Sample Dependence of Magnetism in the Next Generation Cathode Material LiNi 0.8 Mn 0.1 Co 0.1 O 2
P. Mukherjee, J. A. M. Paddison, C. Xu, Z. Ruff, A. R. Wildes, D. A. Keen, R. I. Smith, C. P. Grey, S. E. Dutton
SSensitivity of Bulk Magnetism to Off-Stoichiometry in theNext Generation Cathode Material LiNi Mn Co O P. Mukherjee,
1, 2, ∗ J. A. M. Paddison,
1, 3, 4
C. Xu,
5, 2
Z. Ruff,
5, 2
A. R.Wildes, D. A. Keen, R. I. Smith, C. P. Grey,
5, 2 and S. E. Dutton
1, 2, † Department of Physics, University of Cambridge, JJ Thomson Avenue, Cambridge CB3 0HE, United Kingdom The Faraday Institution, Quad One, Harwell Science and Innovation Campus, Didcot OX11 0RA, United Kingdom Churchill College, University of Cambridge, Storey’s Way, Cambridge CB3 0DS, United Kingdom Materials Science and Technology Division, Oak Ridge National Laboratory, Oak Ridge, TN 37831, USA Department of Chemistry, University of Cambridge, Lensfield Road, Cambridge CB2 1EW, United Kingdom Institut Laue-Langevin, CS 20156, 38042, Grenoble Cedex 9, France ISIS Neutron and Muon Source, Rutherford Appleton Laboratory,Harwell Campus, Didcot OX11 0QX, United Kingdom (Dated: April 23, 2020)We present a structural and magnetic study on two batches of polycrystalline LiNi Mn Co O (commonlyknown as Li NMC 811), a Ni-rich Li ion battery cathode material, using elemental analysis, X-ray and neutrondiffraction, bulk magnetometry, and polarised neutron scattering measurements. We find that the samples,labelled S1 and S2, have the composition Li x Ni x – y Mn y Co O , with x = 0 . , y = 0 . for S1and x = 0 . , y = 0 . for S2, corresponding to different concentrations of magnetic ions and excessNi in the Li + layers. Both samples show a peak in the zero-field cooled (ZFC) dc susceptibility at 8 K but thetemperature at which the ZFC and FC (field-cooled) curves deviate is substantially different. Ac susceptibilitymeasurements indicate a frequency-dependent transition in S1 and a frequency-independent transition in S2.Our results demonstrate the extreme sensitivity of bulk magnetic measurements to off-stoichiometry of themagnetic transition metal ions in Li NMC 811 and indicate that such measurements can be used to benchmarksample quality for Ni-rich Li ion battery cathode materials. I. INTRODUCTION
LiNiO , a layered transition metal (TM) oxide with S = / Ni ions on a triangular lattice [Fig. 1(a)], has been widely in-vestigated as a quantum spin liquid candidate. However, thenature of its magnetic ground state remains controversial afterdecades of investigation [1–6]. This is because LiNiO is ex-tremely prone to off-stoichiometry and excess of Ni in Li + layers. Studies have shown that it is not possible to synthesiseperfectly stoichiometric LiNiO , and instead the formula isLi x Ni x O with x ≈ . in the best quality samples [7–11]. This results in 2 x S = 1 Ni spins in order to maintaincharge balance. Additionally, due to the similarity in Li + andNi radii, x Ni tend to migrate from the transition metal(TM) to the Li + layers. These factors have two crucial con-sequences for the magnetism. First, the presence of differentmagnetic species ( S = / and S = 1 ) in amounts depen-dent on the degree of off-stoichiometry results in the mag-netic ground state being highly sample-dependent. Second,the excess Ni in the Li + layers changes the competing inter-actions: in addition to the intra-layer ( J ) and inter-layer ( J (cid:48) )interactions, there are interactions between the spins in theLi + layers and those in the TM layers ( J (cid:48)(cid:48) ) [Fig. 1(b)]. Thismay cause the spins to order magnetically or freeze insteadof remaining in a dynamic liquid-like state [6, 12]. A previ-ous muon spin relaxation ( µ SR) and magnetometry study onLi x Ni x O with x = 0 . , . found static ferromagnetic ∗ [email protected] † [email protected] order for x = 0 . below ∼
50 K and a disordered antiferro-magnetic state for x = 0 . below ∼
20 K [13]. Another µ SRstudy on quasi-stoichiometric LiNiO ( x = 0 . ) found adisordered, slowly fluctuating state below 12 K [14]. Thus themagnetic ground state varies significantly with slight changesin the off-stoichiometry x .LiNiO has also been investigated as a Li ion battery cath-ode material as it is cheaper and less toxic than the commer-cially established LiCoO [15–18]. However, safety issuesdue to thermal instability severely limit its practical applica-bility [19]. Additionally the inherent off-stoichiometry andmigration of Ni to the Li + layers leads to irreversible capac-ity loss on long-term cycling [18, 20]. Therefore the focushas shifted towards Ni-rich compositions from the family ofLi ion TM oxides with the general formula LiNi x Mn y Co z O , x + y + z = 1 , commonly known as Li NMC oxides [20–22].Being Ni-rich, these materials inherit the tendency for off-stoichiometry and Ni excess in Li + layers associated withLiNiO , which can significantly affect their performance inbatteries [23].In this paper, we focus on the next-generation Li ion bat-tery cathode material LiNi Mn Co O , commonly knownas Li NMC 811. We present an investigation of the bulk mag-netic properties, crystal structure and chemical composition oftwo different samples of Li NMC 811 using elemental anal-ysis, powder X-ray and neutron diffraction, bulk magneticmeasurements, and polarised neutron diffraction. We showthat the magnetic susceptibility is extremely sensitive to slightchanges in the stoichiometry, consistent with previous reportson LiNiO [7, 9, 24, 25]. This indicates that bulk magneticmeasurements can be a powerful tool to benchmark samplequality for industrial-scale production of Ni-rich Li ion bat- a r X i v : . [ c ond - m a t . m t r l - s c i ] A p r FIG. 1. a) Crystal structure of layered Li TM oxides, showing O asred spheres, TM polyhedra in burgundy, and Li polyhedra in green.b) Competing magnetic interactions J , J (cid:48) , and J (cid:48)(cid:48) in LiNiO andNi-rich Li TM oxides. tery cathode materials. II. EXPERIMENTAL METHODS
