High-throughput Li plating quantification for fast-charging battery design

Fast charging of most commercial lithium-ion batteries is limited due to fear of lithium plating on the graphite anode, which is difficult to detect and poses considerable safety risk. Here we demonstrate the power of simple, accessible and high-throughput cycling techniques to quantify irreversible Li plating spanning data from over 200 cells. We first observe the effects of energy density, charge rate, temperature and state of charge on lithium plating, use the results to refine a mature physics-based electrochemical model and provide an interpretable empirical equation for predicting the plating onset state of charge. We then explore the reversibility of lithium plating and its connection to electrolyte design for preventing irreversible Li accumulation. Finally, we design a method to quantify in situ Li plating for commercially relevant graphite|LiNi0.5Mn0.3Co0.2O2 (NMC) cells and compare with results from the experimentally convenient Li|graphite configuration. The hypotheses and abundant data herein were generated primarily with equipment universal to the battery researcher, encouraging further development of innovative testing methods and data processing that enable rapid battery engineering. Lithium-ion batteries are prone to unpredictable failure during fast charging, known as lithium plating. Now, innovative testing protocols can quickly quantify lithium plating and inform battery design strategies to mitigate it.

The urgent need to combat climate change has sparked extreme growth in demand for lithium-ion batteries (LIB). Rapid innovation in battery materials and cell design is critical to meet this demand for diverse applications, from electronics to vehicles and utility-scale energy storage. Composite graphite electrodes remain a universal component of LIB and are expected to dominate anode market share through to 2030 despite the introduction of silicon and lithium-based materials 1 .
The design space for graphite electrodes is immense, with parameters such as loading, porosity, particle size, binder composition and electrolyte being carefully selected to meet requirements for the lifetime, operating temperature, charge time and manufacturing. Regardless of design and application, the lithium plating reaction on graphite is a performance and safety concern due to the formation of non-cyclable 'dead' lithium metal and salts. While recent studies have focused on Li plating during fast charging, the phenomenon is also pertinent to other operating extremes, such as low temperature 2 , overcharge 3 or system malfunction 4 .
Electrochemical (EChem) modelling is an important tool for understanding design trade-offs that improve graphite performance while avoiding plating. Over decades, Newman-based models that relate cell current density, voltage, temperature and material properties to graphite intercalation have been enhanced to also estimate lithium plating [5][6][7][8][9][10] . This has led to initial insight into the effect of charge rate, electrode loading and temperature on lithium plating amount, but simulations rely on debated parameters, such as the plating exchange current density or reversibility, and are frequently not verified with direct experimental measurements 11 , such as Li gas evolution titrations 12,13 . EChem models also have limited ability to Article https://doi.org/10.1038/s41560-023-01194-y decreasing cell voltage in the Li|Gr half-cell configuration. We previously demonstrated this 'SOC-sweep' approach to study plating with differential voltage analysis 22 , and a similar stepwise capacity cycling has been used for Gr|cathode full-cells 23 . Here, however, we first focus on half-cells due to the stable potential of the Li counter electrode and the desire to isolate graphite anode degradation effects.
The CE for each of the cycles is shown versus charge capacity in Fig. 1b. To estimate the irreversible Li, a high-efficiency baseline CE (dashed line) is first assigned to the data points at low SOC, where we attribute the non-unity values to continued SEI formation or slow cell degradation processes rather than to Li plating. CE data are then subtracted from these baselines, ranging from 99.85 to 99.98% (Supplementary Figs. 1-3 and Supplementary Table 1), to yield a Coulombic inefficiency (CIE) from Li plating-related degradation. The CIE multiplied by the SOC for each cycle gives irreversible plating capacities as a percentage of each cell's experimental capacity, which is shown in Fig. 1c for various rates, with the result for each cell represented by a set of connected data points. Throughout this work, we point out ways that cycling data, modelling and titrations further confirm the reliability of CIE for irreversible Li plating quantification. To start, Fig. 1c Fig. 6 voltage relaxation data further supports the link between CIE and irreversible plating.
Increasing the charge temperature is a well-known operating control to avoid lithium plating, but, to the best of our knowledge, no work has simultaneously quantified the effects across charge rates (C-rates), loadings and SOC, all of which are relevant for battery design. the averaging process is illustrated by comparing Fig. 1c and Fig. 2a. The technique fidelity is supported by experimental trends that are universally consistent with the expectation that the starting SOC of lithium plating should be postponed with decreasing current rates (left to right within panels), decreasing loadings (left to right across panels) and increasing temperature (top to bottom), as seen by the shifting irreversible Li curves in the x direction. predict the chemical compatibility and interphasial properties for novel electrolytes. High-throughput modelling advances for battery materials and interfaces could fill this void, but they too lack commensurate validation 14 .
Challenges to high-throughput battery testing can include limited access to expensive equipment, slow multiweek cycling tests, limited material availability, high labour costs of cell assembly, complex analysis methods and inefficient data handling. There are promising solutions to some of these problems. To conserve newly synthesized electrode materials, it is common practice to determine charge rate capabilities by testing multiple rates on a single cell 15,16 . High-precision coulometers have been developed to improve early performance prediction 17,18 . Data-driven models that predict cycle life from minimal data 19 can be used to quickly optimize charge protocols 20 , although large datasets are difficult to obtain in most laboratory settings 21 .
Here we demonstrate the power of simple, quantitative and accessible cycling protocols to inform battery design for Li plating-free charging. The trade-offs between energy density, charge rate, charge temperature and lithium plating are experimentally quantified and used to refine a mature electrochemical model. We then explore the reversibility of lithium plating under varied fast-charging conditions, and apply our understanding towards development of electrolytes and interfaces that limit dead Li formation. We emphasize that the hypotheses and abundant data presented herein were generated primarily with equipment universal to the battery researcher, enabled by strategic data handling, while the sophisticated modelling and titration techniques were reserved for secondary support of the findings.

