Visualizing nanoscale 3D compositional fluctuation of lithium in advanced lithium-ion battery cathodes

The distribution of cations in Li-ion battery cathodes as a function of cycling is a pivotal characteristic of battery performance. The transition metal cation distribution has been shown to affect cathode performance; however, Li is notoriously challenging to characterize with typical imaging techniques. Here laser-assisted atom probe tomography (APT) is used to map the three-dimensional distribution of Li at a sub-nanometre spatial resolution and correlate it with the distribution of the transition metal cations (M) and the oxygen. As-fabricated layered Li1.2Ni0.2Mn0.6O2 is shown to have Li-rich Li2MO3 phase regions and Li-depleted Li(Ni0.5Mn0.5)O2 regions. Cycled material has an overall loss of Li in addition to Ni-, Mn- and Li-rich regions. Spinel LiNi0.5Mn1.5O4 is shown to have a uniform distribution of all cations. APT results were compared to energy dispersive spectroscopy mapping with a scanning transmission electron microscope to confirm the transition metal cation distribution.

T he portable consumer electronics revolution has driven the development of Li-ion batteries for efficient energy storage over the last decade 1,2 . Currently, there is also a strong interest in developing cost-effective, rapidly recharging Li-ion batteries suitable for long range electric vehicles 3 . As cathodes constitute a substantial portion of the volume and cost of a battery significant effort has been focused on the development of next-generation cathode materials 4,5 . Designing materials that can retain structural integrity after repeated cycling is a substantial challenge 4 . Fast ionic transport during electrochemical cycling of a material depends critically on the initial structure and crystal stability. The presence and stability of channels for fast Li-ion diffusion in cathode materials is an important design criterion for developing next-generation cathode materials for Li-ion batteries with higher capacity and long-term energy storage performance. Understanding nanoscale distribution of all of the elements that makeup Li-ion battery cathodes-especially Li ions-as a function of different synthesis procedures and extents of electrochemical cycling is a critical step towards developing new materials.
Li-rich layered cathode materials with the general formulae Li[Li 1/2-2x/3 Ni x Mn 2/3 À x/3 ]O 2 where 0oxo1/3 and more specifically, Li 1.2 Ni 0.2 Mn 0.6 O 2 -have been demonstrated with capacities 4250 mAh g À 1 : significantly higher than the 140-mAh g À 1 capacity of the best LiCoO 2 cathodes, used widely in consumer electronics [6][7][8][9] . Structurally layered Li 1.2 Ni 0.2 Mn 0.6 O 2 is considered to be a phase mixture of the trigonal LiMO 2 (R-3m) and monoclinic Li 2 MO 3 (C2/m ) phases (M ¼ Ni, Mn). Both of these structures can be represented as repeating layers of transition metal ions, O and Li. Recently, compositional segregation of Ni to surfaces and grain boundaries within some particles and partitioning of Mn away from Ni-rich regions in layered Li 1.2 Ni 0.2 Mn 0.6 O 2 has been shown by energy dispersive spectroscopy (EDS) tomography [10][11][12] . By comparison of high-angle annular dark-field scanning transmission electron microscopy (HAADF-STEM) images to multislice simulations, the Ni-rich regions were shown to be consistent with an R-3m phase and the Ni-deficient regions a C2/m phase 12 . Electron energy-loss spectroscopy (EELS) analysis has indicated a 7:9 Ni:Mn ratio (0.77) in the Ni-rich regions, best matching Li(Ni 0.5 Mn 0.5 )O 2 , and a 5:42 Ni:Mn ratio (0.12) in the Ni-deficient regions, best matching Li 2 MnO 3 (ref, 12). X-ray diffraction from as-prepared Li 1.2 Ni 0.2 Mn 0.6 O 2 predominantly matched an R-3m structure, with the exception of three peaks between 20 and 25°matching the C2/m structure 8,13,14 . However, no detailed information was found on the distribution of Li, or of its correlation to other elements in the lattice, within the published literature.
