Skip to content
LumenaryAcademy
Frontier 85 min

Battery materials

The lithium-ion cell as a materials problem — host crystals and why their structure sets voltage, rate and life; capacity, swelling and cell energy density computed from chemistry; how cells age; solid electrolytes, sodium-ion and other chemistries; and the minerals and recycling behind them, as of late 2026.

Code tested · 6AI-draftedAwaiting review by a materials scientist

Your guides: Chloe · The Materials Scientist, Margaret · The Tester, Gideon · The Failure Analyst

MotivationH

A cell is a materials problem

Lumenary Chemistry has already shown how a lithium-ion cell works as chemistry: electrochemistry derives cell potentials and Faraday's law, and the chemistry of energy computes the theoretical capacities of the main electrode materials and why a whole cell reaches only about half of what its active materials could store. This lesson looks at the same cell as a materials scientist does. Why does lithium cobalt oxide give about 4 V and lithium iron phosphate about 3.4 V? Why is one voltage curve sloping and the other flat? Why does a nickel-rich cathode crack, silicon crumble, and a ceramic electrolyte short-circuit along its grain boundaries? Each answer is a fact about a crystal, an interface or a microstructure, and each has been decided in the last decade by materials work.

The stakes are large: lithium-ion cells now store solar power on grids from California to Inner Mongolia to South Australia, and their materials are industrial policy on every continent. The physics you met in Crystal structures, Diffusion and Phase diagrams is exactly the physics of a battery.

Note

Everything in this lesson is computation and description. A lithium-ion cell stores a great deal of energy, and a damaged, crushed, overheated or short-circuited one can enter thermal runaway, venting flammable, toxic gases and catching fire. Never open, puncture, heat or short a cell; stop using any battery that is swollen, hot or damaged; and take used batteries to a collection point. Battery laboratories work with cells in dry rooms and gloveboxes, behind fire-rated enclosures, for good reasons.

IntuitionH

Hosts, chemical potentials and a window

A lithium-ion cell has two intercalation hosts, crystals whose sites lithium can enter and leave while the framework stays intact. On discharge, lithium leaves the negative electrode as Li⁺ plus an electron; the ion crosses the electrolyte, the electron goes round the circuit, and they meet again in the positive electrode, where the electron reduces a transition-metal ion. Because both hosts survive, the reaction can run backwards thousands of times.

Definition— Open-circuit voltage of an intercalation cell

The open-circuit voltage is set by the difference in the chemical potential of lithium in the two electrodes: V=−[μLipos(x)−μLineg]/eV = -\left[\mu_\text{Li}^\text{pos}(x) - \mu_\text{Li}^\text{neg}\right]/e, where xx is the lithium content of the positive host. Averaged between two lithium contents x1<x2x_1 \lt x_2, and with lithium metal as the reference, Vˉ=−E(Lix2H)−E(Lix1H)−(x2−x1)E(Li)(x2−x1) e\bar V = -\dfrac{E(\mathrm{Li}_{x_2}\mathrm{H}) - E(\mathrm{Li}_{x_1}\mathrm{H}) - (x_2 - x_1)E(\mathrm{Li})}{(x_2 - x_1)\,e}, where H is the host and the energies can be measured or computed with DFT.

That second formula is why density functional theory became a battery-design tool in the late 1990s: an average voltage needs only three total energies. It also explains the chemistry of voltage. Lithium's chemical potential in the positive host is low when the electron it brings goes into a deep, low-energy state: a Co³⁺/Co⁴⁺ couple in an oxide sits low, so LiCoO₂ gives about 3.9 V against lithium. In LiFePO₄ the Fe²⁺/Fe³⁺ couple is lower in energy than in a simple iron oxide because the strongly covalent P–O bonds pull electron density away from the Fe–O bonds (the inductive effect), which is why a phosphate of iron, a cheap and modest metal, reaches 3.4 V.

electron energylowest unoccupied levelhighest occupied levelelectrolyte stability windowe·Vnegative electrode (graphite, Li)above the window: SEI formspositive electrodea 5 V cathode: below the window,oxidises the electrolyte
The energy-level picture of a cell, after Goodenough. The electrolyte is stable only between its lowest unoccupied and highest occupied levels. A graphite or lithium negative electrode sits above that window, so the electrolyte is reduced on its surface until a passivating solid–electrolyte interphase (SEI) stops it; a positive electrode below the window would oxidise the electrolyte. The cell voltage is the gap between the two electrodes' electrochemical potentials.

