Stoichiometrically driven disorder and local diffusion in NMC cathodes

University College London, Department o E-mail: t.ashton@ucl.ac.uk Rutherford Appleton Laboratory, ISIS, Didc TRIUMF, Centre for Molecular and Materia University of British Columbia, Stewar Vancouver, BC, Canada Ibaraki University, Frontier Research Cen Ibaraki, Japan High Energy Accelerator Research Organisa † Electronic supplementary informa 10.1039/d1ta01639c Cite this: J. Mater. Chem. A, 2021, 9, 10477


Introduction
In response to the ever growing environmental and societal demands for decarbonised electrication, much effort has been afforded to the development of lithium-ion batteries (LIBs). Despite LIBs offering a reliable and efficient method of storing energy for portable electronics, transport and grid-based energy, the underpinning materials chemistry employed in such devices continue to be of paramount importance. Lithiated nickel manganese cobalt oxide (NMC) cathodes currently remain one of the most promising cathode materials families for commercial LIBs. 1,2 The attraction of NMC cathodes lies in their demonstrated stability over several thousand cycles and the promise of high energy densities. 3 Furthermore, the electrochemical attributes of these cathodes can be tuned by altering the metal ratios of Ni, Mn, and Co. Recent attention has been focused on high-nickel NMC cathodes (>80% of the metals being Ni) due to their increased capacity ($200 mA h g À1 ) versus lower Ni concentration NMCs. 4 This increase in Ni also allows higher maximum voltages to be accessed (approaching 5 V versus Li/Li + ) during electrochemical cycling, increasing energy density. 5 However, increasing the Ni concentration may also lead to increased structural instability, both during synthesis and electrochemical cycling. 6 Thus, signicant challenges still remain.
The electrochemically attractive crystal phase of NMC adopts a layered R 3m structure, typical of LiCoO 2 . Alternating slabs of Li atoms on the 3a sites are sandwiched between layers of MO 6 octahedra (M ¼ Ni, Mn or Co), where M are located on the 3b sites ( Fig. 1). 7 Isolation of a defect-free R 3m phase has proven to be a sensitive process due to thermodynamic limitations on structural evolution during heating. [8][9][10] Typically, syntheses are carried out in a two-step reaction. The rst reaction consolidates the constituent transition metals (Ni, Mn and Co) into a single precursor, usually by coprecipitation at temperatures <350 C. 11,12 The second step requires a solid-state reaction between a lithium source (such as LiOH or Li 2 CO 3 ) and the aforementioned precursors, typically in a furnace at >800 C for 5 to 48 h. 11 Due to these high temperatures and long times, a ne interplay between lithium incorporation into the metal oxide and evaporative lithium loss arises. 13 Thus, an excess of the Li source is commonly used, but consequently risks the formation of additional 'lithium-rich' phases. 14 In combination, these reports also suggest that lithium incorporation becomes increasingly more difficult with greater Ni concentration, requiring longer times or higher synthesis temperatures.
Suppression of lithium-nickel cation mixing (Li 3a has also been highlighted as a major synthetic challenge. 15 Reportedly, this exchange can lead to further structural instability and lower intrinsic Li mobility (at elevated concentrations). Li 3a + /Ni 3b 2+ mixing is purportedly more probable in highnickel NMC due to the larger availability of Ni, and thus a smaller amount of R 3m templating Co 3+ , and R 3m stabilising Mn 4+ . Li 3a + /Ni 3b 2+ mixing in NMC has commonly been assessed using the peak intensity ratios of the I (003) and I (104) by X-ray diffraction, however this analysis is limited and presumes the only contribution to these intensities is Li 3a . Other contributions may include bifurcation of element rich and poor phases, Li deciency, or the presence of impurity phases whose peaks overlap such as Li rich C2/m. Suggestions have also been made that the exchange of Li + and Ni 2+ are not necessarily equivalent. 16 Furthermore, the diminished X-ray attenuation of Li versus other elements (due to its lower relative mass) reduces the accuracy of such measurements. 17 Therefore, advanced techniques such as Neutron Powder Diffraction (NPD) are key for gaining a more thorough understanding of structural chemistry.
Typically, Li ions migrate through the 2D Li plane and are reversibly removed and inserted from the crystals' surface. As the diffusion of lithium ultimately dictates the operation of intercalation cathodes such as NMC, these systems have been widely studied by a range of methods including cyclic voltammetry (CV) electrochemical impedance spectroscopy (EIS), galvanostatic intermittent titration technique (GITT) and solidstate lithium nuclear magnetic resonance (Li-NMR). [18][19][20][21] Among these, muon spin relaxation (mSR) remains a unique and powerful tool to probe the diffusive properties of lithium ions (among others) on a local scale. 22 In contrast to bulk electrochemical methods, mSR is not limited by grain boundaries or electrode preparation. Thus, mSR can establish diffusion coef-cients (D) and activation energies (E a ) specic to the crystal structures present, irrespective of the particle-or mesostructure. mSR has previously been employed on isostructural materials such as the parent LiCoO 2 and sister Li n Ni x Co y Al z O 2 materials, and have been in good agreement with computational predictions of D Li and E a values. [23][24][25][26] Although many studies have been performed to characterise electrochemical performance, there is still a lack of understanding of the effect of increasing nickel concentration on the structure and its relationship with local lithium diffusion in NMC cathodes. Herein, we present the need for improved structural scrutiny and an evaluation of increasing Ni content on local structure by NPD and local diffusion by mSR.

