With the sustained growth of the electric vehicle and electric power industries, the global energy storage market has entered a new phase where the innovative design of cathode materials is inevitably desired beyond commercial energy materials1. As the oxidation number of redox-active transition metals (TMs) limits the theoretical specific capacity in a conventional charge compensation system2–6, the possibility of incorporating novel anionic redox chemistry into the conventional charge compensation system has been extensively explored. The cumulative cationic and anionic redox processes enable discharge capacities exceeding 250 mAh g−1 with a high upper cut-off potential of 4.6 V, delivering high-energy densities of up to 1,000 Wh kg−1.2,7 However, such a high-voltage redox (hereafter denoted as HV-redox), commonly known as oxygen-redox (hereafter denoted as O-redox), triggers remarkable voltage hysteresis, voltage fading, and sluggish kinetics accompanied by irreversible structural transformations such as oxygen evolution and phase transition8–10. Previous studies have addressed these issues by (1) involving 4d-TM Ru and 5d-TM Ir to stabilize the oxygen state via stronger M-O bonding than that in 3d-TM Mn11,12 and (2) doping inactive TMs to enhance the anionic redox reversibility13–15. Despite the positive outcomes, the interpretation of the origin of O-redox still remains challenging in these new types of cathode materials for high-energy Li-ion batteries.
Typical Li-excess layered oxides (xLi2TMO3∙(1-x)LiTMO2) possess honeycomb-type atomic arrangements of Li1/3TM2/3, where four Li-ions and two TM ions coordinate oxygen to form an OTM2Li4 octahedron (Fig. 1a). Considering the orbital orientation of TMt2g and one of the O2p orbitals along the direction where oxygen ions linearly bond to two Li-ions (Li-O-Li configuration), they do not hybridize with each other via sigma bonding, as shown in the inset in Fig. 1a, thus retaining nonbonding electrons around the oxygen anions2. From a quantitative molecular orbital perspective (Fig. 1b), the O2p nonbonding (denoted as ONB) energy states are located close to the original energy levels of oxygen and higher than the energy levels of bonding states from the hybridized bond. The general perception is that these ONB states enable anionic oxidation, O2−/O−, prior to the depopulation of the M–O bonding state, causing a permanent occurrence of highly reactive oxygen radicals16. We previously defined this type of oxidation as the oxidation of the orphaned oxygen O2p state along the Li-O-Li configurations3. First-principles calculations demonstrated that the hole density around the O anions resided along with the Li-O-Li local configuration after extracting labile electrons from ONB states in the highly charged state. However, this theoretical modeling should be discussed assuming that the ONB orbital does not overlap with adjacent TM orbitals such as Ni-O, Co-O, Mn-O, etc. Considering this, in the case of Ru (4d) and Ir (5d) with more spatially expanded orbitals than Mn (3d)11,17, the t2g orbitals of TM are more likely to interact with ONB orbitals via π-type interactions15. Therefore, a deeper study of the “O-redox” process is necessary for 4d- and 5d-based Li-excess materials. The orbital hybridizations between the t2g and ONB orbitals lead to energy splitting into the b1* antibonding state and b1 bonding state; the splitting degree is affected by the covalence of the TM-O bond (Fig. 1c)18. Li-excess oxides with hybridization than one without hybridization facilitate the oxidation of hybridized t2g and ONB orbitals, allowing further oxidation beyond conventional TM oxidation. However, oxidation of the O-dominant b1* state causes structural distortion followed by the formation of peroxo-like O-O species by unstable HV-redox19,20. Although highly covalent Ru and Ir can sustain peroxo-like O-O species via ligand-to-metal charge transfer (LMCT), the so-called reductive coupling mechanism, such structural distortion induces O-O bond shortening from approximately 2.5 Å in peroxo-like O-O species to 1.5 Å in O-O dimers, eventually resulting in the decoordination of dioxygen from TMs. Several studies have demonstrated that Sn-substituted materials (Li2Ru1-ySnyO321 and Li2Ir1-ySnyO315) serve as reversible redox hosts in the form of MLi-VM antisite-cation-vacancies even at a high voltage, inducing stable 1.8 Å TM=O oxo species and short 1.4 Å O-O dimers, but the underlying redox chemistry remains elusive.