Under battery charge-discharge cycling, layered Li 1.2 Ni 0.2 Mn 0.6 O 2 cathodes were observed to develop a thin surface reconstruction layer, featuring structural transformation, Mn and Ni enrichment, oxygen vacancy formation and Li depletion [15][16][17] . Due to the structural and chemical change of the thin surface layer, it is believed to be a likely main contributor to voltage fading 18 . Furthermore, it has been proven that the surface layer continues to grow in thickness during continuous cycling 15,19 . EDS analysis of cycled layered Li 1.2 Ni 0.2 Mn 0.6 O 2 has indicated composition variations even within the thin surface reconstruction layer 15 . To understand the structural degradation mechanism and seek a means to suppress the surface reconstruction, requires monitoring the evolution of all elements quantitatively at different states of cycling. However, despite substantial characterization efforts in previous studies, quantifying lithium depletion and subtle changes in oxygen concentration in the surface layer have been proven to be elusively challenging.
High-voltage spinel LiNi 0.5 Mn 1.5 O 4 , is considered one of the most promising candidates for hybrid electric vehicle batteries 2,20,21 . Stoichiometric LiNi 0.5 Mn 1.5 O 4 is known to have a spinel structure with ordered P4 3 32 or disordered Fd3m, depending on the post-synthesis annealing temperature [22][23][24] . This order-disorder phase transformation is expected to occur during higher temperature annealing as a result of generation or elimination of oxygen vacancies, by affecting the presence of Mn 3 þ in the lattice 23,24 . The disordered phase of LiNi 0.5 Mn 1.5 O 4 is shown to have better electrochemical performance than the ordered spinel, owing to its higher electronic conductivity in the presence of increased disordered phase and/or Mn 3 þ concentration 22,[24][25][26][27] .
The common obstacle to using these materials is capacity and voltage fading, believed to be closely related to a gradual structural evolution, governed by the spatial distribution of Li ions and their correlation with other ions in the lattice. Therefore, one of the great challenges facing the development of these high-voltage cathode materials for Li-ion batteries is to locate the spatial distribution of ions with sub-nanometre-scale spatial resolution. Aberration-corrected scanning and conventional transmission electron microscopy (S/TEM) [28][29][30][31] and soft X-ray imaging/spectroscopy 32 have been used to spatially map the transition metal cations within Li-ion battery cathodes, but these techniques do not have sufficient sensitivity to map Li at sub-nanometre scales, especially in three-dimensional (3D). Atom probe tomography (APT) is uniquely capable of providing quantitative 3D, sub-nanometre-scale compositional characterization of oxides and composites [33][34][35][36][37][38]  Oxygen is known to be deficient during laser-assisted APT analysis of oxides except at extremely low laser energies 33 . To verify the accuracy of the APT quantification, the measured concentrations of Li, Mn and Ni were renormalized independently of O and compared with the expected composition. In stoichiometric Li 1.2 Ni 0.2 Mn 0.6 O 2 , the Li fraction of the total cations, Li/(Li þ Mn þ Ni), should be 0.6; Mn, 0.3; and Ni, 0.1. The measured cation fractions were 0.647, 0.258 and 0.095, respectively. The concentration these specimens will be a function of the volume fraction of Ni-rich and Mn-rich regions sampled within the reconstruction, which may account for the minor variation from the expected stoichiometry. The APT reconstruction from layered-LNMO is shown in Fig. 1f (Fig. 1h). A 13 at % Ni isocomposition surface image, shown in Fig. 1i, highlights regions enriched in Ni distributed throughout the reconstructed volume. The compositional partitioning across the Ni-and Mn-rich regions was quantified using a proximity histogram calculated perpendicular to the 13 at % Ni isocomposition surface 51 (Fig. 1j). From the steady-state regions on either side of the interface in proximity histogram, the concentration of Ni-rich region was estimated to be 21. ARTICLE composition through the thickness of the LMNO particle. The truly 3D measurements made by APT would therefore be expected to yield a higher maximum Ni:Mn ratio in 3D Ni-or Mn-rich regions.