Results and discussion
Structural and elemental characterisation NMC-433, NMC-622 and NMC-811 were all collected as ne black powders aer solid state lithiation and analysed by inductively coupled plasma mass spectrometry (ICP-MS) to elucidate the bulk elemental composition (Fig. 2a). It was found  Ordering of Li into a hexagonal arrangement to give concentrations exceeding unity has previously been reported as a 'lithium-rich' phase, adopting a monoclinic C2/m unit cell similar to Li 2 MnO 3 . [27][28][29] It is therefore suggested that it is the inclusion of increasing amounts of Mn, rather than decreasing Ni, would promote the higher concentration of Li observed in the samples as this phase is oen observed in Mn rich NMC. 30,31 A representation of the MO 6 slab for both R 3m and C2/m is shown in Fig. 2b where MO 6 polyhedra are shown in grey and Li + in green. It can be seen that the only signicant difference between the two structures is the replacement of 20% of the M sites are occupied by Li + to produce a 'honeycomb' structure.
Scanning electron microscopy (SEM) was used to further investigate particle size and morphology of the NMC samples. SEM of NMC-433, NMC-622 and NMC-811 ( Fig. 3a-c, respectively) showed similar granular morphologies with typical diameters z2 mm. It is interesting to note that NMC-433 was signicantly smaller than the other two materials. This may be due to the increased levels of Mn and/or Co providing resistance to particle sintering (as the same solid state lithiation conditions were used).
To corroborate the presence of a C2/m phase, powder X-ray diffraction (PXRD) and subsequent Rietveld renement was employed to evaluate the structure of the NMC samples (Fig. 4). All samples showed well dened, sharp peaks indicating a high degree of crystallinity and a rhombohedral R 3m space group typical of LiCoO 2 layered compounds (no contribution from a C2/m phase was identied). However, previous reports have suggested that an increase in cobalt content (relative to the other metals present) can decrease the size of the C2/m domains, causing them to be essentially amorphous to PXRD. 32 The peak intensity ratio of the I (