The figure shows why every commercial lithium-ion cell depends on a film nobody deposits on purpose. Graphite at about 0.1 V against lithium is far outside the stability window of the carbonate electrolyte, which is reduced on the first charge into a few-nanometre solid–electrolyte interphase: lithium carbonate, lithium fluoride, alkyl carbonates and polymers, an electronic insulator that still lets Li⁺ through. A good SEI makes the cell possible; its slow growth is the main way cells age on the shelf.

Worked examplesH

Cathode families and anodes

Three crystal structures carry almost all commercial lithium-ion positive electrodes, and their structure decides how lithium moves.

layered: LiCoO₂, NMC, NCAspinel: LiMn₂O₄, LNMOolivine: LiFePO₄, LMFPLi in 2D sheets3D network of paths1D channelsgold circles: lithium
The three cathode structure families, schematically. Layered oxides (left) hold lithium in sheets between transition-metal oxide slabs: it moves in two dimensions. Spinels (centre) have a three-dimensional network of paths. Olivines (right) have one-dimensional channels, which an antisite defect or an impurity can block.

Layered oxides, LiMO₂ with M = Co, Ni, Mn, Al, stack close-packed oxygen layers with lithium and transition metals in alternate layers of octahedral sites. Lithium diffuses quickly within its sheets. Goodenough's group showed LiCoO₂ in 1980; only about half its lithium can be removed at moderate voltage before the structure degrades, giving about 140 mAh/g (coated, doped LiCoO₂ charged to about 4.5 V now reaches 180–190 mAh/g in phones). Replacing cobalt by nickel — NMC (Ni, Mn, Co) and NCA (Ni, Co, Al) — raises capacity and cuts cost, and nickel-rich grades such as NMC811 deliver about 200 mAh/g. Nickel has a price: Ni²⁺ is close in size to Li⁺ (ionic radii about 0.69 and 0.76 Å), so some nickel sits in the lithium layer and blocks it (cation mixing), and highly delithiated nickel-rich oxides are less stable, losing oxygen at lower temperatures.

Spinels, LiMn₂O₄ (Thackeray and colleagues, 1983), have a three-dimensional network of diffusion paths and high rate capability, and use cheap manganese; but Mn³⁺ is Jahn–Teller active and manganese dissolves into the electrolyte, especially when warm. Substituting nickel gives LiNi₀.₅Mn₁.₅O₄ (LNMO) at about 4.7 V, a cobalt-free high-voltage cathode whose obstacle is the electrolyte, oxidised at that potential.

Olivines, LiFePO₄ (Padhi, Nanjundaswamy and Goodenough, 1997), hold lithium in one-dimensional channels, are poor electronic conductors, and looked impractical until nanosized particles with thin carbon coatings, developed by groups in Canada, the United States and elsewhere in the 2000s, made them fast. Their strong P–O bonds keep oxygen locked in even when abused, they need neither nickel nor cobalt, and by the mid-2020s LFP had taken most of China's electric-vehicle market and most stationary storage worldwide. Adding manganese (LMFP) raises the voltage to about 4.1 V for part of the capacity.

On the negative side, graphite holds one lithium per six carbons in a sequence of ordered "stages" and swells about 10 per cent along its stacking axis. Silicon stores nearly ten times as much per gram, as an alloy, but swells enormously; today it is blended with graphite as silicon oxide or nanostructured silicon–carbon composites. Lithium titanate, Li₄Ti₅O₁₂, a spinel, barely changes volume and lasts tens of thousands of cycles, at the cost of a 1.55 V operating potential. Lithium metal is the limit. The block computes what these choices mean per gram and per cubic centimetre, taking the swelling of the negative electrodes from crystallography.

python
F = 96485.0     # C/mol

def capacity(n, M):
    """Specific capacity in mAh/g: n electrons per formula unit of molar mass M (g/mol)."""
    return n * F / (3.6 * M)

# Positive electrodes: (Li actually cycled per formula unit, molar mass, crystal density g/cm^3)
cathodes = {
    "LiCoO2 (0.5 Li, to 4.2 V)":   (0.50, 97.87, 5.06),
    "NMC811 (0.73 Li, to 4.3 V)":  (0.73, 97.28, 4.75),
    "LiMn2O4 spinel (1 Li)":       (1.00, 180.81, 4.28),
    "LiFePO4 olivine (1 Li)":      (1.00, 157.76, 3.60),
}
print(f"{'positive electrode':28s} {'mAh/g':>6s} {'mAh/cm3':>8s}")
for name, (n, M, rho) in cathodes.items():
    q = capacity(n, M)
    print(f"{name:28s} {q:6.0f} {q * rho:8.0f}")

