Use of Asymmetric Average Charge-and Average Discharge-Voltages as an Indicator of the Onset of Unwanted Lithium Deposition in Lithium-Ion Cells

Unwanted lithium-metal deposition on the negative electrode of a lithium-ion cell causes capacity loss due to poor lithium deposition and stripping efﬁciency and the possibility for internal short circuits. Internal short circuits may cause thermal runaway, which is especially dangerous in applications requiring many individual cells. This article proposes a method capable of identifying the onset of unwanted lithium deposition in-situ using cycles to full depth of discharge in a cell of any form-factor. The two most important factors affecting the average voltage of a Li-ion cell under load are the internal resistance increase and loss of lithium inventory. Increasing internal resistance increases average charge voltage and decreases average discharge voltage. Loss of lithium inventory, which occurs rapidly during unwanted lithium deposition and stripping, increases both average charge and average discharge voltage. Increasing internal resistance and loss of lithium inventory have a linearly additive effect on average voltage; therefore, tracking the average of average charge-and average discharge-voltages versus cycle count allows one to determine where rapid changes to lithium inventory onset, indicative of the onset of unwanted lithium deposition.

Lithium-ion batteries have become an important part of the world energy storage market with diversifying applications from small electronics and home appliances, to electric vehicles, to home-and gridenergy storage. Larger applications require more individual battery cells, more energy density, longer cycle life, and longer calendar life. It is becoming important in some applications to maintain safety and preserve capacity for thirty years or more.
Deposition of metallic lithium on the negative electrode is an undesirable and potentially dangerous process whereby lithium-ions are reduced on the surface of the negative electrode to form a metallic film instead of intercalating into the host material. Lithium deposition occurs if the charging rate of the cell exceeds the rate at which lithium can be intercalated into the graphite negative electrode. The film of deposited lithium may be uniform, or localized randomly, [1][2][3] or concentrated in the overhang region. 4,5 Deposited lithium may have planar, mossy, or dendritic morphology. 6 Probability for lithium deposition is higher at low temperature, high charge rate, and high state of charge. Additionally, the maximum charging rate prior to lithium deposition generally decreases as cells age due to increased internal impedance. The consumer market demands low temperature and fast charge, therefore strategies must be found to extend the lifetime of lithium-ion batteries without deposition. Solutions have included suitable electrolyte additives and solvent systems, appropriate negative to positive electrode ratios, and cell design. 7,8 Work on in-situ detection of lithium deposition is ongoing, varied, and has generated recent review articles. 9,10 Previously, Downie et al. 11 used isothermal microcalorimetry to show a thermal signature associated with lithium deposition. This technique was sensitive to small changes in upper charging potential. Burns et al. 1 used Ultra High Precision Coulometry (UHPC) to show a decrease in coulombic efficiency (CE) when cells lost lithium inventory to lithium-metal deposits. In mere weeks this technique can determine maximum charge current without lithium deposition for a given temperature. Petzl et al. showed that inspection of capacity retention curves can yield information about lithium inventory. 12 They tested cells at −22 • C, and 1C CCCV charge with a C/2 discharge. They concluded that the inflection point in the capacity retention curves corresponded to the maximum rate of lithium deposition. This group conducted another study 13 where they concluded that lithium deposition creates a characteristic feature in incremental capacity or incremental voltage plots.
The work in this article proposes a simple method using inexpensive equipment under standard test conditions to detect onset of unwanted lithium deposition. It uses inspection of average charge and discharge voltages for the in-situ identification of the onset of lithium deposition compatible with commercial form-factor cells.

Experimental
Cell build.-Size 402035 machine-made wound pouch cells were obtained heat sealed and without electrolyte from LiFun Technologies (Zhuzhou City, China). Cells had a single crystal LiNi 0.5 Mn 0.3 Co 0.2 O 2 (NMC532) positive electrode material that was well characterized by Li et al. 14 The positive electrode was made up of active material: conducting carbon: polyvinylidene fluoride (PVDF) binder in a 94:4:2 weight ratio. Positive electrodes had a loading of 21.1 mg/cm 2 , and a density of 3.5 g/cm 3 . Negative electrodes used artificial graphite with 13.6 mg/cm 2 loading and 1.55 g/cm 3 density. The negative electrodes used active material: conductive carbon black (CB): carboxymethyl cellulose (CMC) and stryene butadiene (SBR) binders in a 95. 4  Cell formation.-Filled cells were clamped between two rubber blocks to force any gas evolved out of the electrode stack and into the pouch bag. Cells were held at 1.5 V for 24 h at room temperature for proper wetting. They were then moved to a temperature controlled box at 40. • C connected to a Maccor 4000 series cycler (Maccor Inc.). Cells were charged at C/20 (11 or 12 mA) to 4.2 or 4.3 V respectively and held at top of charge for one hour. Cells were discharged to 3.8 V and held for 2.5 h. Cells were returned to the argon-filled glove box where they were cut open to release gas evolved during formation and resealed using the same conditions as described previously. Cells were removed from the glove box and re-clamped between rubber blocks.
