Application-driven design of non-aqueous electrolyte solutions through quantification of interfacial reactions in lithium metal batteries

Preparation of the electrolyte solutions and SLS pouch cells

LiFSI (99.99%, Kaixin), DME (99.95%, Guotai), TTE (99.5%, Aladdin), BTFE (99%, Aladdin) and BZ (99.9%, Aladdin) were used for the preparation of electrolyte solutions. All chemicals were used as received and the water content was determined to be <20 ppm by Karl Fischer titration. All solutions were gravimetrically prepared and magnetically stirred in glass scintillation vials in a dry room (4 m × 5 m) with relative humidity <1% at 25 °C using Pasteur pipets (for liquids) and a 4-digit analytical balance. Graphite||NMC811, Cu||NMC811 and Li||NMC811 SLS pouch cells were fabricated in a CATL pilot line (technical specifications of the electrode formulations cannot be disclosed as they are covered by an industrial non-disclosure agreement) and used for cycling after electrolyte injection. NMC811-based positive electrodes are 42 mm × 49.5 mm and double-side coated, with an active material loading of 17.1 mg cm⁻² and an areal capacity of 3.53 mAh cm⁻² for a 0.2 C (28 mA) current on each side. Cell assembly was performed in the same dry room mentioned above. Electrolyte injection was done gravimetrically after sealing three sides of the pouch cell; the final side was heat-sealed under vacuum (−90 kPa) immediately thereafter. Parameters for the pouch cell are shown in Supplementary Table 8. All the cells were set at 0.21 MPa (30 psi) initial pressure and cycled under a fixed-gap condition (that is, securing the pouch cell between two aluminium plates with an initial 0.21 MPa pressure using four screws, such that the clamped cell maintains a fixed thickness) using a Neware CT-4008Tn-5V6A-S1 testing system in a temperature-controlled room set to 25 °C. All cells were cycled at 0.2 C (28 mA)–1 C (140 mA) between 2.8 V and 4.3 V without any prior formation cycles, and the charging followed a constant current–constant potential (CC–CP) protocol with a cut-off current of 0.1 C (14 mA). For simplicity, we abbreviate this cycling process as ‘0.2–1 C’.

T-DEMS

Preparation of cycled electrode samples

The Cu||NMC811 pouch cells were first cycled for 1 cycle, 20 cycles, 40 cycles, 60 cycles, 80 cycles and 100 cycles under 0.2–1 C. A deep discharge procedure was then applied on each cycled Cu||NMC811 pouch cell. The deep discharge procedure refers to the repeated discharging of the cell to 2.8 V using successively smaller currents (7 mA, 3.5 mA and 1 mA) to fully strip the active Li from the negative electrode such that it will not be mistaken as ‘dead’ Li. The deep discharge procedure usually extracts an additional 5–20 mAh of discharge capacity. After deep discharge, the cell was disassembled in an argon-filled glove box (H₂O <0.1 ppm, O₂ <0.1 ppm). Either the Cu electrode or the NMC811-based electrode was then placed into a titration vessel for subsequent tests (Supplementary Fig. 1).

Quantification of ‘dead’ Li, LiH and Li₂CO₃

The quantification of ‘dead’ Li and LiH was performed with deuterated ethanol (CD₃CD₂OD) as the titrant. The quantification of Li₂CO₃ was performed with 10 M H₂SO₄ as the titrant. The titration was performed in an in-house developed Teflon container. The gas generated was collected and measured using a differential electrochemical mass spectrometry system (DEMS, Shanghai LingLu Instruments; Supplementary Fig. 1).

