An Experimental Method to Determine the Measurement Error of Reference Electrodes within Lithium-Ion Batteries
Graphical Abstract
An experimental approach to determine the potential deviation of a perforated reference electrode within a NMC811/graphite pouch cell in-operando is presented. By provoking Li plating via the graphite potential and detecting it via cell dilatation, it is possible to evaluate reference potentials more accurately without the need for post-mortem analysis.
Abstract
Reference electrodes (REs) play a crucial role in the accurate assessment and control of battery potentials, but their confidence is overestimated. Researchers have tracked the source of the error to the RE design that blocks the lithium-ion path between anode and cathode. These errors or potential deviations are mostly modeled or less-frequently estimated after observing Li plating post-mortem. This is the first study to showcase an experimental method that allows a more precise error quantification in-operando of a RE. The key idea is to relate the error-affected reference potential to an unaffected quantity, such as the cell dilatation. Although our experimental setups are special, this approach can also be applied to different setups and REs. Using the presented method, we provoked Li plating in NMC811/graphite pouch cells and determined the potential deviation of our perforated RE to be 12 mV under fast charging conditions. In contrast to previous studies, we found the error to be positive, offering a new explanation of the error mechanism of REs.
Introduction
The use of reference electrodes is a common approach for studying electrode materials in lithium-ion cells, providing a valuable extension to conventional two-electrode configurations. Most researchers mention them in their work, but often more as a side note without any experimental details; a powerful tool that can be trusted for its capability. Unfortunately, this confidence is usually overestimated, one can speak of an error-plagued research.1 The reasons for this are quickly identified, but complex in their origin: Elemental lithium is still the most used reference electrode (RE) in non-electrolytic, beaker-less, non-aqueous, sealed cells since it is easy to implement and requires no conversion (vs. Li/Li+).2-5 The issue with Li is that the measured potentials vitally depend on the electrolyte composition6, 7 and concentration8 and its cleaning procedure,9 limiting their comparability considerably. An equally constraining factor is the encountered variation of the design and size of REs, delivering different potentials even if the RE material is the same.2, 3, 10, 11 Finally, the position of the RE is another aspect, that can yield totally different results.2, 11-14 Understandably, the development of REs has been driven by some researchers and has received considerable critical attention.2, 3, 11-17 It was found that one major error source is connected to the blocking of Li-ion paths, which some of these studies approached theoretically. So whether a measured reference potential can be considered accurate or trustworthy is not a trivial question. To the best of our knowledge, the only experimental approach to validate a RE is by trying to provoke Li plating at different C-rates and then observe it post-mortem.3 However, this is an elaborate and destructive process. Unsurprisingly, the errors have barely been quantified experimentally. Since reliable REs and reference potentials are paramount for the entire field, we developed a method to experimentally determine the error and applied it to our setup as an example. This work is therefore based partly on our previous studies, where we fabricated and tested a perforated reference electrode (PRE), basically a mesh-like design that we encourage to try.10, 18 It must be emphasized that the core idea of this work is not confined by its experimental characteristics, but should be understood as a general approach; a method to inspire researchers to empirically validate their REs.
Theory
In order to understand the key aspects of this work, the reader needs basic knowledge about reference electrodes and overpotentials in lithium-ion cells and the correlation between cell dilatation and State of Charge (SoC). We will only cover these basics briefly. For further understanding, the reader is referred to reviews and papers on these topics.4, 19-23
Charging via anode potential
Previous studies have shown how the anode potential can be used to perform charging protocols.24-29 In contrast to a regular constant current constant voltage (CCCV) protocol, the CV phase is not adjusted to the cell voltage but to the reference potential (CVref). During the CC phase, a high C-rate can be applied until a fixed cut-off reference potential is reached. The further the anode cut-off voltage is selected in the direction of Li plating, the longer the CC phase lasts. Once the target anode voltage is reached, it is kept constant by gradually decreasing the current until either the cell voltage exceeds a threshold or the current drops to a certain limit.
Using this charging technique, it is theoretically possible to find the ideal current characteristic for fast charging a cell without damaging it by Li plating.30 However, potentials measured in Li-ion batteries via internal electrodes are generally deviating from the actual electrode potentials. These differences are modeled, using overpotentials.
