End-of-Charge Temperature Rise and State-of-Health Evaluation of Aged Lithium-Ion Battery

Abstract

An increasing demand to repurpose used lithium-ion batteries in secondary applications is driving the need to develop methods of evaluating the state-of-health of used batteries. In this paper, we discover a self-terminated end-of-charge temperature rise (ECTR) phenomenon in 18650 lithium-ion cells, both recycled from the field and aged under controlled conditions in the lab. ECTR is characterized by an additional temperature rise near the end of the charging process and is accompanied by low coulombic efficiency. A higher charge rate and longer inactive time at low state-of-charge appear to increase the occurrence of ECTR. The intensity of ECTR is found to closely correlate with the excess charge capacity but is less affected by the charge current or cell impedance. ECTR is weakly dependent on the remaining cell capacity in recycled cells, and the controlled aging study shows that aging condition, not remaining capacity or internal resistance, determines the presence and intensity of ECTR behavior, which indicates that usable capacity or internal resistance should not be the single criterion to effectively evaluate the state-of-health of used cells intended for repurposing. We hypothesize that the origin of the ECTR is due to the formation of an internal lithium metal short that forms near the end of the charge process and self-terminates over time. The investigation of ECTR in this work provides a new criterion and approach to evaluate the state-of-health of cells required to safely handle aged/recycled cells.

1. Introduction

The rapid rise in demand for electric vehicles, power tools, and portable electronic devices is driving an ever greater need for lithium-ion batteries [1,2,3,4]. By 2025, the global factory capacity for lithium-ion battery production is predicted to reach 1 TWh [5] and the global lithium-ion battery market is predicted to reach ~USD 140 billion [6]. Depending on the application, the life of a lithium-ion battery is usually 1 to 5 years for consumer electronics and expected to be 10 to 20 years for electric vehicles [7,8,9]. That means megawatt-hours of lithium-ion batteries will enter their end-of-life in the near future [1,3,10]. As a result, there is a growing market for repurposing batteries that are no longer suitable for their original application [1,3,6,10,11,12]. Particularly, an increasing number of companies are repurposing aged batteries for less demanding applications such as battery banks for back-up power or grid energy storage, or collecting and reselling used batteries directly to consumers as a cheap energy storage option [3,11,12,13,14,15]. A good state-of-health screening of retired cells is critical for safe repurposing. Many different methods have been developed to evaluate the state-of-health of lithium-ion batteries, particularly at their end of life, including the simple measurement of remaining capacity and internal resistance as well as more sophisticated analyses such as dQ/dV, equivalent circuit model analysis, and big data-based machine-learning modeling, etc. [16,17,18,19] However, there are currently no universal and reliable criteria for safely repurposing batteries from their original application. Batteries are retired from their original application for many reasons (only some of which relate to battery performance), and most purchasers of used batteries have no knowledge of the cell’s history [11,14,15]. Without properly screening and assessing the aged batteries, battery reuse programs will be risky and costly, which must be resolved for them to become a practical solution [3,12,13].

In this paper, we report systematic and comprehensive analyses on used 18650 lithium-ion cells purchased from an online vendor and on cells subjected to controlled laboratory aging. An abnormal temperature rise accompanied by excess charge capacity was observed near the end of charge in most of these recycled aged cells, which we refer to as end-of-charge temperature rise (ECTR). ECTR events were found to be self-terminating and triggered by elevated charge rate and increased resting time at low state-of-charge. The magnitude of the temperature rise was found to be independent of the charge current and cell impedance. We attribute ECTR to a lithium short which self-terminates during the charge or subsequent rest. We also reproduce ECTR on cells aged under controlled lab conditions. The controlled aging study shows that cells aged under different conditions can have vastly different ECTR behavior even when they have the same remaining capacity or internal resistance, suggesting that remaining capacity or internal resistance are not reliable criteria on their own to effectively determine the health status of the aged cells. Although we did not observe catastrophic failure during any of the ECTR events in this study, the possibility of such failures cannot be excluded and presents a significant risk for the reuse of retired cells. The study of ECTR phenomenon provides a new criterion and approach to evaluate the state-of-health of cells and to safely manage the reuse of retired lithium-ion batteries.

