Layered Oxide, Graphite and Silicon-Graphite Electrodes for Lithium-Ion Cells: Effect of Electrolyte Composition and Cycling Windows

Table I lists details of the cell materials used in our experiments. Both positive and negative electrodes are from CAMP. The NCM523 electrode contains Li_1.03(Ni_0.5Co_0.2Mn_0.3)_0.97O₂ as the active material, C-45 carbons for enhanced electron conduction and polyvinylidene fluoride (PVDF) binder. The Gr and Si-Gr (15 wt% silicon) electrodes contains C-45 carbons and partially-lithiated polyacrylic acid (LiPAA) binder (pH ∼ 7); the silicon particles are nanosize, in the 50 to 70 nm range. Our baseline electrolyte (Gen2) contains 1.2 M LiPF₆ in ethylene carbonate (EC): ethyl methyl carbonate (EMC) 3:7 by weight; FEC, VC or a mixture of FEC and VC are added to the baseline electrolyte to examine their effects on cell performance. Note that the electrolyte density and salt concentration changes because of these additions (see Table II).

The electrochemical tests were conducted either in 2032-format stainless steel coin cells or in three-electrode cells containing a Li-metal reference electrode (RE). All cells were assembled in a glove box under argon atmosphere (O₂ < 1 ppm). Before cell assembly the electrodes were dried overnight in a vacuum oven; the NCM523 and Gr electrodes were dried at 110°C and the LiPAA containing Si-Gr electrodes at 150°C. The Celgard 2325 separators were dried for approximately three hours in a vacuum oven set at 75°C.

In a half-cell configuration the coin cells contained either NCM523, Gr, or Si-Gr electrodes (1.6 cm² area) and a Li-metal counter electrode; the cells were cycled at a low rate (< C/10) to examine potential profiles of the active materials. In a full-cell configuration the coin cells contained NCM523 as the positive electrode and either Gr or Si-Gr as the negative electrode; a single layer of Celgard 2325 separated the 1.6 cm² area electrodes. All coin cells were contained in a Cincinnati Sub-Zero MicroClimate Model MCH-3 temperature chamber at 30°C and cycled with a Maccor Model 2300 battery test system. Full cell cycling performance was evaluated in various voltage windows between 2.5 and 4.4 V. The cell aging protocol consisted of three slow cycles at an approximate C/20 current rate (0.05 mA cm⁻²), 94 faster cycles at an approximate C/3 rate (0.33 mA cm⁻²), and then three additional slow cycles (0.05 mA cm⁻²) resulting in a total of 100 cycles. To examine the effect of voltage hysteresis, DC impedance of some full cells was obtained by applying a 3 C, 10 s charge or discharge pulse at predetermined cell voltages along both the charge and discharge curves; at each voltage, the cell was allowed to equilibrate for 1 hour prior to the 10 s pulse. In separate measurements AC impedance data were also obtained at these voltages with a Solartron Analytical 1470E cell test system.

The three-electrode cells were used to examine potential profiles of the positive and negative electrodes while the cell was cycled in various voltage windows between 2.5 and 4.3 V. The experimental set-up is described in more detail elsewhere.^25,28 In brief, the electrode area in this set-up is 20.3 cm² for both positive and negative electrodes, and two Celgard 2325 separators are used. The Li-metal RE, in contact with the electrolyte at the side of the electrode sandwich, provides accurate values of the electrode potentials when the cell is at open-circuit (zero current) and reasonable values at low cycling currents (< C/20). At higher currents, geometrical edge effects due to small electrode misalignments have a larger effect and the assignment of electrode potential contributions to cell voltage becomes increasingly inaccurate, as described previously.²⁵

