Unbiased Quantification of the Electrochemical Stability Limits of Electrolytes and Ionic Liquids

All reagents were used as received without further purification unless otherwise noted. Anhydrous acetonitrile, anhydrous propylene carbonate, tetrabutylammonium perchlorate (NBu₄ClO₄), tetrabutylammonium iodide, phenol (>99%), formaldehyde (aqueous solution, 37 wt %), tetrahydrofuran, polytetrafluoroethylene (PTFE, 60 wt % in water), tetraethylorthosilicate (TEOS, 99%), L-lysine (98%), hydrochloric acid (37 wt %), and sodium hydroxide were obtained from Sigma-Aldrich (St. Louis, MO). The ionic liquids, 1-ethyl-3-methylimidazolium bis(trifluoromethylsulfonyl)imide (EMI TFSI), butyltrimethylammonium bis(trifluoromethylsulfonyl)imide (BuNMe₃ TFSI), and tributylmethylammonium bis(trifluoromethylsulfonyl)imide (MeNBu₃ TFSI), were purchased from IoLiTec (Tuscaloosa, AL). Black Pearl 200 carbon was obtained from Cabot Corporation (Boston, MA). Tetrahydrofuran (HPLC grade, >99.9%) and hydrofluoric acid (48–52 wt%) were purchased from Fisher Scientific (Waltham, MA).

Colloid-Imprinted Mesoporous (CIM) carbon with mesopores of about 24 nm diameter was synthesized according to a previously reported method.^25,26 Three-dimensionally ordered mesoporous (3DOm) carbon was prepared through infiltration of a phenol–formaldehyde resol into an ordered silica template made from sedimented silica spheres, as described previously.²³ A final pore size of 37 nm was determined through nitrogen sorption, and an ordered structure was observed through transmission electron microscopy and small-angle X-ray scattering. CIM and 3DOm carbon films were prepared by finely grinding the carbon powder with an agate mortar and pestle, followed by addition of an aqueous suspension of PTFE (19:1:20 carbon:PTFE:water by mass) and grinding until a homogenous paste was formed. The paste was rolled out with a mechanical press to a thickness of approximately 250 μm and dried at 110 °C for 24 h under vacuum. The carbon films were then cut into 3.0 mm-diameter circles for use in electrodes.

Voltammetry experiments were carried out with a CHI600C Potentiostat (CH Instruments, Austin, TX). A three-electrode set up with a 3.0 mm-diameter glassy carbon (GC) disk working electrode (BAS, West Lafayette, IN), a 0.25 mm Pt wire coil (99.998%, Alfa Aesar, Ward Hill, MA) auxiliary electrode, and a Ag⁺/Ag reference electrode was used (reference solution: 10 mM AgNO₃ and 100 mM NBu₄ClO₄ in acetonitrile). The working electrode was polished on Microcloth polishing pads using 5.0 μm Micropolish II deagglomerated alumina, both from Buehler (Lake Bluff, IL). After polishing, the electrode was rinsed thoroughly, first with deionized water and then with ethanol, followed by drying under a stream of Ar. The reference electrode was prepared as described previously.²⁷ Briefly, a glass tube equipped with a Vycor glass plug was filled with acetonitrile solution of 10.0 mM AgNO₃ and 100 mM NBu₄ClO₄, and an Ag wire was inserted into the tube. The reference electrode was kept in a solution of identical composition for at least 1 week prior to measurements. All potentials reported herein are with respect to 10 mM Ag⁺/Ag. The anhydrous solvents used in this work were stored and transferred under argon. Prior to measurements, all solutions were purged with argon for 15 min while stirring vigorously to remove dissolved oxygen. Linear fits were performed in Microsoft Excel 2010, and statistical analysis was based on a two-tailed T-test at the significance level of 0.95.

For the preparation of the carbon electrodes, carbon films were immersed in EMI TFSI ionic liquid, followed by application of a vacuum for 1 h to ensure infiltration of the ionic liquid into the carbon pores. Electrodes were fabricated using an experimental procedure published earlier.²⁶ Briefly, each carbon film was mounted on a gold disk electrode (CH Instruments, Austin, TX), and a layer of porous polyethylene film (CelGard 3501, Charlotte, NC) was placed on top of the carbon film to increase the mechanical stability of the electrode. To avoid delamination of the carbon film from the underlying gold, the electrodes were mounted into cylindrical bodies custom-made from Dupont Delrin acetal resin. A screw cap at the opposite end of the electrode pressed the carbon film onto the electrode.