The dilemma of error sources
The classification and explanation of overpotentials is not trivial. In general, overpotentials arise mainly due to slow Li+ diffusion within active materials and electrolyte phases of the pores.22 In other words, the rate-limiting process during fast charging is the slow transport of Li+ ions within the electrodes, causing polarization and consequently Li plating.31 Overpotentials depend heavily on the cell components and operating parameters used. Nevertheless, when using a RE, there are effects and circumstances that can distort the reference potential:
-
Blocking effect: If the RE's geometry is blocking the Li+ flow between anode and cathode, electrolyte gradients appear in the vicinity of the RE. As a result, the RE will measure potentials that differ from the real ones. In the example of graphite, this could mean that the lithiation steps are only blurred or do not occur at all.10 The error or potential deviation induced by the blocking effect can be minimized by reducing the RE size and applied current or increasing the electrolytic conductivity.2, 3, 15 Theoretical considerations suggest that the blocking width of the RE should be in the range of at most 50 μm11 (or 140 μm3) to deliver trustworthy potentials, assuming moderate (or low) C-rates.
-
RE position: The RE should be placed at a location where the equipotential lines between anode and cathode are not distorted. Practically, this means a centric location within the anode area. Placing the RE outside or at the edge of the electrodes will likely result in distorted EIS spectra and more positive potentials, underestimating Li plating.13, 14 If the RE is covering a relatively large part of the anode, a capacity drop could be detected depending on the C-rate and RE geometry.32 In the classical view, the error of a RE mainly depends on its distance to the working electrode: A greater distance leads to a larger ohmic drop.
-
RE formation: If the RE is formed within the cell, for example through Li alloying, it is recommended to keep the same procedure for reproducibility reasons. In the case of REs that require a specific SoC to provide a constant potential, such as lithium titanate oxide (LTO) or lithium iron oxide (LFP), the C-rate and charged capacity will have an impact. Moreover, the voltage hysteresis of such REs should also be taken into account. Note that the OCV of REs can generally drift over time depending on each system and procedure, highlighting the necessity of measuring and publishing the OCV curves of REs to pinpoint their potential at a given time. Alternatively, an additional Li reference electrode can help to calibrate non-metallic REs. To distinguish from cell formation, the term ”conditioning” is used for the RE formation throughout this work.
-
Electrolyte: In the case of elemental lithium, the SEI is reported to form differently depending on the type of liquid electrolyte.1, 4, 6 Consequently, the ohmic resistance differs. This is the reason why comparing reference potentials measured in different electrolytes, but via the same Li RE, could prove difficult under certain conditions.
-
Electronic equipment: The internal resistance of the electronic equipment can have significant impact on the performance of SoC-dependent REs. If both, the RE capacity and electrometer resistance of the testing channel are small, then the RE discharges over time. Measuring 3 V, for a 100 μAh RE and an internal channel resistance of 30 MΩ, the time it takes to fully discharge is 1000 h.

In our last work, we tried to tackle the main issue of the blocking effect by establishing a post-processing method to manufacture PREs on the basis of LFP.10 These customdesigned REs were positioned centrically within pouch full cells and showed a stable OCV for over 250 h. Moreover, dynamic cycling tests were performed and validated by cell dilatation measurements, showing almost no hindrance of anode expansion at PRE site at low to moderate C-rates. Nonetheless, the question remains how high the error of the PRE is, especially in a fast charging scenario at high currents. In the following sections, an approach is presented that allows the total potential deviation of any RE to be determined experimentally.
Experimental determination of RE error
It is possible to resolve the problem of potential deviation by recording external quantities, i. e. outside the electrolyte, and comparing these with the anode potential. The challenge is to find a common reference point where the real anode potential can be derived from the external, error-free quantity. Li plating, a phenomenon that takes place thermodynamically at <0 V vs. Li/Li+, represents a possible reference point.33 However, Li plating may occur at a potential higher than 0 V vs. Li/Li+ if the temperature or the temperature deviation within the graphite exceeds a certain onset.34 This phenomenon is known as underpotential deposition and is also influenced by the surface where Li plating occurs:35, 36 In the case of graphite, the underpotential deposition is mainly attributed to an inhomogeneous SEI formation, which can be inhibited by the addition of vinylene carbonate (VC) to the electrolyte.37 In our case, an underpotential deposition can be ruled out since we used VC in our electrolyte and, according to our temperature sensors, we did not reach high temperatures or temperature deviations due to our small cell format. If Li plating is taken as a reference point, we recommend monitoring the cell temperature, ideally within the cell, and the use of VC electrolyte additive to avoid the influence of underpotential deposition.