The first two cycles of the NCM523, Gr, and Si-Gr electrodes, in a cell with a Li-metal counter electrode and 10% FEC added to Gen2 electrolyte, are shown in Figures 1 and 2. The NCM523 potential profile (Figure 1) displays two regions with distinct slopes and a discharge capacity of 198 mAh g⁻¹ when cycled between 3.0 and 4.5 V vs. Li. The coulombic efficiencies are 90.1 and 99.7% for the first and second cycles, respectively; values exceeding 99.9% are obtained in succeeding cycles (not shown here). Electrolyte composition has a negligible effect on the NCM523 potential profiles; i.e., similar profiles are obtained in the Gen2 electrolyte and other electrolyte variations considered in this study.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Figures 2a–2f contain capacity and voltage profiles of the Gr and Si-Gr electrodes for the first two cycles. Note that the Gr electrode profiles are similar for the two cycles, while the Si-Gr electrode lithiation profiles are distinctly different. The 15 wt% Si addition clearly increases electrode capacity, resulting in a specific (delithiation) capacity of 816 mAh g⁻¹ compared to 362 mAh g⁻¹ for the Gr electrode for the second 1.5–0.0 V cycle; this specific capacity is calculated based on the Si, graphite, and C45 carbon weight. The coulombic efficiencies of both first and second cycles are lower for the Si-Gr electrode; the first cycle efficiency is 90.4% compared to 93.6%, and the second cycle efficiency is 98.0% for the second cycle compared to 99.0% for the Gr electrode. Features associated with FEC reduction are seen in both the Gr/Li (Figure 2b, ∼1.2 V) and Si-Gr/Li (Figure 2d, ∼1.1 V) cells; this FEC reduction minimizes reduction of the EC component, which is typically observed at ∼0.7 V for cells with the Gen2 electrolyte. A similar behavior is observed for cells containing significant quantities (10 wt%) of VC added to the baseline electrolyte, which is not surprising because of the similarities in FEC and VC reduction characteristics.²⁹

The Gr electrode (Figure 2a) displays distinct plateaus during the Li-ion intercalation/deintercalaction reactions; the voltage profiles are altered by silicon addition. During first lithiation of the Si-Gr electrode (Figure 2c), the first graphite lithiation plateau is still visible (∼0.16 V) before the characteristic low voltage gently-sloping plateau (around 0.1 V), associated with the initial lithiation and alloying of Si^30,31 becomes dominant. The 0.16 V plateau is at a lower potential than that typically expected for graphite (0.21 V, see later). However, the plateau is not observed in the Si-only (no graphite) electrode data and must, therefore, arise from the graphite component; the lower potential may be a consequence of higher ohmic polarization (higher electrode impedance) during the first cycle. Also, no obvious plateau associated with the Li₁₅Si₄ phase is observed; this plateau may be shifted to values below our lower cutoff potential because of stresses remaining in the nanosilicon particles.^32,33 Following the first lithiation, the addition of silicon increases the average voltage over the electrode capacity compared to the graphite electrode, as expected, both for lithiation and delithiation. A clear contrast between the sharp plateaus of the graphite electrode and the sloping profile of amorphous silicon is observed. Furthermore, a large voltage hysteresis between lithiation and delithiation results from Si introduction.

This hysteresis is clearly evident in Figure 3a, which compares the lithiation and delithiation curves of the Gr and the Si-Gr electrodes obtained at a very slow (<C/100) rate; the capacity values shown are normalized to graphite content in the electrodes. The corresponding differential capacity plots are shown in 3b. The data are from the fifth cycle, which provides features related to structural changes in the active material while minimizing features arising from electrolyte reduction. Two plots are shown for the Si-Gr electrodes; the lower cutoff potential (LCP) for the plots are 0.075 V (magenta) and 0.0 V (blue), respectively.