Two batches of polycrystalline samples of Li NMC 811were obtained from Targray, each being purchased on aseparate occasion. The samples will be referred to as S1(Targray, Batch 1) and S2 (Targray, Batch 2) throughoutthis manuscript. Since Li NMC 811 is known to be sensi-tive to moisture [20, 26], the samples were stored in an Ar-atmosphere (O < O < ∼
10 mg of the as-received Li NMC 811powder in 1 ml of freshly prepared, concentrated aqua regia(3:1 hydrochloric to nitric acid, trace element grade, FisherScientific) overnight and subsequently diluting with deionizedwater (Millipore) to ∼ ◦ ≤ θ ≤ ◦ ( ∆2 θ = 0 . ◦ ) using the I11 beamline at DiamondLight Source ( λ = 0 . ˚A). Room temperature powder neu-tron diffraction (PND) experiments for structural character-isation were carried out on the GEM diffractometer, ISISNeutron and Muon Source, Rutherford Appleton Laboratory,United Kingdom. The absorption correction for the time-of-flight (TOF) PND data was carried out using the Mantidprogram [27] and cross-checked using the GudrunN software[28]. The structural Rietveld refinements [29] were carriedout using the Fullprof suite of programs [30]. The back-ground was modelled using a Chebyshev polynomial and thepeak shape was modelled using a pseudo-Voigt function forthe PXRD data and an Ikeda-Carpenter function for the TOFPND data.Magnetic dc susceptibility measurements were performedon a Quantum Design Magnetic Properties Measurement Sys-tem (MPMS) with a Superconducting Quantum InterferenceDevice (SQUID) magnetometer. The zero-field cooled (ZFC)and field-cooled (FC) susceptibility χ ( T ) was measured in afield of 100 Oe in the temperature range 2-300 K to investi-gate the presence of magnetic ordering. ZFC measurementswere also carried out at 1000 Oe in the same temperaturerange to perform Curie-Weiss fits. In a field of 1000 Oe,the isothermal magnetisation M ( H ) curve is linear at all T and so χ ( T ) can be approximated by the linear relation χ ( T ) ≈ M/H . Isothermal magnetisation measurements inthe field range µ H = 0 - T for selected temperatures werecarried out using the ACMS (AC Measurement System) op-tion on a Quantum Design Physical Properties MeasurementSystem (PPMS). ZFC ac susceptibility measurements in thetemperature range 2-60 K were carried out using the samePPMS option with a dc field of 20 Oe and a driving field of3 Oe at frequencies between 1-10 kHz.Polarised neutron diffraction measurements for the S2 sam-ple were carried out on the D7 diffractometer at the Insti-tut Laue-Langevin, France, with λ = 4 . ˚A. A sample ofmass 7 g was loaded in an annular Al can of diameter 20 mmwith an 18 mm cylindrical insert to minimise the effect of ab-sorption from Li in our natural abundance samples. Scanswere collected at 1.5 K and 20 K for 10 hours each. The datawas processed in LAMP [31] to separate the nuclear coher-ent, nuclear-spin incoherent, and magnetic scattering contri-butions.Li NMC 811 electrodes for electrochemical characterisa- TABLE I. Composition from ICP-OES for samples S1 and S2.Element S1 S2Li 1.06(5) 1.01(3)Ni 0.80(2) 0.80(2)Mn 0.10(2) 0.10(2)Co 0.10(2) 0.10(2) tion of S1 and S2 were prepared by mixing 90 wt% Li NMC811 powder, 5 wt% polyvinylidene difluoride (PVDF) binder,5 wt% carbon black (Timcal SuperP Li) and a desired amountof the NMP (1-Methyl-2-pyrrolidinone, anhydrous, 99.5%,Sigma-Aldrich) solvent in a Thinky planetary mixer at 2000rpm for 10 minutes in total (5 minutes per cycle and two cy-cles). The slurry was coated onto an Al foil and pre-driedat 100 ◦ C for 1 hour in a dry-room ( ∼ -55 ◦ C dew point).Dried Li NMC 811 laminates were punched into circular diskswith a diameter of 14 mm, which were further dried at 120 ◦ C for 12 hours under dynamic vacuum ( ∼ − mbar) ina B¨uchi oven. 2032 coin cells (Cambridge Energy Solution)were assembled in an Ar-atmosphere (O < O < , ethy-lene carbonate (EC)/ethyl methyl carbonate (EMC) 3/7, Soul-Brain MI) was added to each coin cell. Battery cyclings wereconducted on an Arbin battery cycler at room temperature be-tween 3.0 V and 4.4 V at various rates. C-rates were definedbased on a reversible capacity of 200 mAh g − , for instance,for the C/5 rate (5 hours for one charge or discharge process),a current density of 40 mA g − was applied. III. RESULTSA. Elemental analysis
The composition of both samples as determined from ICP-OES is given in Table I and the details of the error analysis aregiven in Appendix A. Both samples were found to have theideal transition metal (TM) stoichiometry within error. Sam-ple S1 is found to have a slight Li excess as compared to thenominal stoichiometry.
