Tailoring Electrolyte Solvation for Li Metal Batteries Cycled at Ultra-Low Temperature

Abstract

Lithium metal batteries (LMBs) hold the promise to pushing cell level energy densities beyond 300 Wh kg⁻¹ while operating at ultra-low temperatures (< −30°C). Batteries capable of both charging and discharging at these temperature extremes are highly desirable due to their inherent reduction of external warming requirements. Here we demonstrate that the local solvation structure of the electrolyte defines the charge-transfer behavior at ultra-low temperature, which is crucial for achieving high Li metal coulombic efficiency (CE) and avoiding dendritic growth. These insights were applied to Li metal full cells, where a high-loading 3.5 mAh cm⁻² sulfurized polyacrylonitrile (SPAN) cathode was paired with a one-fold excess Li metal anode. The cell retained 84 % and 76 % of its room temperature capacity when cycled at −40 and −60 °C, respectively, which presented stable performance over 50 cycles. This work provides design criteria for ultra-low temperature LMB electrolytes, and represents a defining step for the performance of low-temperature batteries.

Introduction

The deployment of rechargeable lithium-ion batteries (LIB) is crucial to the operation of modern portable electronics in extreme environments, where a reduction in cell energy density at ultra-low temperatures (≤ −30 °C) has limited the operations of electric vehicles, subsea operations, military and defense devices, and space exploration.^[1–6] In principle, this can be accomplished by both increasing the baseline energy density of the battery, or mitigating the energy loss at low-temperature via improvement of the charge-transfer kinetics. Intuitively, ideal performance would be produced by a system inherently capable of doing both of these things simultaneously.

To address the former of these improvements, replacing the commonly applied graphite anode (372 mAh g⁻¹) with Li metal (3860 mAh g⁻¹) has been noted as an effective step to push cell energy densities above 300 Wh kg⁻¹.^[7] However, Li metal anodes are known to face poor cycling stability, where large volume change paired with the high reactivity of metallic Li inevitably yields low coulombic efficiency (CE), limiting the cyclability of practical Li metal batteries (LMBs).^[8–10] On the other hand, both the capacity retention and operating voltage of LIBs and LMBs are known to suffer severely at temperatures < −30 °C. This performance decrease has been attributed to a number of factors, including increased impedance from bulk ion transport in the electrolyte and migration of Li⁺ through the solid-electrolyte-interphase (SEI). Most importantly, Li⁺ desolvation is believed to be the dominant impedance contributor, and is correlated to the Li⁺/solvent binding energy at the interphase.^[1,11–13] In general, these resistances have been minimized by employing low melting point and/or low polarizability solvents^{[1–6,13–21]}, novel salt additives^[21–23], and most recently surface functionalization, where Gao et al. demonstrated remarkable Li reversibility down to −15 °C.^[24] Though much progress has been made, methods for directly improving the desolvation kinetics are largely unknown.

This work aims to provide such a method, where the solvation structure of Li⁺ in the electrolyte was found to be of great importance to the reversibility and plating behavior of Li metal at low temperatures. These findings were demonstrated through the systematic comparison of a diethyl ether (DEE), and 1,3-dioxolane, 1,2-dimethoxyethane based (DOL/DME) control electrolyte. After experimental and theoretical investigation of their low temperature Li plating behavior, the investigated electrolytes were applied in LMB full cells with practical electrode loadings, employing a sulfurized-polyacrylonitrile (SPAN) cathode at a loading of 3.5 mAh cm⁻² paired with a limited Li anode (N/P = 1). The cells employing the DEE electrolyte retained 76 % of their capacity when charged and discharged at −60 °C, compared to only 2.8 % in the DOL/DME control system. This study sets a performance standard for the operation of ultra-low temperature batteries and reveals key electrolyte design strategies at the molecular level to do so.

Low Temperature System Design

Low-temperature performance loss of batteries can be mitigated by the addition of warming mechanisms. Recent progress has shown promise to limit the impact to total system mass.^[25,26] Battery warming systems, however, consume non-negligible power, which inevitably reduce overall operating efficiency and energy density of low temperature devices.^[25,26] Hence, it is important to consider the operational conditions of low-temperature batteries during system design, which define the requirements of the warming components. In principle, these operation conditions fall under three schemes as shown in Figure 1, where scheme 3 is required to reduce or eliminate the need for battery warming during continuous operation. Despite this, the low temperature LIB and LMB communities have primarily focused on low temperature discharge after charging under mild conditions (scheme 2).^{[1–6,14,16,18–23]} This focus is likely a result of the kinetic demands of scheme 3, particularly for LMBs due to the challenge for maintaining homogenous Li deposition at ultra-low temperatures, which is a notably difficult proposition even under benign temperatures.^[8–10]

Figure 1.

Operational schemes of low temperature LMBs and the significance of their electrolyte structure for ultra-low Li plating. Scheme 1) Thermal management required during both charge and discharge processes, Scheme 2) Low temperature discharge capability with thermal management required during charge, Scheme 3) Batteries capable of both charge and discharge at low-temperature, free of thermal management.