Irreversible Li mapping and modelling
Past independent titration studies of Li plating on copper 12 and graphite 13 show a strong positive correlation between Coulombic inefficiency and inactive Li 0 , with the majority of the irreversible plating capacity being attributed to H 2 -evolving dead Li species (Li 0 , Li x C 6 ) for liquid carbonate-based electrolytes. This observation, combined with the high-throughput, precise nature of Li|graphite (Li|Gr) cell Coulombic efficiency (CE) measurements, motivated the protocol shown in Fig. 1 to estimate irreversible Li plating as a function of charge length. We define irreversible Li as the sum of irreversibly formed species during Li plating, such as isolated metallic lithium and Li + -containing solidelectrolyte interphase (SEI). After formation cycling (Methods), the 4C charge capacity is increased stepwise by 5% state of charge (SOC), or normalized graphite capacity for each cycle, from 10% to 55% SOC (Fig. 1a) Fig. 2a-f. The lithium plating reaction is modelled using the formulation proposed by Ren et al. 8 , with a plating exchange current density of 10 A m −2 and fixed plating reversibility of 70%, both estimated using titrations and voltage profiles from an experiment with similar Li plating conditions (electrode, charge rate, electrolyte and Li plating capacities) 13 to the present study. Specific parameters for these electrodes and electrolyte transport properties have been extensively reported and are given in Supplementary Table 3, along with experimental and modelled voltage profiles in Supplementary Figs. 7 and 8. The excellent qualitative agreement in irreversible Li curve shape between model and experiment increases confidence in the model plating reversibility (Supplementary Figs. 9 and 10), the exponential Butler-Volmer kinetic expression used for lithium plating/stripping and the assumption that experimental capacity loss is mostly due to irreversible Li instead of other slower degradation processes.
From this dataset we extract the SOC at which irreversible Li starts to form, or the 'plating onset', as a metric to inform safe charge durations and assess the quality of our experiment-model agreement. Here we define the plating onset threshold as 0.05% irreversible Li, or 1.0-1.5 μAh cm −2 for the respective electrode loadings, which is represented by the horizontal lines in Fig. 2a-f. This is the lowest value after which clear plating increases are observed, and also avoids uncertainty from experimental noise at low SOC (Fig. 2d, low SOC). The SOCs at which plating begins for all 20 conditions are shown in Fig. 2g. Reasonably linear relationships between onset and C-rate are observed at a given temperature and loading. Additionally, experiment uniquely shows that temperature has nearly double the effect on plating onsets for the higher loading electrodes than for the lower loading electrodes (3.1 versus 2.1 mAh cm −2 ), as indicated by the larger vertical shift in the curves. Physically, this could mean that for thin electrodes with onsets above 50% SOC, the accumulation of bulk Li 1 C 6 and its low open-circuit potential throughout the electrode promote lithium deposition, regardless of improved Li transport or intercalation kinetics at higher temperatures. In the thicker electrodes with plating at low SOC, the strong temperature effect suggests that porous electrolyte Li + transport determines Li plating by controlling the uniformity of graphite lithiation and therefore the SOC at which Li 1 C 6 forms at the graphite|separator interface 24 . These explanations are consistent with optical microscopy that shows Li plating first appears on top of gold-coloured Li 1 C 6 particles 27 .
In general, the EChem model (Fig. 2g, dashed lines) accurately captures the onset of lithium plating with less than 5% SOC error. Model predictions matched experiment best by slightly modifying graphite properties from those previously reported 9,24 , such as lowering the activation energy for solid-state diffusion from 30 kJ mol −1 to 15 kJ mol −1 ( Supplementary Fig. 11). We believe this indicates a need to explicitly determine the diffusion coefficient as a function of lithiation and temperature. At high temperatures, loadings and charge rates (Fig. 2g, light blue data), the model predicts lithium plating 5-10% SOC earlier than measured, and for low loadings (Fig. 2g, 2.1 mAh cm −2 ) the model predicts larger temperature sensitivity than measured. This could be related to changes in SEI composition/resistivity with elevated temperatures, complex graphite phase behaviour or diffusion     Fig. 12). Additionally, the experiment uniquely shows that the higher loading electrode tends to promote faster accumulation of irreversible Li, which is probably due to higher local current densities near the separator that promote faster, more dendritic (and thus more irreversible) Li plating. The low graphite lithiation (SOC) at these onsets could also promote more rapid Li metal dissolution, which supports Li + re-intercalation into the graphite after charge 29,30 , making the remaining Li deposits more susceptible to electrical isolation. Given that the plating onset varies almost linearly with changes in other variables, we propose an empirical equation to calculate the plating onset, separate from the physics-derived EChem model, as a step towards data-driven Li plating models. The plating onset SOC, y, is written as a linear function of the C-rate (c), loading (x) and temperature (T), with coefficients α, β and γ * respectively, and intercept ε (equation (1)). The (1 − y) correction for γ within γ * was added to account for the variable temperature effect with loading in Fig. 2g, noting that T has a smaller impact for plating onsets at higher onset SOC. Rearrangement to solve for y yields equation (2). Applying the empirical fitting to the 20 (y, c, x, T) plating onset pairs with four parameters unsurprisingly gives a much-improved onset prediction compared to the Newman model (Fig. 2h), and nearly 60% reduction in the residual sum of squared errors (SSE). Table 1 highlights the benefits for interpreting the data using an analytically differentiable equation, which can provide heuristics for how Li plating should vary with design parameter changes. For example, starting from 30 °C, 3.1 mAh cm −2 and 4C rate, a 1C rate increase would cause a 9% SOC earlier plating onset and a 1 °C increase would postpone the onset 0.7% SOC. This analysis complements recent work that found a linear correlation between the plating onset and electrode ionic resistance, elucidating the effects of electrode structure and loading 31 . We also investigated the model's predictive capabilities by studying a graphite electrode with identical composition but 3.75 mAh cm −2 loading, well above the previous experimental range, and observe that it impressively predicts the plating onset within 4% SOC at moderate rates and temperatures ( Supplementary Fig. 13). Finally, the equation is useful for visualizing battery design trade-offs, and Supplementary  Fig. 14a shows the charging temperature required to avoid plating for a constant current (CC) charge to 40% SOC for various combinations of rates and loadings. Additional visualizations of the empirical fitting and a discussion of its limitations are in Supplementary Fig. 14b-d and Supplementary Note 1.