B. Room temperature PXRD and PND
The room temperature PXRD data for S1 and S2 indicatedthat the samples were phase pure and adopt the crystal struc-ture of Li NMC oxides (space group R ¯3 m ) with no indicationof lowering of symmetry [23]. The room temperature PNDdata were consistent with this; however, closer inspection ofthe data plotted on a logarithmic intensity scale indicated avery small Li CO impurity peak for S1, while no such peakwas visible for S2. This is consistent with the higher Li con-tent seen in elemental analysis for S1. The amount of Li CO TABLE II. Structural parameters for samples S1 and S2. All refine-ments were carried out in the space group R ¯3 m , with Li on the a sites (0,0,0), TM (Ni, Mn, Co) on the b sites (0,0,0.5), and O on the c (0,0, z ) sites. The Mn composition y and Ni excess on the Li + site x were allowed to vary subject to the constraints discussed in thetext. Parameter S1 S2 a ( ˚A) 2.8719(2) 2.8727(3) c ( ˚A) 14.199(2) 14.207(3) c/a x a y a z ) 0.24095(9) 0.24103(8) B iso ( ˚A )Li/Ni (0,0,0) 0.82(16) 0.79(9)TM (Ni/Mn/Co) (0,0,0.5) 0.28(3) 0.20(2)O (0,0, z ) 0.68(4) 0.66(2) χ a in Li x Ni x – y Mn y Co O for S1 from the refinement was found to be 0.2(3) wt% andhence was not considered in further structural analysis.The crystal structure was refined using a combined Rietveldrefinement with the room temperature PXRD and TOF PNDdata using a structural model based on LiNiO , space group R ¯3 m [32]. PXRD is sensitive to the TMs (mainly Ni becauseof its higher concentration) while PND is much more sensi-tive to the contrast between Li (coherent scattering length b = − . fm) and Ni ( b = 10 . fm) as well as Mn ( b = − . fm)and O ( b = 5 . fm) [33], so a combined PND+PXRD re-finement gives accurate information about the crystal struc-ture. While carrying out the refinement, the weighting of thePXRD data was adjusted to satisfy the following two condi-tions simultaneously: a) the total weighting of the PXRD and5-bank PND refinements summed to 1 (each PND bank wasassigned the same weighting); b) the weighted residual of thePXRD refinement was equal to that of the PND refinement,such that the PXRD and PND data contributed equally to therefinement. The refinement of the chemical composition isdiscussed later in Section III D.Representative fits to the PXRD and TOF PND data areshown in Figure 2 and the refined structural parameters arecompiled in Table II. Li-deficient LiNiO with the formulaLi x Ni x O , x > . , crystallises in a cubic rock salt struc-ture ( c/a = 2 √ . ) with Li/Ni disordered on the a site; however, as the quantity of Li increases, it transforms toa hexagonal structure consisting of alternating layers of LiO and NiO octahedra and the c/a value increases with the lay-ering of the material [18]. The c/a ratio for our samples ofLi NMC 811 (4.944(3) for S1 and 4.946(2) for S2) is con-sistent with a layered hexagonal structure. Though less thanthat of a well-layered compound like LiCoO ( c/a = 4 . )[34, 35], it is consistent with the typical values for the parentcompound LiNiO ( c/a = 4 . ) [17, 34]. FIG. 2. Room temperature PXRD + PND Rietveld refinementfor S2. PXRD data (upper panel) were collected on I11, Diamond,and PND data (lower panel) were collected on GEM, ISIS. Data areshown as red points, fits as black lines, and difference (data–fit) asblue lines. The peak marked with * in the PND data is from thevanadium sample holder.
C. Bulk magnetic measurements
Dc susceptibility χ ( T ) and isothermal magnetisation M ( H ) measurements for S1 and S2 are shown in Figures 3and 4, respectively. At low temperatures, both samples showa peak in the ZFC dc susceptibility at T g = 8 . K [Fig. 3]and deviation in the ZFC-FC curves, indicating glassy be-haviour. However, the temperature T ZFC-FC at which theZFC-FC curves deviate is very different: 8.0(2) K for S1 and122(2) K for S2. The isothermal magnetisation at T = 2 K( T < T g ) shows a slight hysteresis for both samples, consis-tent with a disordered ground state, whereas no hysteresis isobserved at T = 15 K (
T > T g ) [Fig. 4].In a field of 1000 Oe, the reciprocal susceptibility χ − ( T ) is linear above 200 K and was used to fit to the Curie-Weisslaw, χ = CT − θ CW , where C is the Curie constant and θ CW is FIG. 3. ZFC-FC dc susceptibility χ ( T ) in a field of 100 Oe for S1(blue squares) and S2 (red circles). Inset: reciprocal susceptibility χ − ( T ) in a field of 1000 Oe.FIG. 4. Isothermal magnetisation M ( H ) at 2 K and 15 K for S1(blue squares) and S2 (red circles). Inset: M ( H ) at 2 K showing theslight hysteresis at 2 K more clearly.TABLE III. Bulk magnetic properties for both batches ofLi NMC 811 and comparison with the Li NMC 811 sample inRef. 