# Negative electrodes swell as they take up lithium. Get the swelling from crystallography:
# graphite's layers move apart from 3.354 A to 3.706 A in LiC6 (the layer area barely changes);
# silicon (diamond cubic, a = 5.431 A, 8 atoms per cell) becomes Li15Si4 (cubic, a = 10.63 A, 16 Si per cell).
swell_graphite = 3.706 / 3.354
swell_si = (10.63**3 / 16) / (5.431**3 / 8)
anodes = {   # (Li per formula unit, molar mass of the host, density of the host, volume ratio)
    "graphite -> LiC6":   (1.0, 72.06, 2.26, swell_graphite),
    "silicon -> Li15Si4": (3.75, 28.086, 2.33, swell_si),
}
print(f"\n{'negative electrode':20s} {'mAh/g':>6s} {'mAh/cm3':>8s} {'swelling':>9s} {'mAh per lithiated cm3':>22s}")
for name, (n, M, rho, s) in anodes.items():
    q = capacity(n, M)
    print(f"{name:20s} {q:6.0f} {q * rho:8.0f} {100 * (s - 1):7.0f} % {q * rho / s:16.0f}")
q_li = capacity(1, 6.94)
print(f"{'lithium metal':20s} {q_li:6.0f} {q_li * 0.534:8.0f}   (the electrode is the lithium itself)")
g = capacity(1, 72.06) * 2.26 / swell_graphite
s = capacity(3.75, 28.086) * 2.33 / swell_si
print(f"\nsilicon/graphite: {capacity(3.75, 28.086) / capacity(1, 72.06):.1f} times per gram, "
      f"but {s / g:.1f} times per lithiated volume")
Runs entirely in your browser, in an isolated worker. Nothing is sent to a server.

The volume per silicon atom grows from 20.0 ų in silicon to 75.1 ų in Li₁₅Si₄, a 275 per cent expansion, so silicon's tenfold advantage per gram (3578 against 372 mAh/g) shrinks to 2.9 times per cubic centimetre of fully lithiated electrode — and a real electrode must also leave room, in pores or in a compliant binder, for the swelling. Silicon particles larger than about 150 nm tend to crack on the first lithiation, each crack exposing fresh surface on which new SEI forms and consumes lithium. That, not the capacity, is the silicon problem.

ComputationC

Sloping and flat: voltage curves from a lattice gas

Discharge a layered-oxide cell and its voltage falls steadily from about 4.2 to 3.0 V; discharge an LFP cell and it sits near 3.4 V almost the whole way, then drops. The difference is thermodynamic, and the regular-solution model of Phase diagrams explains it. Treat the host's lithium sites as a lattice gas with occupancy xx; with an interaction parameter Ω\Omega between neighbouring lithium ions and the ideal entropy of mixing, the chemical potential of lithium is

μLi(x)=μ0+Ω (1−2x)+kBTln⁡x1−x,V(x)=−μLi(x)/e.\mu_\text{Li}(x) = \mu_0 + \Omega\,(1 - 2x) + k_\text{B}T \ln\frac{x}{1-x}, \qquad V(x) = -\mu_\text{Li}(x)/e .

If Ω\Omega is negative (lithium ions repel, and mixing is favourable), VV falls monotonically with xx: a sloping curve. If Ω>2kBT\Omega > 2k_\text{B}T, the free-energy curve has two minima and the system splits into a lithium-poor and a lithium-rich phase; between them, the common tangent fixes μLi\mu_\text{Li}, so the voltage is flat — the Gibbs phase rule at work, with two phases of a two-component system at fixed temperature and pressure leaving no degree of freedom. The block uses illustrative parameters chosen to mimic the two families.