Here, we propose a metal-to-metal charge transfer (MMCT) process driven by substituting species (TM3d), where internal charge transfer is activated from electropositive (TM3d) to electronegative species (TM4d), as shown in Fig. 1d. Such charge transfer induces a significant difference in TM-O covalency within the asymmetric TM3d-O-TM4d backbone, rectifying orbital overlap between TM4d and ONB with a shifted redox potential of b1* (Fig. 1e). Independent of its oxidation number, denoted by the integer number according to the octet rule, TM with a large charge density affords metallic characteristics to the b1* antibonding states, thereby enabling reversible HV-redox (Fig. 1f). We explore the intrinsic HV-redox chemistry in Li1.22Ni0.17Ru0.61O2 (denoted as LNRO) in contrast to that in Li2RuO3 (denoted as LRO) using bulk-sensitive operando X-ray absorption spectroscopy and first-principles calculations. The redox chemistry of LRO consists of a stable TM-dominant Ru4+/5+ state and an unstable O-dominant Ru5+/6+ state11,22. We demonstrate that Ni substitution induces polarized electrons toward electronegative Ru away from electropositive Ni, such that Ru ions pull the partial electron density away from Ni ions, leading to a stable TM-dominant Ru5+/6+ redox state. In particular, it is well known that bond breaking is much more likely to occur with weak covalent TM-O bonds under bias voltage23, leaving highly reactive dangling oxygen. We reveal that the reactive Ru-O part is stabilized via π back-bonding between O (Lewis base) and Ru (Lewis acid), preferentially over the formation of O-O peroxides, whereas Ni species partially reversibly migrate to tetrahedral sites, and then part of them further migrate toward octahedral sites in the Li layer. It is known that such migration often induces irreversible phase transformation from layered to spinel, which is a primary cause of voltage hysteresis and capacity fading10,24,25. We confirm how additional Al doping thermodynamically suppresses Ni migration through electrochemical tests, advanced analytical tools (synchrotron XRD, HR-TEM, TOF-SIMS, and operando XAFS), and first-principles calculations. The results of this paper shed new light on ternary Li-excess oxides as potential candidates for ultrahigh-energy-density batteries.
Electrochemical evaluation and phase transition in L(N)RO
LRO and LNRO were synthesized by a conventional solid-state reaction26 described in the Methods section. The structures were confirmed by synchrotron XRD combined with Rietveld refinement analysis of LRO and LNRO (Supplementary Fig. 1), confirming that both are well fitted to the C2/c space group with a slight peak broadening due to defects, which are similar to already reported Ru-based Li-excess layered oxides22,27.
The effects of Ni substitution on the redox chemistry in Ru-based Li-excess oxides were investigated using galvanostatic electrochemical testing at a current density of 0.1 C (20 mA g-1) between 2.0 and 4.6 V. The voltage profiles exhibit remarkable gravimetric capacity beyond its theoretical capacity for both LRO and LNRO, as shown in Fig. 2a,d. Theoretically, it is well-known that the stable redox couple of Ru is Ru4+/5+, which corresponds to 164.5 mAh g−1 for LRO.22 Fig. 2a demonstrates that LRO delivers 299.6 mAh g−1 reversible capacity, which corresponds to 1.82 Li+ per formula unit, indicating that Ru4+ is overoxidized to Ru>5.5+. The following cycle supplies 1.71 Li+, i.e., 94.4% of reversible Li+ is supplied on the first cycle. As differential electrochemical mass spectrometry (DEMS) analysis (bar chart in Fig. 2a,d) runs simultaneously to detect gas evolution, a significant amount of oxygen evolution was detected beyond Ru5+, attributed to lattice-oxygen dimerization. Meanwhile, in Fig. 2d, LNRO exhibits high reversible capacities (228.1 mAh g−1, corresponding to 0.96 Li+) beyond the theoretical capacities (226.7 mAh g−1, corresponding to 0.95 Li+), calculated from Ni2+/4+ and Ru4+/5+ redox (Fig. 2e). Moreover, LNRO exhibits 226.2 mAh g−1 reversible capacity (corresponding to 0.95 Li+) on the second cycle with 99.2% capacity loss after the formation process. Noticeably, the DEMS result for LNRO contrasted with that for LRO, showing almost no oxygen evolution. The comparison of synchrotron XRD patterns also supports the results before and after the formation process (Supplementary Fig. 2). The superstructure peaks ranging from 19° to 34° represent the typical Li1/3TM2/3 honeycomb-type ordering mentioned earlier8,9,15. LRO shows overall peak broadening and dissipation, as estimated by a 36.8% peak area decrease, whereas LNRO maintains its original peak shapes with only a 10.8% peak area decrease. The results apparently show that Ni substitution effectively relieves the structural disorder.