Determination of calibration equations

To construct the calibration equations (equations (7) and (8)) used to quantify ‘dead’ Li, Li metal with different known masses was placed into a titration vessel connected to DEMS. After the argon (99.999%, Fuzhou Zhongming Qiti) inflow stabilized, ethanol-d₆ (CD₃CD₂OD, 99%, Aladdin) was injected into the vessel, and the generated gas was flushed into DEMS for analysis. Multiple-ion mode was used to record the ion current of mass/charge ratios: m/z = 3 for hydrogen deuteride (HD) and m/z = 4 for deuterium (D₂) gas. Afterwards, calibration equations were obtained through linear regressions of the areas of HD and D₂ signals against the Li metal masses (with the origin included; Supplementary Fig. 2a,b). Similarly, calibration equations of LiH were obtained through CD₃CD₂OD titration on LiH (97%, Macklin) samples (Supplementary Fig. 2c,d).

The reactions of CD₃CD₂OD with Li metal and LiH happen as follows:

However, weak yet observable signals of HD and D₂ were detected for Li metal and LiH, respectively. We attribute this to the non-ideal deuterium abundance in the CD₃CD₂OD titrant and the random recombination of hydrogen and deuterium radicals during the titration experiments. For the accuracy of subsequent quantifications, we performed linear regressions on both HD and D₂ signals for Li metal and LiH (Supplementary Fig. 2a–d).

A calibration equation of Li₂CO₃ (equation (9)) was obtained through 10 M H₂SO₄ titration on Li₂CO₃ (99.5%, Macklin) samples with CO₂ as the generated gas⁴² (Supplementary Fig. 2e).

The reaction of 10 M H₂SO₄ with Li₂CO₃ happens as follows:

Quantification of ‘dead’ Li, LiH and Li₂CO₃ on a sample

To quantify ‘dead’ Li and LiH, a cycled Cu electrode sample was placed into a titration vessel connected to DEMS. After the argon inflow stabilized, CD₃CD₂OD was injected into the vessel, and the generated gas was flushed into DEMS for measurements. The masses of ‘dead’ Li and LiH on the cycled electrode were set as x and y, respectively, and the areas of peaks attributed to HD and D₂ from the DEMS result was set to be A and B, respectively. The values of x and y were determined through the following equations:

To quantify Li₂CO₃, a similar process was applied with 10 M H₂SO₄ as the titrant. The mass of Li₂CO₃ on the sample was set to be z, and the area of the peak attributed to CO₂ from the DEMS result was set to be C. The value of z was determined through the following equation:

Here \({k}_{\mathrm{HD}-{\mathrm{Li}}\; {\rm{metal}}}\), \({k}_{\mathrm{HD}-\mathrm{LiH}}\), \({k}_{\mathrm{D}_{2}-{\mathrm{Li}}\; {\rm{metal}}}\), \({k}_{\mathrm{D}_{2}-\mathrm{LiH}}\) and \({k}_{{\mathrm{CO}}_{2}}\) are the slopes of the calibration curves (Supplementary Fig. 2a–e).

In this work, we further standardized the masses of ‘dead’ Li, LiH and Li₂CO₃ into equivalent Li capacities (C_Li) through the following calculations:

In equations (11) and (12), the masses of LiH and Li₂CO₃ (y and z, respectively) are converted into mole-equivalent masses of Li through their corresponding molar masses. These, along with the mass of ‘dead’ Li (x) in equation (10), are then converted into equivalent Li capacity using the specific capacity of Li (3,860 mAh g⁻¹).

E-G&IC

E-G&IC was performed on Cu||NMC811 and 50 μm Li||NMC811 cells with 0.3 g of electrolyte after cycling for a fixed number of cycles (for example, 1 cycle, 50 cycles, 100 cycles, …, 300 cycles for 50 μm Li||NMC811 cells) under 0.2–1 C.

Preparation of the extraction agent

1,2-Diethoxyethane (DEE, 99.9%, Aladdin, <20 ppm H₂O) was used as the internal standard. Diethylene glycol dimethyl ether (diglyme, 99.9%, Aladdin, <20 ppm H₂O) was used as the extracting solvent. An extraction agent was prepared by mixing 4 g of DEE with diglyme in a 200 ml volumetric flask at 25 °C in a dry room. The extraction agent was then sealed with parafilm in a scintillation vial and stored under the same conditions for later use.