Long term cycling.-Cells were cycled using a Neware BTS3000 series charger (100 mA/5V, Shenzhen, China) at room temperature (approx. 23 • C). Cells were cycled at C/3 (66.6 mA (4.2 V upper cutoff) or 70.0 mA (4.3 V upper cutoff)) in constant current-constant voltage (CCCV) mode. Cells were held at top of charge until the current reached C/20. After every 50 cycles, a constant current (CC) cycle at C/20 (10.0 or 10.5 mA) was performed to monitor low rate performance. Cells were removed from test near 80% original C/3 capacity or as necessary for destructive experiments. Figure 1 illustrates the origins of the potential -capacity curve of a full cell. The full cell potential is the difference between the potential of the positive electrode and the negative electrode. The experiments in this article specify that the full cell potential shall not exceed an upper cutoff potential (UCV) of 4.2 or 4.3 V, or fall below a lower cutoff potential (LCV) of 3.0 V. To maintain constant cutoff potential, the locations of the UCV and LCV may move relative to the capacity axis depending on impedance growth, loss of lithium inventory, and loss of active material. Figure 2 uses illustrative curves in arbitrary units of graph-paper squares to convince the reader that impedance growth and loss of lithium inventory are the processes capable of most change in average voltage. Change in average voltage that is attributable to impedance growth will hereafter be referred to as resistance voltage, or RV. Change attributable to loss of lithium inventory will be referred to as shift voltage, or SV because loss of lithium inventory causes the positive and negative electrode V-Q curves to shift their relative alignment. Figure 2, panels D-G show the negative electrode shift by 0, 1, 2, and 4 capacity units respectively. Graphite stages have been heavily exaggerated relative to positive electrode slope for clarity.

Results and Discussion
Recall that average voltage is capacity-weighted and can be calculated by, where V av is the average voltage, Q T is the total cycle capacity, and the integral is the area under the V-Q curve of the full Li-ion cell. The V-Q curve of the full Li-ion cell is calculated as the difference between the positive electrode curves and the negative electrode curves in each of the panels of Figure 2. The area between the positive and negative electrode curves has been shaded on the figure so the readers may count the squares to convince themselves that the stated areas are correct.
A perfect lithium-ion cell should have a constant average charge voltage (V av,c ) and a constant average discharge voltage (V av,d ) during its lifetime. However, normally over long times, V av,c increases and V av,d decreases. Average charge-and discharge-voltages are nearly mirror images of each other when plotted versus cycle number.
Mirrored V av,c and V av,d versus cycle number suggest that impedance growth is the most significant parameter in average voltage change. This is reflected by the common use of the parameter V, the difference between average charge-and average discharge-voltages. It is expected that cell impedance will increase during life, as SEI layers thicken, parasitic reaction products deposit on one or both electrodes, electrolyte is consumed, or salt is depleted. 15 The left column of Figure 2 (panels A, B, C) shows an illustrative data set that has been manipulated with an ohmic impedance term, where η is overvoltage due to the applied current, I is applied current, and R is internal resistance. Magnitudes of the IR terms increase red < blue < green. It is assumed that impedance does not change depending on the direction of charge movement. Charge and discharge currents of equal magnitude introduce a mirrored offset in the voltage-capacity plot. When voltage limits are fixed (3 and 8 a.u.), impedance growth limits the available capacity near the beginning and end of charge as shown in panel C. Capacity decreases from 9.9, to 9.2, to 8.0 as impedance growth causes endpoints to be reached sooner. Figure 2 shows that any individual cycle has nearly identical charge capacity (panel A) and discharge capacity (panel B) but significantly different areas under the V-Q curve, which results in charge always having larger average voltage than discharge. Increasing the IR term increases average charge voltage, decreases average discharge voltage, and decreases total capacity. Figure 2, panels D-G, shows the impact of imperfect alignment between positive (blue) and negative (red) electrode capacities. 16,17 As cyclable lithium is consumed from a cell's inventory, the electrode capacities 'shift' relative to each other which changes the overlapping areas. Shift voltage (SV), unlike resistance voltage, is unaffected by charging direction. Figure 2 shows that if a Li-ion cell begins its life in a positive electrode-limited alignment (panel D), average voltage (area between positive and negative curves normalized by total capacity) will decrease until electrodes are in perfect alignment (panel E) while capacity remains the same. This means average voltage will decrease. As the negative electrode slips past the positive due to lithium inventory loss (panels F, G), the area between the curves decreases more slowly than the loss of capacity. This results in an increasing average voltage. Figure 2, panel G shows how previously unused positive electrode capacity may be accessed to achieve the required UCV.