To quantify the extent of phase separation, frequency distribution analysis was performed on the APT results. The entire APT data set was divided in-to 200-atom bins, the composition of each bin was calculated and a histogram was plotted. For a random solid solution, a binomial frequency distribution is expected. Any deviation from randomness will lead to a deviation from binomial distribution. The frequency distribution analysis of Li, Ni, Mn and O in layered-LNMO is shown in Fig. 2a-d. Pearson coefficient test can be used to measure the statistical relevance of the observed deviation from randomness [52][53][54]  Li, Ni, Mn and O distribution in Cycled layered-LNMO. On charge-discharge cycling of layered-LNMO cathodes with pre-existing Ni-rich regions, an additional Ni-rich surface reconstruction layer (SRL) has been found to form and grow as a function of cycling [15][16][17][18] . By detailed STEM, EDS and EELS measurements, a compositional partitioning of Ni and Mn has been observed within the SRL. As the Li concentration is expected to vary between the SRL and bulk of the cathode nanoparticle as a function of cycling, layered-LNMO cathodes cycled for 45 cycles were examined by APT. Needle-shaped APT specimens were fabricated from four different cycled layered-LNMO particles and during the needle preparation the needle specimen apex was kept very close to the top surface of the particles. The overall composition for each of the APT results from the four cycled layered-LNMO particles and two as-fabricated layered-LNMO particles are given below in Table 1.  A reduction in Li concentration and increase in the Mn and Ni concentration is observed in the cycled layered-LNMO specimens, in agreement with previous literature postulating a loss of Li as a function of cycling 15,56 . Two-dimensional composition plots of Li, Ni, Mn and O, as shown in Fig. 3d-g, show that Li, Ni and Mn partially segregate. A proximity histogram plotted across a Ni 14 at% isocomposition surface, shown in Fig. 3e, illustrates the interface between the Ni-rich and Ni-depleted regions within the reconstruction. The proxygram highlights that Li is depleted where Ni is enriched, and that Mn is enriched at the interface between Ni-and Li-rich regions. A second proximity histogram across a Li-rich region indicates a very high Li concentration reaching nearly 100 at %. This apparently facetted region may correspond to a void formed inside the cathode nanoparticle during cycling, with an accumulation of Li.   Fig. 5), all indicating a very close to uniform distribution. The Pearson coefficient estimated for all the molecular species were also well below 0.1. A Pearson coefficient value close to 0 is indicative of a uniform distribution of elements 52,55 . The P values estimated at a 95% confidence interval for Li and O was observed to be below 0.001 and the P value for Mn and Ni were 0.095 and 0.568. These P values in combination with the Pearson coefficient values when compared with layered-LNMO results indicate that there is only a minor deviation from random distribution for Li and O, but Ni and Mn are distributed rather uniformly in the spinel-LNMO APT result.

Discussion
The Li segregation to Mn-rich, Ni-depleted regions in as- the Li distribution. It is plausible that the local depletion of Li leads to partial deactivation of a particle with cycling, which may contribute to the capacity and voltage fading. APT results from the cycled LNMO provide evidence for Li loss. Partial segregation of Li, Mn and Ni to different regions in the cycled layered-LNMO is shown along with an increased extent of phase separation of Li, Mn and O. Prior high spatial resolution studies of Li-ion cathode materials have similarly been left to infer the behaviour of Li by studying the changes in the local crystallography and the transition metal cation concentration. This study conclusively demonstrates that laserassisted APT can be used to not only quantify the Ni and Mn composition in 3D but also the Li, and with sufficient accuracy to postulate the phase. There do not appear to be any barriers to studying the spatial distribution of Li for different synthesis methods, common cathode materials and for varying extents of electrochemical cycling of the cathode material. Quantifying the Li distribution by APT can impact the optimization of cathode synthesis procedures to achieve the highest performance, provide key insights toward the atomic-scale mechanism of capacity decay as a function of cycling, and aid in the effort to create novel Li-ion materials with prolonged lifetimes.
In summary, by comparing the as-fabricated and cycled layered-LNMO we have demonstrated a cycling-induced increased segregation of Li, Mn and O. The APT results of cycled layered-LNMO represent one of the first instances of direct evidence for Li loss in cycled cathode materials, consistent with previous TEM studies and typical explanations for irreversible capacity loss upon cycling of layered cathode materials. Comparison with compositionally uniform spinel-LNMO unambiguously establishes that the laser-assisted APT can differentiate Li segregation in battery-relevant materials at sub-nanometre-scale, in 3D. We anticipate significant application of APT analysis for understanding elemental distribution not just in the as-fabricated cathode materials, but also in electrochemically cycled materials to obtain important insight towards understanding capacity degradation in the cathode materials as a function of extent of cycling.