36. Curie-Weiss fits were carried out in the temperature range200-300 K in an applied field of 1000 Oe. Assuming ideal stoichiom-etry LiNi Ni Mn Co O , the theoretical magnetic moment performula unit (f.u.) µ th = 2 . µ B /f.u. .Parameter S1 S2 Sample in Ref. 36 T g (K) . . T ZFC-FC (K) . θ CW (K) − − − µ eff ( µ B /f.u. ) . . . FIG. 5. Real component of ac susceptibility χ (cid:48) ( T ) for S1 (upperpanel) and S2 (lower panel) with a dc field of 20 Oe and a drivingfield of 3 Oe at different frequencies (labelled on the panels). Inset:imaginary susceptibility component χ (cid:48)(cid:48) ( T ) . the Curie-Weiss temperature. Parameters for the Curie-Weissfits are summarised in Table III. The Curie-Weiss tempera-tures are negative for both samples, indicating net antiferro-magnetic interactions. The calculated moment per formulaunit (f.u.) for S1 of 2.26(3) µ B /f.u. is greater than the the-oretical value of 2.10 µ B /f.u. , indicating a higher concentra-tion of high spin species (Ni with S = 1 and Mn with S = / ) as compared to the nominal stoichiometry. By con-trast, the moment for S2 of 2.08(2) µ B /f.u. is consistent withthe theoretical value, indicating that S2 is almost stoichiomet-ric.We carried out ac susceptibility measurements to probe thedynamics of the magnetic transition in both samples. The realcomponent of the ac susceptibility, χ (cid:48) ( T ) , shows a peak at8 K for both samples [Fig. 5], consistent with the dc suscep-tibility measurements; it also shows an additional shoulderat ≈ K, which could be indicative of a second transitionat lower temperatures. The transitions in sample S1 show a clear frequency dependence, ∆ T g T g ∆(log ω ) = 0 . , consistentwith spin-glass-like behaviour [37]. However, the transitionsin sample S2 show no frequency dependence. The imaginarycomponent of the ac susceptibility, χ (cid:48)(cid:48) ( T ) , shows additionalpeaks at 20 K and 35 K in both samples [Fig. 5 inset]. A pre-vious study on Li NMC 811 reported a frequency-dependentpeak at < T < K, consistent with our measurements[36]. Such additional peaks in χ (cid:48)(cid:48) ( T ) at frequencies up to1 kHz have also been reported for other Li ion battery cathodematerials such as LiNi Mn O and LiNi Mn Co O ,and have been attributed to spin reorientation transitions [38]. D. Determination of chemical composition
We now discuss refinement of the chemical compositionof the two Li NMC 811 samples. The refinement was sub-ject to the following constraints: a) charge balance; b) thethe a , b and crystallographic sites (corresponding to Li,TM and O respectively) were fully occupied; c) the magneticmoment was consistent with µ eff , the magnetic moment ob-tained from the Curie-Weiss fit [Table III]; d) the composi-tion of the TM ions were consistent with the values obtainedfrom elemental analysis within error. We further reduced thenumber of free parameters by noting that, since Co has S = 0 , its composition cannot be constrained using magne-tometry. Hence the Co composition was fixed to the nominalvalue 0.1, consistent with elemental analysis. Previous neu-tron diffraction studies on LiNiO have examined the possibil-ity of Li/Ni site disorder such as { Li x Ni x } a [Ni x Li x ] b O and { Li x Ni x } a [Ni y Li y ] b O and ruled it out for near-stoichiometric samples [32, 39]. Our refinements also indi-cate the absence of Li + in the TM layers and so a single pa-rameter x was used to refine the Ni excess in the Li + layers.The composition Li x Ni x – y Mn y Co O was refined for arange of ( x , y ) values consistent with the above constraintsand the final values were chosen corresponding to the refine-ment with the minimum χ .Both samples of Li NMC 811 are slightly Li-deficient,consistent with previous studies on LiNiO ; however, theircompositions are different. Sample S1 has the formulaLi Ni Mn Co O while S2 has the formulaLi Ni Mn Co O , corresponding to Ni ex-cess in the Li + layers of 2.5(2)% and 0.2(2)%, respectively. E. Polarised neutron diffraction
Neutron scattering experiments using XYZ polarisationanalysis on the D7 instrument at the ILL enable separation ofthe nuclear coherent, nuclear-spin incoherent, and magneticscattering contributions from the sample. Thus they are idealfor investigating diffuse scattering in disordered magnetic sys-tems [40]. Our ac susceptibility measurements indicated astatic magnetically-disordered state for S2 and so polarisedneutron scattering measurements were carried out to investi-gate the nature of the transition at 8 K. Figure 6(a) shows the
FIG. 6. a) Magnetic scattering of S2 at 1.5 K (black squares) and20 K (red diamonds). b) Nuclear coherent scattering of sample S2at 20 K plotted on a logarithmic scale. Impurity peaks belonging toLi CO are marked with a *. magnetic scattering as a function of momentum transfer Q forS2 at 1.5 K (below T g = 8 K) and 20 K (above T g = 8 K).The negative intensity at Q ≈ . ˚A − is an artifact fromthe subtraction of the nuclear Bragg peak and does not haveany physical significance. No magnetic Bragg peaks are ob-served, consistent with the absence of long-range magneticorder; however, there is a broad diffuse feature at low Q in-dicative of short-range spin correlations. A previous inelasticneutron scattering study on Li x Ni x O , x = 0 . , re-ported a decrease in the inelastic channel and an increase