Though reversible LMBs have been recently demonstrated at −15 °C^[24], there has been very few demonstrations of full cell LMBs at temperatures < −30°C ^[14–16], where only Dong et al. provided low temperature charge and discharge cycling.^[15] This earlier work revealed the significant role of ionic solvation in extending the operational potential window while maintaining ionic conductivity at ultra‐low temperature. This work also served to provide the first ever LMB capable of cycling at ultra-low temperature, with a reported CE of 89 % at −40 °C. Thus, design strategies for high Li CE at ultra-low temperatures are yet to be established, and the mechanistic understanding of Li plating dynamics at such temperature extremes are largely unknown. These factors had fundamentally limited the deployment of high energy rechargeable LMBs in extreme environments. In this work, we observe that enabling LMBs at ultra-low temperatures is heavily dependent on optimization of the electrolyte solvation structure, which was found to define the interfacial ion desolvation mechanics and the corresponding Li deposition morphologies (Figure 1). This evidence was shown by electrolytes based on DEE and DOL/DME solvents, which provided vastly divergent Li metal performance at −40 °C and below owing to their differing solvation structures, without any correspondence to their SEI composition and ionic conductivity at low temperature.

Ultra-Low Temperature Li Metal Performance

Electrolytes utilizing lithium bis(fluorosulfonyl)imide (LiFSI) are well known to produce some of the highest reported CE values for Li metal anode cycling.^[27–30] Hence, 1 M LiFSI was paired with the solvents of interest (Table S1) in order to examine their Li metal performance at ultra-low temperature. 1 M LiFSI DOL/DME (1:1 volume) was selected as the control electrolyte due to the large volume of work previously conducted with similar formulations in the field, in addition to the low melting points of both DOL and DME (Supplementary Table 1).^{[28, 31–33]} These electrolytes were first applied to Li||Cu cells, where the Li plating/stripping CE was determined via the method proposed by Adams et al.^[34] As shown in Figure 2a, despite providing an efficiency of 98.9 % at room temperature, the 1 M LiFSI DOL/DME system CE was found to sharply decrease to 45.4 % and 27.5 % at −40 °C and −60 °C, respectively. The 1 M LiFSI DEE system, however, was found to maintain CE values of 98.9 %, 99.0 %, and 98.4 % at 23 °C, −40 °C, and −60 °C, respectively (Figure 2b). Furthermore, it was found that the DEE system yielded smooth Li deposition/stripping profiles with stable voltage outputs, which was not shared by the DOL/DME system, indicative of soft shorting events. Additional data demonstrating the superiority of the DEE electrolyte in terms of long term Li||Cu cycling efficiency, critical current density at different temperatures, and faster kinetics at low temperatures as measured by Tafel plots can be found in Supplementary Figure 1–3.

Figure 2.

Li metal performance and characterization at benign and ultra-low temperatures. Plating/stripping profiles for CE determination in a) 1 M LiFSI DOL/DME, and b) 1 M LiFSI DEE at 0.5 mA cm⁻² with SEI formation steps omitted. Li deposition profiles for characterization in c) 1 M LiFSI DOL/DME, and d) 1 M LiFSI DEE at 0.5 mA cm⁻². Optical photographs of Cu current collector after the corresponding deposition experiments in e) 1 M LiFSI DOL/DME, and f) 1 M LiFSI DEE. The significantly reduced amount of visible Li at −40 and −60 °C in DOL/DME were attributed to shorting. SEM images of Li plated in g) 1 M LiFSI DOL/DME, and h) 1 M LiFSI DEE.

To confirm the presence of shorting in the DOL/DME system and provide further characterization of the Li metal plated at ultra-low temperatures, high capacity deposition was conducted, where 5 mAh cm⁻² was deposited at 0.5 mA cm⁻² in both electrolytes at the temperatures of interest (Figure 2c–d). The cells were then disassembled to observe the morphology of the plated Li. The photographs taken of the Cu electrodes (Figure 2e) clearly show that the observable amount of deposited Li in 1 M LiFSI DOL/DME undergoes a severe reduction from 23 °C to −40 °C, where almost no deposits are visible at −60 °C. We believe this is a clear sign of soft shorting, in which the fast growth of dendritic Li serves to form a pathway for electrons in the circuit, rendering further Li⁺ migration and conversion unnecessary to balance the charge in the electrochemical circuit. On the other hand, the DEE system yielded visibly uniform silver Li metal depositions down to −60 °C (Figure 2d). The scanning electron microscopy (SEM) images of the DOL/DME system are shown in Figure 2g, where the image of plated Li at 23 °C reveals dense Li “chunks” up to 10 μm in size. The same images were also taken for the sparse Li plated at −40 °C and −60 °C of the red-boxed regions indicated in Figure 2e, where extremely porous Li was observed at −40 °C, and a single localized dendrite was observed at −60 °C. However, the Li plated at 23 °C, −40 °C, and −60 °C in 1 M LiFSI DEE reveal the same dense, dendrite-free structure, whose chunk size was found to reduce from ~10 μm to ~1 μm from 23 °C to −60 °C.

