Series-Parallel Testing of Laboratory Lithium-Ion Cells: Not All Coin Cells Are Created Equal

Abstract

This work evaluates parallel battery testing as a research tool for assessing both the reproducibility of laboratory-fabricated cells and the effects of cell-to-cell variation during pack-relevant operation. Eight hand-assembled CR2032 Li||LFP half-cells were tested using both parallel differential battery testing and conventional linear battery testing. Although all cells were prepared from the same nominal materials and procedures, SEM/EDS and electrochemical results showed measurable variability arising from microscale heterogeneity in the hand-cast cathodes and lithium metal anodes. These differences became more pronounced at higher C-rates, where weaker cells delivered lower capacity and contributed less effectively during coupled operation. Parallel testing revealed unequal current sharing, post-discharge equalization behavior, and broader performance differences than linear testing, while impedance measurements confirmed that intrinsic cell-to-cell differences persisted throughout cycling. Overall, the results show that parallel battery testing provides information beyond conventional single-cell measurements and is an effective method for evaluating reproducibility and pack-relevant cell behavior in laboratory-scale batteries.

Introduction

Batteries are central to the modern energy transition, enabling the shift from fossil-fuel-dependent systems toward more sustainable transportation and electrical infrastructure. They power portable electronics, support electric vehicles (EVs), stabilize electrical grids, and provide backup energy for critical infrastructure. Among these applications, EV batteries impose especially demanding performance requirements. From an automotive perspective, EV battery systems are expected to deliver high specific and volumetric energy density, rapid charging, long service life, low cost, and high safety under a wide range of operating conditions. Widely cited U.S. Department of Energy and USABC targets for advanced EV batteries include cell-level values near 350 Wh/kg and 750 Wh/L, fast charging to 80% state of charge in about 15 min, long cycle life, and battery costs below 100 U.S. dollars/kWh. Achieving these metrics also requires robust thermal management, since temperature strongly influences power capability, degradation, charge acceptance, and safety. Commercial EV packs are therefore typically based on lithium-ion chemistries implemented in cylindrical, prismatic, or pouch formats and operated with battery management systems that maintain acceptable voltage and temperature windows during use. − ,

Because a single electrochemical cell cannot usually satisfy the voltage and energy demands of an EV on its own, practical battery systems are assembled from many cells connected in series and parallel. In these architectures, cell-to-cell variation becomes a pack-level problem rather than a purely materials-level issue. Small differences in capacity, internal resistance, self-discharge, reaction kinetics, interconnection resistance, or thermal history can propagate through the module and pack, altering current distribution, usable energy, local heating, and degradation rate. As a result, the electrochemical behavior of each individual cell directly influences the performance, reliability, and safety of the overall battery system, and even nominally identical cells may not behave identically once incorporated into a pack. ,,

The origin of lower-performing cells is multifactorial. Performance differences can arise from variations in active material loading, electrode thickness, porosity, tortuosity, calendering pressure, electrolyte filling and wetting, tab or weld resistance, and interfacial evolution during formation and cycling. In addition, lithium inventory loss, impedance growth, nonuniform solid electrolyte interphase development, and temperature gradients can progressively amplify initially small manufacturing differences. Thus, underperforming cells are not defined by a single failure mode, but instead may emerge from coupled mechanical, chemical, and transport nonuniformities that reduce accessible capacity or increase polarization relative to neighboring cells. −

In series-connected configurations, cell mismatch primarily manifests as voltage and capacity imbalance. Because the same current passes through every cell in series, the weakest cell constrains the usable capacity of the full string and reaches cutoff limits first. This behavior reduces pack energy utilization and can expose individual cells to overcharge or overdischarge conditions if balancing and control are inadequate. Such imbalance accelerates degradation, increases heat generation, and raises safety risk, which is why voltage monitoring and cell balancing are fundamental functions of battery management systems in EV packs. ,