All voltammetry simulations were performed using COMSOL Multiphysics version 5.1 software (COMSOL, Burlington, MA). Calculations were performed by utilizing both secondary current distribution physics with transport of diluted species and tertiary current distribution in two-dimensional space. The electrode was assumed to be 0.1 mm in length, with a 0.6 mm distance between the working and reference electrodes. Simulations assume a generic one-electron transfer reaction in a two-electrode system with an ideal reference electrode. Diffusion coefficients used were 0.5 × 10⁻⁹ m²/s for the oxidized and reduced species and 1.0 × 10⁻⁹ m²/s for the counter ion. The exchange current density was 100 A/m² and the scan rate was 1 V/s for all simulations unless noted otherwise.

Values of J_cut-off ranging from 0.01 to 5.0 mA/cm² have been used in the literature (e.g., 0.01,⁷ 0.05,^3,8 0.1,^9,10 0.2,¹¹ 0.3,¹² 0.5,¹³ 1.0,^14–20 1.5,¹² 2.0,²⁸ 2.5,¹² 3.0,¹² and 5.0¹⁹ mA/cm²), although 1.0 mA/cm² is the most frequently used value. Clearly, the selection of the cut-off current density significantly influences the electrochemical limit. While this conclusion is not new, prior reports with the same conclusion were based on single current–potential polarization measurements and did not include a quantification of the error associated with the determination of the electrochemical limit^4,8,12,19 or were based on the comparison of electrochemical limits of the same electrolyte but from multiple studies with different J_cut-off criteria.^3,5 However, the effect of J_cut-off cannot be readily distinguished from the effects of different experimental conditions, such as different concentrations and scan rates.

To quantify the effect of J_cut-off on the electrochemical limits systematically, five replicates of the current–voltage curve were measured in this work for a variety of electrolytes and ionic liquids. They were chosen because they are all commonly used in electro-analytical devices^{22,23,29–31} and electrochemical capacitors^3,6,7,32 due to their wide electrochemical windows. The electrochemical limits were determined based on J_cut-off values of 0.5, 1.0, and 5.0 mA/cm² and are listed in Table I. The use of 0.01 and 0.2 mA/cm² J_cut-off values was attempted, but these values were too close to the background current density to give meaningful results. For the electrolyte solutions, propylene carbonate was used as solvent due to its wide electrochemical window.^2,3,6 Due to solvent-electrolyte interactions, the choice of solvent can slightly affect the electrochemical stability of electrolytes, and, therefore, the solvent in which the cathodic or anodic limits were measured must be reported. In this study, unless noted otherwise, all the electrochemical limits were measured in propylene carbonate.

Table I confirms that J_cut-off has a large effect on the electrochemical stability limits. For example, a change of J_cut-off from 0.5 to 1.0 mA/cm² significantly decreased the cathodic limit of NBu₄I at all concentrations from 75 to 600 mM, giving 40 to 90 mV changes in the cathodic limit. Further increases of J_cut-off to 5.0 mA/cm² resulted in 250 to 400 mV changes in the cathodic limit. Similarly, increases in J_cut-off also affected the cathodic limit of MeNBu₃ TFSI and EMI TFSI (both in propylene carbonate) by as much as 380 mV. The effect of J_cut-off on the electrochemical stability of the ionic liquids in absence of a solvent was even higher. Changing J_cut-off from 0.5 mA/cm² to 1.0 mA/cm² resulted in a 300 mV decrease in the cathodic limits of MeNBu₃ TFSI, BuNMe₃ TFSI, and EMI TFSI. The potential at which J_cut-off reached 5.0 mA/cm² for the ionic liquid MeNBu₃ TFSI was as high as −5.930 V, which is 2.3 V more negative than the analogous value at 0.5 mA/cm². These data highlight the key advantage of the proposed linear fit method, which is the independence from a cut-off value.

The electrochemical limits determined by the cut-off current density method are strongly influenced by electrolyte mass transport. In this section, we demonstrate that the linear fit method corrects for such effects. The total transport or flux of any chemical species to an electrode is described by the Nernst–Planck equation, which combines the individual contributions from diffusion (resulting from concentration gradients), migration (due to electric fields), and convection (caused by hydrodynamic velocity). For mass transport in one dimension only, the flux, J, of a species is given by³³