Non-Correlation Between SEI or Bulk Transport and Performance

Though desolvation has been previously attributed to be the dictating step of charge transfer in intercalation-based LIBs, the SEI composition and ionic conductivity was first assessed to determine the presence of any correlations that may relate to the low temperature performance deviation.^[12] Through the application of X-ray photoelectron spectroscopy (XPS), it was found that there was little deviation between the interfacial chemistry of Li cycled 10 times the DOL/DME and DEE systems, which were largely composed of CO₃, C-O, S-O, Li-O, and Li-F, and is in close agreement to Li cycled in other DME/LiFSI based systems.^[28] The chemical similarity of the SEI produced in these two systems corroborates the identical CE at room temperature, and indicates the SEI composition is not likely related to the shorting behavior found in the DOL/DME based system. (Figure 3a – c). The SEI composition produced by the DEE system at varying temperatures is also provided in Supplementary Figure 4.

Figure 3.

Lithium SEI and ionic conductivity study of electrolytes. Ex-situ XPS of Li anodes after 10 cycles for 1 mAh cm⁻² at 0.5 mA cm⁻² in 1 M LiFSI DOL/DME and 1 M LiFSI DEE at room temperature. a) C 1s, b) O 1s, and c) F 1s spectra with assigned peaks from Gaussian/Lorentzian product peak fitting. d) Measured ionic conductivity of the investigated electrolytes at different temperatures.

Additionally, the salient deviation of Li metal performance at low temperatures between the DOL/DME and DEE electrolytes cannot be attributed to the reduction of bulk ion transfer in the electrolytes at reduced temperature, where the electrolytes both remain in a liquid state (Supplementary Figure 5). As shown in Figure 3d, the measured ionic conductivity of the 1 M LiFSI DOL/DME electrolyte was found to be consistently higher at all measured temperatures, still retaining a remarkable 3.41 mS cm⁻¹ at −60 °C, compared to only 0.368 mS cm⁻¹ in 1 M LiFSI DEE. It is noteworthy that there was also a deviation in the Li⁺ transference number between the DOL/DME and DEE systems, which were measured to be 0.314 and 0.512, respectively (Supplementary Figure 6). While this discrepancy is significant, the ten-fold difference in ionic conductivities is expected to overwhelm the difference in transference number in the determination of Li⁺ transport limitation, as shown in the calculated Sand’s capacity for the DEE and DOL/DME electrolytes using the model proposed by Bai et al. in Supplementary Figure 7. ^[35] As maintaining high ionic conductivity at low temperatures has been a historical and intuitively rooted focus of the low temperature electrolyte field, this case study serves to challenge this trend. Hence, understanding the underlying electrolyte features that yield such salient deviation of Li metal performance in these two systems is vital not only to studying these systems, but to elucidate the molecular design requirements for the future of low temperature LMBs.

Electrolyte Solvation Structure and Performance Impact

Given the non-correspondence of the ionic conductivity and SEI compositions, it is clear that both the improved CE and morphology provided by the DEE system in stark contrast to the catastrophic failure of the DOL/DME system are a result of charge-transfer reactions at the interphase. As previous studies have indicated that Li⁺ desolvation dominates this process at low-temperature, we hypothesize that the desolvation behavior is a direct result of the inherent solvation structure of the electrolytes, which holds the key to describing the performance discrepancy between the DOL/DME and DEE systems.^[11,12] These solvation structures were investigated using both computational and experimental methods.

First, classical MD simulations were conducted with the resulting data shown in Figure 4a – d. Analysis of the RDF data revealed that 1 M LiFSI in DOL/DME displayed a characteristic solvent-separated ion pair (SSIP) structure, in which the Li⁺ coordination is dominated by DME molecules, with an average coordination number of 4.6 DME oxygens per Li⁺. The SSIP structure is well known to persist in similar glyme/Li⁺ electrolytes, and is characterized by solvation shells dominated by interactions between cation and solvent (Figure 4e).^[35–38] By contrast, the 1 M LiFSI DEE solution was found to display a characteristic contact-ion pair (CIP) structure, in which the Li⁺ solvation shell comprises both FSI⁻ and DEE molecules, with average coordination numbers of 1.8 DEE oxygens and 2.0 FSI⁻ oxygens per Li⁺. This CIP structure is well-known to exist in electrolytes with a high salt/solvating solvent ratio, and is balanced between both ion/solvent and cation/anion binding (Figure 4g).^{[15,27–30,39]} The factors that dictate the formation of SSIP and CIP structures have yet to be proposed quantitatively, however in thermodynamic terms, solubilities resulting in CIP structures are more entropically driven than the SSIP structures. An expanded discussion of these considerations is provided in Supplementary Note 1.

Figure 4.

Theoretical and experimental analysis of electrolyte structure. a) Snapshot and b) Li⁺ radial distribution function obtained from MD simulations of 1 M LiFSI DOL/DME. c) Snapshot and d) Li⁺ radial distribution function obtained from MD simulations of 1 M LiFSI DEE. e) Most probable solvation structure extracted from MD simulation of 1 M LiFSI DOL/DME. f) Raman spectra obtained from electrolytes of interest and their components. g) Most probable solvation structure extracted from MD simulations of 1 M LiFSI DEE.