Neutron powder diffraction (NPD)
To elucidate the crystal structure of the NMC materials further, neutron powder diffraction (NPD) was employed. Fig. 5 shows  the data collected for the three samples with respect to their dspacing, and Rietveld renement thereof. A full account of the tting parameters can be found in the ESI. † While NMC-811 and NMC-622 were adequately t with a single R 3m phase (similar to the PXRD renements), a small inclusion of Li 2 O (0.5 mol%) and LiOH (4.9 mol%) was identied in NMC-622 and an unknown impurity (0.3 mol%) in NMC-433 denoted by an asterisk. These observations led to a more correct stoichiometry being dened for NMC-622 (Li 0.96 Ni 0.62 Co 0.19 Mn 0.19 O 2 ). Similar to the PXRD renements, a (Li 1Àd Ni d ) 3b (Li g Ni xÀg Mn y Co z ) 3a O 2 model was employed with the determined values given in Table  1.
The values in Table 1 show that increasing [Ni] from 0.6 to 0.8 did not lead to a signicant change in the absolute presence of Ni 2+ 3b (NMC-811 ¼ 3.1%, NMC-622 ¼ 3.6%). However, these values do become signicant when comparing them to the relative total Ni content of the material; 3.97% of the Ni in NMC-811 is located on the 3a site in contrast to 9.42% of the Ni in NMC-622. Uniquely, although NMC-811 exhibits a low concentration of Ni 3b 2+ mixing, it is not accompanied by an equivalent exchange of Li 3a + due to a signicant concentration (z7%) of vacancies on the 3b site. This is an important observation that is only made possible through the use of NPD, as Li vacancies have a relatively small impact on the I (003) /I (104) peak intensity ratio due to their low Z. A representation of this effect can be seen in Fig. 6, where values of I (003) /I (104) from simulated PXRD patterns are given versus the concentration of mixing (Fig. 6a) and vacancy defects (Fig. 6b). In both cases, the location of Li makes the smallest contribution to the magnitude of I (003) /I (104) with the smallest contribution attributed to Li vacancies. Thus, Li vacancies may be more easily concealed in PXRD than any other defect and supporting techniques, such as NPD or elemental analysis, are crucial. In comparison, the structural renement of NMC-433 was unable to converge with a single R 3m phase. NMC-433 was best t by a combination of three phases: (i) a conventional, ordered R 3m NMC (58.8 mol%); (ii) a disordered R 3m NMC phase (33.4 mol%) and; (iii) a C2/m Li 2 MnO 3 phase (7.8 mol%). As shown in Fig. 2, the C2/m phase can be thought of as a highly disordered R 3m with 20 mol% of Li on the metal 3b site. Thus, the C2/m phase is "lithium rich" compared to the R 3m phase, which agrees well with the overabundance of Li in NMC-433  A representation of the crystal structures for each of the major phases in NMC-811, NMC-622 and NMC-433 can be seen in Fig. 7 35 Previous reports on similar oxides containing only Mn and Co have also shown phase separation of C2/m and R 3m in the absence of Ni. 36 The increase in [Co] can also be ruled out for  instigating phase separation or cation mixing as Co 3+ is known to nucleate and stabilise the R 3m phase, and alleviate the magnetic frustration responsible for Ni 2+ formation. 37,38 Increased cation mixing has also been directly observed from combinatorial experiments when Ni is replaced for Mn. 39 Thus, it is likely that increasing Mn may both aggravate cation mixing, and encourage the formation of the C2/m phase, if sufficient Li is available during heat treatment.

Local lithium diffusion (mSR)
Since the structures of the NMC materials differed signicantly, it is important to establish the effect of these changes on lithium diffusion. Previously, R 3m structures have been predicted to exhibit two intraplanar diffusion pathways through the LiO 6 slab; the Oxygen Dumbbell Hop (ODH) and the Tetrahedral Site Hop (TSH). 40 Previous investigations on isostructural materials have found the TSH mechanism to occur on a similar timescale (z10 À10 cm 2 s À1 ), but provides a lower activation energy (E a ) diffusion pathway (E TSH a z E ODH a /4). 26 However, the TSH mechanism is only available where [Li] < 1 (due to the requirement for a Li divacancy). In comparison, the C2/m structure has previously been reported to exhibit additional interplanar diffusion pathways between the Li and MO 6 slabs, which may be expected to provide enhanced Li ion conductivity. However, density functional theory (DFT) studies suggest that the average E a of Li diffusion lies between 510 and 840 meV in C2/m LiMn 2 O 3 , similar to values estimated by DFT for the ODH mechanism ($800 meV). 41 To understand the impact of nickel concentration and crystal structure on the local lithium diffusion mechanics, muon spin relaxation (mSR) experiments were performed on all three samples of NMC. Briey, the powders were loaded into a titanium sample holder and exposed to a spin-polarized beam of muons in a vacuum using the EMU instrument at the STFC ISIS Neutron and Muon Facility, Harwell, UK. Muon decay asymmetry data were recorded at 300 K using a transverse magnetic eld of 20 G to normalise the initial decay asymmetry arising from the sample geometry. Data were then recorded between 160 K and 450 K, applying three longitudinal elds of 0, 5 and 10 G at each temperature to decouple the muon spin relaxation from any nuclear magnetic contributions from Ni, Mn or Co (Fig. 8a). The data from the three elds at each temperature were t simultaneously to a modied Kubo-Toyabe relaxation model to describe the dynamic diffusion processes multiplied by a Gaussian relaxation function to account for temperature independent uctuations in Li + diffusion, with a constant background component (eqn (1)).
Two key parameters were extracted from the ts; (i) the eld distribution width (D) which is a measure of the magnetic eld distribution that the muon experiences at the implantation site and, (ii) the uctuation rate (n) which is a measure of the rate of muon decay perturbation from Li + diffusion, where n f D Li . Fig. 8b shows the extracted values for D for the three NMC samples. It is observed that as temperature increases there is a corresponding decrease in the eld distribution width due to Li + diffusion becoming more rapid and leading to a motional narrowing effect similar to that seen in previous investigations. 26

