Ni-rich NMC layered oxides — compositions in the NixMnyCozO₂ family with x ≥ 0.6 — have been the dominant focus of high-energy cathode development for the past decade. The appeal is straightforward: higher Ni content correlates with higher practical capacity and lower average charge voltage, which translates directly to energy density. The challenge is also well-known: Ni-rich compositions are prone to structural disorder under cycling, cation mixing (Ni²⁺/Li⁺ scrambling at high states of charge), and surface oxygen release that accelerates electrolyte decomposition.
The questions that drive Ni-rich NMC research in 2025 have shifted from "can we make Ni-rich materials work?" — they clearly can — to "what is the optimal Ni fraction, and which Co/Mn ratios at each Ni level produce the best combination of energy density, thermal stability, and cycle life?" These are quantitative composition optimization questions, and they are well-suited to computational screening.
The composition space
For a Ni-rich NMC screening campaign focused on x ≥ 0.6, the relevant composition space spans Ni fractions from 0.6 to 0.9, with the remaining transition metal sites occupied by Co and Mn in varying ratios. At each Ni level, the number of distinct (Co, Mn) combinations in a fixed-composition approximation is small — but the number of symmetry-inequivalent cation arrangements in a computational supercell is not.
In a 48-atom supercell of the R3̄m layered structure, the 16 transition metal sites can be occupied by Ni, Co, and Mn in a large number of distinct arrangements for any given overall stoichiometry. Many of these arrangements are symmetry-equivalent and collapse to the same structure under space group operations. The AFLOW-POCC enumeration for a representative Ni₀.₇Co₀.₁₅Mn₀.₁₅ composition at the 48-atom supercell level produces approximately 60–90 symmetry-distinct configurations, depending on the exact stoichiometry and the target space group.
For a campaign spanning Ni = 0.6, 0.65, 0.7, 0.75, 0.8, 0.85, and 0.9 with Co/Mn ratios at each level, the total candidate pool reaches 400–500 structures — within the Lab-tier capacity of MaterSynq's pipeline.
What the screening reveals about Ni fraction
The formation energy landscape across the Ni-rich NMC space has a consistent structure. Formation energies become less negative (less stable) as Ni fraction increases from 0.6 to 0.9, reflecting the increasing Ni²⁺ content and the tendency toward Li/Ni disorder. The convex hull distance increases as Ni fraction increases — the materials become progressively less thermodynamically stable relative to competing oxide phases.
This is not surprising; it matches the experimental observation that Ni₀.₉Co₀.₀₅Mn₀.₀₅O₂ (NMC90) requires more careful synthesis control than NMC622. What the computational screening adds is quantification: the hull distance at NMC90 is approximately 40–60 meV/atom above the hull on average across the symmetry-distinct configurations, compared to 15–25 meV/atom for NMC622 compositions. The accessible low-hull configurations at NMC90 (those within 20 meV/atom) are fewer and require specific cation arrangements that enforce more ordered Ni placement.
Thermal stability proxy: oxygen release onset
One of the most practically important outputs for Ni-rich compositions is the thermal stability proxy score — an estimate of oxygen release onset derived from the energy cost of creating oxygen vacancy defects at the dominant surface termination. This is not a direct simulation of the thermal decomposition cascade; it is a first-principles indicator of which compositions and terminations present lower barriers to oxygen loss.
In the 400-candidate screening campaign, oxygen release tendency correlates strongly with surface energy anisotropy: compositions with large differences between low-index surface energies (high γ(110)/γ(001) ratios) tend to show lower oxygen release onset temperatures, consistent with the experimental observation that NMC particles with high (110) exposure are more prone to oxygen loss under heating.
The computational result identifies specific compositions — particularly those with Ni ≥ 0.8 and Co/(Co+Mn) ≥ 0.5 — as having elevated oxygen release risk based on their surface energy profiles. This matches the experimental literature's observation that high-Co, high-Ni compositions are particularly prone to oxygen evolution at elevated temperatures. The computational screening did not discover this relationship; it reproduced it from first principles, which provides confidence in its predictive power for novel compositions outside the well-studied range.
The composition-structure relationship for Li/Ni mixing
Li/Ni disorder is a structural defect mode specific to the layered oxide family: Ni²⁺ ions (ionic radius 0.69 Å) can occupy Li⁺ sites (ionic radius 0.76 Å) in the Li layer because of the similarity in ionic radius. This cation mixing reduces electrochemical performance by blocking Li⁺ channels.
In a computational screening campaign, the tendency toward Li/Ni disorder manifests as a distribution of formation energies across the symmetry-distinct configurations at each composition: some arrangements place Ni²⁺ predominantly in the transition metal layer (ordered), while others distribute it partially into the Li layer (disordered). The ordered configurations have lower formation energies — the ordered R3̄m structure is the thermodynamic ground state at low temperature — but the energy differences between ordered and disordered configurations are small (5–30 meV/atom for Ni-rich compositions), which explains why disorder is accessible under synthesis conditions.
The key screening finding: the energy gap between ordered and disordered configurations narrows as Ni fraction increases. At Ni = 0.6, ordered configurations are consistently 20–30 meV/atom lower than disordered; at Ni = 0.85, this gap shrinks to 5–10 meV/atom. This quantifies the well-known empirical observation that Ni-rich materials require slower cooling rates and lower synthesis temperatures to achieve ordered structures.
Practical implications for experimental program design
A 400-candidate computational screening campaign of this type reduces the experimental program in two ways. First, it eliminates compositions with unfavorable bulk stability profiles — those consistently above the hull or showing no ordered ground state configuration. Second, it provides specific structure files (VASP POSCAR, CIF) for the top-ranked compositions, enabling targeted synthesis of the most promising cation arrangements rather than an undifferentiated survey of the stoichiometry space.
The computational output does not replace electrochemical testing, thermal abuse testing, or full-cell cycling. These require experiment. What it replaces is the undirected synthesis exploration phase — the months spent making compositions that computation could have flagged as problematic before the furnace was turned on.
MaterSynq does not operate a wet lab. We produce computational predictions that require experimental validation before any conclusions about real battery performance can be drawn. The outputs from this type of campaign are a starting point for targeted experiments, not a substitute for them.