In our last paper, we have already discussed the principle of cell thickness change during charging.10 Its external mechanical property allows detecting Li plating without the direct involvement of disturbances within the cell.38-40 As soon as an irreversible cell thickness increase (Li plating) is detected, the simultaneously measured anode potential represents the error of the RE. In this work, a conditioned LFP PRE was used. To reference the measured potentials versus Li/Li+, they need to be corrected by the LFP half-cell potential; at 50 % SoCLFP that would be ULFP vs. Li/Li+=3.428 V.41 We determined this value in our last publication, after measuring the OCV of the PRE within half-cells for >250 h.10 Since we conducted the experiments within 2 weeks, we chose this value to convert the measured reference potentials, knowing that there might be an uncertainty of a few millivolts.
To provoke Li plating, the graphite potential is gradually decreased towards 0 mV vs. Li/Li+. Figure 1 illustrates this method with a hypothetical example: Within a CCCVref charging procedure, the graphite cut-off voltage UGraphite vs. Li/Li+ is successively decreased by 2 mV per cycle. At the same time, the cell thickness change is logged at RE-site. From the 4th cycle, the cell dilatation increases irreversibly (Li plating). Accordingly, the total error ΔURE measured via RE is 10 mV. The same approach is used in the experimental section to determine the potential deviation of the PRE.

Hypothetical example of a CCCVref protocol where the graphite cut-off potential is decreased by 2 mV per cycle while recording the cell dilatation at the end of each cycle. By detecting Li plating starting at cycle 4, the measured anode potential represents the RE error ΔURE at this time.
Results and discussion
The idea to quantify the error of the PRE presented in Figure 1, and its experimental evaluation is not trivial since it combines two sensor techniques. It is appropriate to examine the raw data of each sensor technique first and then collate all data and take a closer look at each individual cycle. Before proceeding, it will be necessary to rule out any pressure inhomogeneity of the setup itself which can accelerate Li plating.42 Thus, the first section provides a brief report on the pressure distribution of the cell.
Pressure distribution mapping
The pressure on the cell seems homogeneous, with an average of 25 kPa (Figure 2a). Some pressure peaks (≈70 kPa) are detected around the edge of one reference electrode and at the corners of the cell stack (Figure 2b). The latter originates from adhesive tapes that were used during stacking to hold the stack tight. The pressure at the outer rim of the cell is higher (>100 kPa) because the force accumulates at the edge of the stack. In summary, there should not be a major pressure disturbance in the relevant electrode area that would falsify the cell thickness measurement. Having confirmed the setup's pressure homogeneity, the next section focuses on the graphite potential.

Pressure distribution mapping of cell. (a) Homogeneous pressure distribution with an average pressure of 25 kPa. (b) Slight pressure peaks around the edge of the PRE and tapes. The pressure accumulates at the rim of the cell, leading to higher pressure.
Graphite cut-off voltage evolution
The temporal evolution of the graphite cut-off potential during the CCCVref fast charging protocol is displayed in Figure 3.

Excerpt from the CCCVref procedure where the graphite cut-off potential is changed towards Li plating after every third cycle.
Primarily, it illustrates one key aspect of the method and represents the raw, uncorrected data, whereas UGraphite vs. Li/Li+ is converted by the half cell potential of LFP at 50 % SoCLFP (ULFP=3.428 V vs. Li/Li+). Note that the values of UGraphite in Figure 3 are the theoretically expected values, since they are not yet corrected by ΔURE. Starting from a value of 28 mV vs. Li/Li+ towards the expected Li plating threshold, the regulated graphite cut off potential UGraphite(co) is changed gradually by 2 mV after every third cycle, provoking Li plating. There are no outliers or artifacts visible. By carefully examining the data, it is found that the CVref phase requires less time over the course of the experiment; a feature that will be discussed in the next sections.