Methods
Material synthesis. Li 1.2 Ni 0.2 Mn 0.6 O 2 was synthesized by wet chemical process as described briefly here. Nickel sulfate hexahydrate (NiSO 4 6H 2 O), manganese sulfate monohydrate (MnSO 4 H 2 O), sodium hydroxide (NaOH), and ammonium hydroxide (NH 3 H 2 O) were used as the starting materials to prepare Ni 0.25 Mn 0.75 (OH) 2 precursor. The precursor material was washed with hot water to remove residual sodium and sulfuric species, then filtered and dried inside a vacuum oven set at 80°C for 24 h. Ni 0.25 Mn 0.75 (OH) 2 was mixed well with Li 2 CO 3 and then calcined at 900 C for 15 h to form the cathode materials. Detailed experimental setup for the synthesis of the materials was reported in Wang et al. 58 . A facile solidstate reaction method, which is easy to scale up for mass production, was adopted to synthesize the spinel LiNi 0.5 Mn 1.5 O 4 . In detail, LiNi 0.5 Mn 1.5 O 4 was prepared by ball milling a mixture of Li 2 CO 3 , NiO and MnCO 3 (all from Aldrich) in stoichiometric amount for 4 h followed by calcination at 900°C for 24 h in air with the heating rate of 10°C min À 1 and cooling rate of 5°C min À 1 .
Electrochemical cycling. The electrochemical cycling of layered-LNMO particles were conducted using coin cells configuration with metallic Li as counter electrode, separator of Celgard K1640 monolayer polyethylene membrane, with 1:2 volume ratio, 1 M Lithium hexafluorophosphate (LiPF6) dissolved in ethyl carbonate and Dimethyl carbonate (DMC) electrolyte in an argon-filled MBraun glovebox. The 45-cycle sample studied in this work was cycled at a rate of 0.1 C between 2.0-4.7 V versus Li/Li þ at room temperature. The first cycle charge/discharge profile and 'charge and discharge' capacities as function of cycle numbers at 2.0-4.7 V versus Li/Li þ are given in Supplementary Fig. 2. The cycled coin cells were disassembled and the cycled electrode was immersed in DMC for 12 h followed by washing by DMC for three times. The washed electrodes were dried in vacuum for 12 h. The cathode material was removed from the Al-foil and grounded to fine powders and deposited on a lacey carbon TEM grid for TEM imaging. The nanoparticles for APT specimens were lifted out from the TEM grid using lift-out method described below.
STEM/EDS mapping. The STEM and EDS mapping was performed using an FEI Tecnai Osiris microscope at 200 kV. The samples were dispersed onto a holeycarbon TEM grid and imaged accordingly. The FEI Tecnai Osiris microscope is equipped with a state-of-the-art Super-X EDS detector system, which allows efficient and fast mapping at nanoscale.
Atom probe specimen preparation. Needle-shaped specimens for APT analysis were prepared by lift-out procedure using an FEI Helios 600 Nanolab focused-ionbeam/scanning electron microscope (FIB/SEM). Li 1.2 Ni 0.2 Mn 0.6 O 2 nanoparticles were dispersed on a Si substrate; individual nanoparticles were lifted out by contact with OmniProbe nanomanipulator and transferred onto a Si microtip array. Once the nanoparticles were placed on top of Si microtips, electron-beam-assisted Pt deposition was used to coat the individual nanoparticles. All manipulation of nanoparticles and Pt deposition were done using only the electron beam without Ga-ion beam imaging. After Pt deposition, the nanoparticles were subjected to annular milling using Ga-ion beam to form the final needle specimens of the nanoparticles attached to the Si microtip array. If cavities were observed between a nanoparticle and the Si microtip, electron-beam-assisted Pt deposition was also performed during annular milling. Initial annular milling was conducted at 30 kV and final milling was performed using 2 kV to minimize Ga contamination in the final needle specimen. A schematic of the specimen preparation method is given in Supplementary Fig. 3.
Atom probe tomography. Laser-assisted APT analysis was conducted using a CAMECA LEAP4000 Â HR atom probe tomography system with a 355-nm ultraviolet laser, 20-pJ laser pulse energy, 40 K specimen temperature and evaporation rate maintained at 0.005 atoms per pulse. APT data were reconstructed and analysed using IVAS 3.6.6 software.