In parallel-connected configurations, the dominant issue shifts from voltage addition to current sharing. Cells connected in parallel experience the same terminal voltage, but they do not necessarily carry equal current. Instead, branch current distribution depends on differences in internal resistance, open-circuit voltage, connection resistance, capacity, and temperature. Cells with lower resistance or more favorable electrochemical states can accept or deliver disproportionate current, while thermal gradients can further intensify divergence in current distribution and aging. Consequently, parallel modules may develop hidden heterogeneity even when the pack appears electrically stable at the external terminals. −

Conventional battery evaluation is still dominated by linear battery testing (LBT), in which individual cells are charged and discharged on independent channels while voltage and current are monitored with high precision. This approach is highly valuable for obtaining unbiased single-cell performance metrics, but it does not reproduce the electrochemical coupling that occurs when cells operate together under parallel electrical bias. By contrast, parallel differential battery testing (PDBT) makes it possible to observe current redistribution, self-balancing tendencies, branch-level deviations, and the statistical consistency of cells cycling together under shared voltage conditions. A further challenge in parallel testing is the accurate measurement of branch currents without significantly perturbing the system being measured, since added measurement resistance can itself distort current-sharing behavior. ,

These considerations are particularly important for laboratory-scale battery development, where handmade or newly developed cells may exhibit greater variability than mature commercial products. In such cases, conventional single-channel testing may not fully capture how differences in electrode quality, electrolyte distribution, or interfacial resistance translate into pack-relevant behavior. Therefore, evaluation of individual cells in a configuration that more closely reflects real battery pack operation is critically important. In this work, Arbin Instrument’s patented Parallel Technology cycler (PDBT) is used to evaluate small lab-scale coin cells in a parallel configuration and compare their behavior with conventional linear single-channel testing (LBT). The broader objective is to determine whether parallel differential battery testing can provide a more realistic and statistically informative method for assessing cell consistency and predicting challenges relevant to the scale-up of conventional Li-ion chemistries and emerging battery chemistries. ,,,

Experimental Section

Materials

Carbon-coated lithium iron phosphate powder containing 3 wt % carbon (LFP/C), lithium–metal discs (12 mm diameter and 0.5 mm thickness), Super P conductive carbon, polypropylene separators (25 μm thickness), and aluminum foil (18 μm thickness) were purchased from MSE Supplies (Tucson, AZ, USA). Poly­(vinylidene fluoride) (PVDF; average molecular weight, M_w , approximately 530,000 g mol^–1) was purchased from Sigma-Aldrich. A lithium hexafluorophosphate (LiPF₆) 1M solution in ethylene carbonate and diethyl carbonate (EC:DEC, 1:1 by volume) and N-methyl-2-pyrrolidone (NMP) were purchased from Sigma-Aldrich. All materials were used as received without further purification.

Electrode Preparation and Characterization

LFP/C, Super P, and PVDF were combined in a mass ratio of 90:5:5. PVDF was first dissolved in NMP, after which Super P was dispersed in the PVDF/NMP solution. LFP/C powder was then added, and the mixture was ball-milled at 400 rpm for 1 h to obtain a homogeneous slurry containing 40 wt % solids. The slurry was cast onto 18 μm-thick aluminum foil using a laboratory film coater operated at a coating speed of 5 mm s^–1 with a 100 μm doctor-blade gap. The coated electrodes were initially dried in an oven at 120 °C, punched into 12 mm-diameter discs, and subsequently vacuum-dried at 120 °C for 24 h. The resulting LFP active-material loading was approximately 4–5 mg cm^–2.

The surface and cross-sectional morphology and elemental composition of the cast LFP cathodes were examined using a Helios 5 CX DualBeam high-resolution scanning electron microscope (Thermo Fisher Scientific) equipped with an energy-dispersive X-ray spectroscopy detector.