Identifying Li plating
The raw data of all four cell thickness sensors (ΔCT1-4), together with the capacity, are plotted in Figure 4. Cell capacity is constant until 35 h; at that point UGraphite(co) is −3.410 V vs. LFP. From then on it is decreasing continuously with a major drop at 55 h. As explained in the next section, this drop is due to the cell voltage termination criterion. For all cell thickness sensor channels, the characteristic amplitude of each full cycle can be read. They match the capacity profile, including the major drop starting at 55 h. The most striking and significant observation in this section is the irreversible increase of cell thickness of channel ΔCT1 starting at 45 h. After several unremarkable cycles, this is a clear indication of Li plating at PRE site. The second most important finding is a similar behavior for ΔCT4, starting at 50 h. The increase is not as pronounced as for ΔCT1, but it exists. This means, Li plating occurs roughly in the middle of the graphite anode; a place where it would be expected.43 The reversible stripping of plated Li can be expected but it overlays with our measurement protocol in such a weak manifestation that our setup is not able to detect it. Altough the formation of SEI/CEI overlays during cell dilatation measurement, its cell expansion slope is very different compared to the one of lithium plating. The slope would be much smaller, showing a constant drift of the cell thickness curve, which we do not observe in Figure 4. Furthermore, it should be noted that gas evolution can be ruled out as a cross-influence, since our pouch cell format had an extra gassing space and we did not observe any bloathing of the cell. For the other two channels ΔCT2-3, there is no sign of Li plating.

Temporal course of capacity and cell thickness change at channels ΔCT1-4 (positioning see Figure 9b) during CCCVref cycling. Channel ΔCT1 and ΔCT4 show an irreversible increase of cell thickness starting at 45 h and 50 h.
Another interesting feature is the slightly lower amplitude (≈0.5 μm) of ΔCT1 compared to the other channels. It seems that the PRE is slightly blocking the anode area at that location, preventing a complete expansion. Moreover, unlike the other channels, ΔCT1 does not decrease in the range of 40 h to 45 h, further supporting the hypothesis of a blocked anode: If the total amount of lithiated graphite particles is lower, then the total expansion will also be lower. Taken together, the CCCVref fast charging protocol provokes Li deposition at the PRE site and at the center of the anode. Nonetheless, it appears that the PRE exhibits a slight blocking behavior.
Quantifying PRE error
In the previous sections, we observed Li plating that was purposefully provoked by CCCVref fast charging. In this section, all the relevant sensor and cell data are collated per cycle or cut-off potential UGraphite(co) to quantify the PRE error. First, we focus on the third cycle or first graphite cut-off value by plotting only both voltages (Ucell and |UGraphite| vs. LFP), as well as the current I in Figure 5. The graphite potential is given as an absolute value to better visualize both voltages in one figure. Cell dilatation data are left out, since they do not show any remarkable features in the beginning. During the CC phase a current of 10.85 A is applied (≡7C), reaching ≈12.5 % SoC in 1.1 min. After that, the current drops relatively linear until the cut-off of Ucell (4.300 V) is attained. Ucell rises during the CC phase and then exhibits a plateau-like progression during the CVref phase. These features and their progression over cycles is considered as a secondary finding and will be discussed later.

First graphite cut-off voltage or third cycle of the CCCVref protocol showing exemplary the course of current and voltages. UGraphite is given as an absolute value for visualization reasons.
For now, the priority lies on the summarized data at four different cut-off voltages of graphite, representing the relevant phases of the CCCVref procedure (|UGraphite(co)|=3.414 to 3.420 V vs. LFP). They are depicted one above the other in Figure 6. Considering the cell thickness data, two key features to quantify the PRE error become apparent that will be discussed chronologically:

Evolution of cell thickness (ΔCT1-4) and cell voltage (Ucell) depending on graphite cut-off voltage. The raw data of the cell dilatation are plotted in Figure 4.
-
|UGraphite(co)|=3.414 V vs. LFP (Figure 6a): All channels show similar behavior without any remarkable features. This is the last graphite cut-off voltage before Li plating is detected.
-
|UGraphite(co)|=3.416 V vs. LFP (Figure 6b): Channel ΔCT1 rises clearly compared to the other channels, signaling Li plating at RE-site. This significant feature would suggest a PRE potenital deviation of 12 mV.
-
|UGraphite(co)|=3.418 V vs. LFP (Figure 6c): 2 mV further, channel ΔCT4 also starts slowly differing at 30 % SoC, with a substantial irreversible increase from 55 % SoC till the end of the cycle. Together with the raw data from Figure 4, this is a striking sign of Li plating at the center of the anode.
-
|UGraphite(co)|=3.420 V vs. LFP (Figure 6d): Although this charging step only reaches 40 % SoC due to the cell cut-off voltage, the irreversible thickness increase of ΔCT1 (and ΔCT4) is even stronger (remains).