inthe elastic line on cooling through T g = 15 K, consistent withspin freezing [6]. The elastic scattering at 1.7 K in [6] showedno magnetic Bragg peaks, only broad magnetic diffuse scat-tering at low Q . The feature observed in our magnetic scat-tering for S2 is qualitatively similar to this previous report onLiNiO ; however, it was not possible to carry out quantita-tive modelling of the magnetic interactions due to the weakmagnetic scattering and contributions from multiple magneticspecies. The nuclear coherent scattering intensity at T = 20 K isplotted on a logarithmic scale as a function of Q in Fig-ure 6(b). Several very weak peaks from an impurity phase areobserved along with the single nuclear Bragg peak from themain phase; these were identified to be from Li CO . ThisLi CO phase was beyond the detection limit of the PXRDand PND data for S2 used for our structural Rietveld refine-ments; however, it is observed here due to the separation ofthe coherent and incoherent nuclear contributions, which im-proves the signal-to-noise ratio in the coherent Bragg scatter-ing. Due to the limited Q range and the presence of only asingle nuclear Bragg peak from the main phase, it was not pos-sible to carry out a structural refinement to quantify the exactamount of Li CO ; however, based on our structural analysisfor S1 (for which the Li CO impurity was visible in the PNDdata plotted on a logarithmic scale), we can place an upperbound of 0.2(3) wt%. IV. DISCUSSION
We now discuss the key features of our bulk magnetic mea-surements and structural refinements on the two commercialsamples of Li NMC 811.The ZFC transition temperature and ZFC-FC irreversibilityin our Li NMC 811 samples is consistent with previous reportson LiNiO samples with similar values of x in Li x Ni x O ( T g = 9 K) [6, 7], as well as other Ni-rich Li ion cathode ma-terials such as LiNi Co Al O (Li NCA) ( T g = 6 . K)[41] and LiNi Mn Co O (Li NMC 622) ( T g = 7 . K)[42]. This indicates that the magnetic properties are still dom-inated by the S = / Ni and S = 1 Ni spins. The differ-ence in T ZFC-FC in our samples can be attributed to the differ-ence in composition, which alters the relative number of Ni ,Ni , and Mn , and hence the magnetic interactions. A previ-ous study on a sample of Li NMC 811 had reported T g ≈ Kand T ZFC-FC ≈ K; however, the Ni excess in the Li + lay-ers (calculated using XRD) was 3.9% [36], which is greaterthan both our samples, and the Mn composition was not re-fined. Previous investigations on Li x Ni x O have shownthat even a slight change in composition can dramatically al-ter the transition temperature; for example, T g = 7 . K for x = 0 . and 8.6 K for x = 0 . [10]. The transition tem-perature for our samples, T g = 8 K, is consistent with lowervalues of x (2.5(2)% and 0.2(2)% for S1 and S2, respectively)as compared to the previous study.Our bulk magnetic measurements demonstrate that thesample dependence of the magnetic properties widely re-ported for the parent material LiNiO persists in Li NMC 811.The origin of the sample dependence in LiNiO is theoff-stoichiometry Li x Ni x O and x , the Ni excess inthe Li + layers. However, the situation is more complexin Li NMC 811 due to the presence of multiple magneticspecies. For ideal stoichiometry, the formula can be writ-ten as LiNi Ni Mn Co O , the magnetic species being S = / Ni , S = 1 Ni and S = / Mn , whereas Co with S = 0 plays the role of non-magnetic site dilution inthe TM layers. However, the composition for our samples isLi x Ni x – y Mn y Co O , with x = 0 . , y = 0 . for S1, and x = 0 . , y = 0 . for S2. Thus theamount of each magnetic species present depends on the com-position: ( . − x ) Ni ( S = / ), ( y + 2 x ) Ni ( S = 1 ) and y Mn ( S = / ). Since x and y are both different for S1and S2 (S2 is closer to nominal stoichiometry), the relativenumber of magnetic species is also different in these sam-ples. Additionally, x Ni migrate to the Li + layers, intro-ducing a competition between the inter-layer and intra-layerinteractions dependent on x , analogous to LiNiO . Further, ithas been observed that a higher deviation from nominal stoi-chiometry in Li x Ni x O corresponds to a greater tendencyfor spin-glass-like behaviour [6]. Thus S1, which is more off-stoichiometric, exhibits a frequency dependent spin-glass-liketransition whereas S2, close to ideal stoichiometry, shows nosuch frequency dependence in the ac susceptibility. Thus thedifferences in the bulk magnetic properties of S1 and S2 canbe explained.We find that a combination of elemental analysis, bulkmagnetic measurements, and diffraction is essential to pro-vide an accurate quantitative estimate of the composition forLi NMC 811. Previous studies have often set the compositionto the values determined from elemental analysis and carriedout structural refinements using X-ray diffraction data only[42–44]. Our results indicate that this approach may need tobe treated with caution for Ni-rich compositions, which tendto be Li-deficient. Elemental analysis