of the inorganic-rich SEI derived from the CIP and AGG clusters of WSEs. One can conclude that PC, as a strongly solvating solvent, leads to less free solvent and strong solvation, while the WSSs with a Li binding energy higher than −1.55 eV show weak solvation and enable electrolytes with a high content of free solvents.

Before investigating the gas production from the WSE oxidation, it is necessary to eliminate the influence of the parasitic reaction of the electrolyte at the anode. Five weight percent fluoroethylene carbonate (FEC) was added to the electrolyte (S-1, where S is the solvent) to reduce the interference of the anode-involved gas. FEC is selected as an additive for silicon anodes primarily because it decomposes preferentially to form a stable, flexible, and LiF-rich SEI. As shown by the in-depth XPS profiling of the GSiO anode after 2 cycles in MA-0 and MA-1 (Fig. 1e and Supplementary Fig. 13), the LiF (684.6 eV) proportion in the SEI increased due to the FEC decomposition. Due to its lower LUMO energy level, FEC can decompose preferentially over LiFSI and other solvents at the anode side. Meanwhile, the strongly electron-withdrawing fluorine atom in FEC acts as a nucleophile, attacking the sulfur atom in the S–F bond and promoting its cleavage via a nucleophilic substitution mechanism, thereby generating more stable LiF^28,29. Similarly, the weakened S–F peak in the F 1s XPS spectrum of the MA-1 anode confirmed that the parasitic reaction was effectively restricted because the film-forming properties of the FEC were superior to those of PC, EC, and WSSs. The morphologies of the SEI formed in WSE with and without FEC also differed. The cryo-transmission electron microscopy (cryo-TEM) images revealed a relatively uniform SEI with a thickness of ~17 nm in MA-1 compared to that in MA-0 at −20 °C (Fig. 1f). This robust SEI has several advantages, such as adequate Li-ion flux and good mechanical properties; thus, the interfacial stability of the anode in MA-1 was high (Supplementary Figs. 14 and 15), which could reduce the tendency of the anode-involved gas production (it will be discussed later).

The distribution of relaxation times (DRT) analysis was performed on the Li metal||GSiO half-cells to examine the responsiveness of charge transfer kinetics to temperature in PC-1, MA-0, and MA-1 (Supplementary Fig. 16). As depicted in Fig. 1g, four peaks appeared in the DRT profiles of the PC-1 cell at 25 °C, in which the third peak (0.023 s) corresponded to the R_ct process in the inorganic phase and was used as a basis for inferring the desolvation behavior. As a consequence of the strong Li⁺−PC interaction, the relaxation time (τ) for the inorganic R_ct process (denoted as τ_Rct) in the PC-1 cell gradually increased with decreasing temperature, revealing the sluggish Li-ion desolvation at low temperatures. The τ_Rct of the MA-0 cell slightly changed in the range of 25 to −20 °C. However, a more drastic change was observed when the temperature reached −40 °C (Fig. 1h). This result may be related to the cracking and subsequent reformation of the SEI due to stress concentration under low temperatures, which can lead to a significant increase in interfacial impedance. Cryo-TEM images of cycled anode confirm that the addition of FEC greatly suppresses both the cracking and growth of the SEI (Supplementary Fig. 17). In contrast, the τ_Rct of the MA-1 cell at −40 °C was comparable to that at 25 °C (Fig. 1i), without an evident shift. These results showed that the reaction kinetics were less susceptible to temperature under a weak solvation environment, together with the FEC-derived SEI. As reflected in the τ_Rct shift and the corresponding impedance intensity (Supplementary Fig. 18), other S-1-type WSEs also had similar kinetic characteristics to MA-1 (Supplementary Fig. 19) at low temperatures.