where D and C are the diffusion coefficient and the concentration of the species in consideration, Φ is the electrostatic potential, and V is the hydrodynamic velocity. Typically, voltammograms of redox-active molecules are measured in unstirred solutions and in the presence of high concentrations of supporting electrolyte; this minimizes convection and the migration of the redox-active molecule, making diffusion the main mode of mass transport.³³ The shape of the diffusion-controlled cyclic voltammogram (CV) for a fully reversible single-electron transfer reaction and diffusion along one coordinate is shown in Figure 2A (labeled with "Model without distortion from resistance and migration"). Very different voltammetric responses are observed in the case of electrolyte reduction or oxidation (see Figure 1 for examples). Even though electro-decomposition of electrolytes has been experimentally studied for several decades, surprisingly, there has been no theoretical work to date that specifically discusses linear sweep voltammograms that exhibit electrolyte oxidation or reduction. Here, we used COMSOL Multiphysics to perform numerical simulations to illustrate the parameters that affect such voltammograms and the electrolyte electrochemical limits determined therefrom.

Zoom In Zoom Out Reset image size

A first correction that has to be applied to linear sweep voltammograms dominated by electrolyte decomposition is the inclusion of resistive distortion. Since there are no ionic species that can carry current other than the redox-active electrolyte itself, solution resistance, R, decreases the potential applied to the solution–electrode interface by a term directly proportional to the current, i. This decreases the slope of the J vs. E curve, resulting in resistive distortion of the voltammetric response.³⁴ The typical effect of resistive distortion on the numerically simulated voltammetric response is shown in Figure 2A, again illustrated for the simplest example of a fully reversible single-electron transfer. Increasing the solution resistivity decreases the height of the peak current, increases the peak separation, and widens the sections of the voltammogram in which the current increases approximately linearly with the applied voltage. Yet, resistive distortion does not fully explain the observed voltammetric response shown in Figure 1.

As a second correction, migration as a mode of mass transport needs to be considered. Unlike in most conventional voltammetry experiments, when electrolyte electro-decomposition occurs the concentration of the reacting species is high compared to the total ion concentration, and diffusion is no longer the sole mode of mass transport. The simulated voltammetric response that takes into account both migration and diffusion agrees well with the experimental observations. Figures 2 illustrates how parameters such as standard redox potentials, concentrations, and diffusion coefficients affect the voltammetric response. For a kinetically facile electro-decomposition reaction, the onset of reduction occurs close to the standard redox potential (see Figure 2B). Figure 2C demonstrates that the electrolyte concentration does not influence the onset of reduction but affects the slope of the linear portion of the voltammogram, where higher concentrations result in steeper slopes. This is consistent with the experimentally observed effect of electrolyte concentration, as evident from Figure 3A, which shows that for 75, 150, and 300 mM NBu₄I solutions, the slope of J vs. E increased with increasing NBu₄I concentration. Importantly, the slope of the J vs. E curve obtained from the simulation (see inset in Figure 2C) exhibits for all electrolyte concentrations a plateau starting a few hundred mV beyond the standard redox potential, which matches well with experimental observations shown in Figure 1C and Figure 3. The effect of the diffusion coefficients on the simulated voltammograms (see Figure 2D) are completely analogous to those of the electrolyte concentration; larger diffusion coefficients result in faster mass transport and a steeper slope for the linear portion of the voltammogram, but cause no change in the onset of the redox reaction. The parallel effects of diffusion coefficients and concentrations can be readily explained by the direct proportionality of both the diffusion and the migration terms of the total flux on both the diffusion coefficient and the concentration.

Zoom In Zoom Out Reset image size

A key difference between the simulations that take migration into account and those that do not are the wider linear ranges of the slopes of the J vs. E curve, as can be seen in Figure 2A. As illustrated in the electrolyte concentration profiles of Figure 2E (which closely resemble the well-known concentration profiles for linear diffusion systems in which migration does not play a significant role), electrolyte depletion at the electrode surface proceeds quickly and increases over time. Because of the concentration dependence of the flux, this lowers the rate of the redox reaction, and it also increases the iR drop.

To further compare the linear fit and J_cut-off method, electrochemical stability limits were also determined from the simulated voltammograms. As shown by Figure 2F, the electrochemical limits obtained with the linear fit method were not affected by the electrolyte diffusion coefficient or concentration, while quite the opposite was true for electrochemical limits determined with the J_cut-off method, see Figure 2F. In conclusion, numerical simulations confirm that the linear fit method is not biased by electrolyte mass transport, demonstrating the advantages of this method over the J_cut-off method.

The effect of electrolyte mass transport was further studied experimentally. When some of us previously introduced the linear fit method in an attempt to minimize the concentration dependence of the electrochemical limit, we gave examples of cathodic limits for NBu₄ClO₄ decomposition. However, those early examples did not include a statistical analysis of the error associated with this new method.² Here, we quantitatively show the effect of electrolyte mass transport on the electrochemical limits obtained from both the J_cut-off and the linear fit method.