Electrolyte discovery to reduce irreversible Li
Lithium plating is harmful because the reaction is poorly reversible, which causes loss of cell lithium inventory, capacity fade and accumulation of reactive metallic lithium. While the impact of electrolyte on reversibility is at the forefront of Li metal battery research 32 , few have considered electrolyte engineering as a plating control strategy for graphite anodes under fast charging. If the reversibility of plating could be improved from 70% to 90%, for example, then the amount of irreversible plating would be decreased by a factor of 3 (30% to 10%), drastically reducing the impact on performance and safety. In this section, we quantify irreversible Li for different electrolytes using the SOC-sweep of Figs. 1 and 2, demonstrate a rigorous method to estimate plating reversibility on graphite and argue that plating reversibility is an important electrolyte design criterion for fast charging. Figure 3a shows the effect on irreversible plating of swapping ethylene carbonate (EC) for varied weight percent (wt%) fluoroethylene carbonate (FEC). These compositions were inspired by Li metal battery studies that repeatedly show FEC can decrease dead Li formation 12,33 . The notable shift in the curve from 0% to 5% FEC indicates a delayed onset of lithium plating, and the decreasing slopes with increasing FEC suggest a beneficial concentration effect for reducing dead Li. This observation arises despite decreasing bulk electrolyte conductivity with increasing FEC ( Supplementary Fig. 15), which led us to hypothesize that enhanced interfacial properties or fractional plating reversibility may alternatively explain this result.
To explore whether Li plating reversibility plays a role in improved performance with FEC, we sought a rigorous high-throughput method to quantify the value at conditions relevant to fast charging. The estimation of plating reversibility on graphite at standard SOC (below 100%) and ambient temperatures is challenging due to the rapid dissolution of reversible Li deposits, which supports Li + re-intercalation into the graphite 27,29,30 . A workaround to this is to study plating during graphite overcharge (above 100% SOC) 34,35 , which has also emerged in the context of hybrid graphite/lithium anodes [36][37][38] , but reversibility estimates have only been reported at low current rates (<0.5C) and/or are deduced from qualitative voltage plateau transitions.
The framework we apply to carefully estimate the reversibility of Li plating (η) on graphite during fast charge is summarized in Fig. 3b. After formation cycling, the first step is to estimate the Coulombic efficiency for graphite intercalation (CE int ) without lithium plating (Fig. 3b, 'baseline cycle', value ~99.7%; Supplementary Table 4 and Supplementary  Fig. 16). Next, that same cycle is repeated with an added overcharge step to induce a known capacity of lithium plating, P. The capacity lost due to lithium plating is then isolated by subtracting the baseline capacity loss of intercalation from the total irreversible capacity from the overcharge cycle (Q irrev ; Fig. 3b), which allows the calculation of η from equation (3). Repeating the overcharge cycle four times on the same cell gives reproducible calculated reversibility values for the first three overcharge cycles, increasing confidence in the method and allowing error bar estimation with a single cell ( Supplementary Fig. 17).
(3) Figure 3c shows the calculated lithium plating reversibility for various FEC-containing electrolytes when the overcharge amount is varied at a fixed 4C rate (left) and the deposition rate is varied at 20% overcharge (right). The plating overcharge amount is defined as the percentage of total graphite capacity (here, 3.1 mAh cm −2 ) that the electrode is charged beyond complete lithiation. For all conditions, FEC-free electrolyte exhibits the lowest η, ranging between 74 and 91%, and for all electrolytes the expected trends of decreasing η with increasing plating amount and rate are apparent. The beneficial concentration effect in FEC-containing electrolytes from Fig. 3a is again observed, with the exception of low-rate or low-amount conditions, as circled in Fig. 3c. We ascribe this observation to plating occurring SOC/1 ºC a Partial derivatives of equation (2) are evaluated at 30 °C, 3.0 mAh cm −2 and 4C with fitted parameters reported in Fig. 2h caption.
Article https://doi.org/10.1038/s41560-023-01194-y primarily beneath the graphite SEI 39 , which we believe has a similar composition across concentrations due to overlapping differential capacity curves during the first graphite intercalation (Fig. 3d), when the majority of SEI is formed. Finally, we try to connect these η values determined from overcharge experiments to the true η range observed during fast charging. In the latter, plated lithium is observed within micrometres of the graphite/separator interface 25,40 due to concentration and potential gradients, but the overcharge protocol differs because it begins without gradients and thus should initially yield more uniform Li deposition, as imaged at low rates 37 . Consequently, Fig. 3e is a sketch of how lithium plating may accumulate during 4C overcharge, a hypothesis consistent with intuition about gradient development, effective porosity decreasing as Li deposits grow and the observed decrease in η with plating amount, as these effects lead to higher local current densities and non-uniform deposits near the separator interface. To better understand the effect of location on η, an incremental plating reversibility Δη for each subsequent 10% of plating is calculated directly from data in Fig. 3c (Methods) and is shown in Fig. 3f. An interesting feature arising from this analysis is that for 30% overcharge, the reversibility for the final segment of plating Δη 20-30 is drastically lower for the 0 and 5% FEC electrolytes, suggesting that the η for the 10% overcharge experiment, equivalent to Δη 0-10 , is artificially high due to uniform plating deposition throughout the electrode. The 10-15% FEC samples, in comparison, show less performance decline with plating amount, perhaps due to bulk electrolyte effects, such as enhanced Li + solvation by FEC 41 . As depicted by the diagram in Fig. 3e, this last plating segment may occur in a planar manner after protruding through the graphite SEI, with growth constrained by the separator. Thus, we might expect comparable reversibility for Li plating on a planar substrate, such as copper foil, and indeed similar trends are observed using identical plating amounts and current densities ( Despite the illustrative range of possible plating reversibilities, it remains unclear which is most representative of plating under standard charging conditions, that is, which can best predict irreversible Li with models or quantify electrolyte improvements. Leveraging our comprehensive experimental and modelling datasets for the 0% FEC electrolyte of Fig. 2, we determine the single η that minimizes the plating onsets error across all conditions ( Fig. 3g and Supplementary Figs. 9 and 10). The η from this analysis is 80%, but, most importantly, the SSE divergence above 90% provides strong evidence that the plating reversibility does not exceed this value in practice, highlighting the need for careful interpretation of overcharge plating data. Looking at the 0% FEC data in Fig. 3f, η = 80% is between the values for Δη 0-10 and Δη [10][11][12][13][14][15][16][17][18][19][20] Fig. 17). g, The sum of squares error (SSE) for Li plating onsets between experiment (exp) and the electrochemical model (Fig. 2g) across all conditions versus the plating reversibility assumed by the model.  Fig. 3b to be useful for characterizing additional electrolytes and assessing innovative methods to mitigate irreversible plating, such as separator design 42 .

