Quantification of PF₅ and POF₃ from Side Reactions of LiPF₆ in Li-Ion Batteries

Thermogravimetric analysis of lithium hexafluorophosphate (LiPF₆, BASF SE, Germany) was conducted with a Mettler Toledo TGA coupled to a mass spectrometer (Thermostar TGA-MS, Pfeiffer Vacuum, Germany). Prior to the experiments, the LiPF₆ was dried at 70°C for 18 h under dynamic vacuum in a glass oven (Buechi, Switzerland) without exposure to air after the drying step. Then, ∼150 mg LiPF₆ were filled into a sapphire crucible inside an Ar-filled glove box (MBraun, Germany) and transferring the crucible into the TGA chamber. For the "dry" experiment, the sample was heated at 10 K/min to 120°C and then held for 1 h at 120°C in a dry argon flow (5.0 purity, 60 mL/min) to remove any traces of H₂O. The sample was then ramped up to 350°C at 10 K/min, where the temperature was held for another 10 min to ensure a complete decomposition of LiPF₆. For the "wet" experiment, the argon gas (60 mL/min) was saturated with water at room temperature and purged over the LiPF₆ sample. After resting for 1 hour at room temperature under the wet argon flow, the sample was heated at a rate of 10 K/min to 350°C, where the temperature was again kept constant for 10 min.

To measure the decomposition of LiPF₆ in our on-line electrochemical mass spectrometer (OEMS),³⁷ we designed a modified cell setup (see Figure 1). As pre-tests showed that typical polymer sealings were either not tight at high temperatures or not stable against the evolving PF₅ gas, we used a Swagelok T-fitting with two metal end caps. The third opening of the T-cell was equipped with the crimped capillary leak that connects the T-fitting to the mass spectrometer (this is also part of our regular OEMS system and has been described previously).³⁷ All cell parts were dried at 70°C under dynamic vacuum prior to assembly. The modified cell was equipped with a thermocouple to monitor the cell temperature and was wrapped with a heating cord (Horst GmbH, Germany); the entire assembly was thermally insulated by fiber cloth (see red shaded area in Figure 1). Tightness of the modified Ar-filled cell was tested in pre-runs of the actual experiments (i.e., holding the cell at temperatures >200°C during several hours), validating that the mass traces m/z = 28 (N₂) and m/z = 32 (O₂) remained negligible. For the three here shown LiPF₆ decomposition runs, 0.26, 0.96, or 1.08 mg LiPF₆ (± 0.04 mg) was weighed into a TGA crucible inside an Ar-filled glove box. The crucible was then transferred into the modified OEMS cell. The cell was closed inside the Ar-filled glove box and connected to the OEMS. After a rest period of 40 min at 25°C, the temperature was increased to 225°C and held there for ∼4 h. The m/z signals were evaluated by dividing the ion current of the respective channel by the ion current of the ³⁶Ar isotope and subtracting the background during the initial rest period at 25°C.

Zoom In Zoom Out Reset image size

Hydrolysis and electrochemical oxidation of 1.5 M LiPF₆ in ethylene carbonate (EC, BASF SE, Germany) was investigated in our regular OEMS cell setup. Note that this system is unfortunately not able to trace HF, as the mass signal for hydrofluoric acid (m/z = 20) superposes with the signal of the ²⁰Ar isotope. For hydrolysis experiments, 5000 ppm H₂O or 5000 ppm methanesufonic acid (99.5%, Sigma-Aldrich, USA) were added to the EC/LiPF₆ electrolyte. After stirring for 30 s, 240 μL of the electrolyte were transferred into an empty OEMS cell (without any electrodes or separators present). The cell was then placed into a temperature chamber set to 10°C and connected to the OEMS. In this way, any hydrolysis reactions occurring at room temperature before the start of the experiment were minimized. The temperature chamber was then set to constant temperature hold steps of 10°C, 25°C, 40°C, and 60°C, each for 3 h, as previously described by Metzger et al.³⁸ Finally, the temperature was set to 80°C for 12 h to follow the long-term reactions of LiPF₆ decomposition products.