python
import numpy as np

k = 8.617e-5          # Boltzmann constant, eV/K

def mu(x, mu0, omega, T):
    """Chemical potential of Li (eV, relative to Li metal) in a regular-solution lattice gas:
    x is the fraction of host sites filled, omega the interaction parameter."""
    return mu0 + omega * (1 - 2 * x) + k * T * np.log(x / (1 - x))

def miscibility_edge(omega, T):
    """Lower edge x1 of the two-phase region (symmetric model): solve kT ln(x/(1-x)) = omega (2x - 1)."""
    if omega <= 2 * k * T:
        return None                      # above the critical temperature: one phase at all x
    lo, hi = 1e-12, 0.5 - 1e-9
    f = lambda x: k * T * np.log(x / (1 - x)) - omega * (2 * x - 1)
    for _ in range(200):
        mid = 0.5 * (lo + hi)
        lo, hi = (mid, hi) if f(mid) < 0 else (lo, mid)
    return 0.5 * (lo + hi)

T = 298.15
layered = dict(mu0=-3.90, omega=-0.50)   # Li-Li repulsion: a sloping, single-phase curve
olivine = dict(mu0=-3.45, omega=+0.12)   # Li-Li attraction: phase separation, a flat plateau
x1 = miscibility_edge(olivine["omega"], T)
print(f"olivine-like model: two phases coexist for {x1:.3f} < x < {1 - x1:.3f} at 25 C\n")
print("  x     layered-like V   olivine-like V (equilibrium)   (one-phase, unstable)")
for x in np.linspace(0.04, 0.96, 12):
    v_lay = -mu(x, T=T, **layered)
    v_homog = -mu(x, T=T, **olivine)
    v_eq = -olivine["mu0"] if x1 < x < 1 - x1 else v_homog   # common tangent: a horizontal plateau
    print(f" {x:4.2f}      {v_lay:5.3f}            {v_eq:5.3f}                       {v_homog:5.3f}")

print("\nwidth of the two-phase region as temperature rises (olivine-like model):")
for T_c in (25, 150, 300, 400, 450):
    e = miscibility_edge(olivine["omega"], T_c + 273.15)
    print(f"  {T_c:3d} C: " + (f"{e:.3f} to {1 - e:.3f}" if e else "single phase at every x"))
print(f"critical temperature omega/2k = {olivine['omega'] / (2 * k) - 273.15:.0f} C")
Runs entirely in your browser, in an isolated worker. Nothing is sent to a server.

The layered-like curve falls from 4.44 V at x=0.04x = 0.04 to 3.36 V at x=0.96x = 0.96. The olivine-like model separates into two phases for 0.010<x<0.9900.010 \lt x \lt 0.990 at room temperature, so its equilibrium voltage is a flat 3.450 V across almost the whole range, while the single-phase curve it would follow if it could not separate bends the wrong way (rising from 3.410 to 3.490 V), the signature of an unstable solution. Heat it and the gap narrows — 0.049 to 0.951 at 150 °C, 0.344 to 0.656 at 400 °C — and closes above the critical temperature, 423 °C for these parameters. Real LiFePO₄–FePO₄ has a miscibility gap of the same kind at room temperature, which is why its curve is flat; the narrow solubility limits at each end and the gap's shrinkage in nanoparticles were measured in the 2000s.

Phase separation also sets mechanics and kinetics: the two phases have different lattice parameters (FePO₄ is about 6.5 per cent smaller in volume than LiFePO₄), so a moving phase boundary carries coherency strain, and at high rates small particles can transform through a non-equilibrium solid solution instead. The same mismatch, in a nickel-rich layered oxide charged above about 4.2 V, is a sudden contraction along the stacking axis, and it is what cracks those particles, as the next sections show.

ComputationC

From crystal to cell: energy density from the bill of materials

A cell's energy density is the active materials' energy diluted by everything else: current-collector foils, separator, electrolyte filling every pore, binder and conductive carbon, and finally the can or pouch. The block builds one repeating unit of a stacked cell per square centimetre — a positive electrode coated on both sides of aluminium foil, a separator, a negative electrode on both sides of copper foil, a second separator — from rounded, typical design values (not any particular commercial cell), and computes its energy per kilogram and per litre.

python
# Energy density of a cell stack from its materials, per cm^2 of electrode. One repeat unit: a positive
# electrode coated on both sides of aluminium foil, a separator, a negative electrode coated on both sides
# of copper foil, and a second separator. A layer t um thick of density rho g/cm^3 weighs 0.1*rho*t mg/cm^2.
# Inputs are rounded, typical design values, not data for any particular commercial cell.