2. Materials and Methods

2.1. Electrical Measurements

The new and used 18650 lithium-ion cells were purchased from two online vendors, respectively, with a nominal capacity of 3.1 Ah and a working voltage range of 3.0 V to 4.2 V. The cathode material was LiNi_xMn_yCo_zO₂ (NMC) and the anode material was graphite. All cells were of the same brand and model. The cells were cycled using a MACCOR Series 4000 battery and cell tester. During the cycling, the temperature at the surface of the cell can was monitored via a thermocouple. All cells underwent an initial screening, during which the as-received cells were first discharged to 3.0 V at 0.5 A, then charged and discharged at 0.5 A for two cycles, followed by another two cycles at 1.25 A. Each charge process had a voltage upper limit of 4.2 V and a constant-voltage current taper cutoff at 0.025 A. Each discharge process had a voltage cutoff at 3.0 V with no current taper. A one-hour rest period was programmed between each charge and discharge for the cells to relax and cool down. Impedance was measured using the Maccor prior to the start of each discharge.

Fresh 18650 cells were aged using a controlled process by cycling under two different cycling protocols: three cells were cycled between 3.0 V and 4.2 V at 2.5 A at 25 °C for 500 cycles, and another three cells were cycled between 3.0 V and 4.2 V at 1.25 A at 60 °C for 500 cycles. The cycling was paused for every 50 cycles, rested at room temperature for 24 h, then subjected to two 0.5 A cycles and two 1.25 A cycles (following the same test procedure on used cells mentioned above).

2.2. Cryo-Resistance Measurements

To determine whether a cell has an internal short circuit, a cryo-resistance method that separates ionic and electronic resistance was used. The cell was submerged in liquid nitrogen for 20 min, which cools the cell to a low temperature where ionic resistance is suppressed. A DC multimeter was used to measure the resistance. This cryo-resistance can be attributed to an internal electrical short if present.

2.3. Reference Electrode Test

One recycled cell and one new cell were randomly chosen for reference electrode (RE) testing. The cells were opened in an argon-filled glovebox, and the cell windings were removed from the metal cans and then directly immersed into a reservoir of electrolyte (LP47 from Gotion, Fremont, CA, USA). A piece of lithium metal foil (from Alfa Aesar, Haverhill, MA, USA) was immersed into the same reservoir of electrolyte as the RE, in a location close to the cell winding. The positive tab of the cell was used as the counter and counter sense electrode, while the negative tab of the cell was used as the working and working sense electrode. A thermocouple was taped to the middle of the cell winding to monitor the temperature evolution. The whole assembly was enclosed into a gas-tight enclosure and removed from the glovebox for further testing. The assembled RE cells were first cycled at 0.5 A for one cycle and then at 1.25 A for another cycle. The cells were then charged at 1.25 A to 100% state-of-charge for the electrochemical impedance spectroscopy (EIS) measurement. Lastly, cells were unwound while still at 100% state-of-charge for physical inspection. Each charge process has a voltage upper limit of 4.2 V and a current taper cutoff at 0.01 C (0.025 A). Each discharge process has a voltage cutoff at 3.0 V with no current taper. A one-hour rest period was programmed between each charge and discharge procedures for the cells to relax and cool down.

2.4. SEM Characterization

Regions of interest in recycled cells were characterized using a JEOL Model 6390LV scanning electron microscope (SEM) at 20 kV equipped with an energy dispersive X-ray spectroscopy (EDS) detector.