In the same context, galvanostatic intermittent titration technique (GITT) analysis was performed to compare the voltage hysteresis of LRO and LNRO for the 2nd cycle (Fig. 2b,e). GITT measurements show that the overpotential of each relaxation step gradually increases in the early stage of charging but rapidly increases during HV-redox. Relaxation behaviour at a high voltage shows that both samples are kinetically hindered by different parameters (insets in Fig. 2b,e). The relaxation step of LRO proceeds during the steady-state voltage change, whereas that of LNRO is largely contributed by the IR drop, which corresponds to the ohmic resistance and charge transfer. Recently, it was reported that the steady-state voltage change is affected by phase transition or local atomic rearrangement28. Despite similar voltage relaxation, our results show that HV-redox species would be structurally stabilized due to Ni substitution, consistent with significantly reduced voltage hysteresis as confirmed by equilibrium voltage curves (dashed curves in Fig. 2b,e). Fig. 2c,f shows the comparison of voltage profiles at 25 and 60 ℃, demonstrating that the kinetic hindrance during HV-redox is temperature-dependent, and that reduced delivery capacities are sufficiently compensated for the addition of the CV step.
Redox mechanism during the first cycle
To understand the underlying redox mechanism, we employed operando synchrotron transmission X-ray absorption near edge structure (XANES) at the Ru and Ni K-edges during the formation cycle, as shown in Fig. 3a–c. Ru K-edge XANES spectra consist of three representative peaks: i) pre-edge: 1s → hybridized 4d-5p transition, whose peak intensity stands out as the TMO6 symmetry is more distorted, ii) whiteline (WL) edge: 1s → 5p transition, and iii) back peak: 1s → 5p transition by single scattering, following the dipole transition rules29–31. At the voltage increase to 3.9 V, as shown in Fig. 3a, the Ru WL energy in LRO shifts from 22140.0 to 22141.5 eV (Δ = 1.5 eV corresponding to 1.0 e− per f.u. of Ru), with the corresponding WL intensity decrease and back peak intensity increase, taking the typical shape of the Li1Ru5+O3 plot31. However, as the cell potential remains beyond 4 V, where oxygen dimerization evolves, the pre-edge peak protrudes due to structural distortion, and the back peak weakens due to the loss of the O neighbor to scatter. The results consistently imply that the HV-redox of LRO involves progressive structural degradation with fluctuating lattice oxygen. In contrast, LNRO shows different redox behavior, as confirmed by the differential crossline intensity (ΔI: Inth – I1st) in the inset of Fig. 3b (dotted circle), showing no definite propensity related to the structural distortion and oxygen loss in the vicinity of 22120 eV. Remarkably, the Ru WL energy in LNRO increases from 22139 eV, which is lower than that in LRO (22140 eV), indicating that Ru4+ in LNRO holds a higher charge density than Ru4+ in LRO. After charging, a peak shift of 2 eV (corresponding to 1.3 e− per f.u. of Ru) occurs in LNRO. In the meantime, a shift in the WL energy of Ni K-edge spectra from 8350 to 8353.6 eV proceeds until charging up to 4.6 V (Fig. 3c,f). Considering that the Ni K-edge XANES spectra of LNRO are shifted to larger energy with respect to the Ni2+O reference25, Ni2+ in pristine LNRO seems to have a lower charge density than that in NiO. As mentioned earlier, the results show that MMCT occurs from electropositive Ni to electronegative Ru via the O-bridge. The phenomena were also identified through computed average Bader