Electrochemical Cell Assembly and Testing

CR2032 coin cells were assembled in an argon-filled glovebox maintained at O₂ and H₂O concentrations below 0.5 ppm. Each cell consisted of an LFP cathode, a lithium–metal counter/reference electrode, a 25 μm-thick polypropylene separator, and 75 μL of the LiPF₆-based liquid electrolyte. Immediately before cell assembly, the lithium–metal surface was gently cleaned using a nylon brush to remove the visibly passivated surface layer.

Electrochemical impedance spectroscopy measurements were performed using a Solartron 1260 impedance analyzer over a frequency range of 10 MHz to 10 mHz with a sinusoidal perturbation amplitude of 10 mV. Galvanostatic charge–discharge measurements were conducted using Arbin parallel differential battery testing and linear battery testing cyclers over a voltage range of 2.5–4.2 V versus Li/Li⁺. All electrochemical measurements were conducted at room temperature, maintained at approximately 23 °C.

Results and Discussion

The objective of this work was to evaluate parallel battery testing as a tool for assessing both the reproducibility of laboratory-fabricated cells and the behavior of individual cells when electrically coupled under shared-voltage conditions. Eight CR2032 Li||LFP half-cells were prepared using cast LFP/C–Super P–PVDF cathodes, polypropylene separators, liquid electrolyte, and lithium–metal counter/reference electrodes. Because electrode fabrication and coin-cell assembly were performed manually, variability could arise from multiple components and processing steps, including cathode composition and loading, electrolyte distribution and wetting, lithium–metal surface condition, electrode alignment, and interfacial contact. The SEM/EDS results were therefore used to identify cathode-level heterogeneity as one plausible contributor to the electrochemical dispersion rather than as evidence that the cathode was the sole source of variation.

Figure establishes the structural basis for the electrochemical results. In the surface-view SEM image in Figure a, the cathode surface appears continuously and fully covered by fine particulate material, indicating that the slurry-cast electrode formed a coherent composite. The associated elemental maps in Figure a show that Fe, P, and O are broadly distributed across the surface, consistent with uniform coverage of the active LFP phase. The carbon signal is also present throughout the cathode, confirming that the conductive additive is distributed across the coating. However, carbon is not perfectly homogeneous and appears partially segregated into larger agglomerated regions as seen at higher magnification. This feature is important because it suggests that, although the conductive carbon is broadly incorporated throughout the cathode, the electronic network is likely not spatially uniform. Regions enriched in carbon may provide more efficient local conduction pathways, whereas neighboring LFP-rich regions may experience less favorable electronic connectivity and more uneven active-material utilization.

1.

Surface and cross-sectional morphology of the hand-cast LiFePO₄ (LFP) cathodes with corresponding elemental maps, (a) shows the surface-view SEM image of the LFP/Super P/PVDF cathode coating, together with EDS maps for Fe, P, O, C, and F, (b) shows the cross-sectional SEM image of the same cathode on the Al current collector, together with EDS maps for Fe, C, and F.

The carbon agglomeration observed in Figure a may also contribute to mass-loading deviations from cathode to cathode. Because the conductive additive occupies both volume and mass within the composite coating, nonuniform carbon-rich regions can lead to slight differences in how much active LFP is effectively incorporated into each individually punched or assembled electrode, even when the nominal slurry composition is unchanged. In practice, this means that two manually cast cathodes produced from the same slurry may still contain slightly different amounts of electrochemically active material. Such differences would be expected to contribute directly to the cell-to-cell variation in performance.

The cross-sectional SEM and EDS maps in Figure b further show that the composite cathode is continuous through its thickness and remains clearly separated from the Al current collector. Fe, P, O, C, and F are all distributed throughout the electrode layer, indicating that the LFP, Super P, and PVDF phases are incorporated across the full coating thickness, while the Al signal remains localized in the substrate. Taken together, Figure indicates that the hand-cast cathodes are compositionally integrated and structurally continuous but still contain the microscale heterogeneity expected from manual slurry mixing and casting. These non-idealities likely include local differences in connectivity, porosity, and effective active-material loading, all of which provide a plausible microstructural origin for the electrochemical variability observed across the eight cells.