As has been described, average charge voltage is increased by the cumulative effects of SV and RV. Average discharge voltage is increased by SV but decreased by RV. Therefore,  Figure 2. The impact of increasing internal resistance (RV) and losing lithium inventory (SV) on average voltage are shown using illustrative curves. Average voltage is the area under the full cell curve, or the area between the positive and negative electrode curves, normalized by the total capacity. Shaded boxes may be counted to confirm values for area. As shown by the left panels, increased internal resistance will increase average charge voltage and decrease average discharge voltage. Three values of R are shown with R red < R blue < R green . The black dashed curve shows a cell with no internal resistance. As shown by the right panels, loss of lithium inventory will increase both average charge and discharge voltage. A curve representing the positive electrode is given in blue, and the negative electrode is in red. If a cell begins its life positive-limited, it is possible for average voltage to decrease as shown. Capacity remains unchanged until the anode capacity is to the right of the cathode, but area between the curves decreases.
It is a trivial rearrangement of Equations 3 and 4 to show, SV = 1 / 2 V av,c + V av,d [5] RV = 1 / 2 V av,c − V av,d [6]  It is observed that SV begins to increase faster as discharge capacity begins faster decrease. Throughout the work that follows below, RV and SV have been zeroed to cycle 20. Values that have been zeroed will be called shift voltage change, SVC, and resistance voltage change, RVC. The initial value of RVC (before zeroing) is captured by V. The initial value of SVC is less informative than clearly seeing the relative rankings from a matrix of cells.
Shift voltage change is shown to be useful in analysis of cells showing rollover failure. As discussed by Burns et al. 18 rollover is characterized by high capacity retention for many cycles and then abrupt, rapid capacity loss over few cycles. Figure 4 highlights the strength in the SVC:RVC method for accurately determining the onset of unwanted lithium-metal deposition as a cause of rollover failure. The left graphs show absolute capacity loss, SVC, and RVC for four cells at 1200 cycles. Each of the cells had an initial capacity near 220 mAh. The cell in green triangles, FEC-DTD_40MA, is considered "far from rollover". It has lost less than 5 mAh capacity (2.4%), and shows minimal change in impedance. The cell in black crosses, FEC, is "not at rollover". It shows 20 mV of impedance growth, but is similar to FEC-DTD_40MA in capacity retention. The cell in blue circles, FEC_20MA is at the "brink of rollover". Capacity loss, while still at a modest 5%, has the appearance that it may begin to accelerate. Impedance growth is high. The cell in red diamonds is "at rollover". Capacity loss per cycle has increased dramatically; 8 mAh of capacity was lost during the first 900 cycles and the next 300 cycles showed a loss of an additional 27 mAh. Impedance growth is high. Note that SVC and RVC have been plotted on the same scale to show their relative magnitudes. Resistance voltage change is the larger effect.
All cells look identical in SVC except for FEC_40MA, which diverges sharply at 850 cycles. The dashed vertical line at 850 cycles indicates that sharply increasing SVC correlates well with rollover. There is some curvature visible in capacity loss before SVC begins its sharp increase. This may indicate that there is another failure mode preceding lithium deposition. Notably, FEC_20MA does not show any change in SVC despite having nearly identical RVC.
As discussed in Figure 2, loss of cyclable lithium leads to increasing SVC. Lithium deposition rapidly removes lithium from the inventory, which rapidly increases SVC. It was decided that these four cells should be terminated to verify the presence of metallic lithium on the surface of the negative electrode. The right side of Figure 4 shows photographs of the negative electrodes inside an argon-filled glove box. The photographs are labelled and colored to match the graph legend. The electrodes of FEC-DTD_40MA (far from rollover) and FEC (not at rollover) are uniformly red, indicating stage 2 lithiation throughout. 19 The electrodes of FEC_20MA (brink of rollover) and FEC_40MA (at rollover) are red near the edges of the electrode with patchy blue regions (dilute stage 2) down the center of the electrode. Poor lithiation is occurring at the center of the jelly roll, far from the electrolyte reservoir at the edges of the jelly roll. As predicted by the non-destructive in-situ SVC analysis technique, only FEC_40MA shows the presence of metallic lithium. Patchy metallic lithium is located at the center of the jelly roll. Figure 5 shows the cells from Figure 4 within the context of the larger electrolyte and voltage matrices. The left column shows cells cycled to 4.2 V, and the right column cells cycled to 4.3 V. The cells tested to 4.2 V had initial capacities near 210 mAh and the cells tested to 4.3 V had initial capacities near 225 mAh. Each panel shows cells containing FEC in crosses and FEC-DTD in triangles. Colors are used to indicate MA content: blue is no MA, red is 20 MA, and black is 40 MA. This cell matrix was previously presented by Li et al. 20 with accompanying discussion on capacity retention and V growth. Briefly, cells cycled to 4.2 V had better capacity retention and smaller impedance growth than cells cycled to 4.3 V, higher MA content led to earlier cell failure in long term tests, and DTD significantly mitigated impedance growth and extended lifetime.