charge, indicating that the more Ni is located near Ru, the higher charge value Ru has (Supplementary Fig. 10). Notably, Ni is oxidized up to approximately 3+ during charging as the redox potential of Ni increases with decreasing charge density32–34, in contrast to the perception that Ni can be oxidized to 4+ on typical NCM cathode materials6,30. The same is identified through first-principles calculations in Supplementary Table 2, indicating that full oxidation (Ni2+/4+) of partial Ni is thermodynamically hindered depending on the TM sites. Thus, we emphasize that the Ru4+/5+ and Ru5+/6+ redox reactions via MMCT are the main charge-compensating factors, whereas Ni species play a role in auxiliary electron donors for charge redistribution in LNRO. The following section investigates how such electrochemistry is coupled with structural changes.
High-voltage redox stability coupled with local structural change
To monitor the local structural change caused by HV-redox, we measured the operando extended X-ray absorption fine structure (EXAFS) at the Ru and Ni K-edges. Radial distribution functions of Fourier-transformed (FT) k3-weighted EXAFS provide local structure information on specific TM species and interatomic distances, corresponding to TM-O bonding (~2.05 Å, denoted as O) and TM-TM bonding (~3.55 Å, denoted as T)25. As shown in Fig. 4a, EXAFS contour maps demonstrate a drastic contrast between LRO and LNRO, especially in the TM-O peak range corresponding to the first coordination shell. In the charged state, the FT intensity of the O peak in the LRO-Ru spectra, which is an indicator of atomic disordering (Debye–Waller factor, σ) and coordination number31, decreases abruptly with no complete recovery at the end of the formation cycle, which is direct evidence for irreversible oxygen evolution. In contrast, the FT intensity of the O peak in the LNRO-Ru spectra remains constant during the 1st cycle, providing convincing proof that Ni aids the local structural stability of Ru, which is the main redox species. At the same time, Ni spectra exhibit highly fluctuating but reversible spectral changes. Empirically, we determined that Ni ceaselessly migrates along its thermodynamically preferential routes, as verified through operando EXAFS results and calculations of relative site energies. The 2D contour map tracking migrated Ni species relative to fixed Ru species shows that the T-peak at ~3.4 Å splits into two peaks, ~3.2 Å peak for the Li layer, and ~3.4 Å peak for the TM layer, as shown in Fig. 4a, right panel and Supplementary Fig. 13. Considering that Ru is fixed at its original sites, such a peak split represents Ni migration toward its thermodynamically preferential neighboring site depending on the stoichiometry of delithiated LNRO. Additionally, the FT intensity of the O peak in the Ni spectra initially decreases, followed by a reciprocal tendency at the end of the charging process (marked by the white arrow in Fig. 4a), indicating that Ni undergoes out-of-plane migration from OhTM to TdLi sites and further to OhLi sites, simultaneously changing its coordination number (6 à 4 à 6) (Supplementary Fig. 14). The energy landscapes also provide consistent proof of whether Ni is the most stable at the TdLi sites or OhLi sites, depending on the state of charge (SOC) (Supplementary Fig. 16). This is in agreement with a previous report that TM tends to trap the Li layer with a thermodynamic preference8. These results highlight the importance of the research addressing such an irreversible cationic migration, which will be discussed in the next section.
