The influence of electrochemical cycling protocols on capacity loss in nickel-rich lithium-ion batteries

The transition towards electric vehicles and more sustainable transportation is dependent on lithium-ion battery (LIB) performance. Ni-rich layered transition metal oxides, such as NMC811 (LiNi0.8Mn0.1Co0.1O2), are promising cathode candidates for LIBs due to their higher specific capacity and lower cost compared with lower Ni content materials. However, complex degradation mechanisms inhibit their use. In this work, tailored aging protocols are employed to decouple the effect of electrochemical stimuli on the degradation mechanisms in graphite/NMC811 full cells. Using these protocols, impedance measurements, and differential voltage analysis, the primary drivers for capacity fade and impedance rise are shown to be large state of charge changes combined with high upper cut-off voltage. Focused ion beam-scanning electron microscopy highlights that extensive microscale NMC particle cracking, caused by electrode manufacturing and calendering, is present prior to aging and not immediately detrimental to the gravimetric capacity and impedance. Scanning transmission electron microscopy electron energy loss spectroscopy reveals a correlation between impedance rise and the level of transition metal reduction at the surfaces of aged NMC811. The present study provides insight into the leading causes for LIB performance fading, and highlights the defining role played by the evolving properties of the cathode particle surface layer.

.   Taking the derivative of the voltage with respect to capacity yields The expression in Equation S2 shows that dV/dQ contributions from the cathode and anode electrodes sum linearly to give the dV/dQ of the cell. Therefore, DVA involves fitting half-cell anode and cathode dV/dQ data to the full cell dV/dQ data and allowing the relative capacity alignment to change. Additional parameters can also be introduced, as necessary, to account for other processes that affect the voltage-capacity profiles. For example, capacity loss terms for the cathode and anode can be included to quantify the loss of active sites for ion insertion at each electrode. These terms apply a scalar multiplier to the half-cell capacity, effectively compressing the dV/dQ curve horizontally. Therefore, the results from the DVA fitting quantify the capacity loss attributable to degradation at the cathode, anode, and from electrode slippage, decoupling the contribution of each from the measured full cell capacity loss.
To illustrate the approach taken in this work, a full cell voltage profile during charge at C/20 is plotted versus capacity in Figure S3, together with the capacity normalized differential voltage (Q dV/dQ). The full cell data shown in Figure S5 is for a cell aged by constant currentconstant voltage cycling (CYC) in a 2.5-4.2 V window. Half-cell data collected separately in Li/NMC811 and Li/graphite half-cells are also shown. A linear combination of the dV/dQ halfcell data as per Equation S2 generates the full cell model profile, which is fitted to the full cell raw data using a least squares approach. Three parameters are allowed to refine during the data fitting: electrode slippage, cathode capacity loss, and anode capacity loss. By comparing the fitted parameters for the diagnostic cycle after aging with those in the formation cycle before aging, the amount of capacity loss arising from each of the above listed processes can be quantified. For the cell shown in Figure S5, the fitted parameters for before and after aging are listed in Table S1, together with their conversion into capacity loss terms with units in mAh g -1 NMC . Finally, based on the fitting result, the features in the full cell dV/dQ profile in Figure   S5b are assigned as originating from either the cathode (c n ) or anode (a n ), which highlights which electrode(s) dominate(s) the voltage profile in a given region. differential voltage versus capacity for the third C/20 diagnostic cycle after aging by CYC 2.5-4.2 V (black), together with the modelled profiles from differential voltage analysis (red), the difference between the model and data (brown), and the NMC811 cathode (green) and graphite anode (blue) half-cell potential profiles used to generate the full cell model profiles. The derivative, (dV/dQ), in (b) is multiplied by the cell capacity (Q) in the given cycle, which normalizes the derivative based on the cell capacity. In (b), features in the dV/dQ profile are assigned to cathode (c n ) or anode (a n ). In this case, Gr deg is determined to be 2.3 mAh g -1 , however, this apparent decrease does not affect the full cell capacity since the storage capacity of the graphite anode is not limiting. It is also worth noting that in all other conditions (e.g.    Figure 3 and in Figure S6 are given. The capacity lost is measured between the third C/20 formation cycle and after aging in the third C/20 diagnostic cycle. .0 † Capacity from electrode slippage and NMC deg were determined from DVA. NMC deg is the fitted capacity lost attributable to degradation of the NMC cathode. Capacity lost due to impedance is taken as the difference in the CV charge capacity for the formation and diagnostic cycles, which both have a C/40 cutoff current. ‡ NMC HVdeg is the NMC cathode capacity loss that occurs specifically at NMC potentials >4.1 V vs Li/Li + , and is in addition to NMC deg which is lost evenly across the entire SOC. More details on how NMC HVdeg is calculated are provided in Supplementary Note S2. * Measured capacity loss is the difference in the charge capacity at the end of the CC segment for the formation and diagnostic cycles, at a C/20 rate.