3. Results

In total, 6 new cells and 25 used cells were subjected to the initial screening cycles of two 0.5 A cycles followed by two 1.25 A cycles. Representative temperature, current, and voltage profiles during charge and discharge of new and recycled cells at 0.5 A and 1.25 A rates are shown in Figure 1. For new cells (Figure 1a), no clear difference was observed between the first and the second 0.5 A cycles, nor between the two 1.25 A cycles. In all cases, the temperature steadily rose from the beginning of the charging process before reaching a relatively stable temperature. Once the constant-voltage (CV) phase was reached near the end of charge, the temperature decreased as the current tapered down. This indicates the heating and temperature rise during charging of new cells was dominated by Joule heating. Some recycled cells demonstrated substantially different heating pattern from the new cells. Figure 1b shows that during the first part of the 0.5 A charge process, the temperature of a recycled cell was close to the room temperature (23 °C) when the total charge capacity was below 2.0 Ah (indicating low levels of Joule heating, similar to the new cell during the 0.5 A charge). Once the capacity exceeded ~2.0 Ah, the temperature quickly ramped up to 37.5 °C before gradually cooling back to room temperature at the end of charge (without manual intervention). The charge capacity for this cycle was greater than 3.7 Ah, while the subsequent discharge had a capacity of ~2.3 Ah. The temperature bump and excess charge capacity disappeared in the second 0.5 A cycle. During the following first 1.25 A cycle, we again observed a steady Joule-heating-induced temperature rise during the initial charge process, similar to that observed on new cells during 1.25 A charging. At ~1.5 Ah charge capacity, the temperature rapidly rose, similar in shape and behavior to the temperature rise observed in the first 0.5 A cycle. A slight wave was observed in the current profile during the CV portion of the charge, and the total capacity charge capacity for this cycle exceeded the discharge capacity for both the previous and subsequent discharges. Once again, no temperature rise was observed during the second 1.25 A charge process. In all, 25 used cells were tested with such screening protocols, 12 cells (48%) showed this end-of-charge temperature rise (ECTR) effect in both the first 0.5 A charge and first 1.25 A charge (“low-rate trigger”). Eight cells (32%) showed no clear ECTR in the 0.5 A charge but clear ECTR in the first 1.25 A charge, as shown in Figure 1c (“high-rate trigger”). The remaining five used cells (20%) showed no ECTR in either 0.5 A or 1.25 A charge (“no ECTR”). See Figure 2a. ECTR was not observed in any of the new cells screened. Among all the cells, the maximum temperature caused by such ECTR during the screening was as high as 51.5 °C. It worth noting that these cells contained a positive thermal coefficient (PTC) device, but we did not observe any indications that the PTC was activated during our tests.

Figure 2a presents the maximum temperature during each charge, which shows that cells without ECTR have very similar temperature maxima between the two cycles at each rate. Accordingly, we use the difference in the maximum charge temperature between the two cycles (ΔT_max) to quantitatively identify instances of ECTR in Figure 2. Values of ∆T_max greater than 1 °C are defined as demonstrating ECTR. Figure 2b shows that cells with a low-rate trigger usually show excess capacity on the first low-rate charge, while cells with a high-rate trigger show only a slight excess in capacity during the first high-rate charge. Figure 2c compares the value of ∆T_max during the 0.5 A and 1.25 A cycles, with cells grouped based on which rate ECTR was found to be triggered. Note that all cells exhibiting ECTR at low rate also exhibit ECTR during the subsequent higher-rate charge.

Figure 2d compares the value of ∆T_max with the first discharge capacity at each rate. This shows that cells with very high capacity (new or minimally used cells) or very low capacity (heavily degraded cells) tend not to exhibit ECTR. ECTR was observed in cells with intermediate discharge capacity (between 43% and 79% of new cell capacity), but not all cells in this range exhibited ECTR behavior. Figure 2e compares ∆T_max with the AC impedance measured after the second charge at the given rate. As with discharge capacity, this shows that cells with relatively low impedance (i.e., new or minimally used cells) and very high impedance (heavily degraded) cells do not exhibit ECTR, while some (but not all) cells with intermediate impedance values exhibit ECTR. Finally, Figure 2f compares ∆T_max with the first cycle excess charge capacity (where excess charge is defined as the difference between the first 1.25 A or 0.5 A cycle charge capacity and the second 0.5 A cycle charge capacity). All cells exhibiting ECTR behavior are associated with an excess charge capacity of at least 0.4 Ah. As a specific example, the cell shown in Figure 1b had a 0.5 A charge capacity of 3.72 Ah in the first cycle, but a subsequent charge capacity of just 2.30 Ah in the second cycle, corresponding to an excess charge capacity of 1.42 Ah. Note that some cells exhibited excess charge capacity of as much as 0.8 Ah without exhibiting ECTR. This behavior was observed only in cells with very low overall capacity exhibiting rapid capacity loss on each subsequent cycle.