provides the averagecomposition value for each element; that is, the contributionsfrom the main phase (Li NMC 811) as well as any impurityphases (Li CO ) are both included. PXRD is much less sen-sitive to the presence of light elements like Li, C, and O, andso no Li CO impurity Bragg peaks are visible in the roomtemperature PXRD pattern. By contrast, PND is much moresensitive to the presence of these elements; our room temper-ature PND data for S1 and polarised neutron diffraction dataat 20 K for S2 provide conclusive evidence for the presence ofLi CO . Returning to our elemental analysis results, since theLi composition was equal to 1 (within error) and the neutrondata confirms the presence of Li CO , we can conclude thatthe Li NMC 811 samples are indeed Li-deficient. This is con-sistent with our combined structural refinements. Recent highresolution powder diffraction studies on Li NMC oxides havealso indicated the necessity of using PXRD and PND data todetermine the stoichiometry accurately [23, 45]. By includingan additional constraint on the total magnetic moment fromour bulk magnetic measurements, we are able to increase theaccuracy of our refined compositions for Li, Ni, and Mn inLi NMC 811. Techniques such as Li nuclear magnetic reso-nance (NMR) could also be used to quantify the Li compo-sition in Li NMC 811 more accurately as the signal would bewell separated for paramagnetic (Li NMC 811) and diamag-netic (Li CO ) Li-containing phases.Another key result of our study is to demonstrate the re-markable sensitivity of bulk magnetic measurements to vari-ations in the composition Li x Ni x – y Mn y Co O , partic-ularly changes in y , the amount of S = / Mn , and x ,the amount of Ni in the Li + layers. The presence of Ni excess in Li + layers in LiNiO has been repeatedly linked to deterioration in cycling performance as the Ni significantlyhinder Li + mobility [18]. Measurements comparing the elec-trochemical performance of S1 and S2, Appendix B, suggestthat the degree of Ni excess in Li + layers ( = 2.5(2)% for S1and 0.2(2)% for S2 respectively) has no significant influenceon the rate performance for Li NMC 811 samples with lowlevels of off-stoichiometry (upto 2-3% of Ni excess in Li + layers). Our results indicate that bulk magnetic measurementscan serve as a powerful laboratory tool for benchmarking such‘good quality’ samples of Li ion TM oxide battery cathodematerials prior to carrying out long-term electrochemical cy-cling, particularly for Ni-rich systems which are prone to off-stoichiometry and migration of Ni into the Li + layers. V. CONCLUSION
We have carried out elemental analysis, room temperatureX-ray and neutron diffraction, bulk magnetic measurements,and polarised neutron scattering measurements on two pow-der samples of Li NMC 811 from a commercial supplier. Ourcombined PXRD and PND structural refinements using con-straints from all these techniques show that the samples havethe composition Li Ni Mn Co O for S1 andLi Ni Mn Co O for S2 respectively. Bulkmagnetic measurements reveal a transition at 8 K for bothsamples, but the ZFC-FC curves deviate at very different tem-peratures: 8 K for S1 and 122 K for S2. The nature of thetransition is also different: dynamic in S1 with a frequencydependence consistent with spin-glass-like behaviour, and fre-quency independent in S2. This is attributed to the fact thatS1 is more off-stoichiometric and so shows a greater ten-dency for spin freezing, analogous with the parent compoundLi x Ni x O .Thus a combination of elemental analysis, diffraction, andbulk magnetic measurements can be used to distinguish highquality samples of Li NMC 811 and other Ni-rich Li ion bat-tery cathode materials prior to cycling them in batteries. ACKNOWLEDGMENTS
We thank the Science and Technology Facilities Council(STFC) for provision of ISIS Xpress Access beam timeon GEM and the Institut Laue-Langevin for allocation ofEASY beam time on D7. P.M., C.X., Z.R., C.P.G. and S.E.D.acknowledge funding support from the Faraday InstitutionEPSRC Grant EP/S003053/1. Magnetic measurements werecarried out using the Advanced Materials Characterisa-tion Suite, funded by EPSRC Strategic Equipment GrantEP/M000524/1. We thank Cheng Liu for support with theMPMS and PPMS equipment. J.A.M.P.’s work at Cambridge(contribution to neutron data reduction) was supported byChurchill College, University of Cambridge. J.A.M.P.’s workat ORNL (contribution to data analysis) was supported byORNLs LDRD program (
Appendix A: ICP-OES error analysis