Full-cells Li plating quantification
This section first shows that previous half-cell plating onset and electrolyte studies are valuable for informing commercial full-cell design, which instead use a porous, high-voltage cathode material with limited lithium inventory. Next, ex situ titrations are used to verify Li plating and identify cycling data features from full-cells that are quantitative predictors of plating. Finally, the insights are applied to design a validated, highly sensitive, in situ method for Li plating quantification.
To compare lithium plating behaviour across electrolyte compositions, a graphite|LiNi 0.5 Mn 0.3 Co 0.2 O 2 (NMC532) cell was cycled 140 times, alternating five moderate 1C constant-current constant-voltage (CCCV) charging cycles to 4.2 V, holding until C/5 current, with two 6C CCCV fast-charging cycles to 4.2 V, holding until 80% capacity. We selected this protocol to help isolate fast charging-related capacity loss, which was expected only during the 6C cycles, from other cell aging effects, such as FEC degradation 43 . Figure 4a shows that the 5-15% FEC full-cells, similarly to half-cells, outperform the FEC-free electrolyte, undergoing on average only about 30% of the capacity fade over the 100 1C cycles, with similar 6C CCCV charge times compared to 0% FEC (Supplementary Figs. 19 and 20). The 2% FEC electrolyte, included because of the common use of FEC as an additive, performed only slightly worse than the higher concentrations. From our Li reversibility analysis, the lack of a clear concentration effect on performance may indicate small amounts of plating occurring, mostly beneath the FEC-derived SEI. It may also indicate the importance of the SEI in delaying the plating onset SOC, which is seen in Fig. 3a and has been suggested by others 44 to explain the better rate performance with an artificial graphite SEI coating. Considerable sample variability is expected due to the heterogeneous nature of lithium plating, and is depicted by the representative error bars obtained from replicate trials on multiple cells.
We then use the cycling data to quantify degradation from fast charging and compare the results with Li titrations of the extracted electrodes. Others have reported that irreversibly plated lithium is linearly correlated to cell capacity loss 45 , so we expect the abrupt capacity changes after the 6C cycles (Fig. 4a, box) to correlate with titrated Li capacity from mass spectrometry titration (MST). MST 13 accurately quantifies the combined H 2 -evolving species on graphite, such as isolated 'dead' Li 0 and inactive Li x C 6 , with exceptional resolution (Methods and Supplementary Figs. 21-24). However, the titrated Li slightly exceeds the capacity loss for most of the 0% FEC samples despite controls that show minimal Li x C 6 contribution, suggesting that plating is not fully quantified by this metric (Supplementary Figs. 25 and 26). The source of this error may be visualized in the 1C charging profiles for a representative cell shown in Fig. 4b, recalling that 1C cycle 5 is followed by 6C cycles 1 and 2 then 1C cycle 6, and that 1C cycle 10 is followed by 6C cycles 3 and 4, and so forth. The profiles show that for the first few fast-charging cycles, the voltage segment corresponding to early graphite lithiation shifts to the right (dashed box). Physically, the shift indicates a change in the electrode potential windows during charge and the removal of additional cyclable lithium 46 from the graphite to compensate lithium losses from plating. Thus, we believe this graphite SOC shift should estimate losses not captured by the 1C capacity loss, which conversely manifests by the high-voltage capacity shifting to the left (solid box).
The SOC shift (ΔX) and capacity loss (ΔC) are reported for each pair of 6C fast-charging cycles in Fig. 4c (see Methods for detailed calculation and Supplementary Figs. 27 and 28). The combined loss for each pair of cycles is about the same, which is reasonable because (1) the amount of loss per pair is small, ~1% of the total capacity, and (2) the cell aging that might promote increased plating over time is counterbalanced by increasing CCCV charge times, which lowers the average C-rate ( Supplementary Figs. 19 and 20). The graphite SOC shift contribution decreases from about 50% of losses for 6C cycles 1 and 2 to ~0% for cycles 25   Article https://doi.org/10.1038/s41560-023-01194-y accurate early plating quantification. Supplementary Fig. 29 shows the same details as given in Fig. 4c for all cells, an impressive visualization that indicates accurate loss quantification with single-cycle resolution. Figure 4c shows markedly less titrated Li for the FEC electrolyte cells compared to the 0% FEC cells, as anticipated from electrochemical measurements shown in Figs. 3 and 4a and the electrode images ( Supplementary Fig. 30). There is also a strong correlation between the sum of the 6C losses (from Fig. 4c) and titrated Li. For the 0% FEC electrolyte, the fraction of the loss accounted for by titrated Li is about 81% (Fig. 4c, inset), which is comparable to other studies of dead Li using similar electrolytes 12,13 . This leads us to suspect that the majority of the 6C losses are indeed due to irreversible Li plating, but note that this metric may include losses from other fast-charging degradation, such as SEI formation or electrode active material loss. The Li fraction with FEC is lower and decreases slightly from about 40% to 20% with increasing concentration, again highlighting the potential FEC advantage for avoiding metallic Li buildup during cell malfunction. Still, these values are notably higher than the ~10% fractional dead Li that others have observed for slow Li deposition on Cu for similar FEC electrolytes 12,33,47 , emphasizing phenomena unique to fast charging and the need to understand loss mechanisms besides dead Li formation.
Finally, the titration results unveil a route for estimating irreversible Li as a function of SOC in full-cells to allow direct comparison with half-cell results. The combined 1C capacity loss and 1C graphite SOC shift (ΔC + ΔX) was a strong predictor for titrated Li for the 0% FEC electrolyte, so we then designed a protocol alternating two 1C charge cycles with two fast-charging cycles to X% SOC, where X is increased by 5% for each iteration (Fig. 5a and Supplementary Fig. 31). Two cycles of each step were performed to benefit the technique sensitivity and reliability ( Supplementary Figs. 32 and 33). The 1C capacity changes between fast-charging steps (ΔC) correspond to losses from only those X% SOC cycles. Similarly, the 1C graphite SOC shift (ΔX) is calculated for each X% SOC fast charge step, and the combined loss is shown in the bottom of Fig. 5a, as in our previous analysis. This metric is shown for representative cells at various C-rates, and the x axis denotes the SOC cutoff of the previous two fast charge cycles that are analysed. For the full-cells, the rates and SOC are defined with respect to the nominal 3-4.2 V C/10 charge capacity, and were selected so that identical graphite current densities are applied for comparison with 3C-6C rates in the half-cells (Supplementary Note 2).
We then transform the data in Fig. 5a to estimate irreversible Li plating in full-cells and provide a direct comparison with the half-cells in Fig. 5b. The transformation entails (1) subtracting baseline losses observed for fast charging at low SOC before the onset of plating, as in Fig. 1b (Supplementary Figs. 34-36), (2) normalizing the loss to the active graphite capacity, as in Fig. 1c, and dividing by 2 to account for two cycles to each SOC, and (3) converting the x axis from full-cell SOC to graphite lithiation (average x in Li x C 6 ) by differential voltage profile analysis ( Supplementary Fig. 37). We assume that 100% of the baselined ΔC + ΔX data corresponds to irreversible Li plating capacity. A striking similarity is the shape of the Li|Gr and Gr|NMC curves, which extend our hypothesis from Supplementary Fig. 12 of universal physics for Li plating regardless of counter electrode selection. Another interesting observation is that the spacing of the Li|Gr curves have a similar C-rate dependence to those of Gr|NMC, which reveals a route for empirically scaling the half-cell data to predict full-cell behaviour with limited full-cell measurements ( Supplementary Fig. 38). Even without this adjustment, however, the half-cells show average Li plating onset SOC (defined again as 0.05% irreversible Li) within 3% of full-cells for the 20.1 mA cm −2 rate and within 6% for 13.4 mA cm −2 , suggesting that the Li|Gr measurements at the higher current densities (4C and above) in Fig. 2a,b are the most translatable for full-cells. We also offer some physical explanations for the plating onset differences, on the basis of previous modelling, in Supplementary Note 3.
As the final step of technique verification, the graphite electrodes were titrated for comparison with the cumulative irreversible Li estimated for each cell (Fig. 5c), determined by summation of the ΔC + ΔX values of Fig. 5a for each curve after the described baselining. The strong linear correlation (R 2 = 0.991) with near-unity slope further suggests that the method accurately predicts plating amounts, and estimates that, on average, 94% of irreversible Li plating exists in the form of electrically isolated Li 0 and other titration Li, with the remaining 6% as Li + -containing SEI species. The application of this protocol for electrolyte engineering in full-cells should be investigated in future works, but we note that the 1.2 M LiPF 6 in 3:7 EC:EMC (ethyl methyl carbonate) electrolyte offers favourable Li detection properties and is well-suited for immediate subsequent studies. One application is to   Article https://doi.org/10.1038/s41560-023-01194-y quantify the sensitivity of proposed onboard plating detection technologies, such as cell pressure monitoring 48 . Another is to quantify the effects of electrode porosity, loading, temperature, composition and heterogeneity on Li plating, to inform cell manufacturing.