For electrochemical oxidation measurements, 500 mg carbon black (C65, Timcal, Imerys, Switzerland) was mixed with 500 mg polyvinyl difluoride (PVDF, Kynar HSV 900, Arkema, France) and 10 mL N-methyl-pyrrolidone (NMP, anhydrous, Sigma-Aldrich, United States) in a planetary mixer (Thinky Corp., USA) for 15 min at 2000 rpm and coated onto a PET separator (FS 24316, Freudenberg, Germany). After drying at 50°C, 15 mm electrodes (loading 1.8 mg_C/cm²) were punched out, dried at 120°C under dynamic vacuum for 12 h, and then transferred to an Ar-filled glove box without exposure to air. The OEMS cell was assembled with a lithium metal counter electrode (450 μm thickness, Rockwood Lithium, USA), the carbon black coating as working electrode, and two 28 mm diameter PET separators (also dried at 120°C) soaked with 150 μL EC + 1.5 M LiPF₆. Electrochemical measurements were performed by applying a linear sweep from open circuit voltage (∼3 V) to 5.5 V at a rate of 0.1 mV/s using a VMP-300 potentiostat (Biologic, France) after an initial 4 h OCV period.

To understand which conditions are required to produce PF₅ or POF₃ in quantitative amounts, we first investigate the thermal decomposition of LiPF₆ under dry and wet conditions by TGA-MS. The possible fragments of PF₅ and POF₃ and their corresponding m/z signals are listed in Table I; to evaluate the experiments, we chose m/z = 107 as a unique signal for PF₅ and m/z = 85 as a unique signal for POF₃.

Figure 2 shows the TGA-MS run of LiPF₆ with a dry argon gas flow. As a sample transfer from the glove box to the TGA instrument under inert gas was not possible, we added a 1 hour isothermal period at 120°C (see 10–70 min in Figure 2) to the ramp of 10 K/min from room temperature to 350°C. Any physisorbed trace water from the sample transfer should be removed during this step. In fact, at the initial ramp from room temperature to 120°C (0–10 min), a small amount of water is released (purple line corresponding to m/z = 18 in Figure 2c) concomitant with a slight endothermic heat flux (orange line in Figure 2b). Both sample mass and heat flow remain constant during the end of the isothermal step at 120°C (10–70 min), meaning that all physisorbed water has been desorbed. Once the temperature in the subsequent ramp from 120 to 350°C (70–93 min) reaches 165°C (see red dotted lines), a significant endothermic mass loss of 83% referenced to the original mass occurs (see Figure 2b). At the same time, a peak in the mass trace m/z = 107 (blue line in Figure 2c) is observed. Both the formation of PF₅ and the mass loss of 83% are consistent with the quantitative thermal decomposition of LiPF₆ according to Equation 1. Interestingly, the found decomposition onset is 30–60°C higher compared to previous reports with a comparable setup.^2,4,5 This could be due to the large sample size (150 mg) used in the present study, as the resulting higher concentration of PF₅ leads to a thermodynamic shift of the equilibrium toward LiPF₆.^2,3

Zoom In Zoom Out Reset image size

Besides the mass fragments for PF₅, we also observe a signal at m/z = 85 (green line in Figure 2c), which belongs to the POF₂-moiety of POF₃ (see Table I). Furthermore, a slight decrease of the m/z = 18 background can be seen. This suggests that part of the PF₅ reacts with trace water in the instrumentation to form POF₃ (see Equation 4), despite the initial drying step and the use of dry argon (5.0 purity), illustrating the high reactivity of PF₅ with trace water:

Figure 3 shows the thermal decomposition of LiPF₆ in the presence of water. To achieve a significant "wetting" of the LiPF₆, water-saturated argon gas (dewpoint of ∼20°C) is flown over the sample before the temperature ramp is started (0–60 min in Figure 3). During this time, a mass increase of ∼24% is observed (black line in Figure 3b), consistent with the weight gain expected for the formation of an adduct with the nominal stoichiometry of LiPF₆ · 2 H₂O (note that in this experiment the background of the H₂O signal (m/z = 18) was too high to observe any changes in the water concentration). It is to note that the sample weight still increases after the initial 60 min, which implies that a further uptake of water would have been possible if the wetting time was extended; however to stay in the same time scale as in the "dry" experiment shown in Figure 2, we limited the pre-wetting in Figure 3 to one hour. The subsequent ramp from room temperature to 350°C (10 K/min) correlates with a mass loss of only 49% referenced to the original LiPF₆ mass, accompanied by a strong endothermic heat flow. This time, no m/z = 107 signal belonging to PF₅ was observed (see blue line in Figure 3c); instead, only POF₃ (m/z = 85) is detected (green line in Figure 3c). Both the mass loss and the POF₃ evolution start around 90°C (see red dotted lines), i.e., 75°C lower compared to the experiment with dry argon gas flow in Figure 2, which is in good agreement with previous reports on the decomposition of intentionally wetted LiPF₆.^4,5 In contrast to these studies, however, the mass loss in our wet experiment (49%) is much lower than the theoretical mass loss based on a complete conversion to LiF according to Equation 2 (83%). This means that another thermally stable species besides LiF must be produced. It is possible that during the course of the TGA experiment, HPO₂F₂ is formed according to Equation 3, which then reacts with remaining LiF in the following equilibrium:⁴⁰

With a melting point of 360°C,⁴⁰ LiPO₂F₂ will not decompose but contribute to the remaining mass at the end of the TGA experiment; if the conversion of LiPF₆ to LiPO₂F₂ were to be quantitative (the sum of Reactions 2, 3, and 5), the overall mass loss referenced to LiPF₆ would be 29%. Thus, if we assume that LiF and LiPO₂F₂ are the only thermally stable products, the mass loss of 49% would correspond to a composition of 37% LiF and 63% LiPO₂F₂, based on the molar masses of M(LiF) = 25.9 g/mol and M(LiPO₂F₂) = 107.9 g/mol. However, a detailed chemical analysis of the sample residue would be necessary to verify this hypothesis.

Zoom In Zoom Out Reset image size

To obtain a quantitative relationship between the ion current measured in our OEMS system and the concentration of PF₅ or POF₃ in the cell headspace, we aimed to thermally decompose LiPF₆ inside a cell connected to our OEMS system. Therefore, we assembled a modified OEMS cell which can be heated up to 250°C (see Experimental section and Figure 1 for details). Once the cell would be heated to 225°C, i.e., well above the decomposition temperature of dry LiPF₆ at 165°C (see Figure 2), we expected LiPF₆ to decompose stoichiometrically to PF₅ according to Equation 1. As the expected partial pressure of PF₅ is less than 20 mbar, a shift of the equilibrium to the left of Equation 1 can also be neglected. Figure 4 shows the cell temperature and the characteristic mass traces for PF₅ (m/z = 107, blue line in Figure 4b) and POF₃ (m/z = 85, green line in Figure 4b) during the decomposition of LiPF₆ under regular dry conditions in our modified OEMS cell. Surprisingly, only mass traces belonging to POF₃ were detected, although all cell pieces and the LiPF₆ salt had been dried carefully and the cell assembly was done in an Ar-filled glove box. We repeated the experiment twice with different amounts of LiPF₆ (see Figure 5), which however still did not lead to detectable amounts of PF₅ on the mass channel m/z = 107. This clearly suggests that the reaction of PF₅ produced in the cell with either trace water in the tubing of the high-vacuum side of the capillary or with the native oxide on stainless steel surfaces leads to a quantitative conversion to POF₃.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