Recall that RVC is 1 2 V, zeroed to an initial cycle. The practice of zeroing means that information about the magnitude of the cell impedance is lost, but trends in impedance growth become more obvious. Li et al. in Figure 9 20 show that cells containing higher amounts of MA as co-solvent have lower V when cells are fresh. Impedance growth quickly becomes important in cells cycling to 4.3 V with no DTD. The bottom row of Figure 5 shows that cells follow two groupings: the crosses, cells without 1DTD, follow a steeper slope than the triangles, cells with 1DTD. The authors were very surprised that a 1% change in electrolyte (adding DTD) should have a larger impact on suppressing impedance growth than changing up to 40% of the solvents in the electrolyte! There is little difference in impedance growth rates for different amounts of MA up to 400 cycles to 4.3 V or 800 cycles to 4.2 V. After 400 or 800 cycles, higher MA content begins  It is perhaps surprising that onset of lithium deposition does not necessarily correlate with rate of increase in resistance voltage. In Figure 4, the authors suggested FEC was "not at rollover" and FEC-DTD_40MA was "far from rollover" where the primary difference between the two was larger RVC in FEC at 1200 cycles. It was therefore assumed that FEC should fail before FEC-DTD_40MA. Pair cells of both chemistries continued cycling and are now over 2000 cycles. The left panel of Figure 5 SVC shows that FEC-DTD_40MA testing to 4.2 V has lithium-metal deposits and FEC testing to 4.2 V does not. This shows that the SVC:RVC method has limited application in the prediction of onset of lithium deposition, but at very early time may confirm the presence of metallic lithium once deposition has started. Figure 6 shows SVC with standard long term cycling metrics, capacity and V. Both high rate (C/3 CCCV) and low rate (C/20 CC, every 50 cycles) cycles are shown. The low rate, constant-current cycles show very similar trends to high rate cycles in all three metrics. Shift voltage change shows a sharp increase at 1100 and 1200 cycles for FEC and FEC-DTD_20MA respectively. The increase is attributed to lithium-metal deposition (rapid loss of lithium inventory). These cycle counts are indicated with a dashed line. Both cells show slow, smooth capacity loss, and a near-linear increase in V after 300 cycles. There is small curvature before shift voltage increases sharply. Shift voltage change clearly identifies the mechanism of failure (lithium deposition) and the cycle count where lithium deposition begins. The standard metrics cannot identify either the mechanism or the cycle count where failure begins. This demonstrates the importance of the SVC method.

Conclusions
Analysis of the shift voltage change was introduced as an analysis technique suitable for a wide variety of applications and cells. Asymmetries between the average voltages during charge and discharge were used to make inferences about the lithium inventory of a cell while it was in operation. The sharp increase in shift voltage change was shown to be useful in very precisely identifying the onset of unwanted lithium deposition. Common degradation metrics including capacity loss and V have no obvious feature to indicate lithium deposition has begun. Higher RVC does not necessarily indicate earlier lithium deposition. Shift voltage change suggested that the FEC-DTD-MA matrix of cells studied at 20 • C, C/3 CCCV cycling will likely all rollover for reasons related to lithium deposition. Higher MA concentrations lead to shorter lifetime, but the lifetime penalty can be mitigated by the addition of 1% DTD.
This technique opens many new questions. Future work should address the question of why SVC initially has a negative slope for all cells presented in this work. Further work should investigate the effect of different temperatures, and different electrode materials. Effort should be spent investigating if all cells that show lithium deposition . Shift voltage change and standard long term cycling parameters, capacity and V. Low rate C/20 cycles (occurring every 50 cycles) show performance parallel to high rate cycles. Shift voltage change shows sharp increase at cycle number 1100 and 1200 for FEC and FEC-DTD_20MA respectively.Sharp SVC increase is caused by rapid loss of lithium inventory, which is attributed to lithium-metal deposition. This is indicated by dashed lines. There is no indication of lithium deposition in the standard metrics. Near-linear increase in V is observed up to 2000 cycles. Cell capacity fades smoothly with small curvature. Onset of lithium deposition is not easily assigned to a cycle number without considering SVC.
show sharply increasing SVC, or if the cells studied in this matrix were a special case.