Zoom In Zoom Out Reset image size

During lithiation of the Gr electrode distinct plateaus are observed at 0.21 V, 0.12 V, and 0.083 V (Figure 3a), which are seen as peaks in Figure 3b. The Si-Gr electrode lithiated to 0.075 V also contains these features, and additional capacity in broad features above 0.21 V and as a broad background to the graphite between 0.21 and 0.083 V (Figure 3b). An additional feature at 0.1 V, between the plateaus at 0.12 V and 0.083 V, arises from the lithiation of silicon that continues during the stage transitions in graphite. On delithiation the Gr electrode shows features at 0.096 V, 0.136 V, and 0.22 V; these features are also observed for the Si-Gr electrode. On lithiation down to 0.0 V, the Si-Gr electrode shows a feature at 0.053 V (blue curve, Figure 3), and another feature at approximately 0.4 V vs Li/Li⁺ in the subsequent delithiation cycle. These features are indicative of the crystalline Li₁₅Si₄ phase, which forms (at 0.053 V) during lithiation of amorphous α-Li_xSi and dissipates (at 0.4 V) on delithiation.^34,35 The onset of the crystalline phase is believed to negatively impact the cycling behavior of Si electrodes,²⁰ and limiting the capacity to avoid its formation has been suggested as a viable strategy to enhance cycle life.²⁰ Note that the 0.4 V feature during delithiation does not appear when the lower cutoff potential is restricted to 0.075 V (magenta curve, Figure 3). However, the potential hysteresis is evident even without formation of the crystalline Li₁₅Si₄ phase. The difference in potential characteristics resulting from limiting the lower cutoff potential is known for Si electrodes, for example;³⁴ it is simply presented here for completeness.

An important consideration when blending multiple active materials is the utilization of each material at different potentials, and the relationship between potential and electrode state-of-charge (SOC). Due to the large potential hysteresis of silicon (which is not present for graphite), and the shift of electrode capacity to higher potentials when silicon is utilized, it follows that during lithiation, much of the silicon capacity is accessed prior to graphite, i.e. >0.2 V. Below this potential both silicon and graphite (see plateaus) are active. Similarly, during the delithiation cycle, graphite is delithiated prior to silicon as the potential is increased. More notably, this is the case not only for the 0.075 V cutoff but also for the 0.0 V cutoff (magenta and blue curves in Figure 3); even with additional lithiation capacity arising from the silicon component below 0.053 V, lithium is extracted from the graphite first during the following delithiation. The large potential hysteresis of the silicon, with lithiation at <0.053 V and delithiation at ∼0.4 V for the Li₁₅Si₄ phase, means that at a given electrode potential below 0.4 V, one cannot discern state-of-charge (SOC) of the electrode components by simply monitoring the potential, without a knowledge of prior cycling conditions.

For capacity matching in full cells, the electrode loading and areal capacities (mAh cm⁻²) needs to be considered in addition to gravimetric capacities of the active materials. Our electrodes were designed to provide a surplus (initial) areal capacity at the negative electrode; the Si-Gr electrode delivers 2.5 mAh cm⁻² in the 1.5–0.0 V voltage range, while the NCM523 electrode delivers 2.1 mAh cm⁻² in the 3.0–4.5 voltage range. To determine the actual electrode capacity match after formation cycling, which causes electrode potential window shifts because of the SEI-forming side reactions, a three-electrode set-up was used. Figure 4 displays the positive and negative electrode potentials when the cell is cycled between different voltage limits; both the lower cutoff (2.5 V and 3.0 V) and the upper cutoff limits (4.1 V, 4.2 V, and 4.3 V) have been varied. The electrode potentials are monitored simultaneously with a Li-metal RE during the galvanostatic cycling while the cell voltage is varied. The experiments are conducted with low currents (<C/20) to reduce measurements errors introduced from RE placement, and whose absolute potential error depends on the size of the cycling current.²⁵