Extraction of electrolytes in LMBs

Inside an argon-filled glove box (H₂O <0.1 ppm, O₂ <0.1 ppm), an incision was made at the edge of a cycled LMB pouch cell, through which 5 ml of the extraction agent was injected. The cell was then re-sealed by heat-sealing the pouch along the cut edge. The contents of the sealed cell were thoroughly mixed by first allowing the liquid to diffuse during storage at 25 °C for 3.5 days and then vertically inverting and storing the cell for another 3.5 days. After 7 days, the liquid mixture contained in the pouch cell was extracted via a syringe and syringe-filtered (0.22 μm) for subsequent measurements.

IC and quantification of LiFSI

Determination of calibration equation

LiFSI (20 mg, 40 mg, 60 mg, 80 mg and 100 mg) was dissolved in 1,000 ml of deionized H₂O (18.5 MΩ cm at 25 °C, Milli-Q IQ 7000) to form five standard solutions. Each standard solution (5 ml) was measured with IC (Dionex Aquion RFIC, Thermo Scientific) and the area of peaks attributed to the FSI⁻ anion was calculated. Afterwards, a linear regression was performed between the peak area and the LiFSI concentration, serving as the calibration equation (equation (13) and Supplementary Fig. 2f).

Quantification of LiFSI

The extracted liquid of a cycled LMB cell was diluted 200-fold with deionized H₂O. The diluted solution (5 ml) was measured with IC. The concentration of LiFSI in the diluted solution was set to be c_LiFSI, and the area of the FSI⁻ anion peak from the IC result was set to be S_LiFSI. The value of c_LiFSI was determined applying the following equation:

Here \({k}_{\mathrm{LiFSI}}\) is the slope of the calibration curve (Supplementary Fig. 2f). The absolute mass of residual LiFSI in the cycled cell m_LiFSI was further calculated as follows:

where 200 is the dilution factor and 5.25 ml comes from 5 ml of extraction agent and 0.25 ml from 0.3 g of the electrolyte used in this work.

GC and quantification of DME and TTE

Determination of response factors

The freshly prepared (uncycled) electrolyte (0.3 g) was mixed into 5 ml of extraction agent. The mixture was further diluted fivefold with diglyme to obtain a standard solution. The standard solution (1.5 ml) was measured with GC (Nexis GC-2030, Shimadzu). The masses of DME, TTE and DEE in the standard solution were known (m_DME = 32.7 mg, m_TTE = 210.6 mg and m_DEE = 100.0 mg). The areas of peaks attributed to DME, TTE and DEE were collected from the GC results (S_DME, S_TTE and S_DEE). The response factors for DME (f_DME) and TTE (f_TTE) were calculated as follows:

Quantification of DME and TTE

The extracted liquid of a cycled LMB cell was diluted fivefold with diglyme. The diluted solution (1.5 ml) was measured with GC. The areas of peaks attributed to DME, TTE and DEE were collected from the GC results (S_DME-exp, S_TTE-exp and S_DEE-exp). The absolute masses of residual DME (m_DME-exp) and TTE (m_TTE-exp) in the cycled cell were calculated as follows:

Additional physicochemical characterizations

NMR measurements were carried out using a Bruker AVANCE NEO 500 MHz digital FT-NMR spectrometer. After Cu||NMC811 was cycled under 0.2–1 C for 100 cycles, 5 ml of diglyme was added into the cell. The cell was sealed and stored for 7 days at 25 °C, and then the mixture of cycled electrolyte and diglyme was extracted for the NMR test. The whole sampling procedure was conducted in a dry room.