To understand the error in the ICP-OES measurement, errorcalculations were made from the two largest sources of uncer-tainty: the error associated with the sample measurement andthe error in the linear calibration. From these values, errorcalculations were performed at the 95% confidence level asdiscussed in [46]. The standard deviation of the sample mea-surement is simply the standard deviation of the three repli-cate measurements and represents how repeatable the individ-ual measurement is for a given sample. The prediction inter-val, s ( x ) , represents how accurate the instrument responseis based on a linear relationship between the sample concen-tration and intensity at a given wavelength. The predictioninterval can be calculated using equation A1. s ( x ) = RSDb (cid:115) N + 1 n + ( ¯ y − ¯ y ) b Σ ni =1 ( x i − ¯ x ) (A1)whereRSD = residual standard deviation of y with xn = number of calibration points = 4 N = number of repeat calibration points / replicates = 3 b = slope of linear calibration ¯ y = mean of N measurements of y -value (intensities) for thesample ¯ y = mean of the y -values of the calibration standards x i = x -value (concentration) of the standards ¯ x = mean of the x i values of the samplesTo combine the error at each wavelength at the con-fidence level, the confidence interval was calculated forboth the standard deviation of the sample measurement( µ λ n ,s = t N − StDev λs √ N ) and for the prediction interval( µ λ n ,s ( x ) = t n − s ( x ) ) using a two sided t-statistic.In our case, both t-statistics have 2 degrees of freedom( t N − = t n − = 4 . ). The standard deviation of thesample measurement was then combined with the predictioninterval to obtain the total error of the measurement at agiven wavelength µ λ = (cid:112) µ λ n ,s + µ λ n ,s ( x )2 ). The errorsare given in Table IV. Since the concentrations of eachelement were measured at different wavelengths (2 for Co, 3 for Mn and 4 for Ni) and averaged to obtain the meanconcentration in the measurement, the confidence interval foreach wavelength was combined using µ element,k = ¯ µ j √ No.ofλ .Finally, the confidence interval for each element wascalculated using the expression µ element,composition = (cid:113)(cid:112) µ Ni,k + µ Mn,k + µ Co,k + µ element,k since thetransition metals were assume to have a total fraction of 1 andthe Li composition was calculated by dividing the number ofmoles of lithium by the number of moles of transition metal. Appendix B: Electrochemical performance
We evaluated the electrochemical performance of S1 andS2 at various charge/ discharge rates as the Ni excess in theLi + layers is expected to influence the Li ion diffusivity, andtherefore impact the rate capability of the cathode material[47]. The experiments were carried out in half-cell configura-tion, i.e. with Li metal as the anode, and three cells per samplewere tested. The average discharge capacities at various ratesare show in Figure 7. FIG. 7. Discharge capacities of S1 and S2 at various cycling ratesbetween 3.0 V and 4.4 V vs. Li. The capacity is normalized tothe mass of Li NMC 811, and the C-rate is calculated based on areversible capacity of 200 mAh g − . The error bars are calculatedbased on three cells for each sample. Both samples show a discharge capacity of 210 mAh g − at C/20 rate, which is in good agreement with literature that LiNMC 811 cathodes typically show capacities above 200 mAhg − at slow rates [48]. S1 and S2 exhibit good rate capability,with no major capacity decreases until extremely high rates(i.e. 3C and 5C) and the capacity-rate profiles of the two sam-ples are also very similar. [1] K. Hirakawa, H. Kadowaki, and K. Ubukoshi, Journal of thePhysical Society of Japan , 3526 (1985).[2] K. Hirakawa, K. Takeda, R. Osborn, and A. D. Taylor, Journalof the Physical Society of Japan , 3081 (1990). [3] M. Rosenberg, P. Stelmaszyk, V. Klein, S. KemmlerSack, andG. Filoti, Journal of Applied Physics , 6813 (1994).[4] Y. Kitaoka, T. Kobayashi, A. K¯oda, H. Wakabayashi, Y. Niino,H. Yamakage, S. Taguchi, K. Amaya, K. Yamaura, M. Takano, TABLE IV. ICP-OES error analysis for samples S1 and S2.Element Li Ni Ni Ni Ni Mn Mn Mn Co CoWavelength (nm) 610.362 216.556 221.647 230.3 231.6 257.61 259.373 260.569 228.616 237.862S1Mean 1.057 0.802 0.796 0.796 0.795 0.099 0.100 0.102 0.102 0.103StDev (sample)(%) 1.7 0.4 0.5 0.5 0.5 1.1 1.0 1.1 0.5 1.0 µ λ,s (%) 4.3 0.9 1.2 1.2 1.2 2.7 2.5 2.7 1.2 2.6 µ λ,s ( x ) (%) 0.5 0.7 0.9 2.8 0.9 3.1 2.6 2.6 2.8 3.4 µ λ (%) 2.1 2.9 3.7 12.2 3.7 13.4 11.4 11.0 12.2 14.8S2Mean 1.007 0.803 0.800 0.798 0.797 0.098 0.099 0.100 0.102 0.101StDev (sample)(%) 2.2 0.2 0.2 0.1 0.1 1.4 1.6 1.6 0.2 2.1 µ λ,s (%) 5.4 0.5 0.4 0.3 0.2 3.4 3.9 4.1 0.6 5.3 µ λ,s ( x ) (%) 0.6 0.8 1.0 0.6 1.0 4.1 3.5 3.3 3.7 4.6 µ λ (%) 2.7 3.4 4.4 2.8 4.4 17.6 14.8 14.4 15.9 19.7A. Hirano, and R. Kanno, Journal of the Physical Society ofJapan , 3703 (1998).[5] F. Reynaud, D. Mertz, F. Celestini, J.-M. Debierre, A. M. Gho-rayeb, P. Simon, A. Stepanov, J. Voiron, and C. Delmas, Physi-cal Review Letters , 3638 (2001).[6] J. P. I. Clancy, Neutron Scattering Studies of Spin Glass Be-haviour in LiNiO , Ph.D. thesis (2006).[7] J. N. Reimers, J. Dahn, J. Greedan, C. Stager, G. Liu, I. David-son, and U. Von Sacken, Journal of Solid State Chemistry ,542 (1993).[8] A. Rougier, Journal of The Electrochemical Society , 1168(1996).[9] K. Yamaura, M. Takano, A. Hirano, and R. Kanno, Journal ofSolid State Chemistry , 109 (1996).[10] V. Bianchi, D. Caurant, N. Baffier, C. Belhomme, E. Chap-pel, G. Chouteau, S. Bach, J. P. Pereira-Ramos, A. Sulpice, andP. Wilmann, Solid State Ionics , 1 (2001).