Cycling and gas production of the WSE pouch cells

The cycling tests were carried out on a 1.2 A h GSiO||LiNi_0.5Co_0.2Mn_0.3O₂ (NCM523) pouch full-cell. Supplementary Table 4 provides detailed information on the cell assembly. Supplementary Fig. 20 shows the voltage profiles for the charge–discharge process at 0.2 C (1 C is 1.2 A). The room temperature capacities (RTCs) of the S-1 cells were nearly 1.23 A h. Due to strong desolvation barriers, the PC-1 cell demonstrated a 59.5% RTC during cycling at −20 °C (Fig. 2a). In contrast, greater than 70% RTC was retained at −20 °C by using WSEs, with the exception of EA-1. This “abnormal” result was likely induced by the parasitic reaction of EA-1 on the cathode. A comparison of the cycling of the cells at 0.2 C and −20 °C is shown in Fig. 2b. The PC-1 cell merely delivered an average of 0.7 A h reversible capacity (Supplementary Fig. 21). However, its capacity retention was the highest among the tested cells due to the stable electrolyte/electrode interface. In this sense, the degeneration of WSE cells was largely engendered by the unstable cathode interface. Based on the 3D reconstruction of time-of-flight secondary ion mass spectrometry (TOF-SIMS) (Supplementary Fig. 22), it can be observed that the decomposition of more free solvent in MA-1 leads to an increase in organic components (C₂H⁻) at the cathode interface, whereas the organic components in the PC-based electrolyte are significantly reduced. Furthermore, it was observed that after cycling in the MA-1 electrolyte using atomic force microscopy-nano-infrared (AFM-nano-IR), the distribution of the key organic component in the SEI—lithium ethylene dicarbonate (LEDC) (1640 cm⁻¹) wavenumber signal on the cathode was significantly stronger than that in the PC-1 system, which is consistent with the TOF‑SIMS results (Supplementary Figs. 23 and 24)^30,31. One exception is the ether-based WSEs using DEE, DPE, IPE, and DBE. The cell with those ether electrolytes delivered not only low initial capacity but also serious cycling fading even with FEC additive, which the performance is incomparable to that of the cells using ester-based electrolytes (Supplementary Fig. 25). This could be attributed to the co-intercalation and serious parasitic reactions of the ether-based WSSs (Supplementary Fig. 26). Therefore, ether-based WSEs were dropped in the following research since they were currently incompatible with the graphite-blended anode and high-voltage charging^32,33.

Fig. 2. LIB failure under WSE-derived gas generation.

a Capacity retention of the pouch cells with S-1 and S-2 electrolytes at −20 °C relative to their capacity at 25 °C. The cycling performance of the pouch cells with b S-1 and c S-2 electrolytes at 0.2 C and −20 °C (The 1 C is 1.2 A). Digital photos of the pouch cells with d S-1 and e S-2 electrolytes after cycling at 0.2 C and −20 °C. Nano-CT images of the pouch cells with f MA-2 and g MA-1 after 100 cycles at −20 °C. h Three-electrode tests of the pouch cells with MA-1 and MA-2 at 0.2 C and −20 °C. Digital photos of the separator extracted from pouch cells with i MA-1 and j MA-2 after 100 cycles at −20 °C.

With these data in hand, constructing a stable cathode electrolyte interphase (CEI) could further improve the low-temperature performance of the WSE cells. Lithium difluoro(oxalato)borate (LiDFOB), which is a well-known cathode film-forming agent, was then introduced into S-1 electrolytes³⁴. The S-1 with 5 wt% LiDFOB added was recorded as S-2. As shown in Fig. 2a, the capacity retention of the S-2 cells was nearly identical to that of the S-1 cells after tuning the temperature from 25 °C to −20 °C. However, the cycling stability was quite different between the two sets of WSE cells. Despite the PC-2 cell showing negligible cycling improvement with respect to the PC-1 cell, the cycling of the S-2-type WSE cells was significantly improved (Fig. 2c and Supplementary Fig. 27).

Comparing the volume of the pouch cells after cycling in S-1 and S-2 can provide more direct evidence for understanding the performance differences. Notably, the gas derived from the activation cycles at 25 °C was completely removed before the tests. As shown in Fig. 2d, expansions of the S-1 cells were all evident after cycling at −20 °C (Supplementary Fig. 28a). The S-1-type WSE cells exhibited deformations that were close to reaching the rupture limit of the battery package. With LiDFOB, the tendency of gas evolution in S-2 cells all decreased, as indicated by the cell volume being stable over 100 cycles (Fig. 2e and Supplementary Fig. 28b). In addition, nano-CT images of the cycled MA-2 cell revealed the tight electrode stacking structure and no newly formed gaps and twists inside the cell (Fig. 2f and Supplementary Movie 1). For the MA-1 cell, severe gas production led to the stretching and contact loss of the stacked electrodes (Fig. 2g and Supplementary Movie 2). These results initially verified our deduction that WSE also preferred to be oxidized to produce gas when cycled at low temperatures, similar to its behavior at 25 °C (Supplementary Fig. 29).