Electrolyte solutions for Raman measurements were prepared in a dry room and sealed in glass scintillation vials for transfer to a separate laboratory for sample loading. The liquid sample was drawn via capillary action by submerging one end of a quartz capillary tube into the liquid sample under atmospheric conditions. The two ends of the capillary were then sealed with ultra-light clay to prevent sample evaporation and contamination. The sealed capillary was then loaded into a Renishaw InVia Qontor Raman spectrometer. Spectra were acquired at 25 °C using an excitation wavelength of 785 nm.

For dynamic viscosity and ionic conductivity measurements, electrolyte solutions were prepared in a dry room and sealed in glass scintillation vials for transfer to a separate laboratory for measurements. Dynamic viscosities were measured with a Brookfield DV2T viscometer using the SC4-18 spindle at 25 °C under ambient atmospheric conditions. After levelling and autozeroing the equipment, an 8 ml aliquot of the solution (enough liquid to fully submerge the spindle) was transferred to the instrument sample holder and equilibrated at 25 °C for 10 min. Measurements were conducted with periodic stirring. Ionic conductivities were measured with a Shanghai Leici DDSJ-318 conductivity meter at 25 °C under ambient atmospheric conditions. An ~10 ml aliquot of the solution was transferred to and sealed in a 50 ml Falcon tube, and then equilibrated at 25 °C for 10 min in a water bath. The calibration of the meter was verified using low and high ionic conductivity standard samples. The probe head was cleaned with ethanol and DI water in between uses.

XPS measurements were carried out using a Shimadzu Axis Supra+ imaging X-ray photoelectron spectrometer. An Al Kα X-ray (1,486.7 eV) was used as the excitation source, and the data were collected in an area of 700 × 300 µm by using a hemispherical electron energy analyser at an emission power of 195 W. Sputtering was performed on a 3 × 3 cm region with a 5 keV argon ion source and an incident angle of 45°. The electrode samples were washed with DME solvent and dried inside an Ar glove box, and then transferred within an airtight vessel from the glove box to the XPS sample chamber. The sputtering time increments were 0 s, 60 s, 120 s, 180 s and 300 s.

For the ICP-OES measurement, a Cu||NMC811 cell was cycled under 0.2–1 C for 100 cycles, followed by a deep discharge process. Afterwards, the cell was disassembled inside an argon-filled glove box (H₂O <0.1 ppm, O₂ <0.1 ppm). The solid rSEI formed on the Cu electrodes was collected using a scraper into a glass vial. The collected powder was then soaked in DME for 60–120 min. After removing the supernatant, the powder sample was dried in a vacuum chamber at 25 °C. The soaking and drying procedure was repeated five times in dry room before the sample was measured with a ThermoFisher iCAP PRO ICP-OES. These preparation procedures ensured full removal of active Li and residual electrolyte components from the sample.

For FT-IR measurements, the FT-IR spectra for the cycled Cu electrode were collected with a Thermo Scientific Nicolet iS50 spectrometer. A Cu||NMC811 cell with LiFSI–1.2DME–3TTE was cycled under 0.2–1 C for 100 cycles and deep discharged (deep discharge refers to the repeated discharging of the cell to 2.8 V using successively smaller currents (7 mA, 3.5 mA and 1 mA) to fully strip the active Li from the negative electrode such that it will not be mistaken as ‘dead’ Li). Afterwards, the cell was disassembled within an argon-filled glove box (H₂O <0.1 ppm, O₂ <0.1 ppm). The Cu electrode was sealed with tape to prevent corrosion in ambient air and was transferred immediately for FT-IR measurement.