Zoom In Zoom Out Reset image size

In Figure 4 consider the cells cycled in the 2.5–4.1 V, 2.5–4.2 V and 2.5–4.3 V ranges. At the upper cutoff voltages (UCV) of 4.1, 4.2 and 4.3 V, the NCM523 electrode potentials (Figure 4b) are 4.21 V, 4.3 V and 4.39 V and the Si-Gr electrode potentials (Figure 4c) are 0.11 V, 0.10 V and 0.09 V vs. Li/Li⁺, respectively. In all cases, the Si-Gr electrode potential remains above 0.053 V; hence formation of the Li₁₅Si₄ crystalline phase is not expected, and accordingly not observed in the following cell discharge cycle (i.e., no 0.4 V plateau in Figure 4c). On discharge to 2.5 V, the NCM523 electrode potential is 3.62 V and the Si-Gr electrode potentials is 1.12 V, respectively. On discharge to 3.0 V the NCM523 electrode potential is 3.69 V, cell capacity is reduced by ∼10%, and the Si-Gr electrode potential reaches only 0.69 V. The likely consequences of these differing Si-Gr electrode potentials (1.12 V vs. 0.69 V) during full cell discharge will be discussed later. It is interesting to note that the potential profiles of the cell and electrodes are altered when the LCV is changed. This alteration is a consequence of the potential hysteresis inherent to the Si-Gr electrode and from the fact that the full cell voltage results from the difference between the positive and negative electrode potentials.

The initial impedance (after formation cycling) as a function of cell voltage is shown in Figures 5 and 6 for full cells containing the Gen2 + 10 wt% FEC electrolyte. In Figure 5 the DC impedance data was obtained from 10 s 3C pulses during both charge and discharge cycles. Figure 5a contains data from charge pulses, during which Li⁺ ions are extracted from the positive and inserted into the negative electrode. Figure 5b contains data from discharge pulses, during which Li⁺ ions are extracted from the negative and inserted into the positive electrode.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Data from a NCM523/Gr cell are compared with those obtained from a NCM523/Si-Gr cell; the addition of silicon induces hysteresis in the voltage profile, which results in impedance differences between the charge and discharge cycle at the same cell voltage. For example the charge-pulse impedance data for the NCM523/Gr cell (Figure 5a) are very similar during the full cell charge and discharge cycles. The area specific impedance (ASI) values are ∼29 Ω cm² between 4.1 and 3.6 V, and then rise almost linearly to ∼70 Ω cm² at around 3.2 V; the increase at low cell voltages reflects the difficulty of extracting Li⁺ ions from the relatively-full NCM523 oxide. The increases at the lower cell voltages is also seen for the NCM523/Si-Gr cell; however, this cell also shows bigger differences between the values during charge and discharge cycles resulting from the larger difference in electrode state-of-charge (SOC) at the lower cell voltages. The larger differences at low cell voltages for the NCM523/Si-Gr cell are also seen in Figure 5b, which contains impedance data from the discharge pulses.

The effect of voltage hysteresis is also seen when comparing AC impedance data at the same cell voltage for the charge or discharge cycles. In Figure 6, impedance spectra around the voltage loop are displayed, contrasting data from the charge and discharge cycles; for ease of comparison, the spectra are grouped to show low cell voltages (3.3, 3.4 V), intermediate cell voltages (3.5 V–3.8 V), and high cell voltages (3.9, 4.0 V) in Figures 6a, 6b, and 6c, respectively. Again, the largest difference in impedance between charge and discharge at a given voltage is seen at the lower cell voltages; for example, see the 3.3 V and 3.4 V data in Figure 6a. The difference arises because of the voltage hysteresis, as 3.3 V corresponds to a lower positive electrode SOC (higher Li content in oxide) during cell charge than at discharge.