Conclusions
Lithium plating is a nearly universal challenge for battery performance and operation, but the difficulty in detecting it has limited robust experimental studies. We have developed and verified high-throughput cycling techniques to quantify lithium plating in situ in Li|graphite and graphite|NMC cells, and the abundant data have led to physical insights for plating behaviour, electrochemical modelling improvements, cell design heuristics, routes towards data-driven plating models and electrolyte engineering strategies. Going forward, we believe that widespread reporting of irreversible Li plating curves and onset SOC will help quantify the trade-offs of novel battery design or operation approaches for fast charging, as well as lead to improved fundamental understanding. We hope these techniques are employed by academic and industry researchers and continually adapted to further reduce experiment time, consider battery aging effects on plating, transfer effectively to other cell formats and study nascent battery chemistries.
Hohsen CR2032 coin cells were used for all experiments, with 30 μl total electrolyte added quickly in three separate 10 μl aliquots, to ensure uniform wetting while avoiding evaporation. Graphite electrodes were 15 mm diameter punches, paired with either 14 mm diameter Li foil (0.7 mm thickness, MTI) or 14 mm diameter NMC and separated by a single 18 mm diameter Celgard 2500 separator (25 μm monolayer polypropylene). The molar ratio of Li:Gr in half-cells was greater than 30:1 for all loadings (Supplementary Note 4). All assembly/disassembly was performed in an argon-filled glovebox with O 2 < 1.0 ppm, H 2 O < 0.5 ppm. Electrochemical testing used Biologic MPG-200, VMP3 and BCS-810 potentiostats with CCH-8 coin cell holders at temperature control in Thermotron environmental chambers. Coin cell temperature rise from the chamber setpoint was expected to be minimal (<5 °C) during cycling (Supplementary Note 5). Cycling protocols were implemented with Biologic's EC-Lab software.