For the in situ DEMS measurement of gas generation at 25 °C, an airtight electrochemical vessel was used to accommodate a pouch cell. The electrodes of the pouch cell were connected to two binding posts on the electrochemical vessel so that the cell could be cycled. The edge of the pouch cell was incised before being sealed into the vessel. The vessel was connected to a carrier gas system, and the gas generated from the pouch cell was directed into a mass spectrometer for quantitative analysis. The carrier gas system consisted of a carrier gas (Ar, 99.999%, Fuzhou Zhongming Qiti), a 2.0 μm filter (Swagelok), a digital mass flowmeter (Bronkhorst, EL-FLOW Select) and an in-house developed cold trap with a temperature controller. The Ar gas, regulated by a pressure regulator (set to 0.1 MPa), was directed sequentially through a 2.0 μm filter, a quantitative ring in the pulse inlet system, and the cold trap before entering the mass spectrometer for gas analysis. The filter serves to protect the mass flow controller and the mass spectrometer from small particles in the metal tubing and the sample itself. The flow rate of Ar was maintained at 0.6 ml min⁻¹ to ensure high sensitivity for trace gas analysis inside the pouch cell. The cold trap temperature was set to −90 °C to capture the volatile organic species contained in the carrier gas to protect the mass spectrometer and increase sensitivity.

Scanning electron microscope (SEM) images were captured using a ThermoFisher Helios G4 CX dual-beam focused ion beam (FIB)–SEM and ZEISS GeminiSEM 360. Cross-sectional samples were prepared by cutting a small piece of the sample of interest and polishing with a Hitachi ArBlade 5000 under cryogenic conditions in an argon atmosphere. Samples were transferred in an air-free sample holder.

All the STEM characterizations were performed using an aberration-corrected FEI Themis Z electron microscope equipped with a Gatan GIF Quantum 1065 for EELS operated at 300 kV. For STEM HAADF imaging on the NMC811-based positive electrode, site-specific TEM lamellae were prepared by FIB. The Helios FIB–SEM was used for trenching, in situ lift-out and thinning. To reduce the potential surface damage caused by FIB milling, a further low-energy cleaning at 2 kV was performed. HAADF imaging was then carried out with a convergence angle of 26.5 mrad and an angular collection angle between 60 mrad and 120 mrad. For cryo-TEM, STEM, EDS and EELS characterizations on the rSEI sample, a pure Cu TEM grid was mounted on the Cu negative electrode of a Cu||NMC811 pouch cell, which was cycled under 0.2–1 C. The cell was deep discharged after 10 cycles, and then dissembled inside an argon-filled glove box. The TEM grid was transferred to the microscope using a Fischione 2550 Cryo Transfer Holder. The TEM, STEM, EDS and EELS experiments were performed under a temperature of −170 °C. The probe current used for EELS mapping is ~30 pA, and the dose rate is around 7.5 × 10⁴ e/(Ų s).

CO₂ solubility was measured with an Initial Energy Science and Technology (IEST) GVM2200 in situ cell volume analyser. For the gas solubility test, 10 g of LiFSI–1.2DME–3TTE electrolyte was vacuum sealed in an empty pouch in a dry room. CO₂ gas (40 ml) was injected into the electrolyte-containing pouch with a syringe, and the pouch was sealed again with duct tape. The shrinkage of the pouch due to the CO₂ dissolved in the electrolyte solution was measured with the in situ cell volume analyser. The pouch was submerged in silicone oil (at 25 °C), and volume changes were measured in real time by applying the Archimedes principle⁴³.

According to the Archimedes principle, when an object is partially or fully submerged in a fluid, it experiences an upward buoyant force equal to the weight of the fluid displaced by the object.

The volume of the cell could be obtained as follows:

where V is the volume of the cell, \(\Delta m\) is the mass of water displaced by the cell, \(\rho\) is the density of water at 25 °C and \(g\) is the gravitational acceleration.