A closer look at the AC impedance spectra reveals a high-frequency arc, a mid-frequency arc, and a diffusion tail. It is evident that the high-frequency arc is not affected by the cell voltage. The mid-frequency arc arises from NCM523 electrode; at these frequencies the Si-Gr electrode shows a minimum in the Nyquist representation.²⁵ The higher mid-frequency impedance at 3.3 V (charge cycle) is a consequence of the higher Li content in the NCM523 oxide, because Li⁺ ion conductivity is lower for the fully-lithiated oxide compared to the partially-lithiated oxide (at higher potentials).³⁶ As cell voltage is increased above 3.4 V (compare Figs. 6b to 6a) the overall difference in impedance between charge and discharge decreases for potentials 3.5–3.6 V; for 3.7–3.8 V there is only a small variation both between the voltages and between charge and discharge. The variation in the diffusion tail in the full cell spectra originates mostly from the Si-Gr electrode, which was previously shown to have small variations with SOC at high- and mid-frequencies, but a stronger dependency at low frequencies.²⁵

Full-cell data from NCM523/Si-Gr coin cells with different electrolyte compositions, cycled in the 2.5–4.1 V range, are presented in Figure 7. Note that the electrolytes can be grouped into three categories: Gen 2, Gen2 + 5 wt% additive (5 wt% VC, or 5 wt% FEC, or 2.5 + 2.5 wt% of VC and FEC), and Gen 2 + 10(or) 20 wt% (FEC or VC) additives. Figure 7a shows discharge capacity vs. cycle number of representative cells. The cell capacities are bunched together initially but then start diverging. The first cell to show faster capacity fade (around 15 cycles) is the one containing the Gen2 electrolyte; the cells containing the 5 wt% additives are the next to show accelerated fade (between 40 and 50 cycles). The cells with the 10 and 20 wt% additives show similar behavior up to the 100 cycles.

Zoom In Zoom Out Reset image size

Another representation of the cycling data is shown in Figure 7b, which shows the initial discharge capacity (cycle 1) of cells plotted as a function of capacity retention (comparing cycles 3 and 100 capacities, obtained at low rates); the data point and error bars represent the average and 2 standard deviations (2σ) from 2–5 cells. Ideally, for each cell, the initial discharge capacity and capacity retention should be as high as possible. However, Figure 7b shows that the initial discharge capacity decreases when the additive content is increased. For example, the initial discharge capacity is lower (than that of the Gen2 cell) by ∼3 mAh g⁻¹ at 5 wt% additive content, and ∼5–9 mAh g⁻¹ at 10 and 20 wt% additive contents. Note that g (grams) refers to oxide content in the positive electrode, here and henceforth for full cell capacity (or current) data. The lower initial capacity suggests that Li⁺ ions are trapped in the SEI formation products and hence unavailable for further cycling. On the other hand, capacity retention is enhanced by the additives: the retention improves from 8% for Gen2 to ∼15–20% at 5 wt% additive content and further to ∼46–50% at 10 and 20 wt% additive contents.

Both Figures 7a and 7b suggest that cell capacity retention depends on additive content – higher the additive content, better the retention. However, capacity retention does not improve when the FEC content is increased from 10 to 20 wt%. This result may indicate that the 10% FEC is not depleted from the electrolyte during the 100 cycles; further cycling may deplete the FEC and accelerate capacity fade. This observation was made by Jung et al.³⁷ for silicon-carbon electrodes cycled vs lithium metal, who noted that cell cycle life relates to the FEC content and that the capacity drops rapidly as the additive is consumed.^26,37 Therefore, cycling differences between our 10 to 20 wt% FEC electrolytes could become prominent on extended cycling. If FEC is indeed consumed then even higher amounts could further delay capacity fade, though without fully suppressing the problems associated with side reactions at the dynamic silicon surface. Furthermore, for pure FEC, we noted a tendency toward the formation, aggregation, and phase separation of Li⁺ coordination polymers when the LiPF₆ concentration exceeds ∼0.8 M in the electrolyte.³⁸ However, we have shown previously that VC and FEC have other benefits; the compounds improve electrode cohesion thereby preventing loss and isolation of electrochemically active silicon in the composite electrode.²⁵