SOC-sweep testing for Li|graphite cells
Results are shown in Figs. 1, 2 and 3a and Supplementary Fig. 13. One slow formation cycle entails C/10 intercalation to 0.01 V and C/5 de-intercalation to 1.5 V with a 5 minute rest between each step. The experimental graphite capacity was determined from the discharge capacity of the third and final C/10 formation cycle and used to set the C-rates and SOC cutoffs for subsequent cycling. We refer to graphite intercalation as 'charging' and to de-intercalation as 'discharging' for consistency with language used for full-cell commercial lithium-ion batteries, even though the intercalation process is spontaneous in the Li|graphite cell configuration. Next, each cell underwent five fast formation cycles of 4C charge to 10% SOC and C/5 discharge to 1.5 V with 15 minute rest between current steps ( Supplementary Fig. 4). Last, the cell underwent the SOC-sweep cycling in which the charge capacity was increased 5-10% SOC for each subsequent cycle, with each charge step alternated with C/5 discharge to 1.5 V, and a 30 minute rest between current steps. The SOC window and step size were selected on the basis of the expected plating onset SOC; for later expected onsets, a step size of 10% was selected to cover large SOC range while minimizing experiment time (Supplementary Table 2). For high-temperature experiments, the oven temperature was increased from 25 °C to the target temperature during the five fast-charging formation cycles. For Fig. 3a comparing electrolyte compositions, the first formation cycle used C/20 instead of C/10 to clearly articulate dQ/dV features, as seen in Fig. 3d. Typically, three cells were run initially at each condition in Figs. 1, 2 and 3a and Supplementary Fig. 13, but the number of cells reported varied between two and five (for example, see

Electrochemical modelling
Electrochemical modelling is shown in Figs. 2 and 3g. The following are additional notes to supplement the model description in the main text. The universal plating reversibility was previously estimated to be roughly 70% (η = 0.7) under fast charge conditions and modest amounts of plating 13 and is a value previously observed at low temperature (−20 °C) and overcharge rate (C/10) plating conditions 34 . Irreversible lithium plating was determined from multiplying (1 − η) by the modelled plating amount. All electrolyte transport properties were taken from Idaho National Laboratory's advanced electrolyte model (AEM) 49 and used empirical fits as a function of salt concentration and temperature 24 . The anode and separator Bruggeman coefficient were estimated as 2.2-2.3 and 2.0, respectively, on the basis of detailed microstructure characterization/modelling and impedance spectroscopy using a blocking electrolyte 50 . The exchange current density and solid-state diffusion were estimated on the basis of extensive fitting to electrochemical data, including full-cells, half-cells and three-electrode test setups from within the US Department of Energy XCEL fast charge program 9,24,25 . The exchange current density for the lithium working electrode and lithium plating within the graphite anode were both set to 10 A m −2 as in our previous report 13 . The half-cell and full-cell models were written in C ++ and used the SUNDIALS suite of non-linear and differentiable/algebraic equation solvers 51 .