Experimental evidence of these MD results was obtained from Raman spectra of the electrolytes and their components. These spectra are displayed in Figure 4f, where the 1 M LiFSI DOL/DME, and 1 M LiFSI DEE are compared to the individual LiFSI salt, and the DME, DOL, and DEE solvents. It is well-known that upon dissolution, the peaks associated with the anion of the Li⁺ salt undergoes a significant red shift due to the reduced coordination between cation and anion and increased coordination between cation and solvent.^[27,40] As seen in the LiFSI spectra, the S-N-S bending peak of the FSI⁻ at 774 cm⁻¹ undergoes a significant shift to 720 cm⁻¹ when dissolved in the DOL/DME solvents, indicating a strong dissociation of the Li⁺/FSI⁻ interactions, in agreement with the MD SSIP solvation structure. Conversely, the FSI⁻ S-N-S bending peak in 1 M LiFSI DEE undergoes a much smaller shift, from 774 cm⁻¹ in the pure salt to only 748 cm⁻¹ in the electrolyte, indicative of much stronger Li⁺/FSI⁻ interactions that is characteristic of a CIP structure. The Fourier transform infrared spectra (FT-IR) for these systems can also be found in Supplementary Figure 8, where the C-O-C peak of DEE (1130 cm⁻¹) was found to undergo a reduced shift compared to that of DME (1106 cm⁻¹) after the introduction of 1 M LiFSI. A further confirmation of the MD accuracy can be found through the calculated transference numbers, which are close to the experimental values for both systems (Supplementary Table 2). Additionally, it is noteworthy that the CIP solvation structure exhibited by the DEE system may be responsible for the improved oxidative stability shown in Supplementary Figure 9.

The balance between cation/solvent and cation/anion binding in each solvation shell is of particular importance when considering Li⁺/Li charge-transfer mechanics at the anode interphase, where these structures can be expected to undergo different dynamics in the presence of an electric field. Specifically, it has been well documented that the significant negative polarization of the anode surface results in the repulsion of anions^[9,35,41], and Li⁺/anion⁻ binding has frequently been neglected in the desolvation/solvation energy calculations of previous studies.^[16,42] As represented in the proposed mechanisms shown in Figure 5a and Figure 5b, the binding energy of the remaining Li⁺(solvent)_n complexes were assessed via quantum chemistry simulations, which yielded binding energies of −414 and −280 kJ mol⁻¹ for the Li⁺(DME)_2.3 and Li⁺(DEE)_1.8 complexes (average coordination numbers from MD), respectively. It is worthwhile to consider that Li⁺ may undergo significant changes in its solvent coordination number at the interface post anion repulsion, however for the purposes of this study it was determined that the binding energy disparity between DME and DEE persists across comparable Li⁺(DME)_n and Li⁺(DEE)_n complexes for n = 1–3 as shown in Supplementary Figure 10. Hence, the binding energies provided for Li⁺(DME)_2.3 and Li⁺(DEE)_1.8 complexes are meant as a qualitative indication of their divergent desolvation barriers. As the charge-transfer impedance is known to be dominated by Li⁺ desolvation at ultra-low temperatures,^[11,12] we contend that this stark difference in binding energy leads to vastly increased local charge-transfer impedance in the DOL/DME system. Under such severe conditions, it is natural to expect that the Li deposition dynamics would proceed in a tip-driven manner, due to the increased driving force offered by the high-surface area dendritic Li (Figure 5a). Such growth would ultimately result in the rampant shorting observed in the 1 M LIFSI DOL/DME system at −40 and −60 °C, whereas the weakly bound DEE system offers homogenous deposition behavior at these ultra-low temperatures (Figure 5b).

Figure 5.

Proposed relationship between electrolyte structure and desolvation. Proposed desolvation mechanisms and corresponding Li⁺/solvent binding energies obtained from quantum chemistry simulations in a) 1 M LiFSI DOL/DME, and b) 1 M LiFSI DEE.

Further confirmation of the advantages provided by the CIP solvation structure was also observed in additional systems by changing the salt or the solvent. When the LiFSI salt was replaced by LiClO₄ in DOL/DME, evidence of a CIP structure was observed with MD, Raman, ionic conductivity, and transference number measurements, and an improvement of low temperature Li metal shorting behavior was observed at −20 °C (Supplementary Figure 12). LiFSI solutions with dipropyl ether (DPE) and dibutyl ether (DBE) are also expected to yield the CIP structure via MD. These electrolytes show similar improvements in performance at −40 °C with CEs of 97.3 and 98.2 % at −40 °C, respectively (Supplementary Figure 13). Additionally, the influence of bulk ion transport and interphasial kinetics on Li metal shorting was examined through the variation of LiFSI concentration in DOL/DME (Supplementary Figure 14), and DEE (Supplementary Figure 15). The critical short current of the DEE electrolytes were found to be highly dependent on salt concentration and weakly dependent on temperature due to their superior interphasial kinetics and relatively poor mass transport, while the DOL/DME series were found to display the opposite trend (Supplementary Figure 16). All of these observations further confirm the advantage of the CIP structure in enabling low-temperature performance for Li metal anodes. A more detailed discussion of these experiments is provided in Supplementary Notes 2–4.