If the generated gas is not discharged in a timely manner, an increase in polarization at the reaction interface will occur, disturbing the intrinsic lithiation process of the working cell. Three-electrode tests with the Li metal reference electrode clearly indicated that the anode potential quickly reached below 0 V (vs. Li/Li⁺) during the charging process of the MA-1 cell, resulting in Li plating (Fig. 2h, Supplementary Figs. 30 and 31). Other S-1-type WSE cells exhibited the same trend for Li plating as MA-1 cells due to enlarged interfacial resistance and gas production. For the MA-2 cell, the charging process was stable, with a Li plating-free anode interface (Supplementary Fig. 32). Moreover, the mechanical distortion of the electrode caused by gas shock in the cycled MA-1 cell could damage the adhesive network of the active materials such that a portion of the anode powder was easily removed by the separator during cell disassembly (Fig. 2i). In contrast, using MA-2 effectively avoided the above situation and maintained a well-preserved cell structure over cycles (Fig. 2j).

Decoupling the gas evolution of WSEs

The LUMO and HOMO energy levels of free and Li-coordinated solvents were calculated to estimate the redox tendencies of the solvents with different energy states in cell systems. As shown in Fig. 3a, the decreased HOMO energy levels correspond to the enhanced oxidative resistance^35,36. When a solvent molecule forms a coordination bond with the Li⁺, its electron cloud density decreases due to the strong electron-withdrawing effect of Li⁺, leading to a reduction in the energy level of its HOMO. This decrease in the HOMO level is the fundamental reason why the coordinated solvent exhibits higher apparent oxidation stability compared to its free state, demonstrating that free solvent-dominated gas evolution behavior in WSEs was more intense than that in RCEs and PC-based electrolytes.

Fig. 3. Cathode-involved gas production at low temperatures.

a The HOMO/LUMO energy levels of the free and Li-coordinated solvents. b ∆G of the possible reaction for gas formation from solvents at −20 °C. c Reduction and oxidation pathways of MA at −20 °C. Operando DEMS measurements of the pouch cells using d MA-0, e MA-1, and f MA-2 at −20 °C. g Schematic of the gas production derived by the cathode and anode at low temperatures. h Gas volume inside the cells with S-1 and S-2 electrolytes after 100 cycles at −20 °C. i Proportions of various gas components in the cycled cell using the S-1 and S-2 electrolytes at −20 °C. The effect of the j strongly or k weakly solvating solvent on the solvation structure, solvation energy, and gas production.

The operando DEMS was carried out on a pouch-type simulation cell to investigate the gas component (Supplementary Fig. 33). Notably, reactions involving gas production could occur at low temperatures (−20 °C) in relation to the Gibbs free energy (∆G) (Fig. 3b). Using the MA-based WSEs as examples, the DEMS results for the MA-0 cell showed seven main components in the gas phase: CO, CO₂, CH₄, C₂H₄, H₂, C₂H₂, and C₂H₆, in which the outputs of the first four gas were greater during cell charging at −20 °C (Fig. 3d). As for H₂, their sources are generally the reduction of trace amounts of water, with minor contributions from the deep reduction of solvent molecules or conductive salts³⁷. However, in this study, they do not significantly interfere with the quantification of CO₂, CH₄, and C₂H₄, which serve as characteristic markers for reactions at the cathode and anode. After forming a stable SEI by adding FEC, the production of the CH₄ and C₂H₄ in the MA-1 cell was reduced (Fig. 3e), while the signal intensity of CO₂ was stronger than those of other gases generated during the same period. Therefore, it can be inferred that the CH₄ and C₂H₄ belonged to the anode-involved gas, and CO_2, in turn, stemmed from the oxidation of the electrolyte on the cathode. Furthermore, to further differentiate the sources of gas evolution, we assembled DEMS comparison tests for Li||NCM and Li||GSiO half-cells (Supplementary Figs. 34 and 35). In the Li||NCM half-cell, significant CO₂ release was detected only during charging to high voltages. More importantly, after introducing the film-forming additive LiDFOB at the cathode, CO₂ evolution was significantly suppressed, with its proportion in the total gas production sharply decreasing from 38.98% in the MA-1 electrolyte to 15.91%. In the Li||GSiO half-cell, no noticeable CO₂ production was observed regardless of whether the MA-1 or MA-2 electrolyte was used, while CH₄/C₂H₄ were the primary gases evolved (Supplementary Fig. 36). This directly confirms that CH₄/C₂H₄ originates from reductive decomposition reactions at the anode.