We compared the O peak in the EXAFS spectra depending on the cell voltage (Fig. 4b,c) to specify the oxygen stability during HV-redox. The O peaks in the LRO-Ru spectra can be distinguished into three regions corresponding to the short TM-O peak (~1.14 Å, denoted as S), main TM-O peak (~1.61 Å, denoted as M), and long TM-O peak (~2.05 Å, denoted as L). In particular, the interatomic distance of the S peak is similar to that of double bonds15. Two remarkable changes are observed in the EXAFS spectra of LRO upon charging: 1) as the cell voltage increases, the M peak generally contracts with L peak splitting, which is attributed to the distorted Ru octahedron due to the increased interaction between Ru and O originating from the higher oxidation states31. The extended TM-O bond weakens and ultimately gives rise to decoordinated oxygen20. 2) The decrease in S peak intensity and peak broadening also represent a disordered phase, which correlates with an unstable oxygen state upon charging. On the other hand, the O peaks of LNRO-Ru remain unchanged during delithiation, except for the S peak. The S peak gradually evolves while the cell is charged, indicating Ru-O bonds approaching double bonds through additional π-bonds. This terminal oxo species, which is double bound to Ru, can be stably maintained compared with single bound oxyl species35,36, finally leading to a highly reversible HV redox.
We performed first-principles calculations to unveil the energetic effect of Ni on the redox system in the Ru-based Li-excess layered oxides (Fig. 4d,e). First, we visualized the electronic structures calculated on the basis of the Heyd–Scuseria–Ernzerhof (HSE06) hybrid functional37, enabling us to predict the partial density of states (pDOS) of the O 2p and Ru 4d states around the Fermi level (EF) for LRO and LNRO. For both pristine LRO and LNRO, hybridized orbitals consisting of Ru 4d and O 2p located just below the Fermi level indicate the most preferential redox of Ru4+/5+ at a low voltage (top of Fig. 4d,e). However, after all, Ru converts to the 5+ oxidation state, and the behaviour of the redox-contributing energy states of LRO and LNRO is quite different (bottom of Fig. 4d,e). The charge compensation of LRO at a relatively high voltage stems from the unstable Ru5+/6+, which is caused by O-dominant states. In stark contrast, pDOS of LNRO exhibits much more TM-dominant features near EF. The corresponding hybridized orbital states can be identified by visualizing the charge density for the energy range between 0 and -1.64 eV. The labile electrons of delithiated LNRO are contributed from TM-dominant b1* states, whereas those of delithiated LRO are from O-dominant b1* states (inset in the bottom of Fig. 4d,e). Computational analyses such as electron localization function (ELF) and integrated crystal orbital overlap population (ICOOP) also demonstrate a strong covalency between Ru 4d orbitals and O 2p orbitals in LNRO compared with that in LRO (Supplementary Fig. 18). As expected from the XANES analysis, the MMCT process induced by Ni substitution enriches the charge density around Ru, making the redox system more reversible even at a high voltage, where the σ-bonding between p orbitals on neighboring oxygen ligands is inhibited by the competitive π-type interaction between Ru and O38.
Our suggestions regarding oxygen states coupled with the coordinated environment are summarized in Fig. 4f,g. According to crystal field theory39, the orbital energy state of Ru splits into upper eg* states and lower t2g states. In particular, partially filled t2g states, which are affected by O2p neighbors via π-type interactions, split into b1* antibonding states and b1 bonding states. As mentioned earlier, the b1* states of Ru in LRO in Fig. 4f consist of a highly reversible redox of Ru4+/5+ (blue line) and unstable redox of Ru5+/6+ (red line). When the energy states reach unstable b1* states of Ru5+/6+ during HV-redox, the crystal structure suffers from severe distortion, followed by the formation of peroxo-like O-O species. On the other hand, the d states of Ru in LNRO hold higher charge densities by taking the negative charge from the Ni species via the MMCT process, where stronger Ru-O covalency induces larger splitting of b1/b1* states (black dotted line). This shifted energy state of b1* makes both redox couples (Ru4+/5+ and Ru5+/6+) TM-dominant (blue line in Fig. 4g). Additionally, Ru-O bonding can be stabilized by pulling the O ligands close to Ru accompanied by the spontaneous breaking of the Ni-O bond. Here, we conclude that the most critical chemistry inherent in stable HV-redox is the stabilization of high valent redox couples35 along with terminal oxo species through double bonding before RuO6 distortion, which leads to the formation of O-O peroxide.