A comparison of the 10 wt% VC and 10 wt% FEC data (Figure 7b) reveals a slightly better capacity retention for the VC-bearing cell. A similar result was observed during the cycling of NCM523/Si-Gr cells with an LiFSI-electrolyte containing either 10 wt% FEC or VC.¹⁸ This result may simply be a consequence of the difference in molecular weights; VC has lower molecular weight than FEC and hence will have a higher molarity in the electrolyte (see Table II). Alternatively, the better performance of VC could be related to the large amount of graphite in the Si-Gr electrode. In a study of pure silicon electrodes, VC induced no improvement in cycle performance while increasing amounts of FEC substantially reduced capacity fade.¹⁶ However, VC rather than FEC has been shown to improve capacity retention in graphite-containing cells.^14,39

The effect of electrochemical cycling windows were examined in NCM523/Si-Gr coin cells containing the Gen2 + 10 wt% FEC electrolyte; the data are presented in Figure 8. Figure 8a shows the initial discharge capacity vs. capacity retention plot; the costs and gains of varying the cycling voltage limits are evident from this plot. For example, compare data for the 2.5–4.1, 2.8–4.1 and 3.0–4.1 V cycles. It is clear that the initial discharge capacity decreases from ∼154 mAh g⁻¹ to ∼142 mAh g⁻¹ when the LCV is increased from 2.5 V to 3.0 V; however, the capacity retention improves from 47% to 57%. That is, the narrower voltage window provides less capacity (smaller ΔSOC), but leads to improved capacity retention over the 100 cycles.

Zoom In Zoom Out Reset image size

On the other hand, cell capacity (and energy density) can be increased by raising the UCV. For example, compare data for the 3.0–4.1, 3.0–4.2, 3.0–4.3 and 3.0–4.4 V cycles in Figure 8a. The initial discharge capacity of the 3.0–4.2 V cell (∼152 mAh g⁻¹) is similar to that of the 2.5–4.1 V cell. Raising the UCV further to 4.3 V and 4.4 V increases the capacity to ∼162 mAh g⁻¹ to ∼174 mAh g⁻¹, respectively; that is, a gain of 32 mAh g⁻¹ is seen when the UCV is raised from 4.1 V to 4.4 V. The corresponding capacity retention values over the 100 cycles are ∼57%, ∼54%, ∼55%, and ∼52% for UCV's of 4.1, 4.2, 4.3 and 4.4 V, respectively. That is, the increased cycled capacity (wider ΔSOC) in the direction of higher cell voltages has less impact on capacity retention than a wider ΔSOC toward lower cell potentials.

As indicated above raising the UCV to 4.4 V has less impact on capacity retention than lowering the LCV to 2.5 V. Earlier, from the RE study (Initial electrochemical performance - RE cells section), we showed that lowering the full cell LCV from 3.0 V to 2.5 V raises the Si-Gr electrode potential from 0.69 V to 1.12 V. The consequences of this potential rise could include (i) enhanced fracture of the nanosilicon as the particles contract further during delithiation, and (ii) SEI decomposition at high Si-Gr electrode potentials, requiring lithium-consuming side-reactions during the following cycle to form a passivation layer, which increases cell capacity loss.

To further explore these possibilities we plot Figure 8b, which shows total discharge capacity throughput (summed over the 100 cycles) as a function of capacity retention. If particle fracture due to contraction/expansion is the dominant phenomenon then we would expect the cells with a lower discharge capacity throughput (less volume expansion/contraction) to have higher capacity retention. Figure 8b shows that the 2.5–4.1 V cells have a lower capacity throughput but also a lower capacity retention than the 3.0–4.1 V cells. That is, the lower volume expansion/contraction does not correlate to improved capacity retention. Furthermore, the likelihood of particle fracture, during delithiation (contraction) of our nanosize amorphous silicon (50–70 nm), is small due to the small particle size.⁴⁰ Hence, we posit that SEI breakdown at high Si-Gr electrode potentials during cell discharge accelerates capacity loss; additional experiments are underway to confirm this hypothesis.