We further analyzed the decomposition pathways of solvents for both reduction and oxidation by employing the DFT calculation. As shown in Fig. 3c, the overall energies of MA molecules decrease upon oxidation, with the breaking of C–O and C–C bonds and the generation of CO₂. For the reduction process, the C–O bonds in ester groups of the MA molecule first break with the formation of alcohol groups, then the C–C bonds break to derive the CO and CH₄. PC and other WSSs follow similar decomposition rules as MA (Supplementary Figs. 37 and 38). This result is consistent with the DEMS observation that CO₂ was an oxidation byproduct from the WSEs (Fig. 3g), especially from the free WSS³⁸. In addition, since the ∆G of the redox reaction of the solvents remains negative at −20 °C and is less affected by temperature when the temperature change is not large, the difference in temperature within a certain range may merely affect the gas generation rate by influencing the distribution of molecule activity, rather than the thermodynamic tendency of the reaction (Supplementary Fig. 39). As expected, the production of CO₂ was suppressed in the MA-2 cell at −20 °C (Fig. 3f) since the passivation of the cathode by the LiDFOB-derived CEI was similar to that of the anode by the FEC-derived SEI, blocking the oxidation pathway of the WSE (Supplementary Figs. 40–43).

The cumulative volume of the gas inside the S-1 and S-2 cells after cycling at −20 °C was measured by a simple drainage method combined with DEMS detection to quantitatively analyze the true content of each gas in the working cells (Supplementary Fig. 44). As shown in Fig. 3h, the gas volume in the PC-1 cell was the lowest at approximately 14 mL, while the S-1-type WSE cells all had a gas volume of up to 22 mL. Compositional analysis by transferring the gas to the DEMS revealed that the proportion of CO₂ was greater than 50% in the S-1-type WSE cells (Fig. 3i and Supplementary Fig. 45). Using MA as an example, 13.1 mL of CO₂, which accounted for 52.3% of the total gas, was produced inside the pouch cells after cycling. The EP-1, EA-1, EMC-1, and DEC-1 cells also showed similar patterns in the CO₂ evolution to those of MA-1. With LiDFOB in WSEs, the total gas volume significantly decreased and reached 9.2, 8.3, 7.7, 10.3, 6.3, and 8.6 mL for PC-2, MA-2, EA-2, EP-2, EMC-2, and DEC-2 cells, respectively. Moreover, the CO₂ proportion also showed a downward trend, where only 1.1 mL of CO₂ was generated in the MA-2 cell after cycling.

Notably, although the CO₂ level in the MA-2 cell was suppressed, it was not completely eliminated. The generation of CO₂ may also originate from non-electrochemical, purely chemical processes, such as slow chemical reactions occurring during later cycling stages or static storage, involving components like lithium salt decomposition products or the hydrolysis/thermal decomposition of ester-based solvents³⁹. In addition, electrochemical oxidation of inactive materials (carbon, binder) and deep complex reduction of solvents at the anode may also contribute to CO₂ generation, although the amount produced is relatively small compared to that from the primary pathway of high-voltage electrochemical oxidation of solvents^40,41. Due to the involvement of additional uncertainties and potential gas consumption processes within the complex chemical environment of the battery, all gas volumes reported in this work are strictly defined as the observed quantities at the end of the experiment⁴².

Given the understanding toward the gas production of the WSE, the complex relations between the solvent properties and solvation structure/solvation energy/gas production could be somewhat clarified, as described in Fig. 3j, k. Highly polar solvents can form a strong Li-ion solvation and produce a small amount of free solvent in the regular 1 M electrolyte. Therefore, the desolvation barrier (E_a,Rct) in the strongly solvating solvent-based electrolyte is high. The HOMO energy level of the solvent decreases after coordination so that the strongly solvating electrolyte (SSE) with less free solvents presents a high oxidation resistance, suppressing the cathode-involved gas production. For WSSs, the weak solvation structure with fewer coordinated solvents showcases lower E_a,Rct, facilitating an easier desolvation process.

In addition, the WSE with a large proportion of free solvents exhibits a relatively higher HOMO energy level, suggesting that the WSEs are more prone to undergo oxidation reactions, resulting in gas production. From the perspective of ∆G, solvents strongly coordinated with the Li ion will show higher interaction with electrons. Thus, the ∆G for the reduction and oxidation process of the strongly solvating solvent would turn more negative and positive, respectively. Since the interaction between Li ion and WSS is weak, the ∆G of the redox reaction of the WSS and WSE may be unchanged. Our calculation verified that ∆G for the redox process of the WSE solvent was all negative. Therefore, the degree of the electrode-involved gas production could be roughly forecasted by comparing the HOMO/LUMO energy level of the electrolyte under different solvation environments.

Low-temperature cells with modified WSEs

The low-temperature application of modified WSEs was assessed by 1.2 A h GSiO||NCM523 pouch cells. After the activation process, the cells were subjected to conditioning cycling with the constant current–constant voltage (CC–CV) procedure to determine the capacity retention and reversible capacity. Supplementary Figs. 46 and 47 show that the MA-2 cell could retain a reversible capacity of 0.76 A h at 0.7 C and −20 °C. For the MA-1 cells, severe electrolyte decomposition at high rates, coupled with the inherently unstable electrode-electrolyte interface, resulted in prolific gas evolution. This massive gas accumulation, physically disrupting the electrode structure and increasing internal resistance, ultimately led to rapid cell failure when the rate reached 0.4 C.