,42 As [Ni] increases (and thus, [Mn] and [Co] decreases)
there is a consistent decrease in D due to the signicantly smaller nuclear moment present in Ni, when compared to Mn or Co. Fig. 9a shows the extracted values of n versus temperature where a steady increase was observed, indicating the thermal activation of Li diffusion. The data was used to estimate the average D Li of each sample employing eqn (2); N i is the number of accessible Li + sites in the i th path, Z n , i is the vacancy fraction of the destination sites, s i is the hopping distance between Li + sites and n is the uctuation rate at each temperature. 43 Structural values were extracted from the NPD renements, and the calculation for NMC-433 included a weighted average of the three observed phases.
Values of D Li at 300 K for the two samples exhibiting only the R 3m structure were determined to be 2.90 Â 10 À11 cm 2 s À1 for NMC-811 and 4.36 Â 10 À11 cm 2 s À1 for NMC-622. Statistical signicance was conrmed by calculated errors of AE1.75 Â 10 À28 (NMC-811), AE2.65 Â 10 À28 (NMC-622) and AE1.92 Â 10 À20 (NMC-433). Using the Arrhenius relationship over the thermally activated region of each sample (Fig. 9b), E a of muon hopping were found to be 58 meV for NMC-811, 61 meV for NMC-622 and 28 meV for NMC-433. Compared to NMC-622, it is possible that the marginally slower but more facile diffusion of NMC-811 (indicated by both lower D Li and E a ) may be attributed to the Li vacancies. Lower [Li 3b ] causes less electrostatic repulsion between Li aiding diffusion (lowering E a ), but provides less diffusion events for the muon decay to sample (lowering n). The lowering of E a with decreasing [Li 3b ] has been previously predicted in the parent LiCoO 2 by computational means, and observed in isostructural materials by mSR (Table 2). However, this does not preclude the possibility of the observed differences being solely or partly due to changes in transition metal stoichiometry, and the values of E a for NMC-811 and NMC-622 lie within experimental error.
In comparison, while the estimated D Li for NMC-433 was found to be 3.35 Â 10 À11 cm 2 s À1 the estimated value of E a was 28 meV, signicantly lower than that for NMC-811 or NMC-622. Although the structure of NMC-433 is too complex to be certain of the origin of the decrease in E a , it is unlikely to be due to the presence of Li 2 MnO 3 as it was present <10 mol% of the total sample and has previously shown a larger E a versus R 3m structured materials (Table 2). Thus, it is likely that this marked decrease in E a is either due to (i) the elevated concentration of Mn and/or Co or (ii) a contribution from the heavily disordered R 3m phase.

Conclusions
Intimate studies of the three NMC samples (made using identical heat treatments) have provided direct evidence for stoichiometry driven phase formation. Neutron powder diffraction studies on 'Ni rich' NMC-811 has shown that while a single R 3m phase can be obtained; Li incorporation is more challenging (under similar synthetic conditions) compared to the lower Ni content NMC-622. However, lowering the Ni content further in NMC-433 leads to a dramatic structural disruption and the evolution of three phases; an ordered R 3m phase, a disordered R 3m phase with a high degree of Li 3a + /Ni 3b 2+ mixing, and a Li rich C2/m phase. In this case, the authors suggest that this is  In closing, it is of paramount importance to interpret results given from PXRD responsibly. Although no signicant cation mixing was detected with PXRD, this is not proof of its absence; a vastly different story was told by using NPD data that showed not only cation mixing, but multiple crystalline phases. It is also important that I (003) /I (104) is not accurate in solely determining the Li 3a + /Ni 3b 2+ mixing in NMC materials, as vacancies and closely related crystal structures may also have a large impact on the results.