Based on these observations, we propose that ECTR is caused by the formation of an unstable and self-terminating lithium short circuit formed during charge. The proposed short circuit develops when the cell reaches a high state-of-charge resulting in excess heating (via Joule heating through the short circuit) and excess charge capacity. At the point of peak heating, the short is of sufficiently low resistance to conduct the majority of the applied current, but which, by 1 h after the end of the charge, does not result in any detectable reduction in the cell’s impedance. This short circuit behavior is associated with cell degradation (as it is never observed in new cells) and is related to changes in charge rate.

To further test this hypothesis, we selected seven used cells which exhibited low-rate ECTR during screening and interrupted the charge process at the onset of ECTR to examine the potential internal short and the possible corresponding self-discharge. To do this, we first recorded the temperature when the cell was charged to 4.0 V at 0.5 A as the baseline charging temperature; for the remainder of the charge process, any temperature more than 2 °C above this baseline was defined as the beginning of ECTR, at which point the charge was terminated and the cell’s open-circuit voltage and surface temperature were monitored. An example of this behavior is shown on the red line in Figure 3a. All seven cells exhibited ECTR; one cell was allowed to rest at open circuit with active temperature monitoring while the remaining six cells were immediately removed from the battery tester and cooled in liquid nitrogen in order to freeze the electrolyte and eliminate the contributions of ionic conductivity to the cell’s impedance. Doing so allowed the direct measurement of the electrical resistance of any potential short-circuit. As controls, one new cell and one used cell without ECTR were also tested; both cells exhibited open-circuit resistance (i.e., an impedance above the measurement range of the resistance meter) at cryotemperatures, indicating no internal short-circuit. For used cells interrupted at the beginning of ECTR, the initial cryo-impedance varied from 67 Ω to 875 Ω, indicating the existence of an internal short-circuit. This is consistent with the lithium-induced short circuit hypothesis proposed above, as a Li-metal short would provide a current path through the cell at cryotemperature. To observe the evolution of this internal short over time, the cryo-resistance measurement procedure was repeated over the course of 2 days on one of the tested cells, with the cell being allowed to re-warm between measurements. One cell initially showed an impedance of 67 Ω (at 0 h). Over the next two days, the impedance increased to 186 Ω (at 1 h), 365 Ω (at 2 h), 560 Ω (at 3 h), 0.739 MΩ (at 24 h), and open-circuit (at 48 h), see Figure 3c. For another cell with an initial cryo-impedance of 533 Ω at the beginning of ECTR, the cryo-impedance increased to open-circuit after 1 h of room-temperature rest time. These findings are consistent with the self-termination feature of the soft lithium short that caused ECTR, as the metallic lithium dendrite could redistribute into the graphite negative electrode or react with electrolyte to open the short circuit. In addition to the cryo-resistance measurement, for one other cell, after each charge process (interrupted or not) the cell was allowed to rest for 6 h while the voltage and temperature evolution were monitored. As an example, for the cell shown in Figure 3b, after the interruption of ECTR in the first cycle, the temperature continued to rise at the beginning of the rest step and reached a maximum 11.3 min after charging was interrupted. As shown in the second panel of Figure 3b, the interrupted ECTR charge exhibited a much larger drop in open circuit voltage observed in the first ~2 h than the other charges (neglecting the first five minutes of rest after charge to account for differing terminal currents between the interrupted and non-interrupted charges). This behavior is consistent with self-discharge through an internal short-circuit. Over time, the rate of voltage drop slowed down, which is consistent with the self-termination characteristic of ECTR. These observations are consistent with our proposed lithium short-circuit hypothesis where the internal short-circuit acts as a heat source during ECTR, independent from the charge current applied to the cell.