Overall, the detailed analysis indicates a total error ΔURE of the PRE of at least 12 mV under the given parameters. This can be considered one of the key take-aways of this study. However, Li plating is a non-uniform, partly reversible and stochastic phenomenon.44 It probably started a few milli-volts earlier and in such a weak manifestation that our cell thickness setup was not able to detect it. Surprisingly, the PRE potential error was found to be positive instead of negative. This means that our PRE underestimates the actual graphite potential, although studies on the blocking effect suggest an overestimation.2, 3, 11, 12, 15 Consequently, a contradiction arises in the interpretation of the blocking effect and thus error of a RE: In contrast to the familiar explanation of ohmic drop and electrolyte concentration gradients, we link our results mainly to local SoC divergence induced by the geometry of the PRE (Figure 7): The perforations act as Li ion channels that enable lithation of the nearby graphite particles. But there may still be blocked areas close to the 50 μm broad webs. This leads to less lithiated graphite particles, which have a more positive potential (vs. Li/Li+) compared to the particles above the perforation. Assuming that the PRE is sensing the graphite surface at its place, this potential or SoC divergence of graphite particles leads to a total error that is more positive than the actual graphite voltage. This hypothesis is supported by our previous work, where we used the same RE but without any perforations, measuring much higher and unplausible graphite potentials at low currents.

Visualization of the blocking effect during fast-charging. Although the perforations act as Li-ion channels enabling lithiation in the vicinity, graphite particles near the web do not get fully lithiated. This lithiation and thus potential difference leads to a positive error.
Observations regarding evolution of current and cell voltage
Before turning to experimental validation by post-mortem analysis, there are also some findings related to current and cell voltage evolution during the CCCVref procedure: Comparing the cell voltage plateau in Figure 5 and Figure 6 shows an uplifting shift, i. e. the cell cut-off voltage is reached progressively earlier. An equally interesting feature is the development of the current curve from a linear (Figure 5) to a non-linear shape (Figure 6). Apparently, the non-linearity (shift) of current (cell voltage) is connected to the increasing graphite cut-off potential UGraphite(co) and thus longer lasting CC phase. This suggests a kinetical overpotential progression or more precisely an interaction between graphite lithiation (with its stages) and overpotential. However, this hypothesis should be interpreted with caution, since more experiments would be needed to draw a conclusion, which is beyond the scope of this work.
Post-mortem analysis
As seen in Figure 8, the post-mortem analysis confirms the cell thickness results: Li plating can be seen at the locations of ΔCT1 (green) and ΔCT4 (gold), i. e. at PRE-site and the center of the anode (Figure 8a). Interestingly, the Li deposition extends only partially over the area of ΔCT4, which explains why the irreversible cell thickness increase was less pronounced compared to ΔCT1. There is also Li plating at some edges of the electrode, which can be attributed to the overcharging of the anode overhang.45 Li deposition is particularly strong at the southern edge, which was narrowly missed by the thickness sensor at ΔCT2 (pink). In contrast to non-plated regions, spraying these areas with 99.9 % isopropanol resulted in gas evolution, indicating the presence of elemental Li.46 An important observation is that Li plating was at the PRE-site, proving at least a not fully blocked Li-ion path towards the graphite. In addition, the PRE did not show any defects or clogging (Figure 8b). As expected, the unperforated rim of the PRE did block the graphite lithiation, leaving a blurred black ring area as a trace.

Post-mortem analysis. (a) Aged graphite layer where the PRE was located, showing Li plating at ΔCT1 (green) and partially at ΔCT4 (gold) as well as at the outer edges. (b) The PRE exhibits no damages or clogging of the perforations.
Conclusions
This is the first study reporting an experimental method to quantify the error of a reference electrode in-operando within a full cell. The key idea is to relate the reference potential which is affected by the blocking effect, with a correlating, external quantity, such as the cell dilatation. For this, we have used our eddy-current sensor setup and taken Li plating as a common reference point. Li plating was provoked by a CCCVref procedure, that should resemble a real fast charging scenario: A 7C-CC phase until a defined graphite cut off-potential that was cycle-wise changed towards 0 V vs. Li/Li+. Cell dilatation data indicated Li plating at two locations, which was later confirmed by post-mortem analysis. Comparing the onset of Li plating with the graphite potential measured at that time reveals the error of the PRE to be at least +12 mV vs. Li/Li+. This means that Li plating is underestimated. Surprisingly, previous studies suggest a negative error, i. e. overestimating Li plating. Consequently, our interpretation of the blocking effect is a different one: We agree with recent studies that the RE size is one of the dominant factors, blocking the Li-ion path and thus inducing potential errors. However, we assign this error not to the familiar explanation of electrolyte gradients and ohmic drop but to less lithiated graphite particles, hence showing a more positive potential. In principle, the presented method can be used to analyze other RE-types as well, but a few important limitations need to be considered:
-
According to our data, we assumed that the PRE measures at its location, but a general theory of the sensing mechanism is still at stake.