Full Cell Behavior

In order to demonstrate the low temperature performance of the 1 M LiFSI DEE system, a SPAN cathode was selected as the basis of eventual full cell construction due to its high capacity, low cost, and modest voltage, which satisfies the oxidative stability range of most ether electrolytes.^[43–45] The full cells were comprised of a SPAN cathode with the high mass loading of 3.5 mAh cm⁻² paired with a 40 μm Li metal anode, which corresponds to one-fold excess capacity (Figure 6a). Due to the inherent solubility of lithium polysulfides in typical ether solvents, the SPAN cathode is generally discouraged from use.^[44–47] However, as displayed in Figure 6b, DEE does not display the same solubility for polysulfides as the DOL/DME mixture. This reduced dissolution feature is also apparent in the room-temperature voltage profiles found in Supplementary Figure 17 and the cycling data of the full cells (Figure 6c), where the DEE system retained stable cycling performance in contrast to the low initial CE and immediate capacity fade of the DOL/DME system, which can be attributed to polysulfide shuttling.^[45–47] Such polysulfide dissolution was also observed via ex-situ FT-IR (Supplementary Figure 18) and XPS (Supplementary Figure 19) conducted on delithiated SPAN electrodes after 10 cycles in the electrolytes of interest. In both FT-IR and XPS, clear Li-S signals were observed on the electrodes cycled in the DOL/DME system, which is attributed to residual polysulfides and is not found in the DEE cycled electrodes. It is also noteworthy that the formation of polysulfides in the DOL/DME electrolyte during cycling may also interfere with the production of a stable cathode-electrolyte-interphase (CEI), as both C 1s signals and Li-F peaks were found to be diminished in the DOL/DME cycled electrode, whereas clear peaks were still visible in the DEE cycled electrodes, particularly for Li-F.

Figure 6.

1x Li||SPAN full cell performance at benign and ultra-low temperature. a) Schematic of the demonstrated full cells. b) Dissolution photograph of 0.25 M Li₂S₆ obtained by stirring stoichiometric amounts of Li₂S and S for 24 hr. c) Cycling performance of full cells in each electrolyte at 23 °C and 0.333 C rate. d) 0.1 C charge/discharge profiles, and e) cycling performance at −40 °C and 0.2 C. f) 0.1 C charge/discharge profiles, and g) cycling performance at −60 °C and 0.2 C.

A similar cycling trend was also observed in half cells, where the DEE system retained stable performance at both 0.5 mAh cm⁻² and 3.5 mAh cm⁻² cathode loadings, while the DOL/DME system’s poor cycling stability and low CE is clearly evident at 0.5 mAh cm⁻² loading and significantly exacerbated at 3.5 mAh cm⁻² (Supplementary Figure 20). These cells were then subjected to cycling at −40 °C and −60 °C to satisfy the design requirements of operating scheme 3 as previously discussed. The voltage profiles for the DEE and DOL/DME full cells cycled at −40 °C and −60 °C and a 0.1 C rate are shown in Figure 6d and f, where it was found that the 1 M LiFSI DOL/DME electrolyte produced diminished output capacities of 236 mAh g⁻¹ and 13 mAh g⁻¹, which correspond to 38.9 % and 2.8 % of the capacity produced at room temperature, respectively. On the other hand, the 1 M LIFSI DEE system was found to produce improved capacities of 519 mAh g⁻¹ and 474 mAh g⁻¹ at the same conditions, which corresponds to room temperature capacity retentions of 84% and 76%, respectively. The variance in cell performance were found to be primarily attributed to charge transfer at low temperature, which is shown via EIS in Supplementary Figure 21.

The full cells based on these two electrolytes were then cycled at −40 °C and −60 °C in order to provide the wholistic effect of both the Li metal anode and SPAN cathode performance at ultra-low temperatures. As shown in Figure 6e, the 1x Li||SPAN full cell utilizing 1 M LiFSI DOL/DME at −40 °C was found to undergo severe capacity fade after only 2 cycles of operation, where the low CE observed starting on the 3^rd cycle was taken to be clear evidence of the exhaustion of the one-fold excess Li reserve. Furthermore, the same full cell utilizing 1 M LiFSI DOL/DME was found to provide no viable capacity output after 2 cycles at −60 °C (Figure 6g). On the other hand, the 1x Li||SPAN full cell was able to provide reliable cycling performance at both −40 °C and −60 °C (Figure 6e&g).

Comparison with the State-of-the-Art

To put this work in a historical perspective, we calculated the energy density of previously published systems based on the 18650 cell level projection model proposed by Betz et al.^[48] with the details outlined in the experimental section. The fundamental appeal of LMBs are their ability to achieve higher energy densities than that of LIBs, however until very recently there had not been a practical demonstration of a rechargeable LMB exceeding the theoretical limit of a LIB 18650 cylinder cell (~250 Wh kg⁻¹) at the cell level projected using the above model.^[29] Intuitively, state-of-the-art low temperature LMBs are behind ambient temperature LMBs due to the inherently more difficult considerations of system design. For this reason, all of the cell demonstrations to this date have employed Li metal anodes of excess capacity, often paired with cathodes of low mass loading.^[14–16] Moreover, no LIB or LMB has ever demonstrated both charge and discharge at ultra-low temperatures with the notable exceptions of Cho et al.^[49], and Dong et al.^[15]. It is also worth noting that while our previous work has demonstrated a LMB system with practical Li loading capable of ultra-low temperature discharge, the correspondingly high energy densities of this system are not expected to be sustained if the charging was carried out at low temperatures, particularly given the cell shorting considerations outlined above.^[50]

In this regard, this work is a significant step both in diagnosing the challenge of plating Li at ultra-low temperatures and high CE, as well as providing the criteria required to overcome this challenge via solvation chemistry of the electrolyte. In doing so we provided a route to practical LMB full cells with cell level energy densities of 218, 143, and 126 Wh kg⁻¹ when charged and discharged at 23, −40, and −60 °C (Figure 7a), respectively. This performance compares favorably with other LMB and LIB energy densities at the same temperatures, the majority of which were charged at room temperature before low temperature discharge. A quantitative comparison of the practical considerations of low temperature LMBs can be found in Figure 7b corresponding to the metrics listed in Supplementary Table 3.