Improving structural reversibility during extended cycling
Thus far, we have revealed how Ni species affect the redox chemistry of Ru-based Li-excess layered oxides. However, various Ni-related issues have already been considered the primary degradation mechanism, such as Ni trapping and irreversible phase transitions8,10,40. To prevent such irreversible Ni migration, we propose the aluminum doping of stoichiometric Li1.22Al0.05Ni0.12Ru0.61O2 (denoted as LANRO), one of the most effective methods to suppress structural issues in NCM and NCA cathode materials41,42.
Fig. 5a compares the voltage profiles of the first and 50th cycles versus the normalized capacity of LRO, LNRO, and LANRO. The voltage profile of LANRO is similar to that of LNRO, which delivers 286.9 mAh g−1 with Coulombic efficiency (C.E.) of 81.0 % on charging. The low first-cycle C.E. might be attributed to the synergistic effect of the slightly increased O-dominant HV-redox due to the reduced Ni/Ru atomic ratio and irreversible Al migration. Meanwhile, the capacity retention and voltage decay are significantly enhanced starting from the second cycle. LANRO shows a superior stability of 97.5% over LRO (86.5%) and LNRO (89.2%) over 50 cycles at 0.5 C between 4.6 and 2.0 V versus Li0/Li+ at 25 °C (Fig. 5b). To better understand the structural stability, we checked the average potential on each cycle. LRO with irreversible O-dominant HV-redox shows a remarkable voltage decay from 3.29 V for the second cycle to 2.96 V for the 50th cycle (Δ = 0.33 V). LNRO shows a relatively stable average voltage drop of 3.48 V to 3.28 V (Δ = 0.20 V), manifesting less severe Ni-induced structural degradation caused by oxygen evolution. As expected, the LANRO exhibits superior stability with a slight voltage drop from 3.40 to 3.32 V (Δ = 0.08 V), proving that the harmony between Ni and Al brings outstanding redox stability in Ru-based Li-excess materials. These advantages are also observed through ex-situ synchrotron XRD of pristine, 4.6 V charged, 2.0 V discharged materials, and those after 50 cycles (Supplementary Fig. 22). All XRD patterns exhibit well-defined representative peaks of the honeycomb-phase superstructure (shaded region). In particular, these peaks are still clearly seen even at 4.6 V charging and after 50 cycles, which means a well-maintained cationic arrangement with almost no migration within TM layers during HV-redox and subsequent cycling.
Interestingly, HR-TEM data of each Li-excess oxide after cycling reveal how severely the cells are aggravated during cycling and how much Al doping enhances the structural integrity at the atomic level. (Fig. 5c and Supplementary Fig. 25) After prolonged cycles under the 0.5 C-rate, severe vacancies and microcracking were observed in LRO, as shown in Fig. 5c (left), which originate from severe lattice-oxygen-evolution. In addition, atomic arrangements are not recognized on high magnification (high-MAG) HR-images with random cationic disordering and numerous defects and vacancies. In contrast, low magnification (low-MAG) HR-images of LNRO do not reveal microcracking, but numerous voids are still clearly seen, especially near the surface (Fig. 5c, middle). We attributed these voids to Ni dissolution into electrolytes, eventually toward the opposite side, which catalyzes electrolytic decomposition on the anode side and increases the cell resistance43,44. This phenomenon is clearly identified through TOF-SIMS mapping on anode surfaces after cycling, as shown in Supplementary Fig. 26. For the sake of convenience, we conducted full-cell tests with a graphite anode under identical conditions. Heavier Ni dissolution, compared with Ru dissolution, is observed in LNRO, where Ru dissolution might be attributed to the destructive impact of lattice-oxygen-evolution45 and Ni dissolution might be attributed to its intrinsically unstable repetitive out-of-plane migration46. In particular, these migrated Ni atoms are presumably trapped into tetrahedral sites in the Li layer (see high-MAG TEM images in Fig. 5c, middle), in agreement with a previous report indicating that the Ni species are thermodynamically stable on the intermediate Td site of the O3 structure8. In stark contrast with both previous samples, the cycled LANRO shows almost flawless low-MAG TEM images and well-defined TM layers with slight cationic fluctuations in the high-MAG TEM images in Fig. 5c (right). In addition, TOF-SIMS data also confirm that Al doping is notably effective in suppressing Ni dissolution (see Supplementary Fig. 26).