The effect of this fabrication-level heterogeneity becomes evident in the rate capability results shown in Figure a. As expected for Li||LFP half-cells, the specific capacity decreases as the cycling rate increases from 0.1C to 1C and then recovers when the cells are returned to lower c-rates. This behavior confirms that all eight cells remain electrochemically active throughout testing and that the primary capacity loss at higher rates is kinetic in origin rather than due to immediate irreversible degradation. At the same time, Figure a clearly shows that the cells do not behave identically. Most channels cluster within a relatively narrow capacity band, but certain channels consistently underperform across the full rate sequence (Sub Chan. 5 and Sub Chan. 7), and the gap between the highest- and lowest-performing cells becomes more apparent as the rate increases. This indicates that the observed variability is not simply random cycle-to-cycle scatter but rather reflects persistent electrochemical differences among the hand-fabricated cells. While parallel and linear battery testing show very similar trends, there is a noticeable change in the gap between the better performing cells (1, 2, 3, 4, 6, 8) and the worst performing cells (5, 7). The observed capacity dispersion likely reflects contributions from several components of the manually assembled cells. Differences in cathode composition, local carbon distribution, electronic connectivity, porosity, and active-material loading may produce variations in accessible capacity and rate capability. However, the lithium–metal counter/reference electrode may also contribute to the observed differences. Variations in lithium surface passivation, roughness, interfacial contact, electrolyte wetting, and solid-electrolyte interphase formation can affect polarization and impedance, particularly at higher C-rates. Because the lithium–metal surfaces were not independently characterized, the relative contributions of the cathode and lithium electrodes cannot be quantitatively separated in the present study. , The measured dispersion should therefore be interpreted as complete-cell variability rather than being attributed exclusively to the LFP cathodes. In practical terms, this means that mild operating conditions can partially mask manufacturing variability, whereas more demanding conditions expose it more clearly. Thus, Figure a,b show that parallel battery testing can be used not only to increase experimental throughput, but also to quantify reproducibility across a group of nominally identical research cells.

2.

Electrochemical performance of eight hand-assembled Li||LFP coin cells tested in parallel and linear. (a) Specific capacity of Cells 1–8 during stepped-rate cycling at 0.1C, 0.2C, 0.5C, and 1C, followed by return to lower rates on PDBT. (b) Specific capacity of Cells 1–8 during stepped rate cycling on LBT. (c) Direct comparison of Cells 1–8 on PDBT and LBT with solid bars representing the true range of specific capacity and the faded region representing the average ± one standard deviation. (d) Coulombic efficiency of Cells 1–8 compared between PDBT and LBT.

The worst cells in terms of specific capacity (Cells 5 and 7) performed better during linear testing than during parallel testing and the better cells (Cells 1, 2, 3, 4, 6, and 8) performed worse during linear testing. Between Figure a and Figure b, we can see that cells that perform well in a parallel configuration may not necessarily perform as well on linear configuration and vice versa. To better visualize comparison between PDBT and LBT, Figure c utilizes a gain-loss chart where the solid bar represents the full range of values, and a faded region is centered on the overall average and extends to plus and minus one standard deviation. This is performed for both specific discharge capacity and coulombic efficiency. Parallel testing showed a greater range of values and a greater standard deviation with an average that was generally displaced toward the top of the range. The average capacity was overall about 10% higher when tested in parallel. Linear battery testing had smaller ranges of specific capacity with averages centered in the ranges. Coulombic efficiency was found by the quotient of discharge capacity by charge capacity for each cycle. Linear battery testing had notably higher coulombic efficiency which is expected as there will be no losses associated with equalization during rest steps.