In order to remove any trace water within the tubing of the high-vacuum side of the OEMS setup, we conducted another experiment where the corrugated tube which connects the OEMS capillary with the mass spectrometer was heated to 120°C for 48 h under ultra-high vacuum (see yellow shaded area in Figure 1), while the modified OEMS cell was already attached with the capillary leak closed but kept near room temperature. This allowed to subsequently perform the LiPF₆ decomposition experiment with the OEMS inlet perfectly dried prior to initiating the thermal decomposition of LiPF₆ in the OEMS cell. The results of this experiment are depicted in Figure 4c. While in this experiment POF₃ is still the by far dominant species (green line, m/z = 85), a small amount of PF₅ (blue line, m/z = 107) could now be observed. As the partial pressure of PF₅ in the tubes beyond the capillary is ∼10⁸ times lower than in the OEMS cell (10⁻⁵ vs. 10³ mbar),³⁷ very low amounts of adsorbed water or oxides on the stainless steel tubing surface are obviously sufficient to convert most of the thermally formed PF₅ to POF₃ before it reaches the quadrupole. The apparent decrease of POF₃ in the beginning of the experiment (1.0–1.5 h in Figure 4c) is thus most likely because PF₅ emerging through the capillary reacts with initially present trace amounts of water in the OEMS inlet, leading to a relatively higher POF₃ signal at m/z = 85; after depletion of trace water, increasing amounts of PF₅ at m/z = 107 can be observed, concomitant with a simultaneously lower POF₃ signal.

This experiment shows that even if PF₅ were to form during conventional OEMS experiments in our setup, it will react with trace water (see Equation 4) and/or surface oxides on the stainless steel tubes within the OEMS inlet and thus will predominantly be detected as POF₃. Consequently, at least in our OEMS setup, the above experiment shows that we are not able to differentiate between POF₃ and PF₅ in our measurements. For this reason, it is quite likely that at least part of the m/z = 85 signals which previously had been assigned to POF₃ by our group^31,32 may actually have been due to PF₅. While we do not know whether this artefact might also be occurring with other on-line mass spectrometry systems used for the study of lithium ion batteries, the above experiments certainly suggest that it would be worthwhile to examine the extent of the PF₅ to POF₃ conversion for each system.

Although we cannot distinguish between PF₅ and POF₃ in our OEMS setup, we can still establish an at least semi-quantitative calibration factor for the apparent amount of "POF₃", representing the sum of PF₅ + POF₃. This will be done by correlating the amount of thermally decomposed LiPF₆ with the measured ion currents, evaluating the ion currents on all significant mass channels once they stayed constant, which was typically 3–4 h after the start of the heating experiment. We only considered the experiments where all PF₅ was converted to POF₃ (i.e., without heated tubing), as this would be the default case for all OEMS measurements conducted in our laboratory. Figure 5a depicts the ion current on m/z = 85 normalized to the ³⁶Ar isotope (I₈₅/I₃₆, y-axis) vs. the theoretical concentration of "POF₃" (referring to apparent POF₃) from the thermal decomposition of LiPF₆ (x-axis) for the three measurements with 0.26, 0.96, and 1.08 mg LiPF₆ (± 0.04 mg, corresponding to 4200, 15500, and 17400 ppm "POF₃"). Calibration factors were determined by the linear regression slope between the three data points shown in Figure 5a, which was then referenced to 2000 ppm "POF₃". Figure 5b shows the thus determined calibration factors corresponding to 2000 ppm of "POF₃" for the different POF₃ mass channels after normalizing their ion currents to the ion current for ³⁶Ar (I_x/I₃₆) signals. It can be seen that the fragment on m/z = 85 (POF₂) has the highest intensity, corresponding to a calibration factor of 0.167 ( ≡I₈₅/I₃₆ at 2000 ppm "POF₃"), while the fragments with m/z = 47 (PO), m/z = 50 (PF), m/z = 69 (PF₂), and m/z = 104 (POF₃) all have much lower relative intensities. Note that this differs significantly from the fragmentation reported by the National Institute of Standards and Technology (NIST) for the same ionization energy of 70 eV (see Table I),³⁹ which we ascribe to the fact that the ionization chamber in our instrument is heated at 120°C.³⁷ Table II shows the calibration factors for H₂, C₂H₄, CO, and CO₂ (determined by purging the cell with a test gas containing 2000 ppm of each species), along with the newly determined calibration factor for "POF₃", which lies within the same range (0.1–0.7) as the other calibration factors. Thus, when only POF₃ related signals are observed in our OEMS setup, the sum of PF₅ + POF₃ can be quantified, even though their distribution cannot be determined.