Supplementary Note S2: Determining capacity lost due to NMC high voltage degradation
Aging by high voltage cycling (HVC 3.95-4.3 V) leads to a capacity mismatch between the data and the model derived from DVA (see Figure 3e, f and S5), with the model over-estimating the true capacity after aging. The model deviates from the data only at high SOC, where the NMC cathode dominates the voltage and dV/dQ profiles, suggesting a second NMC capacity loss process occurs specifically at NMC potentials >4.1 V vs Li/Li + -see Figure S5. This is capacity loss process is termed NMC high voltage degradation (NMC HVdeg ). Note that this is in contrast to NMC deg , which models capacity lost evenly across the entire SOC range. Thus capacity lost from NMC deg and NMC HVdeg are additive. The amount of capacity lost due to NMC HVdeg is determined by taking the difference in the modelled capacity and the measured CCCV capacity. The modelled capacity accounts for capacity lost from slippage and NMC deg and that on the CV step accounts for capacity lost due to impedance. We attribute the remainder to NMC HVdeg .

Supplementary Note S3: EELS data processing
The spectrum images (SI) were processed using HyperSpy (a Python library for multidimensional data analysis). First, any unexpected spikes (cosmic rays, stray X-rays hitting the detector) in the data were removed. Then, the zero-loss peak position for each pixel of the SI was used to shift both low-and core-loss spectra so the center of the zero-loss peak was at 0 eV energy loss. A power law background subtraction was implemented from the core-loss spectra, by fitting the region before the onset of the oxygen K-edge. The Ni L edge was used to separate bulk and surface contributions to the spectrum image. Briefly, the Ni L edge was fitted at each pixel with: i) a power law background, ii) two Hartree-Slater generalizedoscillator-strength based ionization edges, iii) two Gaussian peaks (for the L 3 and L 2 white line peaks), and iv) their convolution with the low-loss peak to account for the sample thickness.
Maps of Ni L 3 gaussian peak position were produced. The maps were then used to extract two binary masks for each spectrum image: surface (Ni L 3 peak position <855.2 eV) and bulk (Ni L 3 peak position >855.7 eV). Morphological operators were used as necessary to remove single pixels and fill in single-pixel gaps in the masks. The masks were used to average the raw data from the bulk and the surface, to give the data shown in Figure 6. Each of the averaged TM L edges (for all protocols, two particles for each, surface and bulk spectrum) were then fitted using the same approach as for the Ni edge fitting. This way, the positions of the Ni, Mn and Co L 3 peaks for the two regions were calculated (Figure 7). TM L 3 /L 2 edges correspond to transitions between 2p and unfilled 3d orbitals, which manifest as two peaks (L 3 and L 2 ) due to spin-orbit splitting. Several methods can be used to measure changes in the TM oxidation state from the L 3 /L 2 edge. 4,5 In this work, we focus on the chemical shifts of the L 3 peak, as other methods (e.g. peak intensity ratio) require high signal-to-noise ratios, which need higher electron doses, potentially leading to sample damage.
Reduction in the TM oxidation state causes the L 3 peak to shift towards lower electron energy losses. 4,5 Moreover, we use the difference between the bulk and surface positions of the L 3 peak as a way of ensuring an internal reference.
On the other hand, the oxygen K edge consists of two main sets of peaks (individual peaks cannot be resolved with the energy resolution available, which is about 1 eV): (1) pre-edge peaks between 528 and 531 eV and (2) main peaks between 535 and 545 eV. The pre-edge peaks correspond to transitions between O 1s orbitals and the hybridized O 2p and TM 3d orbitals, while the main edge peaks are due to transitions from O 1s to hybridized O 2p and TM 4sp orbitals. 6,7 In layered cathode materials, such as NMCs, it has been shown that the preedge peaks become broader and shift towards higher energies (to >530 eV) with lower average oxidation state of the TMs (also demonstrated by a reference NiO spectrum, Figure S11a). This is due to the fact that the peak at ~528 eV corresponds to transitions from O 1s to hybridized Ni 3+ or Ni 4+ 3d-O 2p orbitals, while the peak at ~532 eV corresponds to transitions to the hybridized Ni 2+ 3d-O 2p orbitals. [6][7][8] The Ni L 3 position maps, used to calculate representative Mn and Co core-loss spectra, were also used to separate O K-edge into surface and bulk regions for each spectrum image.