An improved cycling was also obtained with the MA-2 cells at 0.7 C (Fig. 4a and Supplementary Fig. 48), showing an 86.2% capacity retention over 250 cycles at −20 °C and good repeatability (MA-2₁ and MA-2₂). The capacity output of cells at temperatures ranging from 25 °C to −50 °C is shown in Fig. 4b. Due to the stable interface with low polarization (Fig. 4c and Supplementary Fig. 49), more than 40% and 30% RTC of the MA-2 cells were retained at −40 and −50 °C, respectively. These results were much better than those of the MA-1 cells. The specific energy of the pouch cell was also compared. Based on the calculation mentioned in Supplementary Table 4, the specific energy of the MA-2 cell was 219.1 and 176.6 W h kg^–1, respectively, for operating at room temperature and −20 °C.

Fig. 4. Low-temperature performance of the WSE cells.

a Cycling performance of the cells with MA-1 and MA-2 at 0.7 C and −20 °C (the 1 C is 1.2 A). b Reversible capacity of the cells with MA-1 and MA-2 at 0.2 C under 25 ~ −50 °C (the 1 C is 1.2 A). c Charge−discharge voltage profiles of cells with MA-2 at various temperatures. d Rate performance of the cylindrical cells with MA-1 and MA-2 at −20 °C (the 1 C is 3.5 A). e Capacity proportions of the CC and CV process in the rate cycling of the MA-1 and MA-2 cylindrical cells (the 1 C is 3.5 A). f Reversible capacity of the cylindrical cells with MA-1, MA-2, and RCE at 0.15 C under 25 ~ −60 °C (the 1 C is 3.5 A). g Long-term cycling of the cylindrical cells with MA-1 and MA-2 at 0.4 C and −20 °C (the 1 C is 3.5 A). h Comparison of the low-temperature performance of the cells using MA-2 and other related strategies (room temperature capacity is RTC). (The 1 C is 1.2 A for a pouch cell, and the 1 C is 3.5 A for a cylindrical cell). The source of the literature data shown in this figure can be found in Supplementary Tables 7–9.

Motivated by the enhanced low-temperature stability of MA-2, we further examined a 3.5 A h 21,700-type cylindrical cell by CC–CV charging procedure (see Supplementary Table 5 for assembly information). In contrast to that from the pouch cells, better rate performance was obtained in the cylindrical cells with MA-1 and MA-2 at −20 °C (Fig. 4d). This result was potentially caused by the structural feature of the cylindrical cells, such as high electrode stacking pressure, all-tab design (Supplementary Fig. 50), and low internal resistance (Supplementary Fig. 51 and Table 6). The results indicate that the cell using MA-2 exhibits minimal capacity fluctuation at low temperatures with the rate increases, maintaining a capacity retention of 71.6% at 1 C (3.5 A) and at −20 °C relative to its highest capacity output (HCO) of 3.21 A h tested at 0.15 C. A high capacity of 3.11 A h, around 96.9% of HCO, could be maintained when the MA-2 cell is charged back to 0.15 C, showing good reversibility. In stark contrast, a significant capacity decline was obvious in the MA-1 cell when cycled at −20 °C. Especially at 1 C, the capacity of the MA-1 cell is merely 13.0% of its HCO (2.45 A h) at −20 °C. Furthermore, the CC proportion of the CC–CV process for the MA-2 cell is above 72% at all rates, while that for MA-1 cells is below 70%, indicating the better rate performance of the MA-2 cell. These results can be attributed to the stable interface and smaller polarization of the MA-2 cell even at high rates and under low temperatures (Fig. 4e and Supplementary Fig. 52).

The weak solvation and gas restriction of MA-2 also enabled the cylindrical cell to stably operate at temperatures lower than −20 °C. As shown in Fig. 4f and Supplementary Fig. 53, a reversible capacity of 1.8 A h was retained by the MA-2 cell at −60 °C, while the cells with RCE and MA-1 struggled to survive. Fig. 4g shows the stable cycling of the MA-2 cell at −20 °C for over a year (about 830 days). More than 2100 cycles were demonstrated, with 88% RTC and 81.1% capacity retention. The cycling differences between MA-2 cells from the same batch were relatively small (Supplementary Fig. 54). As revealed by the nano-CT image, the internal structure of the MA-2 cell remained well preserved after cycling (Supplementary Fig. 55). Whereas for MA-1 cells, excessive gas production caused the cell to self-stop after 140 cycles. Compared with the related strategies, MA-2 had more advantages and was competitive in collaborating with the high capacity and stable cycling at low temperatures (Fig. 4h, reference and other specific information were recorded in Supplementary Tables 7–9).