First-principles simulations

All surface calculations were conducted utilizing the density functional theory (DFT) as implemented in the Vienna Ab initio Simulation Package (VASP)^44,45. The electron exchange-correlation energies were determined using the generalized gradient approximation and Perdew–Burke–Ernzerhof functional within the DFT framework⁴⁶. Transition metals were treated using the DFT + U augmented approach with U values of 4 eV, 4.4 eV and 5 eV for Mn, Co and Ni, respectively. The DFT + D3 method, which incorporated dispersion correction, was used to account for weak interactions in the systems under investigation⁴⁷. All calculations were spin-polarized, and a plane-wave cut-off energy of 520 eV was utilized. All surface calculations were performed using a 2 × 2 × 1 k-point within the Monkhorst–Pack scheme, and a 15 Å vacuum layer was added to avoid the interactions between repeated periodic slabs. A five-layer slab of the (110) surface of Li was utilized to investigate the reduction decomposition process, while the charged NCM811 slab, by taking the Li atoms out, was used to study the DME oxidation process. Geometric structure optimizations were performed until the force on all atoms was less than 0.02 eV Å⁻¹, with energy convergence criteria set to be smaller than 10⁻⁵ eV per atom. The climbing image nudged elastic band⁴⁸ and dimer methods⁴⁹ were combined to locate the transition states along the reaction pathways, with all transition states verified to have only one imaginary vibrational frequency along the reaction coordinate.

The energies of the highest occupied molecular orbital and lowest unoccupied molecular orbital were calculated using the DFT method at the B3LYP/6-311G+(d, p) level⁵⁰ implemented in the Gaussian 09 (ref. ⁵¹) software package. The SMD (solvation model based on density)⁵² was chosen to account for the solvent effect.

The conductor-like screening model for real solvents (COSMO-RS) method^53,54 was used to get macroscopic gas solubility data. The BP functional and TZVP basis set from the Turbomole programme⁵⁵ were used for COSMO calculations. The resulting COSMO files were subsequently imported in the COSMOtherm programme⁵⁶ to determine the solubility of gas⁵⁷.

MD simulation

MD simulations were carried out with large-scale atomic/molecular massively parallel simulator (LAMMPS)⁵⁸. As visualized with OVITO⁵⁹ (Supplementary Fig. 12a), the simulation box encompasses two Li metal electrodes separated by a distance of 144 Å and a region of electrolyte. Each Li metal electrode surface is represented by the (100) facet and has a dimension of 37 Å × 37 Å × 10 Å with 500 atoms. About 130 LiFSI, 157 DME and 472 TTE molecules were placed between the two electrodes, and the configuration was obtained through a preliminary MD simulation of the bulk electrolyte under the NPT ensemble at 298 K.

The OPLS-AA force field⁶⁰ was used to treat the interactions between the atoms in the liquid phase. Force field parameters were generated by the LigParGen web server⁶¹. Parameters for Li in the electrode were obtained from Nichol et al.⁶². Interactions between electrode and electrolyte atoms were modelled by the Lennard-Jones potentials based on geometric mixing rules, in addition to the long-range Coulomb forces. Electrode atoms were fixed during the simulation, and only electrolyte atoms were allowed to move within the space confined by the two electrodes. Under the NVT ensemble, the system was simulated using the Nosé–Hoover thermostat⁶³ at 313 K.

To accurately depict the charges held by the electrode atoms, we implemented a constant potential method^64,65,66,67. This involved dynamically assigning a charge to each electrode atom in a way that ensured that all atoms in one electrode were at a single Poisson potential, while all atoms in the other electrode were at a different Poisson potential. The two potentials were then set to differ by a predetermined value, ΔU. The two electrodes bore charges of equivalent magnitude but opposite signs, resulting in a charge-neutral system overall. On the basis of the constant potential method, the charge held by each atom in the electrodes can be determined through the following equation⁶⁷:

where Q is a vector containing the charge for each electrode atom, A is the elastance matrix representing the interactions between electrode atoms, b is an electrolyte vector representing the electrostatic potential caused by the electrolyte atoms, which is a function of the electrolyte atom positions⁶⁷, and v is a vector containing the applied potential (U) for each electrode atom, which depends on ΔU. In this study, one pair of ΔU were used: {−5 V, 5 V}. This corresponds to {bottom electrode potential, top electrode potential} in Supplementary Fig. 12.