Figure 7.

The historical context of this work. a) Cell level energy density of selected low-temperature batteries by publication year. Energy densities were calculated based on the 18650 cylinder cell model proposed by Betz et al.^[48] using the assumptions outlined in the experimental details. b) Comparison of relevant low temperature LMBs as quantified by Supplementary Table 3, where loading values are normalized to the highest reported values in literature and the capacity retention values are normalized to 100%. It is noteworthy that the ostensibly greater energy densities from previous works are not expected to be upheld under low temperature charging.

It is worth noting that depending on the application for a given system the safety should also be considered. To provide a first-step in demonstrating the scalability and potential safety risks associated with LMBs at ultra-low temperature, we have assembled a 160 mAh Li||SPAN pouch cell utilizing a further increased cathode loading of 6 mAh cm⁻² (Supplementary Figure 23a). This pouch cell was able to produce a capacity of 450 mAh g⁻¹ when charged and discharged at −40 °C, which compares favorably to the coin-cell performance given the increased loading and overall cell size (Supplementary Figure 23b). Furthermore, when put under soft-shorting conditions at −40 °C, the cell temperature was found to change negligibly, indicating that this process may not result in catastrophic outcomes at such operating temperatures (Supplementary Figure 23c).

Conclusions

In summary, we have demonstrated that the electrolyte solvation structure is crucial to enable the reversible cycling of Li metal at ultra-low temperatures primarily through the comparative study of 1 M LiFSI DEE and 1 M LiFSI DOL/DME. These insights, as well as the performance metrics provided by the 1 M LiFSI DEE system were leveraged to produce a full cell with practical electrode loadings that promise to enable LMBs cycled at 23 °C, −40°C, and −60 °C. This work represents a significant advancement in the design of low-temperature batteries, both in concept and demonstration, and sets a new performance standard which promises to yield systems that exceed previous energy density limitations while reducing or eliminating the need for battery warming systems due to their universal operating capabilities at ultra-low temperatures.

Methods

Materials.

The electrolyte materials 1,3-dioxolane (DOL), and 1,2-dimethoxyethane (DME) were purchased from Gotion and used as received. Diethyl ether (DEE) was purchased from Sigma Aldrich, and Lithium bis(fluorosulfonyl)imide (LiFSI) was obtained from Capchem. The electrolytes were prepared dissolving predetermined amounts of LiFSI salt into the solvents of interested and stirred.

The SPAN electrodes were prepared by hand milling polyacrylonitrile (Sigma-Aldrich, Mw = 150,000) and elemental sulfur (Sigma-Aldrich) in a mortar with the mass ratio of 1:4 until a homogenous mixture was obtained. The mixed powders were heated in an argon filled tube furnace at 450°C for 6 hours with a ramp rate of 2 °C/min, then cooled down to room temperature. The SPAN cathode was prepared by mixing the synthesized SPAN powder, Super-P and PVDF (KYNAR 2800) in a ratio of 80:10:10 in N-methyl pyrrolidinone (NMP) solvent, cast on carbon-coated Al foil and dried in a vacuum oven at 60 °C overnight.

For electrochemical tests CR-2032 type coin cells were assembled with prepared cathodes and anode separated by a 25 μm Celgard membrane soaked with 75 μL of electrolyte. It is noteworthy that the cells were first assembled dry and electrolyte injection was saved for the last step before crimping in order to minimize evaporation of the DEE solvent. For Li metal performance tests, 250 μm Li metal chips were purchased from Xiamen TOB New Energy Technology Co. LTD. and paired with Cu foil for Li||Cu cells, or an identical Li chip in Li||Li cells. The Li||SPAN half cells were assembled with a 250 μm chip and the prepared SPAN cathodes. Linear scan voltammetry (LSV) stability tests were conducted with 250 μm Li and a blocking working electrode made of Al foil. 1x Li||SPAN full cells were assembled with a 3.5 mAh cm⁻² SPAN electrode and a thin Li electrode obtained from China Energy Lithium Co. that was determined to be 40 μm thick via SEM. Two celgard separators were employed in the full cells in an attempt to mitigate shorting during cycling.

Characterization.