[11] C. Sow and P. S. A. Kumar, Journal of Applied Physics ,07E130 (2012).[12] E. Chappel, M. D. N´u˜nez-Regueiro, S. de Brion, G. Chouteau,V. Bianchi, D. Caurant, and N. Baffier, Physical Review B ,132412 (2002).[13] J. Sugiyama, Y. Ikedo, K. Mukai, H. Nozaki, M. M˚ansson,O. Ofer, M. Harada, K. Kamazawa, Y. Miyake, J. H. Brewer,E. J. Ansaldo, K. H. Chow, I. Watanabe, and T. Ohzuku, Physi-cal Review B , 224412 (2010).[14] T. Chatterji, W. Henggeler, and C. Delmas, Journal of Physics:Condensed Matter , 1341 (2005).[15] J. Dahn, Solid State Ionics , 87 (1990).[16] C. Delmas, J. P. P´er`es, A. Rougier, A. Demourgues, F. Weill,A. Chadwick, M. Broussely, F. Perton, P. Biensan, and P. Will-mann, Journal of Power Sources , 120 (1997).[17] H. Li, N. Zhang, J. Li, and J. R. Dahn, Journal of the Electro-chemical Society , A2985 (2018).[18] M. Bianchini, M. Roca-Ayats, P. Hartmann, T. Brezesinski, andJ. Janek, Angewandte Chemie International Edition , 10434(2019).[19] D. MacNeil, Z. Lu, Z. Chen, and J. Dahn, Journal of PowerSources , 8 (2002).[20] W. Liu, P. Oh, X. Liu, M.-J. Lee, W. Cho, S. Chae, Y. Kim,and J. Cho, Angewandte Chemie International Edition , 4440(2015).[21] A. Manthiram, B. Song, and W. Li, Energy Storage Materials ,125 (2017). [22] J. Xu, F. Lin, M. M. Doeff, and W. Tong, Journal of MaterialsChemistry A , 874 (2017).[23] L. Yin, Z. Li, G. S. Mattei, J. Zheng, W. Zhao, F. Omenya,C. Fang, W. Li, J. Li, Q. Xie, E. M. Erickson, J.-G. Zhang, M. S.Whittingham, Y. S. Meng, A. Manthiram, and P. G. Khalifah,Chemistry of Materials , 1002 (2020).[24] A. Rougier, C. Delmas, and G. Chouteau, Journal of Physicsand Chemistry of Solids , 1101 (1996).[25] P. T. Barton, Y. D. Premchand, P. A. Chater, R. Seshadri, andM. J. Rosseinsky, Chemistry - A European Journal , 14521(2013).[26] L. Dong and J.-H. Kim, Impact of Moisture on the Properties ofLiNi x Co y Mn − x − y O (NMC) Cathodes for Lithium-Ion Bat-teries , Ph.D. thesis (2018).[27] O. Arnold, J. Bilheux, J. Borreguero, A. Buts, S. Camp-bell, L. Chapon, M. Doucet, N. Draper, R. F. Leal, M. Gigg,V. Lynch, A. Markvardsen, D. Mikkelson, R. Mikkelson,R. Miller, K. Palmen, P. Parker, G. Passos, T. Perring, P. Pe-terson, S. Ren, M. Reuter, A. Savici, J. Taylor, R. Taylor,R. Tolchenov, W. Zhou, and J. Zikovsky, Nuclear Instrumentsand Methods in Physics Research Section A: Accelerators,Spectrometers, Detectors and Associated Equipment , 156(2014).[28] A. K. Soper, GudrunN and GudrunX: programs for correctingraw neutron and X-ray diffraction data to differential scatteringcross section , Tech. Rep. (Science and Technology FacilitiesCouncil, 2011).[29] H. M. Rietveld, Journal of Applied Crystallography , 65(1969).[30] J. Rodr´ıguez-Carvajal, Physica B: Condensed Matter , 55(1993).[31] D. Richard, M. Ferrand, and G. J. Kearley, Journal of NeutronResearch , 33 (1996).[32] C. Pouillerie, E. Suard, and C. Delmas, Journal of Solid StateChemistry , 187 (2001).[33] F. S. Varley, Neutron News (1992).[34] T. Ohzuku, A. Ueda, M. Nagayama, Y. Iwakoshi, and H. Ko-mori, Electrochimica Acta , 1159 (1993).[35] J. Akimoto, Y. Gotoh, and Y. Oosawa, Journal of Solid StateChemistry , 298 (1998).[36] J. M. Wikberg, M. Dahbi, I. Saadoune, T. Gustafsson, K. Ed-str¨om, and P. Svedlindh, Journal of Applied Physics ,083909 (2010).[37] J. A. Mydosh, Spin glasses : an experimental introduction (Tay- lor & Francis, London; Washington DC, 1993).[38] N. A. Chernova, M. Ma, J. Xiao, M. S. Whittingham, J. Breger,and C. P. Grey, Chemistry of Materials , 4682 (2007).[39] J.-H. Chung, T. Proffen, S. Shamoto, A. M. Ghorayeb,L. Croguennec, W. Tian, B. C. Sales, R. Jin, D. Mandrus, andT. Egami, Physical Review B , 064410 (2005).[40] J. R. Stewart, P. P. Deen, K. H. Andersen, H. Schober, J.-F.Barth´el´emy, J. M. Hillier, A. P. Murani, T. Hayes, and B. Lin-denau, Journal of Applied Crystallography , 69 (2009).[41] H. Liu, H. Liu, I. D. Seymour, N. Chernova, K. M. Wiaderek,N. M. Trease, S. Hy, Y. Chen, K. An, M. Zhang, O. J.Borkiewicz, S. H. Lapidus, B. Qiu, Y. Xia, Z. Liu, P. J. Chu-pas, K. W. Chapman, M. S. Whittingham, C. P. Grey, and Y. S.Meng, Journal of Materials Chemistry A , 4189 (2018).[42] M. Dahbi, J. M. Wikberg, I. Saadoune, T. Gustafsson,P. Svedlindh, and K. Edstr¨om, Journal of Applied Physics , 023904 (2012).[43] A. Mauger, F. Gendron, and C. M. Julien, Journal of Alloys andCompounds , 42 (2012).[44] X. Zhang, W. J. Jiang, A. Mauger, Qilu, F. Gendron, and C. M.Julien, Journal of Power Sources , 1292 (2010).[45] H. Liu, H. Liu, S. H. Lapidus, Y. S. Meng, P. J. Chupas, andK. W. Chapman, Journal of The Electrochemical Society ,A1802 (2017).[46] J. Miller and J. C. Miller, Statistics and Chemometrics for Ana-lytical Chemistry (Pearson education, 2018).[47] Y. Makimura, T. Sasaki, T. Nonaka, Y. F. Nishimura, T. Uyama,C. Okuda, Y. Itou, and Y. Takeuchi, Journal of Materials Chem-istry A , 8350 (2016).[48] H.-H. Ryu, K.-J. Park, C. S. Yoon, and Y.-K. Sun, Chemistry ofMaterials30