One unique feature, that can be seen in Figure 1 and Figure 2, is that in two continuous cycles with the same charge rate, if ECTR is observed, it usually only appears in the first cycle. One potential explanation is that severe lithium dendrite growth may lead to local depletion of electrolyte via a side reaction between metallic lithium and the electrolyte, which can create a localized electrochemically inactive region, preventing immediate retriggering of ECTR. To test the influence of relaxation time on ECTR, we tested the occurrence of ECTR during a 1.25 A fast charge on six used cells after different rest periods (some are shown in Figure 4). For the first example cell shown in Figure 4a, ECTR was observed in the first cycle, but not in the second cycle with a 1 h rest period after the first discharge. A 1-day rest was inserted after the second discharge, after which ECTR was observed in the third cycle, showing that extended rests are capable of retriggering ECTR without changing rate. A fourth cycle was performed 4 days after the third cycle (resting in the fully discharged state), after which a much more severe ECTR was observed (T_max of 70 °C). Although the temperature naturally peaked during this ECTR event, the cell continued to charge at more than 0.2 A for several days until more than 16 Ah of capacity was charged into the cell and the charging process was manually interrupted. This demonstrates the variability of the self-terminating aspect of ECTR.

The second cell tested exhibited ECTR in all four cycles, including the second cycle that occurred after a 1 h rest, although with a much smaller temperature rise relative to the first cycle. This behavior (ECTR during two consecutive cycles at the same rate) was not observed for any cell during the initial screening test. During the fourth cycle (after a four-day rest while fully-discharged), the maximum cell temperature rose to 80 °C, after which the test was terminated from pre-programmed safety limits. This suggests that an ECTR-lead thermal event cannot be ruled out if appropriate safety measures are not implemented.

Although most of the cells with ECTR showed a larger temperature rise after longer rests (e.g., cells in Figure 4a,b), the cell in Figure 4c exhibited clear ECTR after 1 day of rest but no ECTR after 4 days of rest (despite severe initial ECTR exceeding 70 °C). This means long shelf time does not always lead to more severe ECTR, which might be related to the randomness of lithium dendrite growth and the short-circuit path formation process.

To study the internal cell condition during ECTR, lithium reference electrode (RE) and electrochemical impedance spectroscopy (EIS) measurements were made on one new cell and one used cell (Figure 5). A thermocouple was placed on the surface of the winding roll during the RE test. In the new cell, no ECTR was observed in either cycle. The charging temperature immediately dropped when the current started to taper (Figure 5a), consistent with Joule heating with a fixed internal resistance. In comparison, the used cell showed ECTR in both the 0.5 A and 1.25 A cycles accompanied with a large excess charge capacity in each cycle (Figure 5b). The temperature bump in the 1.25 A cycle was slight, but the temperature remained elevated during the current taper, suggesting it still experienced ECTR. At the end of 0.5 A and 1.25 A charge processes, the negative electrode potential of the used cell was slightly lower than that of the new cell (~0.10 V vs. lithium compared to ~0.13 V vs. lithium), but still much higher than 0 V vs. Li. At the end of the 0.5 A discharge, the negative electrode potentials of the used and new cells were ~0.17 V and ~0.29 V vs. lithium, respectively. These results suggest that neither cell was thermodynamically favored to form metallic lithium during charge (although localized lithium plating remains possible), and that more lithium remained trapped in the negative electrode of the used cell after discharge.

After completing the RE test, the cell windings were removed and unwound at 100% SOC in the fully-charged state under argon in order to examine the physical appearance of the electrodes and separators, as shown in Figure 6b. In addition, a fresh cell and a recycled cell with no ECTR were also opened and unwound at 100% SOC in the glovebox for comparison. As shown in Figure 6a, the fully charged new cell has clean electrodes and separators, with no evidence of discoloration or lithium plating. The golden color of the negative electrode is typical of fully lithiated graphite. In contrast, the recycled cell exhibiting ECTR has large areas of lithium plating (visible as grey deposits) on the negative electrode as well as discoloration of the separator facing the negative electrode (Figure 6b). Under SEM, the gray deposits exhibited a mossy morphology with some areas appearing to contain reacted dendrites (Figure 6d), which is consistent with lithium plating. Please note that during the transfer from the glovebox to the SEM, the sample has been briefly exposed to air. A high oxygen peak was observed in EDS in this region after air exposure (Figure 6e), which is also consistent with localized lithium plating (although lithium cannot be directly detected by EDS, metallic lithium reacts with moisture in air and readily forms LiOH, resulting in strong oxygen signals). Elevated levels of fluorine and phosphorus were also observed in EDS, indicating that the electrolyte degradation is co-located with lithium plating. For the recycled cell with no ECTR observed in the initial screening, see Figure 6c, low levels of lithium plating were observed (limited to a few local small spots).