Current measurements show a major difference between parallel and linear battery testing during cycling as shown in Figure . For linear testing, the only notable difference between cells is the time it takes to complete a cycle where better performing cells will accumulate charge for longer. Parallel testing provides a more comprehensive view of current imbalances and provides another metric to compare cells. Better performing cells are indicated by higher peak currents, particularly at higher C-rates. While peak currents for each cell tend to occur at about the same time near the end of charging or discharging steps, there is some displacement of this peak for different cells, particularly at low C-rates. The lowest capacity cell (Cell 7) shows a pattern of drastic current increase at the end of charging and discharging steps, particularly at lower C-rates. A cell that is modestly weaker in single-cell capacity can become substantially less involved in current carrying when electrically coupled to stronger neighboring cells. Conversely, the strongest cells can become overburdened because they carry a larger share of the total current. Therefore, Figure a–d provides information that cannot be extracted from capacity measurements alone and demonstrates why parallel battery testing is valuable not just for throughput, but for probing pack-relevant behavior in research-scale cells.

3.

Current sharing behavior of the eight Li||LFP cells during parallel and linear operation at different applied rates. Current traces for Cells 1–8 are shown during charge, rest, and discharge at (a) PDBT 0.1C, (b) LBT 0.1C, (c) PDBT 1C, and (d) LBT 1C.

The magnitude and shape of current redistribution during parallel testing are expected to depend on the voltage profile of the electrode material. LFP exhibits an extended and relatively flat voltage plateau, allowing parallel-connected cells at different states of charge to maintain similar terminal voltages over much of the charge and discharge process. As a lower-capacity or more highly polarized cell approaches the end of charge or discharge, its current contribution can decrease while the remaining cells carry a larger fraction of the total applied current. Materials with continuously sloping voltage profiles may exhibit smoother and less pronounced current redistribution because changes in state of charge produce more continuous changes in equilibrium potential. Nevertheless, PDBT remains applicable to these materials because differences in capacity, impedance, open-circuit voltage, temperature, and connection resistance can still produce unequal branch currents. The current-sharing response observed during PDBT should therefore be considered dependent on the electrode chemistry rather than limited exclusively to materials with voltage plateaus.

The present study was designed to evaluate cell-to-cell variability during short-duration rate-capability testing and does not establish whether the observed current peaks affect long-term cycle life. Repeated unequal current loading could potentially increase the polarization and electrochemical stress experienced by the cells carrying the largest fraction of the total current, which may cause aging rates to diverge during extended cycling. Confirming this effect would require long-term parallel cycling with periodic capacity, impedance, and, where possible, temperature measurements. PDBT and LBT should therefore be considered complementary techniques for cyclability studies. LBT provides an independently controlled baseline for intrinsic single-cell degradation, whereas PDBT reveals current redistribution, equalization, and coupled aging during parallel operation.

The electrochemical origin of these differences is further supported by the impedance spectrum in Figure , which compares the same eight cells before cycling, after parallel battery testing, and after subsequent linear battery testing. Before cycling, the cells exhibit variation in semicircle size and position, indicating that the initial interfacial and charge-transfer impedances were not identical across the hand-assembled coin cells. This initial spread is consistent with the microstructural and compositional heterogeneity, including carbon agglomeration, local differences in conductive connectivity, and small cathode-to-cathode deviations in effective active-material loading, as well as differences across the lithium metal anodes. After parallel testing, the charge-transfer impedance of all cells except Cell 5 decreased relative to pristine condition, indicating improved wetting and partial conditioning of the electrode and interfacial regions during early cycling. Following the subsequent linear tester measurements, many cells maintain similarly reduced or further stabilized impedance, consistent with the preserved low-rate capacities and rate-recovery behavior observed in Figure . At the same time, the spectra do not converge to a single common response; instead, measurable cell-to-cell differences remain after parallel and linear testing. This is an important result because it shows that electrochemical conditioning reduces the absolute impedance of many cells but does not eliminate the intrinsic differences introduced during fabrication and assembly. In turn, those residual impedance differences help explain why some cells consistently deliver lower capacity in Figure , and why those same weaker cells are less able to compete for current during the coupled operation shown in Figure . Thus, Figure reinforces the broader conclusion of the study: the unequal current sharing observed during parallel testing is not an artifact of the test configuration, but rather the pack-level manifestation of real cell-to-cell differences in impedance, polarization behavior, and accessible capacity.