Each O K-edge spectrum was fitted using 5 Gaussian peaks, a single Hartree-Slater generalized oscillator strength based ionization edge and their convolution with the low-loss spectrum. The Gaussian peaks used were centered around the following energies: 528 and 531 eV for the preedge peaks, 540, 547 and 562 eV for the main edge peaks to best fit the data. Each Gaussian's center parameter was bound within ±0.5 eV for the pre-edge peaks and ±3 eV for the rest of the peaks.
To quantify the average reduction of the TMs in the surface and bulk spectra for all spectrum images, the difference between center of mass of the main edge peak at about 540 eV and the pre-edge peaks was calculated in terms of center parameters of the preedge peaks , , main peak and their respective total areas and as: Given the low S/N ratio and low energy resolution which prevents resolving each individual peak, this analysis is more robust than a full Gaussian deconvolution.
The results of this analysis for the TM L 3 edges and the O K edge are shown in Figure 7a in  Supplementary Note S4: Capacity loss from impedance and the current rate dependence Friedrich et al. 9 showed that after 1000 cycles in a pre-lithiated graphite/NMC811 cell cycled between 3.0-4.5 V vs Li/Li + at C/2 (cathode potential controlled versus a lithium metal reference electrode), that all the capacity loss attributed to impedance could be recovered in slow C/50 cycles. This is consistent with our own experiments in graphite/NMC811 cells (without pre-lithiation) after 500 cycles between 2.5-4.3 V at a C/2 rate (full cell voltage controlled, see Figure S12), which shows that the capacity at C/20, C/50, and C/100 are equivalent within ±2 mAh g -1 NMC , but 20 mAh g -1 NMC higher than that at C/2. In the analysis above, we quantified capacity "lost" due to impedance as the difference between the capacity on the constant-voltage charge step (C/40 cutoff current) in the formation and diagnostic cycles collected with constant-current at a C/20 rate. With a large number of NMC secondary particles already partially damaged or completely destroyed in calendered, pristine NMC electrodes (Figure 8), it raises the question -how important is particle cracking to the key issues of capacity loss and impedance rise? In the literature it appears that cracking is often investigated by focusing imaging efforts on particles that have (presumably) avoided damage from calendering (Type A). [10][11][12][13][14] However, since the same particle is not generally tracked before and after cycling this cannot be known for certain.
Furthermore, important details of the electrode calendering and porosity are often omitted in the experimental details of literature reports. 10,14,15 Prior literature has shown that NMC electrodes calendered to 30 % porosity show equivalent capacity to uncalendered electrodes (~50 % porosity) in the first cycle at a C/10 rate 16 -the NMC811 electrode in this work was calendered to 33 % porosity. Furthermore, for graphite/NMC811 full cells with calendered NMC electrodes, the impedance after three formation cycles is low (see Nyquist plot in Figure 4c), and only small differences are reported for EIS of uncalendered NMC electrodes versus those calendered to 30 % porosity. 16 Together, these findings indicate that mechanical cracking and/or pulverization of a significant portion of NMC secondary particles does not immediately give rise to capacity loss or impedance rise.
However, it is important to note that secondary particle cracking, induced by either electrode manufacture or the electrochemical protocol, may initiate degradation processes that manifest themselves in terms of capacity loss and impedance rise over the course of longer-term electrochemical aging.
Figure S13. Ni L 3 peak position map (a) and thickness of the same region in t/λ (b) for a FIB lamella of the Formed sample, where λ is mean free path of electrons in the material, which is on the order of 100 nm. Thinner (along the electron beam direction, purple arrow) crystal shows a thicker apparent RSL compared to neighboring, thicker (along the beam direction, green arrow) crystals. Apparent curvature of crystal facets is due to sample drift during spectrum image acquisition. Abrupt change in t/λ at the edges of crystals is due to change in crystallographic phase and local density of the RSL compared to the bulk of the sample.