Based on these results, we performed proof-of-concept investigations on the low-temperature application of MA-2. As Supplementary Fig. 56 shows, a portable power supply (PPS) equipped with two 3.5 A h cylindrical cells using MA-2 was placed in a cryogenic chamber to charge a mobile phone at room temperature. When the depth of discharge of the PPS reached 50%, as revealed in Supplementary Fig. 57, the power of the mobile phone increased by 27%, 26%, 23%, 21%, and 13% at PPS operating temperatures of 25, 0, −20, −40, and −60 °C, respectively. When the MA-2 cell was replaced with the RCE or MA-1 cell, the lowest operating temperature of PPS was approximately −40 °C, and the increase in power provided to the mobile phone at −40 °C was less than 13% (Supplementary Figs. 58 and 59). As a result, the high cycling stability achieved by MA-2 (or other S-2-type WSEs) at low temperatures and our understanding of the fundamental gas evolution behavior of WSEs provide systematic knowledge of the cell degeneration that occurs with WSEs. To further enhance the low-temperature performance of LIBs, future research could focus on the molecular engineering of WSEs to develop novel solvents that combine weak coordination with high oxidation resistance. Simultaneously, using functional additives or localized high-concentration strategies to establish a stable CEI that resolves the trade-off between kinetics and stability is also the key to the development of the LIB with satisfactory cycling stability at low temperatures.

Discussion

In this study, the oxidation-involved gas production of WSEs applied for low-temperature LIBs was investigated. By analyzing the gas evolution behavior on electrodes, CO₂ was used as the identifier gas for WSE oxidation. Since free solvents have higher HOMO energy levels than Li-coordinated solvents, the oxidation of the studied WSEs containing many free solvents was severe at low temperatures, in conjunction with gas production, this led to electrodes with ripple-type structures, living materials falling off, and increased resistance. Therefore, low-temperature LIBs using WSEs exhibited rapid capacity fading, Li plating, and nonthermal runaway failure despite their high capacity output (vs. room-temperature capacity). Based on this, we designed a proof-of-concept MA-based WSE with a LiDFOB additive capable of forming a stable CEI to suppress the parasitic reaction between the free solvents and the cathode and successfully reduce the gas production of WSEs. As a result, 3.5 A h cylindrical cells employing the modified WSEs delivered 88% RTC at −20 °C and maintained a stable cell structure with 81.1% capacity retention after 2100 low-temperature cycles at 0.4 C. These findings facilitate a balance between the Li-ion transport kinetics and oxidative stability of WSEs for developing low-temperature batteries with high capacity and long lifespan.

Methods

Experimental section

Electrolyte preparations

The battery-grade solvents and salts including salt of lithium difluoro(oxalato)borate (LiDFOB, DodoChem Co., Ltd.) and lithium bisfluorosulfonimide (LiFSI, DodoChem Co., Ltd.) and the solvents of fluoroethylene carbonate (FEC, Adamas-beta), propylene carbonate (PC, Adamas-beta), methyl acetate (MA, Adamas-beta), ethyl acetate (EA, Adamas-beta), ethyl propionate (EP, Adamas-beta), ethyl methyl carbonate (EMC, Adamas-beta), diethyl carbonate (DEC, Adamas-beta), diethyl ether (DEE, Adamas-beta), dipropyl ether (DPE, Adamas-beta), dibutyl ether (DBE, Adamas-beta), isopropyl ether (IPE, Adamas-beta) and (dimethyl carbonate, DMC) were used without further purification. Electrolyte preparations were performed in an Ar-filled glove box. Firstly, the 1 M LiFSI electrolyte (S-0) was prepared by dissolving LiFSI in PC, MA, EA, EP, EMC, DEC, DEE, DPE, DBE, and IPE. Secondly, the electrolyte containing FEC (S-1) was prepared by adding 5 wt% FEC into the S-0 electrolyte. Thirdly, the electrolytes (S-2) containing LiDFOB were prepared by adding 5 wt% LiDFOB addition into the S-1 electrolyte. The regular carbonate-based electrolyte (RCE) of 1.0 M LiPF₆ in EC: DEC (1:1 by volume) was used as the reference. All electrolytes were dried by molecular sieve before testing. The water content was less than 2 ppm, as detected by a coulometric Karl Fischer Titrator.