As already mentioned, POF₃ has been observed at high positive potentials on cathode active materials or on carbon electrodes in LiPF₆-containing electrolytes at room temperature, but its amount has never been quantified.^29–32 Therefore, we examined the oxidation reactions of an EC + 1.5 M LiPF₆ electrolyte on a carbon black working electrode vs. a lithium counter electrode, focusing on the evolution and quantification of LiPF₆ decomposition species. Figure 6 shows the current density (a) and the gas evolution (b) during a linear potential sweep from OCV to 5.5 V vs. Li⁺/Li, whereby all signals have been quantified using the calibration factors given in Table II (note that only signals related to POF₃ were observed, so that a quantification of "POF₃" is possible). Around 4.2 V vs. Li⁺/Li, the evolution of "POF₃" (i.e., POF₃ + PF₅, green line in Figure 6b) sets in, together with the formation of H₂ (orange line in Figure 6b). Starting at 4.95 V vs. Li⁺/Li, a significant oxidation current (black line in Figure 6a) and the simultaneous evolution of CO₂ (dark blue line in Figure 6b) are observed, as previously reported from the oxidation of EC-based electrolytes,^35,42,43 along with an enhanced formation of H₂ and "POF₃". Density functional theory calculations by Borodin et al.²⁴ and Li et al.²⁵ have pointed out that upon EC oxidation (i.e., after the first electron transfer), the abstraction of a proton by a neighboring PF₆⁻ anion would occur in LiPF₆-based electrolytes, leading to HF and PF₅ formation, and ultimately to the release of CO₂ and a vinyl alcoholate radical (see reaction pathway (1) in Scheme 1). The produced PF₅ would then detected as "POF₃" in our OEMS setup, while the reduction of HF at the Li metal counter electrode²¹ would result in the evolution of H₂:

Zoom In Zoom Out Reset image size

Scheme 1. (1) Oxidation mechanism of EC in the presence of PF₆⁻, based on Refs. 25,48. (2) Oxidation of ethylene glycol according to Refs. 44,45.

However, contrary to this mechanisms, the evolution of H₂ and "POF₃" between 4.2–4.95 V occurs without the simultaneous formation of CO₂, suggesting that a different process is taking place in this potential range. As trace amounts of water gradually react with EC to form ethylene glycol,^22,38 the latter is a likely impurity in EC present at ppm levels (unfortunately below the NMR detection level). The oxidation of ethylene glycol in aqueous electrolytes leads to the formation of protons,^44,45 which could also react with PF₆⁻ to HF and PF₅ (see pathway (2) in Scheme 1). In fact, the amounts of H₂ and "POF₃" evolved up to 4.95 V would correspond to the oxidation of ∼20 ppm ethylene glycol, which is a probable concentration for this impurity.

For either mechanism (1) or (2) in Scheme 1, the predicted molar ratio of PF₅/HF would be 1/1, so that the experimental molar ratio of "POF₃" (representing PF₅) and H₂ should be 2/1 (since the reduction of 1 HF produces 0.5 H₂; see reaction 6). This is in good agreement with the data in Figure 6, where the amount of H₂ is indeed about 50% compared to that of "POF₃". In summary, the data shown in Figure 6 lead us to the following hypothesis: a) protons or acidic species from EC or ethylene glycol impurity oxidation lead to a fast dissociation of PF₆⁻ to HF and PF₅ already at room temperature, and b) POF₃ observed at oxidative potentials in OEMS experiments on a carbon black electrode is in fact mainly PF₅ rather than POF₃ as we had assumed previously.³²