-
One finding gained from the raw cell thickness data is that our PRE design seems to partly block the anode lithiation and thus expansion under high C-rates. A solution to this would be to further reduce the web width and increase the open area. Ideally, the design of the PRE should exhibit a high open area with web widths that are smaller than the graphite particles.
Despite these limitations, the empirical findings in this study could lead to a new understanding of the blocking effect and serve as a basis for future studies.
Experimental
The reference electrode and cell thickness setup used in this work can be considered special. Their development has been described in our last work, but will be presented briefly again in the following.10 Nonetheless, theoretically, any sufficient-sized RE could be analyzed together with a high-precision cell thickness setup.
Reference electrode preparation
A slurry was prepared with a composition of 92 wt % commercial LFP active material, 5 wt % carbon black (C45, Imerys, France) and 3 wt % polyvinylidene difluoride binder using a planetary mixer. The slurry was coated on aluminum (16 μm) via a roll-to-roll coating system.
The loading of the LFP coating was 6.75 mgLFP cm−2, corresponding to ≈1.0 mAh cm−2 based on a theoretical specific capacity of 155 mAh gLFP. Afterwards the coating was compressed to a total thickness of 51 μm (≡1.98 g cm−3 electrode density). REs were prepared in the same way as shown in our earlier work. In brief, we used a 40 W ultrashort pulse Nd : YAG laser to perforate and cut LFP-electrodes. The perforation design consisted in hexagonal closely arranged circular openings with diameters of ≈50 μm and web widths of ≈50 μm, forming a mesh-like structure. The perforated electrode had 21.6 mm in diameter with a 200 μm unperforated outer edge and a 1.0 mm wide aluminum current collector leading out of the stack. The whole electrode was laser-cut in one piece to fit in our cell format. All PREs were washed with dimethyl carbonate (DMC, ≥99.8 %, Carl Roth, Germany) to remove burn residues and dried at 60 °C under vacuum overnight. The open area was determined gravimetrically to be 25 % (±2 %, excluding unperforated edge). The total discharge capacity of the PRE was 1.8 mAh.
Full pouch cell assembly
Three ≈1.55 Ah full pouch cells with implemented LFP PREs near the anode tab were manufactured. The PREs were sandwiched between two 9 μm separators. In order to keep the same 18 μm distance between all electrodes, two separators of the same type were placed between the remaining layers. While LiNi0.8Mn0.1Co0.1O2 (NMC811) with a loading of ≈16.15 mg cm−2 (≡3.10 mAh cm−2, 96 wt % cathode active material) was used as cathode, graphite with a loading of 9.82 g cm−2 (≡3.30 mAh cm−2, 94.5 wt % anode active material) served as anode. All electrodes and cell components were vacuum dried overnight at 65 °C and placed in an argon-filled glove box (<0.1 ppm H2O, <0.5 ppm O2) without exposure to air for cell assembly. A custom-made holder was used to positon the PRE always at the same location during stacking, i. e. near the anode tab center. To each cell, 6.0 ml electrolyte was added, consisting of 1.4 M lithium hexafluorophosphate (LiPF6) mixed in ethylene carbonate, ethyl methyl carbonate, DMC (2 : 1 : 7 v : v, <1 ppm H2O) with 1.0 wt % vinylene carbonate and 1.0 wt % lithium difluorophosphate. Subsequently, all pouch cells were degassed and sealed at 40 mbar. After compressing the cells to ≈25 kPa via metal pressure plates, the cells were placed in a climate chamber (Memmert, Germany) at 25 °C (±0.1 °C) and connected to a battery testing system (CTS, BaSyTec, Germany).