Cathodic limits of 75 to 600 mM NBu₄I are presented in Table I. At both J_cut-off values of 0.5 and 1.0 mA/cm², the cathodic limits for 75, 150, and 300 mM NBu₄I solutions do not differ significantly from one another, but for the 600 mM solution a lower cathodic limit is determined. (confidence level of 99%). At the higher J_cut-off of 5.0 mA/cm², higher resistive distortions are observed, and thus the cathodic limits at different concentrations of 75, 150, 300, and 600 mM–which are characterized by different solution resistances–are significantly different from each other.

Using the linear fit method, the fits were applied in the −4 to −5 V region, where the current density exceeds 5 mA/cm². Notwithstanding, the cathodic limits of the 75 to 300 mM NBu₄I solutions as determined with this method are not statistically different from one another. While the limit for the 600 mM NBu₄I solution is significantly lower than the ones for 75 to 300 mM NBu₄I, the difference is only on the order of 40 to 60 mV.

The cathodic limits of MeNBu₃ TFSI were measured both for the pure ionic liquid and a 300 mM solution in propylene carbonate (see Figure 3B). At the J_cut-off values of 0.5 and 1.0 mA/cm², the cathodic limits for the ionic liquid form are approximately 500 and 700 mV lower, respectively, than for the 300 mM MeNBu₃ TFSI solution (see Table I). Figures 3C and 3D show that the onset of the reduction occurs at approximately −3.2 V for both the ionic liquid and the solution. There are no noticeable differences in the electrochemical behavior of MeNBu₃ TFSI in the two cases other than the fact that the J vs. E curve is much steeper in case of the 300 mM MeNBu₃ TFSI (see Figures 3B to 3D; note different scales for dJ/dE in Figures 3C and 3D). Due to different conductivities, mass transport characteristics in the pure ionic liquid and solution are drastically different. The cathodic limit of the ionic liquid as determined with the cut-off current density method is 500 mV lower than the one determined for the solution. In contrast, the linear fit method minimizes the mass transport and resistive distortion effects, and gives similar values of −3.494 ± 0.048 and −3.415 ± 0.005 V for the pure ionic liquid and the solution form, respectively.

Similar observations were made when comparing the cathodic limits of the ionic liquids MeNBu₃ TFSI and BuNMe₃ TFSI. The J_cut-off method gives for both J_cut-off values of 0.5 and 1.0 mA/cm² a difference of 600 mV in the cathodic limits for the two ionic liquids (see Table I and Figure 4). While Figures 4B and 4C show similar onsets of the reduction, they show very different plateau values of dJ/dE for the two ionic liquids (note the different y-axis scales), clearly indicating that a 600 mV difference between the cathodic limits of ionic liquids gives an unrealistic representation of the onset of ionic liquid reduction. The specific resistivity of BuNMe₃ TFSI is approximately one order of magnitude lower than that of MeNBu₃ TFSI (4.86 vs. 34.72 Ω·m),^30,35 resulting in a much larger dJ/dE value for the latter and, consequently, the cathodic limit as given by the J_cut-off method. The linear fit method corrects for the effect of the iR distortion and mass transport and gives more representative cathodic limits of −3.494 ± 0.048 for MeNBu₃ TFSI and −3.368 ± 0.019 V (vs. Ag⁺/Ag) for BuNMe₃ TFSI. Similarly, compared to the J_cut-off method, the linear fit method predicts closer cathodic limits for EMI TFSI as a pure ionic liquid and as 300 mM solution in propylene carbonate (see Table I).

Zoom In Zoom Out Reset image size

It is well known that the scan rate can affect current–voltage curves and, consequently, change the electrolyte electrochemical limits determined with the J_cut-off method,^5,4 but in the past this effect was not quantitatively studied. In this work, the cathodic limit of NBu₄I was measured at scan rates of 1, 10, 100, and 1000 mV/s (see Table II). The scan rate had only a minor effect on the slope of the J vs. E curve and the onset of electrolyte reduction (see Figure 5B) but, as expected, a higher capacitive background current is observed at higher scan rates (see Figure 5A).