In addition to the ECTR phenomena observed on recycled cells with varied and unknown history, ECTR was also observed on cells under controlled aging conditions. We aged fresh 18650 cells of the same model under two different conditions: 2.5 A cycling at 25 °C and 1.25 A cycling at 60 °C (using the same current for both charge and discharge in each case), which were the highest allowed charging rates according to the cell specification sheet at 25 °C and 60 °C, respectively. All cells were cycled between 3.0 V and 4.2 V. Three cells were aged under each condition for up to 500 cycles. Plots showing the temperature, current, and voltage profiles during various screening cycles of one cell at each temperature are shown in Figure 7.

After every 50 cycles, the cells were rested for 1 day at 0% SOC at room temperature, then subjected to the screening protocol (also at room temperature) to check for ECTR behavior. Although the three cells cycled at 25 °C exhibited different rates of capacity fade near the end of cell life (Figure 8a), ECTR occurred in all three cells during the first 0.5 A cycle after a 1 day rest when the 0.5 A discharge capacity dropped below ~85% (Figure 8c). In all three cases, a small initial temperature peak was observed in the ECTR screening preceding the first instance of ECTR (blue arrow in Figure 7d), suggesting that preliminary ECTR behavior can occur with as much as ~90% of the original capacity remaining. As the capacity continued to fade to ~65–75%, ΔT_max peaked for cells aged at 25 °C. Further capacity loss resulted in decreases in ΔT_max, indicating the maximum temperature that can be reached during ECTR is limited by both the severity of cell aging and the remaining capacity. In contrast, ECTR was not observed in the 0.5 A cycles for cells aged at 60 °C even when their capacity was in the 75–85% range. ECTR was only observed in the 1.25 A ECTR check cycles in cells cycled at 60 °C, and ∆T_max was lower than for cells cycled at 25 °C. This implies the ECTR is sensitive to the use history of the cell and cannot be determined by a simple measurement of the remaining usable capacity in the cell. This agrees with the varied response observed in used cells with unknown history, as shown in Figure 2d. These findings confirm that ECTR can occur in cells cycled entirely within the cell specification under controlled laboratory conditions.

4. Discussion

To summarize, we find that ECTR events have seven consistent properties. ECTR events: (1) are triggered by extended relaxation at low SOC, (2) are triggered by increases in charge rate, (3) are weakly related to pre-event discharge capacity, (4) are self-terminating, (5) have a severity independent of charge rate or pre-event cell impedance, (6) result in excess charge capacity, and (7) do not influence the discharge capacity. Each of these are discussed in detail in the following paragraphs.

(1)

ECTR events are triggered by extended relaxation at low SOC. ECTR was not observed during the continuous charge/discharge cycling (with 1 h rests between half-cycles). ECTR only occurred when the cell was allowed to rest at low SOC for an extended period before being charged, although no consistent relationship between rest time and ECTR severity was found.

(2)

ECTR events are more likely to occur at high charge rates. If ECTR is triggered in a low-rate charge (e.g., 0.5 A), it can be re-triggered in a higher-rate charge (e.g., 1.25 A) even without extended relaxation. In other instances, cells that do not exhibit ECTR at low rate may still exhibit ECTR when subsequently charged at higher rate.

(3)

ECTR events are weakly related to pre-event discharge capacity. We observed that cells with high discharge capacity (i.e., minimally aged) or very low capacity (i.e., aged too much) do not exhibit ECTR, while the strongest ECTR effect is usually observed in cells with moderate remaining capacity. However, some cells with intermediate discharge capacity do not exhibit ECTR, and the frequency and magnitude of ECTR events appears to be sensitive to the cell’s use history.

(4)

ECTR events are self-terminating. When ECTR occurs, the temperature naturally reaches a maximum before dropping back to the room temperature without manual intervention during the CV charge step, resulting in a temperature bump at the end of charge. This self-termination can be slow sometimes, as in the cell we interrupted during the rest-period dependent test due to reaching the safety limit.