To verify whether the formation of acidic species (e.g., HF or H⁺) can lead to a significant dissociation of LiPF₆ at room temperature within the timescale of an OEMS experiment, we investigated the reaction of an EC + 1.5 M LiPF₆ electrolyte with 5000 ppm methanesulfonic acid (MSA, pK_A = 8.3 in propylene carbonate).⁴⁶ With this electrolyte, we performed a similar experiment as described by Metzger et al.,³⁸ namely monitoring the gas evolution while gradually increasing the temperature of the electrolyte from 10°C to 80°C. Note that for this type of experimental procedure no active electrodes or lithium are present, so that it only probes purely chemical reactions of the electrolyte. The cell temperature and the resulting gas evolution are shown in Figures 7a and 7b. A significant amount of "POF₃" (POF₃ + PF₅, green line in Figure 7b) is already formed at 25°C, leveling off at ∼1000 ppm. Going to higher temperatures, the overall "POF₃" concentration increases further (from ∼3000 ppm at 40°C to ∼11000 ppm at 80°C); however, in contrast to the constant steady-state concentration reached at 25°C, at higher temperatures the "POF₃" evolution reaches a maximum within ∼1 hour at the respective temperature step and thereafter decreases gradually (a more detailed discussion follows below). CO₂ (navy line in Figure 7b) is only evolved above 60°C, but remains at a much lower concentration compared to "POF₃"; at 80°C, the amount of CO₂ increases continuously.

Zoom In Zoom Out Reset image size

Clearly, the detection of "POF₃" at 25°C in the MSA-containing electrolyte demonstrates that protons can rapidly and substantially shift the dissociation of PF₆⁻ toward PF₅ and HF (see Equation 7). This is in good agreement with Freire et al.,⁴⁷ who found that PF₆-based ionic liquids hydrolyze in aqueous solutions with pH = 3 at room temperature, but not under neutral conditions.

At 25°C, the establishment of a steady-state "POF₃" concentration and the absence of CO₂ indicate that at this low temperature no PF₅ is consumed in a reaction with the electrolyte or that this reaction is too slow to be detected within the timescale of the OEMS experiment (∼3 h). In contrast, the decreasing concentrations of "POF₃" during the second half of the temperature step at >40°C are likely a result of PF₅ reacting with the electrolyte to oligomers⁸ or alkyl fluorophosphates.^2,7,9–11 As the rate of chemical PF₅ consumption will depend on both temperature and the concentration of PF₅ dissolved in the electrolyte (which is by Henry's law proportional to the amount of PF₅ in the headspace of the cell), a faster consumption of POF₃ is expected at 60°C and 80°C, where both temperature and overall "POF₃" concentration are higher. Additionally, a stepwise increase of the temperature can cause the desorption of PF₅ or POF₃ from the inner surface of the steel tubing, which leads to a peak in the detected "POF₃" concentration at the initial phase of each temperature step.

Interestingly, the extent of CO₂ formation in Figure 7b is very similar to that observed with an EC/LiClO₄ electrolyte with < 20 ppm water reported by Metzger at el.³⁸ (see Figure 1 in Ref. 36), which means that the direct reaction of PF₅ with EC does not generate additional CO₂. For comparison, Figure 7c shows the same experiment with 5000 ppm of added water instead of MSA. Up to 40°C, neither "POF₃" (green line in Figure 7c) nor CO₂ (navy line in Figure 7c) are observed. Only at temperatures 60°C and above, low amounts (< 2000 ppm) of "POF₃" are detected, in this case likely due to the formation of POF₃ rather than PF₅ (this, however, cannot be discerned in our OEMS experiments). At the same time, the CO₂ concentration increases linearly at 60°C and 80°C to above 12000 ppm, in agreement with previous experiments on the water-driven hydrolysis of EC in LiClO₄-based electrolytes.³⁸

Comparing the results of Figures 7b and 7c, it becomes apparent that only highly acidic species can trigger the formation of PF₅ at room temperature. Accordingly, the oxidation of an EC/LiPF₆ electrolyte (see Figure 6) must be leading to the formation of species which act as proton donors or Brønsted acids, which is consistent with the mechanism proposed by Borodin et al.²⁴ and Li et al.²⁵ (see Scheme 1). Furthermore, the release of PF₅ might explain the strong temperature dependence of the oxidative stability of LiPF₆-based electrolytes.³³ In this context, the use of proton-scavenging additives, i.e., bases,^9,18 could be a successful strategy to prevent the dissociation of PF₆⁻ at high potentials, thereby suppressing HF formation and electrolyte degradation.