Lithium plating reversibility on graphite protocol
The protocol is shown in Fig. 3b. After three C/10 formation cycles and determining the experimental capacity, the graphite was intercalated at C/3 to 0.01 V and held for 1 h or until the current dropped below 10 μA (C/500), followed by immediate C/5 discharge to 1.5 V. This cycle was to determine the Coulombic efficiency for complete graphite lithiation in the absence of lithium plating. The following five cycles were identical except that after the intercalation intentional overcharge (Li plating) occurs at the selected C-rate (0.2C-4C) and capacity (10-30% SOC), and both were specified relative to the experimental full graphite intercalation capacity. A representative voltage profile for this cycling protocol is provided in Supplementary Fig. 16. The specific plating reversibility calculation is detailed in equation (3) and the corresponding text.

Incremental plating reversibility calculation
This is shown in Fig. 3f. The data from Fig. 3c report the plating reversibility for 10%, 20% and 30% overcharge (η 10 , η 20 , η 30 ) collected with separate coin cells, and these values can be algebraically manipulated to estimate the reversibility for Li deposited between 10-20% SOC (Δη 10 Similarly, And error bars were estimated by standard propagation of uncertainty ( Supplementary Fig. 17).

Lithium plating on copper foil
Results are shown in Fig. 3f. Lithium was deposited on 15 mm Cu foil (25 μm, MTI) from a 14 mm Li metal electrode at a current density of 4C with respect to anode A2 capacity (3.1 mAh cm −2 ) for 1.5 minutes (0.31 mAh cm −2 , 10% SOC) to mimic plating at the graphite|separator interface during the graphite plating reversibility experiments. Immediately after Li deposition, an oxidative C/5 current was applied until the cell voltage exceeded 1.0 V. The capacity ratio of the current stripping and plating steps is the reported reversibility. This cycle was repeated five times in total with 10 minutes rest in between, and the reversibility reported is an average value from cycles 3-5 (2+ cells for each electrolyte), which exhibit stabilized CE values relative to the first two cycles ( Supplementary Fig. 18).

Graphite|NMC532 full-cell electrolyte testing
Results are shown in Fig. 4a-c. The experimental full-cell capacity was determined from the discharge capacity of the third and final C/10 formation cycle and used to set the C-rates and capacity cutoffs for subsequent cycling. One slow formation cycle entails C/10 charge to 4.2 V and C/5 discharge to 3.0 V. All full-cell cycles include 5 minute rests between current steps. Next, 20 additional formation cycles were performed with 1C charge to 4.2 V and 1C discharge to 3.0 V, holding until the current dropped below C/5 on discharge (~5 min). Cell performance was analysed from the following sequence: five cycles of (1) 1C CCCV charge to 4.2 V, holding until C/5 (~10 min) and 1C discharge to 3.0 V holding until C/5, alternating with two cycles of, (2) 6C CCCV charge to 4.2 V, holding until 80% SOC (about 12 min total charge), then 1C discharge to 3.0 V holding until C/5. This sequence was repeated 20 times for a total of 100 1C cycles and 40 6C cycles. To prepare full-cells for titrations, the final step was a C/5 deep discharge down to 0.1 V to remove residual active lithium from the graphite. Electrochemical data analysis (Fig. 4b-d). Electrode voltage (V) shifts or capacity (Q) changes in full-cells are often characterized by monitoring the capacity (x position) at which local extrema in differential voltage curves (dV/dQ, y axis) occur 52 . Here, the dV/dQ versus Q curve shift was alternatively calculated from the capacity at which Q 0 dV/dQ = 1.0 V, defined as X, where Q 0 is the initial cell capacity ( Supplementary Fig. 27 and Supplementary Note 6). The graphite SOC shift (ΔX) between cycles 5 and 6, which corresponds to 6C cycles 1 and 2 in Fig. 4c, was calculated with the following equation, where the subscripts denote the 1C cycle number: This equation is used instead of ΔX = (X 6 − X 5 ) to account for transient behaviour of the first 1C cycle after fast charging (here, cycle 6) and to subtract nominal SOC shift that would also occur in 1C cycles, (X 5 − X 3 ), reducing contributions from cell aging unrelated to fast charging. Supplementary Fig. 28 provides thorough justification for this formula. Generalizing to determine ΔX that occurs for the two 6C cycles n and (n + 1) that occur between 1C cycles N and (N + 1) yields: Similarly, the changes in 1C discharge capacity reported in Fig. 4c, ΔC, was calculated by the following, where C is the discharge capacity for the Nth 1C cycle: For both ΔX and ΔC, the values for 6C cycles 39 and 40 are assumed to be identical to cycles 37 and 38 because additional 1C cycles were not performed after the two final fast charge cycles.