The simulation was run for a minimum of 20 ns with a step of 1 fs to allow for equilibration of the solvation structure near the electrode interface. During this time, the initially uncharged electrode gradually acquired charge, and ions with opposite charges approached the electrode to form electric double layers. Following the equilibrium period, samples were taken at 2,000 fs intervals for the final 5 ns, then averaged and analysed. To obtain the distribution of electrolyte species, the space occupied by the electrolyte was segmented into bins with widths of 0.1 Å. Numbers of electrolyte species in each bin were tallied and number densities (Supplementary Fig. 12b) were calculated, which could also be normalized by the corresponding volume density (Supplementary Table 2).

Faradic currents between the electrodes and (electro)chemical reactions were not allowed to happen during this simulation.

Electrochemical simulation

Li||NMC811 cell’s discharge potential profiles (Fig. 3j) and electrolyte concentration distributions (Fig. 3k) were simulated through COMSOL Multiphysics version 6.0. A Li metal electrode was treated as an ideal planar electrode and its surface morphology change during discharging was not considered. Therefore, x = 0 in Fig. 3k represents the interface between Li metal electrode and separator. Parameters of the simulation are listed as follows.

Electrolyte

Diffusion coefficient is 1 × 10⁻¹⁰ m² s⁻¹. The transference number is 0.363. Static molar concentration and ionic conductivity are extracted from Fig. 3h.

Separator

The thickness is 15 μm. The porosity is 0.39. Tortuosity is correlated with porosity following the Bruggeman relationship⁶⁸ with a Bruggeman coefficient of 2.

NMC811 positive electrode

The thickness is 49.6 μm. The porosity is 0.25. Tortuosity is correlated with porosity following the Bruggeman relationship with a Bruggeman coefficient of 2.2. The open circuit potential is experimentally measured for a Li||NMC811 cell (Supplementary Fig. 28). The solid-state diffusion coefficient is 4 × 10⁻¹⁵ m² s⁻¹. The electrochemical reaction rate constant is 8 × 10⁻¹² m s⁻¹.

Li metal negative electrode

The electrochemical reaction rate constant is 6 × 10⁻¹¹ m s⁻¹.

Calculation of lifetime CE for LMBs

Determination of the overall CE during the cycle life of an LMB (hereinto referred to as lifetime CE) was modified based on our previously reported approach³¹. After a Cu||NMC811 or Li||NMC811 cell capacity retention decayed to 50–80%, the cell was stopped from cycling and a deep discharge was performed to strip away all the remaining active Li on the negative electrode. We define the ith cycle charge capacity as C_i-c, the last cycle number as n_EOL, the total discharge capacity of the last cycle, including that during the deep discharge as C_EOL-dc, and the Li capacity of the pristine Li foil as C_{Li foil} (for a Cu||NMC811 cell, C_{Li foil} = 0 mAh). The lifetime Li metal CE is given by

Note that two key experimental operations are necessary:

1. 1.

   The electrolyte amount needs to be excessive.

   This guarantees that the cell failure is due to active Li loss instead of electrolyte consumption.

2. 2.

   Deep discharge needs to be performed at the last cycle.

Although capacity trends could differ between replicate cells, C_EOL-dc results remained consistent (Fig. 4a). This indicated the high repeatability of CE and the impact of rSEIs on Li stripping polarization increase, which led to the discharge capacity variation between replicate cells, especially near EOL. In fact, C_EOL-dc results increased along with Li foil thicknesses and cycle life (Fig. 4a). This was because a thicker layer of rSEIs accumulates after longer cycling, leading to a higher polarization during Li stripping. Nevertheless, C_EOL-dc is still lower than C_1-c for all the cells in Fig. 4a. This guarantees the complete stripping of all active Li on the negative electrode and the accuracy of the CE calculations.