To reveal local structural changes in LANROs during cycling, we analyzed the migration behavior of each atom with 2D contour maps. In contrast to LRO and by analogy to LNRO (Fig. 4a-c), EXAFS at the Ru K-edge of LANRO demonstrates an entirely static propensity on the formation, the second cycle, and the 50th cycle (Supplementary Fig. 27). However, a remarkable feature arises at the end of the charging where the Ru-TM2 peak irreversibly splits into those corresponding to the TM layer (~3.6 Å) and Li layer (~3.2 Å), which is attributed to Al migration. Such irreversible peak splitting does not occur in the next cycle; instead, a new peak reversibly appears at a similar interatomic distance. Therefore, the irreversible out-of-layer migration of Al atoms on the formation step induces a reversible atomic rearrangement within the structural network after the 2nd cycle, suggesting the positive interplay of Ni and Al in Ru-based materials.
A thermodynamic penalty of Ni migration by Al dopants
A previous report proposed that Al atoms located in the tetrahedral sites in Al-doped LiNiO2 may mitigate the phase transformation upon increasing the temperature by disrupting Ni out-of-plane migration42. This statement is also experimentally verified in this work by the change in Ni coordination number during charging from (6→4→6) to (6→4) with Al doping, blocking the final migration toward octahedral sites (Supplementary Fig. 27).
To clarify how Al thermodynamically hinders Ni migration, we conducted a computational study on the site preferences of Ni and Al (tetrahedral or octahedral sites) in Li slabs. Beforehand, we statistically examined the distribution of Ni ions neighboring Al among the 15 most stable LANRO structures based on Ewald summation followed by GGA+U relaxation (Supplementary Fig. 28). In 9 of 15 structures, all Ni ions (3Ni) are located on TM sites adjacent to Al within the second nearest neighboring Oh sites in the TM layer. Five of 15 structures hold two-thirds of Ni in the identical range. Statistically, Al is surrounded mainly by Ni in approximately 93% of modeled structures, i.e., most Ni migration can be controlled by Al ions within the crystal structure, in agreement with a previous paper reporting the same phenomenon through NMR spectroscopy47. The intriguing point is that Al migration to Td sites is thermodynamically and kinetically preferential for Ni migration nearby. The nudged elastic band (NEB) calculation in Fig. 5e shows that the migration barrier for Ni (1.54 eV) is higher than that for Al (1.16 eV), implying sluggish Ni migration around Al dopants.
Furthermore, it is noteworthy that AlTd significantly increases the site energy of both Td and Oh sites in Li slabs (Fig. 5f). Fundamentally, all Ni ions tend to migrate to adjacent Td sites on delithiated LNRO. Among them, the relative site energy of the Oh site near Ni (i) is estimated to be -0.92 eV, which indicates further migration toward the Oh site. As the rest of the Ni ions also have a minor thermodynamic barrier for nearby Oh sites, their further migration is also feasible. In contrast to delithiated LNRO, however, Al prevents Ni migration toward Td sites, changing the site energy from negative to positive, except for Ni (i). Although Ni (i) can move toward the Td site (-0.29 eV), Ni (i) migration is entirely suppressed in comparison with LNRO (-0.88 eV), and it is hard to migrate toward the Oh site (+0.10 eV) further. Thus, it is concluded that Al doping can support the structural integrity not only by physically preoccupying the empty sites toward which Ni tends to migrate but also by raising the relative thermodynamic free energy of neighboring sites.