(5)

ECTR events have a severity independent of charge rate or pre-event cell impedance. The maximum temperature reached during ECTR frequently occurred during the CV charge stage (when the current is reduced) or even during the open-circuit rest after charge (see Figure 3b). This is in contrast to normal cell operation when the temperature during charge is tightly correlated with the applied current. Despite this, there is no correlation between the cell’s impedance 1 h after charge ended and the magnitude of the ECTR effect. The current- and impedance-independency implies the origin of ECTR is not caused by changes to the internal resistance of the cell.

(6)

ECTR events result in excess charge capacity. ECTR is accompanied with high excess charge capacity, which is defined as the difference between the first 1.25 A or 0.5 A cycle charge capacity and the second 0.5 A cycle charge capacity.

(7)

ECTR events do not influence the discharge capacity. As shown in Figure 1b,c, little difference was observed between the first and second discharge after an ECTR event. Discharges also only exhibit a temperature profile consistent with Joule heating from a constant internal resistance. This indicates that active material degradation resulting in loss of lithium inventory is not the root cause of ECTR.

Based on these features and the destructive analyses of cells with and without ECTR behavior, we propose that ECTR is due to an internal short circuit caused by lithium plating that can self-terminate, see the scheme in Figure 9. We propose that, in aged cells, local electrolyte deficiency, electrode delamination, and/or other defects promote the deposition of porous metallic lithium (e.g., dendrites) on the negative electrode [20], as shown in Figure 9a. Metallic lithium is unstable in Li-ion electrolytes and is usually covered by a thick passivation layer due to its reaction with the electrolyte. During continuous cycling, the region surrounding this deposit becomes depleted of lithium ions (Li⁺) and/or electrolyte solvent, which slows the rate of lithium plating, minimizing further lithium deposition and preventing the formation of an internal short circuit. However, if the cell is relaxed for a long enough time, the lithium ions and or solvent will reenter the depleted region, electrochemically reactivating the depleted region and allowing further deposition of metallic lithium. Moreover, because the cell is at low SOC, the region of the negative electrode with plated lithium would be lithium rich, increasing the likelihood that this region becomes fully lithiated, thus promoting further lithium plating (Figure 9b). As a result, during the first charge process after an extended rest, lithium dendrites will grow at a faster pace and be more likely to bridge the electrodes and trigger an internal short circuit. This is most likely to occur towards the end of the charging cycle, as the electrode volume expands during charge and the dendrites are most likely to form and grow at high states of charge.

Once a dendrite reaches the positive electrode, an internal short is formed, as shown in Figure 9c. Short-circuit current then flows through the dendrite from the positive to the negative electrode, heating the cell, and leading to a rise in temperature at the end of charge (ECTR). The parasitic short-circuit current leads to an apparent excess in charge capacity, which explains the ECTR feature (6). The amount of heating is driven by the resistance of the short circuit and the cell voltage, but is independent of the current supplied by the charging circuit, therefore ΔT_max is not related to the total charge current and the cell impedance, as described in the ECTR feature (5). Because metallic lithium is unstable in the electrolyte, the dendrite forming the short circuit will continue to react with the electrolyte, resulting in the formation of a passivation layer on the surface of the short-circuiting dendrite(s). In addition, since the graphite on the negative electrode may not be fully lithiated and the metallic lithium in the dendrites has a higher free energy than lithium intercalated into graphite. Driven by thermodynamic force, the lithium dendrite near the negative electrode will be thermodynamically driven to intercalate into the surrounding partially-lithiated graphite. The reaction and redistribution of lithium metal will lead to the necking of lithium dendrite, increase the resistance of lithium short path, and consequently reduce the short-circuit current, resulting in the temperature bumps shown in Figure 1, Figure 4 and Figure 7. Ultimately, the short circuit will reach a high enough resistance that the cell reaches the CV cutoff current and charging stops, resulting in apparent self-termination (ECTR feature (4)). This is also distinct from lithium short circuits that are often believed to disappear during discharge rather than during charge [21]. Ultimately, the reaction and redistribution of lithium causes the lithium dendrite to become isolated from one or both electrodes (as shown in Figure 9d), terminating the short circuit and leaving dead lithium pieces in the separator.