Zoom In Zoom Out Reset image size

Because the J_cut-off method does not correct for the background current, the electrochemical limits determined with this method are affected by the scan rate more than the electrochemical limits obtained with the linear fit method. For example, at a J_cut-off of 0.5 mA/cm², cathodic limits of −2.240 ± 0.193 and −3.295 ± 0.070 V are determined for the scan rates of 1000 mV/s and 1 mV/s, respectively. The difference of more than 1.0 V in the cathodic stability of NBu₄I at these two scan rates is clearly an unrealistic representation of the electrochemical stability of the electrolyte. This effect is decreased at very high J_cut-off values (e.g., 5.0 mA/cm²), where the Faraday current is much larger than the background current and cathodic limits are less dependent on scan rate. The linear fit method gives similar cathodic limits of −3.513 ± 0.049 V and −3.422 ± 0.016 V at the two scan rates of 1 mV/s and 1000 mV/s respectively. In conclusion, the proposed linear fit method is less affected by the scan rate and provides a more realistic evaluation of the electrochemical stability of electrolytes.

Porous high-surface-area carbon materials commonly used in electrochemical capacitors include commercially available activated carbons,²² carbide-derived carbons with small, disordered mesopores,³⁶ nanofibers,³⁷ and porous templated carbons with large, open mesopores.²³ To obtain a representative selection, three high-surface-area carbon materials were used for this work, that is, Black Pearl 2000 (a commercially available microporous carbon), three-dimensionally ordered mesoporous (3DOm) carbon,²³ and colloid-imprinted mesoporous (CIM) carbon with a disordered pore structure.²⁶

The total current observed at any electrode is the sum of the faradaic (oxidative or reductive) current, and non-faradaic (i.e., capacitive) current.³³ During a linear potential sweep of the potential, E, with the scan rate v (where E = vt) within the potential window of the electrolyte (where there is no significant electrolyte decomposition), the capacitive current is given by νC_d {1 – exp[-t/(R_sC_d)]}, where C_d is the electrode capacitance and R_s the voltage independent solution resistance.³³ As the scan starts, the non-faradaic current rises from zero to reach a steady value of νC_d. Since the electrode capacitance is proportional to the surface area of the electrode, high surface areas result in large capacitive currents. This is clearly shown in Figure 6A, which shows the capacitive current at a non-porous glassy carbon electrode to be much smaller than in the case of the highly porous Black Pearl 2000 carbon. Note that the two electrodes have the same cross-sectional ("geometric") surface area. While the current density at the glassy carbon electrode (defined as current per geometric surface area) is less than 0.05 mA/cm², current densities as high as 1.0 mA/cm² are observed for the Black Pearl 2000 carbon electrode.

Zoom In Zoom Out Reset image size

The electrochemical limit of the ionic liquid EMI TFSI (commonly used in electrochemical devices with high-surface-area carbon electrodes^22,23) at electrodes consisting of non-porous glassy carbon or of a high-porosity carbon (Black Pearl carbon, 3DOm carbon, or CIM carbon) are shown in Table III for several J_cut-off values. While the onset of reduction occurs at approximately −2.5 V (see Figure 6B), the most commonly used J_cut-off value of 1.0 mA/cm² predicts for all highly porous carbons unreasonably small cathodic limits in the range of −0.5 to −1.0 V. This is readily explained by the large capacitive current at these high-surface area electrodes.^3,24,38 This problem was also observed by Jow and co-workers, who recommended an alternative method for assessing the electrochemical electrolyte stability at electrodes with high surface areas.^21,24 They proposed an approach according to which the limit of electrochemical electrolyte stability is defined by the voltage at which the faradaic current associated with electrolyte decomposition reaches one ninth of the capacitive current (i.e., 10% of the total current, also referred to as 1/R_cut-off; see Figure 3 in Ref. 21). Even though this method takes into account the non-faradaic current, introducing an arbitrary cut-off value for R_cut-off opens the door for disagreements on what that value should be. Indeed, Moosbauer et al. already suggested to change the R_cut-off from 0.1 to 0.02 or lower,³⁹ and Weingarth suggested cut-off values of 0.05 for d²R/dV^*2 instead.³⁸ In contrast, the linear fit method provides electrochemical stability limits without the need to define any cut-off value, as shown in Table III.

Note that current densities at porous electrodes are calculated here based on the geometric surface area (i.e., the area obtained by projection of the electrode shape on a plane). Due to the micropores and mesopores in the electrode material, the real surface area that is accessible to the electrolyte is much larger than the geometric surface area. Therefore, it is difficult to assess the real current density at these electrodes accurately. The linear fit method does not require converting currents to current densities as the surface area of the electrode appears as a constant in the linear fit equations and will be omitted during the intersection of the two. Therefore, there is no need to know the real surface area of highly porous electrodes, which adds to the convenience of this method.