Graphite|NMC532 SOC-sweep Li plating quantification
Results are shown in Fig. 5a-c. For these cells, the experimental full-cell capacity, C full-cell , was fixed at 4.30 mAh (100% SOC, 2.80 mAh cm −2 , average of previous experiments) to fix the current density applied to the graphite electrodes for comparison with Li|graphite cells. The cycling protocol and data analysis were as follows.
a. Three times slow formation cycles 3.0-4.2 V as described above. b. Ten times 1C formation cycles CC charge to 4.2 V, 1C CCCV discharge to 3.0 V hold until C/20. Holding until C/20 was selected to minimize the graphite lithiation at the start of charge for the best comparison with Li|graphite cell measurements. c. One cycle of C/10 charge to 4.2 V, 1C discharge to 3.0 V hold until C/20. The charging step is used for dV/dQ analysis to determine the active graphite capacity and graphite lithiation at the start of charge ( Supplementary Fig. 37). d. Cell performance was analysed from the following sequence (see Supplementary Fig. 31 for representative voltage profiles during this protocol): two cycles of (i) 1C CCCV charge to 4.2 V, holding until C/5 (~10 min), and 1C discharge to 3.0 V holding until C/20, alternating with two cycles of (ii) fast charge at the specified C-rate constant current until X% SOC, then 1C discharge to 3.0 V holding until C/20 and (iii) repeating from sequence (i) except increasing the fast charging SOC cutoff of (ii) by 5%. After the final set of fast charging cycles, two additional 1C cycles are performed. (iv) C/5 deep discharge to 0.1 V to prepare for titrations. 2. Data analysis.
a. The graphite SOC shift ΔX and capacity loss ΔC for each pair of fast charging (FC) cycles was calculated by taking the difference of the second cycle of each pair of 1C cycles. Only the second cycle was analysed due to transient capacity and Coulombic efficiency behaviour for the first 1C cycle of each set after fast charge (Supplementary Figs. 32 and 33). Inspired by the analysis described for the 140-cycle full-cell methods above: ΔX fast charge at XC to X% SOC = (X 2nd 1C cycle after FC − X 1C cycle before FC ) −ΔC fast charge at XC to X% SOC = (C 2nd 1C cycle after FC − C 1C cycle before FC ) Article https://doi.org/10.1038/s41560-023-01194-y Note: in contrast to the 140-cycle full-cell equations for ΔX and ΔC, here there is no correction term that subtracts losses for 1C aging. b. This is because for the next analysis step, to estimate irreversible Li plating, the ΔC + ΔX data from part (a) (seen in Fig. 5a) is baselined to subtract losses from aging that are not related to lithium plating. This process is illustrated and discussed thoroughly in Supplementary Fig. 34. c. Finally, to convert full-cell SOC at the end of charge (Fig. 5a, x axis) to graphite lithiation at end of charge (Fig. 5b, x axis), the following equation is used: x in Li x C 6 = x initial + SOC full−cell ⋅ C full−cell C active−graphite (12) where x initial is the initial graphite lithiation at the beginning of charge and C active-graphite is the active graphite capacity, both determined from dV/dQ analysis ( Supplementary Fig. 37). Uncertainty propagation analysis indicates that the error induced by this transformation is no larger than 1% lithiation (Supplementary Note 7).

Electrode extraction, imaging, mass spectrometry titration and titration calibrations
Results are shown in Figs. 4c and 5c. Graphite electrodes from full-cell experiments were extracted with a Hohsen coin cell disassembling tool in the glovebox and imaged with a wireless handheld microscope (TAKMLY) before transferring to individual 6 ml vials (Metrohm). The vials were placed under active vacuum for 5 minutes before crimp-sealing the septum caps. Electrodes were extracted within 24 h of cycling completion and were stored in the glovebox for up to 3 days before titration. Rinsing the electrodes twice with dimethyl carbonate before vial storage was found to have minimal effect on dead Li measurements, so the majority of samples were not rinsed (Supplementary Fig. 26).
The Ar-filled sample vials were removed from the glovebox, quenched with 0.5 ml nitrogen-sparged deionized water, swirled for 10 seconds and then attached to the MST system using a novel syringe needle attachment featuring an adaptor (Valco, part no. ZBUMLPK) from 1/16" stainless steel tubing to Luer-lock ( Supplementary Fig. 21). The MST system draws 2 ml of the vial headspace every 2 minutes, refilling the balance with ultra-high purity Ar, using a constant system pressure of 1,030 ± 10 Torr. After about 40 minutes, or when the H 2 signal (m/z = 2) had decayed to its initial value ( Supplementary Fig. 22), the next vial was attached. This improved vial-swapping design, along with the smaller vial volume, resulted in a threefold throughput increase from our previous work 13 , and the signal strength suggests that 50 ng of Li metal (equivalent to 0.2 μAh total capacity) can be confidently quantified with each headspace sample precise to 10 ng (Supplementary Fig. 23). The calibration process that quantifies the linear relationship between the H 2 signal and the partial pressure of H 2 is detailed in Supplementary Note 8 and Supplementary  Fig. 22. To safely and precisely generate small quantities of H 2 in the 6 ml vials, graphite electrodes were formed and lithiated to known SOC (10-30%) in half-cells, extracted as detailed above, cut into pieces with known mass fractions of the entire 15 mm electrode and titrated, assuming complete conversion of the following reaction: Li x C 6 + xH 2 O → C 6 + 0.5xH 2 + xLiOH (13) The amount of titrated Li in the manuscript is presented as a capacity by converting the moles of H 2 assuming 1 mol oxidizable Li species per 0.5 mol H 2 , and 1 mol e − per mol Li.
Even in the absence of lithium plating, cycled graphite electrodes are expected to have non-zero titrated Li due to the presence of residual Li x C 6 (ref. 13 ) that is either electrically isolated or not fully removed during the deep discharge step. This non-zero amount was quantified with controls for each type of experiment and subtracted from the values reported in Figs. 4 and 5. For Fig. 4 experiments, the value was 0.012 ± 0.002 mAh cm −2 ( Supplementary Fig. 25), and for Fig. 5 experiments it was 0.019 ± 0.001 mAh cm −2 ( Supplementary Fig. 36), both of which are <1% of the total graphite lithiation capacity of 3.25 mAh cm −2 .

Data availability
All data supporting the findings in this study are available within the paper and the Supplementary Information. Source data are provided with this paper.