Measurement of cell thickness change
For the cell thickness measurement, the setup of our last work was used, consisting of four eddy-current sensors in a quadratic formation (Figure 9a).10, 47 Each sensor measures the cell thickness change over a circular area (Ø=14 mm). The sensor is placed in a custom-made holder, forming a flat surface facing the cell. The placement of the sensors is shown in Figure 9b. One of the sensors measures directly above the RE near the anode tab (ΔCT1), while the other three sensors measure the thickness change at the middle regions and edge. Data is collected via a micro-controller outside the setup. To ensure a homogeneous pressure distribution, a polyurethane foam is placed between cell and sensor array. The pressure on the cell is adjusted by the length of four metal spacers that are clamped between two metal plates. To rule out any pressure inhomogeneities that might occur due to the PRE, pressure distribution mapping was previously performed using a pressure measurement film (Prescale, Fujifilm, Japan).

(a) Setup for cell thickness (CT) change measurement. (b) Top view cell – positions of the sensors (ΔCT1-4) for measuring the cell dilatation. Position ΔCT1 is above the PRE. (c) Steps in a CCCVref protocol: Initially, a fixed current I is applied until a specified cut-off voltage anode voltage was reached (step 1). Subsequently, UGraphite is kept constant until the cell voltage UCell hits a predefined limit (step 2).
Testing procedure to quantify total error of PRE
At first, cells were formed in a full cycle at C/10 with potential limits ranging from 4.2 V to 2.8 V. At the end of the formation, they were charged to 30 % SoC. The RE and anode were then connected to the cycler and conditioned at 250 μA over two C/10 cycles (CCCV charge to −3.6 V, discharge to −2.5 V vs. LFP), and a subsequent charging to 50 % SoCLFP. Note that referencing graphite versus LFP results in a negative potential by definition. The full cells were then rewired and underwent a small dynamic CCCV-cycling protocol: Two cycles each at C/10, C/5 and C/3, respectively. This was done to check whether the measured reference voltages were plausible and reproducible. Following the idea to quantify the potential deviation of the PRE described in Figure 1, a special CCCVref fast charging procedure was performed (Figure 9c): During the CC phase, 7C was applied until a fixed graphite cut-off voltage UGraphite(co) was reached. The CVref phase was terminated as soon as the cell voltage hit 4.3 V. After every 3rd cycle, UGraphite(co) was decreased by 2 mV, starting from −3.400 V to −3.426 V vs. LFP. After each fast charging step, the cell paused 20 min and was then discharged at 1C. During the whole protocol, the cell thickness change was recorded simultaneously to identify the start of Li plating. An overview of steps the cells underwent is shown in Table 1.
Step |
Cycles |
Charging |
Discharging |
---|---|---|---|
Formation |
1 |
CC: 0.1 C |
0.1 C |
|
|
CV: 4.2 V until 0.1 C |
|
Conditioning |
2 |
CC: 0.1 C |
0.1 C |
|
|
CV: −3.6 V[a] until 0.1 C |
|
Check-up |
6 |
CC: 0.1–0.3 C |
0.1–0.3 C |
|
|
CV: 4.2 V until 0.1–0.3 C |
|
CCCVref |
40 |
CC: 7C |
1 C |
|
|
CV: −3.400–(−3.426 V)[a] |
|
- [a] Graphite versus LFP.
Post-mortem analysis
All cells were discharged to <1 V and transferred to an argon-filled glovebox (<0.1 ppm H2O, <0.5 ppm O2) for disassembling. Graphite anodes, NMC811 cathodes, separators and PREs were visually examined. When graphite electrodes showed suspected Li deposition, 99.9 % isopropanol was sprayed on these areas to observe any gas evolution.
Author Contributions
Daniel Rutz: Conceptualization, Methodology, Writing – Original Draft, Writing – Review & Editing, Visualization. Felix Brauchle: Methodology, Visualization, Software, Investigation, Writing – Review & Editing. Ingolf Bauer: Supervision, Resources, Writing – Review & Editing. Philipp Stehle: Investigation, Writing – Review & Editing. Timo Jacob: Supervision, Writing – Review & Editing.
Acknowledgments
Special Thanks to Rico Knappe from Requisimus AG for laser modification. This work was partly founded by the Federal Ministry for Economic Affairs and Energy in the context of the project “next generation power batteries” (NEWBIE, Grant No: 01MV21013 A). Open Access funding enabled and organized by Projekt DEAL.
Conflict of interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Open Research
Data Availability Statement
The data that support the findings of this study are available from the corresponding author upon reasonable request.