The morphology of the deposited Li metal at various temperatures was characterized using a FEI Quanta 250 SEM. The samples were obtained from coin cells and washed with either DEE or DME before analysis. XPS (Kratos Analytical, Kratos AXIS Supra) was carried out using Al anode source at 15 kV and all the peaks were fitted based on the reference C-C bond at 284.6 eV on both Li metal and SPAN samples. All XPS measurements were collected with a 300 mm × 700 mm spot size during acquisition. Survey scans were collected with a 1.0 eV step size, and were followed by high resolution scans with a step size of 0.05 eV for C 1s, O 1s, F 1s, and S2p regions. All prepared samples were placed in a heat-sealed bag inside the glovebox before they were transferred to the XPS and SEM. SPAN electrodes before and after 10 cycles in 1 M LiFSI DEE and DOL/DME were analyzed by a Perkin Elmer FT-IR. The polysulfide dissolution tests were performed by immersing stoichiometric ratios of Li₂S and S corresponding to 0.25 M Li₂S₆ in DEE and DOL/DME (1:1) solvents and stirring for 24 hr.

The ionic conductivity of the electrolyte was measured by a customized two-electrode cell, in which the two polished 316 stainless-steel electrodes were spaced symmetrically at a set distance. The cell constant is frequently calibrated by using OAKTON 0.447 to 80 mS·cm⁻¹ standard conductivity solution The electrolytic conductivity value was obtained with a floating AC signal at a frequency determined by the phase angle minima given by electrochemical impedance spectroscopy (EIS) using the following equation:

$$\sigma =\frac{L}{A*R}$$ (1)

where R is the resistance and A and L are the area of and space between the electrodes, respectively. The data points from 40 °C to −60 °C were measured by LabView Software, which was also used to control an ESPEC BTX-475 temperature chamber to maintain the cell at a set temperature for 30-minute intervals during measurement.

Sand’s capacity projections were calculated using the following equation as reported by Bai et al^[35]:

$$C_{\mathit{\text{Sand}}}= \pi \left(\frac{m_{L{i}^{+}}{D}_{FS{I}^{-}}+m_{FS{I}^{-}}{D}_{L{i}^{+}}}{m_{L{i}^{+}}+m_{FS{I}^{-}}}\right)\left(\frac{C_{L{i}^{+}}^{*} e}{2 j}*\frac{1}{\left(1-t_{L{i}^{+}}\right)}\right)$$ (2)

Where m is the mobility and D is the diffusion coefficient of ionic species, C*_Li+ is the bulk Li⁺ concentration, j is the areal current density, e is the elementary charge, and t_Li+ is the Li⁺ transference number.

Electrochemical Testing.

All electrochemical data provided in this work were produced by CR-2032 type coin cells or heat-sealed pouch cells assembled in an Ar filled glovebox kept at < 0.5 ppm O₂ and < 0.1 ppm H₂O. All low temperature data points were obtained from these cells inside SolidCold C4–76A and SolidCold C-186A ultra-low chest freezers for −40 °C and −60 °C tests, respectively. All galvanostatic testing was done on an Arbin LBT-10V5A system and all potentiostatic tests were carried out on a Biologic VSP-300 potentiostat.

For Li metal coulombic efficiency (CE) determinations, the accurate coulombic efficiency test popularized by Adams et al.^[34] was carried out on Li||Cu cells. Prior to the test, a condition cycle was carried out on all the cells, where 4 mAh cm⁻² of Li was deposited onto the Cu foil at 0.5 mA cm⁻², and then fully stripped to 1 V to form the SEI before CE testing. During testing 4 mAh cm⁻² was first deposited followed by 10 cycles of 1 mAh cm⁻² plating and stripping before finally stripping all Li to 1 V. The CE was calculated by diving the total stripped capacity by the total plated capacity. The morphological Li studies at various temperatures were conducted on Cu working electrodes plated with 5 mAh cm⁻² at 0.5 mA cm⁻² at room temperature as well as −40 °C and −60 °C after resting for 2 hours to achieve temperature equilibration.

EIS tests of Li||Li cells were carried out on a Biologic VSP-300 potentiostat with a 10 mV perturbation in the frequency range of 1 MHz to 100 mHz. In the case of biased EIS tests 100 mV was applied to the cell for 2 hours before taking the impedance spectra, and the same bias was applied during the EIS measurement.

The transference numbers of the electrolytes were determined via a commonly applied potentiostatic polarization technique on a Biologic VSP-300 potentiostat in which 5 mV was applied for 2 hr to a Li||Li cell with 5 Celgard separators to obtain the initial current I₀, where the cation concentration is uniform and the current corresponds to both the cations and anions, and the steady state current I_ss, which is only attributed to the cations. EIS was applied before and after the polarization on a in order to obtain the cell impedance, where the transference number was then calculated using the following equation:

$$t_{+}=\frac{I_{SS}\left(\Delta V-I_0{R}_0\right)}{I_0\left(\Delta V-I_{SS}{R}_{SS}\right)}$$ (3)

where ΔV is the applied bias, R₀ is the initial cell impedance, and R_SS is the steady state cell impedance. The oxidative stability of the electrolytes was determined via linear scan voltammetry (LSV) of a Li||Al cell at 1 mV s⁻¹.

Li||SPAN cells were assembled and subjected to galvanostatic cycling at room temperature and −40 °C and −60 °C inside the respective chest freezers after resting for 2 hours to achieve temperature equilibration. Li || SPAN half cells utilized of 250 μm Li counter electrodes with either 0.5 mAh cm⁻² or 3.5 mAh cm⁻² SPAN electrodes. Li || SPAN full cells consisted of 40 μm Li anodes paired with 3.5 mAh cm⁻² SPAN cathodes, which corresponds to a one-fold Li anode excess (N/P = 1) for the SPAN after lithiation.