Synthesis of lithium nickel manganese cobalt oxide (NMC)
To prepare the nanoparticulate NMC precursors a continuous hydrothermal ow synthesis (CHFS) process was used, which has been described elsewhere. 26,44 To summarise the process, three diaphragm pumps (Primeroyal K, Milton Roy, Pont-Saint-Pierre, France) were used to supply feeds of deionized water (P sw ), an aqueous solution of the metal salts (P M ), and a 1 M KOH solution (P base ) at ow rates of 80, 40 Table S1. † The DI water feed from pump P sw was heated in ow to 450 C using a 7 kW custom-built in-line electrical heater. Precursor feeds from pumps P M and P base were separately mixed in a Tpiece in ow (at room temperature) prior to mixing in ow with the combined superheated water feed in the patented (WO2011148121A1) conned jet mixer (CJM). 45 In all cases, the total concentration of metals, P M was xed at 0.5 M and the concentration of base, P base was xed at 1 M. Upon mixing of the feeds in the CHFS process, nanoparticles were formed, which were then cooled to $40 C using a 1.5 m counter-current pipein-pipe heat exchanger, before being passed through a backpressure regulator (BPR) valve at the CHFS outlet. The aqueous nanoparticle slurry exiting the BPR was collected and cleaned by repeated centrifugation and washing with deionized water until the conductivity of the supernatant was consistently below 50 mS, as measured by a conductivity probe (Hanna Instruments, model HI98311, Leighton Buzzard, UK). The concentrated slurry was then freeze-dried (Virtis Genesis 35XL) by gradually heating a sample from À60 to 25 C (over 24 h) in vacuo (<100 mTorr), which yielded free-owing dark brown/ black powders in all cases.
The NMC precursor powders were then mixed with a 30% molar excess of LiOH in a Thinky™ mixer for 30 min at 1500 rpm and heat treated in a box furnace at 800 C for 5 hours in air (ramp rate 5 C min À1 ). The resulting black powders were characterised and tested with no further modication.

Structural characterisation (PXRD and NPD)
Powder X-ray diffraction (XRD) patterns were collected using a STOE Stadi P diffractometer in transmission geometry (Mo-Ka 1 radiation, l ¼ 0.70932Å), equipped with a germanium (111) monochromator and a DECTRIS Mythen 1k silicon strip detector (DECTRIS, Baden, Switzerland). An Yttria (Y 2 O 3 ) standard was used to estimate instrumental peak broadening. Datasets were collected over the 2q range of 2 to 40 with a step size of 0.5 and a count time of 5 s per step.
Time-of-ight (TOF) neutron powder diffraction (NPD) was carried out on the iMATERIA diffractometer at J-PARC. 46 The each as-synthesized powder sample ($2.5 g) was placed into vanadium cell, which was installed to iMATERIA. In this study, a wide d-spacing (0.11 < d < 5.29Å) and a high resolution (Dd/ d ¼ 0.16%) were used for the TOF NPD measurements. The sample environment of iMATERIA was evacuated to prevent incoherent scattering of water vapor (H atoms). All experiments were measured for 4 h at room temperature. The beam power of the J-PARC for NPD measurements was 500 kW. The data was analysed by Rietveld renement using the Z-Rietveld soware package (ver. 1.0.2). 47,48 In the case of both NMC-811 and NMC-622, the Rietveld renements were demonstrated between d ¼ 0.35 and 4.95Å, whereas that of NMC-433 was done the region between d ¼ 0.70 and 5.08Å.

Scanning electron microscopy (SEM)
Scanning electron microscopy (SEM) was performed using a JEOL JSM-6700F microscope. To minimise charging samples were deposited on copper foil tape, mounted on aluminium stubs, aer dispersion in methanol (99.9%, Sigma Aldrich, Dorset, UK) and ultrasonication in a XUBA3 Ultrasonic Bath (Grant Instruments, Cambridge, UK) for 5 minutes. Image analysis was carried out using ImageJ soware.

Elemental analysis (ICP-MS)
Inductively coupled plasma mass spectrometry (ICP-MS) was carried out by fully dissolving each sample in concentrated hydrochloric acid overnight by microwave digestion (CEM Discover SP). These solutions were then diluted with dilute hydrochloric acid prior to analysis by ICP-MS (Agilent 7500ce). Calibration curves were constructed with elemental standards and used to calculate the concentrations of Ni, Mn, Co and Li for each sample.

Local diffusion analyses (mSR)
Muon spin relaxation experiments were performed at STFC ISIS Neutron and Muon Source, Harwell, UK. Approximately 3.0 g of powdered samples of NMC-811, NMC-622 or NMC-433 were packed into recessed titanium sample holders, covered with a titanium window and secured with a titanium bezel. The titanium holder provided a simple background signal from muons that are not implanted into the sample and any signal is easily subtracted from the data. The sample holder was placed into the muon spectrometer (EMU) and evacuated to <1 Â 10 À6 mbar. The instrument was cooled to a temperature of 150 K and instrument asymmetry was measured using a transverse magnetic eld of 20 G. Measurements were taken every 10 K up to 450 K, with longitudinal at elds of 0, 5 and 10 G. Data collected at each temperature for the three applied magnetic elds were t simultaneously using the WiMDA soware. 49

Conflicts of interest
There are no conicts to declare.