Electrode preparation and cell assembly

The LiNi_0.5Co_0.2Mn_0.3O₂ (NCM523), LiNi_0.8Co_0.1Mn_0.1O₂ (NCM811) cathode, and graphite-SiO_x (GSiO) anode were purchased from Beijing Li-Volt Energy Technology Co., Ltd. The mass loading of NCM811 and GSiO for coin cells was about 15.52 and 7.8 mg cm⁻² (calculated capacity according to 181 mA h g⁻¹ for NCM811 and 445 mA h g⁻¹ for GSiO), respectively. Electrochemical tests were carried out using CR2032-type coin cells and Celgard 2325 separators. All electrolytes and cells were prepared and assembled in an Ar-filled glove box (O₂ < 0.01 ppm, H₂O < 0.01 ppm). A 40 µL electrolyte was added to each coin cell for testing. For pouch cells, the 1.2 A h GSiO||NCM523 pouch cells (Li-FUN Technology Co., Ltd.) were injected with 3 g A h⁻¹ electrolytes before sealing. The mass loading of NCM523 and GSiO for pouch cells is the same as coin cells. For cylindrical cells, the 3.5 A h 21,700-type cylindrical cells with NCM811 and GSiO were prepared and assembled in an Ar-filled glove box. The mass loading of NCM811 and GSiO for cylindrical cells were 15.52 and 8.4 mg cm⁻² (calculated capacity according to 181 mA h g⁻¹ for NCM811 and 404 mA h g⁻¹ for GSiO), respectively.

Electrochemical measurements

All cells were both charged and discharged at the same temperature. Galvanostatic cycling and rate tests of the cells were performed using a Neware tester (CT-4008) within a voltage range of 2.5–4.2 V. For coin cells at room temperature, the Li metal||GSiO half-cells were cycled at 0.2 C for two cycles to form a stable SEI. EIS was carried out on the Li metal||GSiO half-cells at the discharge state with an amplitude of 10 mV over a frequency range of 0.01 Hz to 1 MHz using a BioLogic (MPG-2) electrochemical workstation, and 10 data points were collected per decade of frequency. The ionic conductivity was studied by AC impedance spectroscopy in a symmetrical Pt||electrolyte||Pt Swagelok-type cell with an electrolyte volume of 1 mL and an electrode distance of 1 cm using a Solartron 1470E electrochemical workstation with a frequency range of 0.1 mHz to 1 MHz over a temperature range from −25 to 40 °C. Linear sweep voltammetry was carried out from 3 to 5 V on a half-cell using an Al foil electrode to simulate the actual situation in full cells. The Li⁺ transference number (t_Li+) of cells was determined from AC impedance and DC polarization at room temperature. The t_Li+ was harvested from the ratio of steady-state current to initial-state current, as follows,

$$t_{{Li}^{+}}=\frac{I_{\mathrm{s}}\left(△V-I_0{R}_0\right)}{I_0\left(△V-I_s{R}_s\right)}$$ 1

t_Li+: the Li⁺ transference number; ΔV: the potential applied across the cell; I₀: the current at the initial state; I_s: the current at the steady state; R₀: the initial-state resistance; R_s: the steady-state resistance. For pouch cell testing, 1.2 A h GSiO||NCM523 pouch cells were cycled at 0.2 C for the first five cycles at room temperature, and then the cell was resealed to reduce the impact of film-forming agent decomposition on gas observation. After that, the cell was tested at low temperatures (0 ~ −60 °C). For 21,700 cylindrical cell testing, 3.5 A h GSiO||NCM811 cylindrical cells were cycled at 0.2 C for the first five cycles at room temperature, followed by tested cycling at low temperatures (0 ~ −60 °C). 1 C is equal to 1.2 and 3.5 A, respectively, for pouch and cylindrical cell testing.