4.

Electrochemical impedance spectra of the eight Li||LFP cells before cycling, after parallel battery testing, and after subsequent linear battery testing. Nyquist plots are shown for Cell 1–8.

To further compare the performance of the individual cells, the energy deviation during each cycling step was evaluated, as shown in Figure . The energy transferred by each cell was estimated by integrating the measured power over time using Simpson’s rule. The energy deviation was then calculated as the difference between the energy transferred by an individual cell and the average energy transferred by all eight cells during the same step. Values obtained from repeated cycles at the same C-rate were averaged to consolidate the data. Figures a and b compare the energy deviations during discharge for PDBT and LBT, respectively. Cells with more negative energy-deviation values delivered more energy than the group average, whereas cells with more positive values delivered less energy. The relative performance trends were generally consistent between the two testing methods, with the lower-capacity cells delivering less energy than the better-performing cells. However, the magnitude of the energy deviation was greater during PDBT because the current was free to redistribute among the parallel-connected cells. By contrast, LBT subjected each cell to an independently controlled current, resulting in a narrower distribution of discharge energy. Figure c shows the energy deviation during the post-discharge rest period in PDBT. Although the externally applied pack current was zero during this step, the cells remained electrically connected and exchanged current as they equalized toward a common terminal voltage. Consequently, some cells received energy while others supplied energy, producing positive and negative energy deviations respectively among the individual branches. The magnitude of this equalization behavior generally increased with C-rate, indicating that greater performance differences developed among the cells under more demanding operating conditions. A corresponding LBT post-discharge-rest panel was not included because the cells were electrically isolated during LBT and no current sharing or energy exchange occurred during the rest period.

5.

Energy deviation of each cell relative to the eight-cell average during (a) PDBT discharge, (b) LBT discharge, and (c) PDBT post-discharge rest at the investigated C-rates.

The eight-cell configuration used in this study was sufficient to identify persistent differences in capacity, impedance, energy transfer, and current-sharing behavior among nominally identical laboratory-fabricated cells. However, eight cells should not be considered a universal minimum for either PDBT or LBT experiments. The number of replicate cells required depends on the magnitude of the cell-to-cell variability, the electrochemical parameter being evaluated, and the statistical confidence required for the study. PDBT provides an efficient means of comparing multiple cells under the same shared-voltage condition, but it does not eliminate the need for independent cell replicates. To clearly communicate reproducibility, battery studies should report the number of independently assembled cells together with the individual-cell results, mean, standard deviation, and full measured range. When the objective is to establish a population-level confidence interval or detect a predefined performance difference, the required number of cells should be determined using an appropriate statistical power or convergence analysis.

Conclusions

This work demonstrates that parallel battery testing is an effective tool for evaluating both the reproducibility of laboratory-made cells and the practical consequences of cell-to-cell variation during coupled operations. Although the eight Li||LFP coin cells were fabricated from the same nominal materials and procedures, SEM/EDS and electrochemical data showed that small differences in cathode structure and assembly produced measurable variation in capacity, rate performance, current sharing, and impedance. These differences became more pronounced at higher C-rates, where weaker cells contributed less effectively and stronger cells carried a greater fraction of the load. The results also show that parallel testing provides insight beyond conventional linear single-cell measurements. In addition to screening whether a group of cells was fabricated consistently, it reveals how intrinsic cell differences influence current redistribution, equalization behavior, and imbalance when cells operate together in a pack-like configuration. This makes parallel battery testing especially valuable in research settings, where hand-fabricated cells often contain unavoidable variability, and supports its use both for validating lab-scale cell construction and for better understanding how cell-to-cell differences affect overall battery pack behavior.

The authors declare no competing financial interest.