Lithium-Ion Battery Degradation: Measuring Rapid Loss of Active Silicon in Silicon–Graphite Composite Electrodes

To increase the specific energy of commercial lithium-ion batteries, silicon is often blended into the graphite negative electrode. However, due to large volumetric expansion of silicon upon lithiation, these silicon–graphite (Si–Gr) composites are prone to faster rates of degradation than conventional graphite electrodes. Understanding the effect of this difference is key to controlling degradation and improving cell lifetimes. Here, the effects of state-of-charge and temperature on the aging of a commercial cylindrical cell with a Si–Gr electrode (LG M50T) are investigated. The use of degradation mode analysis enables quantification of separate rates of degradation for silicon and graphite and requires only simple in situ electrochemical data, removing the need for destructive cell teardown analyses. Loss of active silicon is shown to be worse than graphite under all operating conditions, especially at low state-of-charge and high temperature. Cycling the cell over 0–30% state-of-charge at 40 °C resulted in an 80% loss in silicon capacity after 4 kA h of charge throughput (∼400 equiv full cycles) compared to just a 10% loss in graphite capacity. The results indicate that the additional capacity conferred by silicon comes at the expense of reduced lifetime. Conversely, reducing the utilization of silicon by limiting the depth-of-discharge of cells containing Si–Gr will extend their lifetime. The degradation mode analysis methods described here provide valuable insight into the causes of cell aging by separately quantifying capacity loss for the two active materials in the composite electrode. These methods provide a suitable framework for any experimental investigations involving composite electrodes.

surfaces other than the base of the cell. The aluminium block at the base of the cell was held at a constant temperature using a PID controller on an Arduino Mega 2560. K-type thermocouples (TCs) provided the feedback for the control loop, which uses a Peltier element (20 W single-stage thermoelectric cooler, European Thermodynamics) to heat or cool the aluminium block based on the output current signal. Excess heat from the reverse side of the Peltier elements was extracted using an aluminium water-circulated cooling plate. The water was pumped through the cooling plate at a constant temperature using a CW-5200 industrial chiller. Thermal paste (10 W m -1 K -1 , Fischer Elektronik) or thermal interface material (12.5 W m -1 K -1 , t-Global Technology) were used at the interfaces between each component of the thermal apparatus to improve heat transfer. Cells were forced downwards onto the thermal interface block by the springs which held the electrical connections in place.
Surface temperature measurements of the cells were made using two K-type TCs per cell: one was recorded using a Pico TC-08 data logger (sampling frequency of 0.2 Hz), the other was recorded by the battery cycler whilst being used as its safety control. The TCs were adhered to the surface of the cell using Kapton tape, positioned halfway along the axial direction of the cell. The test rigs were found to be capable of holding the cells at stable temperatures of 5°C to 45°C, with fluctuations of ± 0.1°C while at rest.

S3
Both the positive and negative electrical connections were made via the top of the cell. The negative connection was made via an aluminium, ring-shaped terminal which sat on the outer circumference of the top of the cell. The positive connection was made using an aluminium pin terminal which made contact with the positive cap. Both of these terminals were encased in plastic to prevent short-circuiting. The aluminium terminals were sanded and cleaned with isopropanol prior to use. Cell terminals were also cleaned with isopropanol. The connections were forced down into the top of the cells using springs which were anchored to the cooling plate at the base of the cells.
The terminals were connected to the battery cycler (Biologic BCS-815) current and voltagesense cables via banana plugs. Cell impedance was tested after set-up by using the iRcompensation function of the BCS-815 at a frequency of 10 kHz and amplitude of 5 mV, averaging over 4 measurements. The total impedance (including ohmic resistance from the cell itself) was consistently found to be 27-30 mΩ; the cell itself was found to be 27 mΩ when measured using a 4-point connection, indicating the electrical connections added < 3 mΩ. The resistance was not compensated for in subsequent tests.

Break-In Cycles
The break-in cycles consisted of cycling cells between the upper and lower voltage limits (4.2 V and 2.5 V) using the standard charge and discharge procedures outlined in the manufacturer's specification sheet. These were a constant-current, constant-voltage (CC-CV) charge (with a Crate of 0.2C until 4.2 V, and a 4.2 V hold until the current dropped below 0.01C), and a CC discharge (with a C-rate of 0.2C). Cells were rested under open-circuit conditions for 2 hours after each charge and 4 hours after each discharge. All break-in cycles were performed at 25°C.

Reference Performance Test (RPT)
As a balance between gaining important information on the performance of the cell and not expending too much time (and charge) outside of the aging sets, two different RPT procedures were used in this study. The longer procedure was performed after each even-numbered aging set, whereas the shorter procedure was used after each odd-numbered aging set. Both procedures were run at BoL after the break-in cycles had been performed. RPTs were always performed at 25°C.
Once the RPT had finished, prior to the next aging set commencing, cells were returned to approximately 50% SoC by charging at 0.3C until the cell voltage reached 3.7 V. The two procedures are detailed in Table S1 & S2, and shown visually for an example cell in Figure S4 & S5.
S8 Table S1. The long version of the RPT procedure, run on every cell after each even-numbered aging set.

Sub-Test
Step  Figure S4. The longer version of the RPT procedure, showing current (top), voltage (mid), and temperature (bottom) versus time. This procedure was performed on every cell after each evennumbered aging set. Colors correspond to the various sub-tests described in Table S1. S12 Figure S5. The shorter version of the RPT procedure, showing current (top), voltage (mid), and temperature (bottom) versus time. This procedure was performed on every cell after each oddnumbered aging set. Colors correspond to the various sub-tests described in Table S2.

Data Analysis
All analyses of the electrochemical data were performed using Python, making use of the Pandas, Numpy, Scipy, and Matplotlib libraries. Measurements of charge throughput were taken from the aging cycle data; this corresponds to the total measured charge passed during the aging cycles (from both charge and discharge sections). Aside from the measure of charge throughput, all other information used in the rest of the analyses utilize data from the RPTs, as described in the main text. S13

Cell-to-Cell Variation
To compare the variability between cells at BoL, the 0.1C discharge capacities for 40 cells were compared. These data were recorded on cells which had undergone break-in cycles before being tested as per the 'long' RPT procedure described above. This revealed an average capacity of 4864.91 mA h, with a standard deviation of 20.52 mA h (0.42% of mean).

Aging Cycles
The conditions used in the aging cycles are described in Table 1 and 2 of the main text. Table   S3 below summarises the different conditions and the number of cells tested under each. Figure   S6 shows the current and voltage profiles for some example aging cycles for the two SoC ranges used in this study.

DM Analysis
To calculate the fractional capacity of graphite (Gr) and silicon in the negative electrode (NE), we compared the measured Si-Gr voltage (V) vs capacity (Q) delithiation profile for the NE (taken from reference [1]) against separate measured V vs Q profiles for Si and Gr delithiation (refs. [2] and [3], respectively). As described in the main text, the capacity of a composite electrode at any given V is the sum of the constituent parts. By altering the Gr:Si ratio for the calculated Q vs V curve, we can obtain a fit for the experimentally measured Q vs V curve for the Si-Gr NE ( Figure   S7). This was done using the curve_fit function of the SciPy library in Python, using a Trust Region