Characterization

The morphologies of electrodes and SEIs were characterized by a high-resolution scanning electron microscope (SEM, JSM 7401F) and cryo-transmission electron microscopy (cryo-TEM, Spectra 300). The chemical structure of the cycled electrodes was analyzed using an X-ray photoelectron spectrometer (XPS, ESCALAB Xi⁺) with Al-Kα radiation. The depth profile was obtained via Ar-ion sputtering at 1 kV. The Ar-ion sputtering rate for the XPS depth-profiling calibrated on the TaO₂ surface was ca. 0.04 nm s⁻¹. The studied electrodes were subjected to a rinse in the DMC solvent, followed by a vacuum dry before XPS measurements. Thereafter, the samples were transferred from the Ar-filled glove box to the instrumental chambers using an instrumental accessory of “sample transfer vessel” without exposure to air. Time-of-flight secondary ion mass spectrometry (TOF-SIMS, GAIA3) was applied to investigate the spatial stereoscopic distribution of SEI. NMR (Avance Ⅲ 400 MHz) measurement was used to detect ⁷Li signals in different electrolytes. The Young’s modulus of the samples was measured by the tapping mode of the atomic force microscopy (AFM, MFP-3D) in the Ar atmosphere. Operando differential electrochemical mass spectrometry (DEMS, QAS 100, Shanghai Linglu) measurements were used to detect the gas evolution in the studied cells using different electrolytes during cycling. The DEMS cells were assembled by using a pouch cell mold inside an argon-filled glovebox. During the DEMS measurement, the prepared cells were cycling at −20 °C for 0.2 C, and pure argon gas was flushed for 3 h to remove the air in the cells. The galvanostatic cycling step was conducted at 0.2 C.

Ionic models and computational methods

Quantum chemistry calculations

The calculations were first performed to optimize the molecular geometries of PC and MA molecules using the Gaussian 16 package at B3LYP/6-311+G(d,p) level of theory⁴³. Atomic partial charges for these molecules were subsequently determined via the ChelpG method at the same theoretical level. All ion and solvent molecules were described using atomistic force field parameters in the AMBER format, adopted from prior work. Cross-interaction parameters between dissimilar atom types were derived following the Lorentz–Berthelot combining rules.

Modeling systems

Two modeling systems were constructed, and the detailed system compositions are listed in Supplementary Table 10. All classical MD simulations were conducted using the GROMACS package under cubic periodic boundary conditions (PBCs)⁴⁴. The equations for the motion of all atoms were integrated using a classic Verlet leapfrog integration algorithm with a time step of 1.0 fs. Short-range van der Waals and real-space electrostatic interactions were truncated at 1.6 nm, while long-range electrostatic contributions were treated using the particle-mesh Ewald method with a fifth-order interpolation and a 0.20 nm Fourier grid spacing. All simulation systems were first energetically minimized using a steepest descent algorithm, and thereafter annealed gradually from 600 K to target temperatures (25, 0, −20, and −40 °C, respectively) within 10 ns. Subsequent equilibration was performed in the isothermal-isobaric (NPT) ensemble for at least 20 ns, with temperature and pressure maintained at target values and 1 atm using a Nosé–Hoover thermostat (coupling constant: 0.4 ps) and a Parrinello–Rahman barostat (coupling constant: 0.2 ps). Production runs were then carried out in the canonical (NVT) ensemble for 40 ns, with trajectory snapshots saved every 100 fs for subsequent structural and dynamic analyses. Representative solvation configurations were extracted from extensive atomistic simulations, and these solvation structures were adopted as starting configurations for additional DFT calculations. Geometry optimizations were first carried out using the Gaussian 16 software package [Gaussian], employing the B3LYP hybrid functional with the 6-311+G(d,p) basis set and Grimme’s D3 dispersion correction with Becke–Johnson damping (GD3BJ). Optimized solvation structures were then subjected to single-point energy calculations at the M06-2X/def2-TZVP level. Implicit solvation effects were incorporated using the SMD (solvation model based on density) continuum model, with solvent parameters for PC and MA applied according to the corresponding solvation structures. From these DFT calculations, we computed molecular electrostatic potentials, HOMO-LUMO energy levels, and desolvation energies.

MD simulations

Additional MD simulations were carried out using the large-scale atomic/molecular massively parallel simulator code, employing the optimized potential for liquid simulations-all atom force field. The initial atomic coordinates were generated with the Packing optimization for molecular dynamics simulations (Packmol) program, while final solvation structures were visualized using VESTA. PBCs were enforced in all three dimensions. A cutoff distance of 12 Å was applied for both van der Waals interactions and the long-range correction of Coulombic interactions.

Author contributions

Z.L. conceived and designed the experiments; Y.C. and X.Y. conducted the experiments; W.H. performed computational tests and calculations; Y.D. analysed the electrochemical data; Y.C. and X.Y. wrote the original draft; Z.L. revised the draft; Z.L. supervised the project. All authors contributed to the scientific discussion of the manuscript.

Peer review

Peer review information

Nature Communications thanks Jianhui Li, Lizhi Sheng, and Tianye Zheng for their contribution to the peer review of this work. A peer review file is available.

Data availability

The datasets generated during and/or analysed during the current study are available from the corresponding author on reasonable request. Source data are provided with this paper. Figshare 10.6084/m9.figshare.32099506 Source data are provided with this paper.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains Supplementary material available at 10.1038/s41467-026-74277-5.