Three electrode pouch cells consisted of a 12 mm diameter SPAN or 15 mm diameter Li working electrode, a 15 mm diameter Li counter electrode, and a Li reference electrode slightly outside the ionic path of the two electrodes sandwiched between 2 Celgard separators. Tabs were sealed to the pouch bag using an MTI MSK-140 compact heating sealer. The selected electrodes were attached to the tabs in the Ar-filled glovebox and were sealed using a table-top impulse heat sealer after electrolyte injection. The Li||Li||Li pouch cells were subjected to LSV measurements at room temperature and −40 °C at 1 mV s⁻¹ down to −100 mV vs. Li. The SPAN||Li||Li pouch cells were subjected to EIS measurements (same conditions as above) at room temperature and −40 °C directly after discharging them to 50% SOC at the selected temperature to avoid encountering nucleation impedance on the anode side. The two electrode SPAN||Li pouch cell was prepared using a 6 mAh cm⁻² SPAN cathode cut to 4.4 × 5.7 cm using an MTI MSK-180 punch and a copper anode to which Li was rolled onto inside the glove box. The sealing procedure matched the three electrode pouch cells. The two electrode pouch was subject to ~ 100 kPa of pressure using two parallel plates fastened together by screws. Temperature monitoring for the two electrode pouch was conducted with a HOBO thermocouple attached directly to the outside of the pouch on the cathode side and recorded using the HOBO data logger software.

MD Simulations.

Molecular Dynamics (MD) simulations were performed in LAMMPS using the OPLS-AA forcefield^[51] with the FSI molecules description from Gaouveia et al.^[52] For electrolyte structure determination simulation boxes containing 20 LiFSI molecules and 192 DEE, or 96 DME + 143 DOL molecules corresponding to 1 M LiFSI DEE and 1 M LiFSI DOL/DME (1:1 volume) electrolytes, respectively. Similarly, 1 M LiClO₄ DOL/DME and 1 M LiFSI DPE simulations were carried out in boxes composed of 20 salt molecules and 96 DME + 143 DOL, and 147 DPE molecules. In all cases the charges of the Li⁺ and FSI⁻ molecules were scaled to the high-frequency dielectric properties of the solvents present in the system according to the method proposed by Park et al.^[36] For each system, an initial energy minimization at 0 K (energy and force tolerances of 10⁻⁴) was performed to obtain the ground-state structure. After this, the system was slowly heated from 0 K to room temperature at constant volume over 0.2 ns using a Langevin thermostat, with a damping parameter of 100 ps. The system was then subjected to 5 cycles of quench-annealing dynamics, where the temperature was slowly cycled between 298 K and 894 K over 0.8 ns in order to eliminate the persistence of any meta-stable states. After annealing, the system was equilibrated in the constant temperature (298 K), constant pressure (1bar) (NpT ensemble) for 0.5 ns before finally being subjected to 5 ns of constant volume, constant temperature dynamics. Radial distribution functions were obtained using the Visual Molecular Dynamics (VMD) software. Snapshots of the most probable solvation shells were also sampled from the simulation trajectory using VMD. Transference numbers were calculated from the MD trajectories through a mean-squared displacement analysis over the duration of the constant volume dynamics.

Quantum Chemistry Calculations.

Quantum chemistry simulations were performed using the Q-Chem 5.1 quantum chemistry package at the PBE/6–31G+(d,p) level of theory, which has previously been shown to produce accurate binding energies.^[42] Binding energies of the Li⁺(Solvent)_x complexes were calculated after geometry optimizations, where the full complexes were optimized with and without Li⁺, representing their separation at infinite distance. The binding energy was calculated as: $E_B=E_{L{i}^{+}{\left(\mathit{\text{Solvent}}\right)}_x}-\left(E_{L{i}^{+}}+E_{x\left(\mathit{\text{Solvent}}\right)}\right)$. The fractional Li⁺(DME)_2.3 and Li⁺(DEE)_1.8 binding eneries were calculated via linear interpolation of the Li⁺(DME)₃, Li⁺(DME)₂, and Li⁺(DEE)₁, Li⁺(DEE)₂ complex binding energies, respectively.

Cell Level Energy Density Calculations.

The cell level energy densities for this and previous works were calculated via the 18650 cylinder cell model proposed by Betz et al.^[48] previously. As the previous LIB works were generally conducted using commercially produced cells, it was assumed that the cathode loading was 2.5 mAh cm⁻² given the lower loadings generally applied from 2000–2010 with an N/P cathode/anode capacity ratio of 1.1. For metal oxide-based cathodes, the active loading was assumed to be 96%, while polymer-based electrodes were assumed to be 90%. The volume change during discharge was determined to be 20 % for LIB chemistries, and −20 % for the Li||SPAN full cell, which are conservative estimates compared the previous work, and align with the differences between LIBs and LMBs.^[48] The macroscopic volume change of the previous LMBs were assumed to be 0 % given the excess anode capacities of the previous LMB works. For all works, it was assumed that a 30 % electrode porosity and an electrolyte loading of 2 g Ah⁻¹ was achievable and were used in the calculation. For the LMB works, the electrode loadings provided in the original publications were used.