Meanwhile, reactions between the electrolyte and dead dendrite consumes the surrounding electrolyte. This process is purely chemical and will not directly influence the normal discharge process except for the loss of a small amount of discharge capacity, which is consistent with the ECTR feature (7). The passivation reaction will locally deplete the concentration of lithium ions and/or solvent, reducing electrochemical activity. Note that this behavior does not rely on a short circuit ever having been formed; cells with high-rate triggered ECTR are likely forming lithium dendrites at lower rates that do not fully bridge the two electrodes. Subsequent charging at the same rate with minimal relaxation at low SOC is therefore less likely to form a dendrite at the same location. Therefore, no ECTR is observed in the second 0.5 A or 1.25 A charge in Figure 1b,c, consistent with ECTR feature (1). If a higher charge rate (such as 1.25 A) is applied, the higher overpotential will cause more severe lithium plating [20]. As a result, cells with ECTR triggered at a low rate will be retriggered with a new short circuit formed forming at a similar location (Figure 9d,e, corresponding to the case in Figure 1b), and cells with no ECTR at a low rate may be triggered at the higher rate (Figure 9a–e, corresponding to the case in Figure 1c). Such rate dependency can explain the ECTR feature (2).

Finally, the formation and penetration of lithium dendrites have some uncertainty and are highly related to the aging history (in particular, cells charged at higher rates and at relatively lower temperatures are more susceptible to lithium plating), therefore the corresponding ECTR will also show some randomness in intensity and variability between cells of differing histories. In addition, cells which are severely degraded may no longer have enough capacity to form dendrites capable of forming a short circuit, which is consistent with the reduced occurrence of ECTR in cells with very low capacity as shown in Figure 2d and Figure 8c. These behaviors are all consistent with ECTR feature (3).

We emphasize that, although we found that ECTR events usually self-terminate before catastrophic cell failure happens, the occurrence of ECTR is still a warning sign of poor cell health. The cells exhibiting strong ECTR in the controlled aging study exhibited rapid capacity loss coincident with the most severe ECTR events, suggesting that cells exhibiting ECTR will rapidly lose capacity even if the risks of ECTR are mitigated. We also cannot exclude the possibility that particularly severe ECTR could lead to catastrophic failure of the cell, which may raise safety concerns for the reuse of retired cells. To reduce the risk of an unpredictable lithium short circuit, we suggest integrating a system to monitor ECTR (by temperature or charge capacity cutoff) into the battery management system to better evaluate and retire aged cells in a timely fashion. We also suggest adding the inspection of potential ECTR into any state-of-health evaluation process before reusing retired cells and disqualifying any cells which exhibit ECTR behavior.

5. Conclusions

In this paper, we discovered and characterized a phenomenon we identified as end-of-charge temperature rise, or ECTR, in previously used and laboratory-aged 18650 lithium-ion cells. This phenomenon presents obvious safety concerns (we observed a maximum temperature during testing of 80 °C) that could be widespread in different types of lithium-ion batteries, which could impede adoption of programs to repurpose batteries after initial retirement. Elevated charge rate and extended storage at low state-of-charge (SOC) can increase the occurrence of ECTR. The intensity of ECTR was found to correlate with excess charge capacity but not with the charge current or cell impedance. Occurrence of ECTR was weakly dependent on the remaining cell capacity or internal impedance in reused cells, while the controlled aging study showed that different aging conditions dramatically altered the severity and occurrence of ECTR. Thus, some commonly-used state-of-health (SOH) evaluation methods such as the measurement of usable capacity or internal resistance are not sufficient to effectively evaluate the health of used cells. Further studies are needed to explore if more sophisticated SOH evaluation methods, such as dQ/dV, equivalent circuit modelling, or machine-learning are capable of identifying cells susceptible to ECTR. We attribute the origin of ECTR to an internal short circuit caused by unstable lithium dendrites that form during charge and self-terminate over time. Further study of this phenomenon is needed, for example, to learn the dependency of ECTR on cell formats and active material types. Systems to detect ECTR events in lithium-ion batteries should be developed to detect and safely retire and dispose of cells that are susceptible to ECTR.