def stack(q_pos, volts, rho_pos, q_neg, rho_neg, areal=4.0, n_to_p=1.10, porosity=0.25,
          active=0.96, al=12.0, cu=8.0, sep=12.0, lithium_metal=False):
    """(Wh/kg, Wh/L) of the electrode stack, excluding tabs and the can or pouch."""
    m_pos = areal / q_pos * 1000 / active                 # mg/cm^2 of coating, one side
    t_pos = m_pos / (0.1 * rho_pos * (1 - porosity))      # um
    if lithium_metal:                                     # dense lithium foil, no pores
        m_neg = areal * n_to_p / q_neg * 1000
        t_neg, neg_porosity = m_neg / (0.1 * rho_neg), 0.0
    else:
        m_neg = areal * n_to_p / q_neg * 1000 / active
        t_neg, neg_porosity = m_neg / (0.1 * rho_neg * (1 - porosity)), porosity
    pore_um = 2 * t_pos * porosity + 2 * t_neg * neg_porosity + 2 * sep * 0.45   # filled with electrolyte
    mass = (2 * m_pos + 2 * m_neg
            + 0.1 * 2.70 * al + 0.1 * 8.96 * cu           # aluminium and copper foils
            + 0.1 * 0.95 * 2 * sep * 0.55                 # polyethylene separators, 45 % porous
            + 0.1 * 1.20 * pore_um)                       # liquid electrolyte, 1.2 g/cm^3
    thickness = 2 * t_pos + 2 * t_neg + al + cu + 2 * sep
    energy = 2 * areal * volts                            # mWh/cm^2: two cells per repeat unit
    return 1000 * energy / mass, 1e4 * energy / thickness, mass, thickness

designs = {   # q_pos mAh/g, average V, density; q_neg mAh/g, density
    "LFP | graphite":              (160, 3.20, 3.60, 350, 2.24, {"porosity": 0.32, "active": 0.94}),
    "NMC811 | graphite":           (200, 3.65, 4.75, 350, 2.24, {}),
    "NMC811 | Si-graphite (500)":  (200, 3.60, 4.75, 500, 2.20, {}),
    "NMC811 | lithium metal":      (200, 3.75, 4.75, 3862, 0.534, {"lithium_metal": True}),
}
print(f"{'stack (4 mAh/cm2 per side)':28s} {'Wh/kg':>6s} {'Wh/L':>6s}")
for name, (qp, v, rp, qn, rn, kw) in designs.items():
    wkg, wl, _, _ = stack(qp, v, rp, qn, rn, **kw)
    print(f"{name:28s} {wkg:6.0f} {wl:6.0f}")

print("\nthicker electrodes, NMC811 | graphite:")
print("areal capacity   Wh/kg   Wh/L   share of mass that is active material")
for areal in (2.0, 3.0, 4.0, 5.0, 6.0):
    wkg, wl, mass, _ = stack(200, 3.65, 4.75, 350, 2.24, areal=areal)
    active_mass = 2 * areal / 200 * 1000 + 2 * areal * 1.1 / 350 * 1000    # mg/cm^2 of active material
    print(f"   {areal:3.1f} mAh/cm2    {wkg:5.0f}  {wl:5.0f}    {100 * active_mass / mass:4.0f} %")
Runs entirely in your browser, in an isolated worker. Nothing is sent to a server.

The stack of an NMC811–graphite cell reaches 328 Wh/kg and 922 Wh/L in this model, LFP–graphite 237 Wh/kg and 586 Wh/L; tabs, the can or pouch and unused space take roughly another 10 to 25 per cent, which is consistent with commercial cells of 2025–2026 at around 250–300 Wh/kg for nickel-rich chemistries and about 160–200 Wh/kg for LFP. Replacing graphite with a silicon–graphite blend at 500 mAh/g adds about 10 per cent; lithium metal, with only 10 per cent excess, would add half again, which is the prize that drives solid-state research. The second table is why electrode makers push to thicker coatings: going from 2 to 6 mAh/cm² raises the active share of the mass from 64 to 77 per cent and the stack's energy from 286 to 345 Wh/kg. The limit is transport. Lithium must diffuse through the whole depth of the electrode's electrolyte-filled pores, so thick electrodes lose power and charge unevenly, plating lithium at the separator side during fast charging — a diffusion problem with a time constant that grows as the thickness squared.

At pack level, "cell-to-pack" designs that put large cells straight into the pack without modules recovered much of LFP's deficit in the early 2020s. Lumenary EE's renewables and storage and Lumenary ME's vehicles and propulsion take the pack and the system from there.

Worked examplesH

How cells age

A cell ages along two clocks: calendar ageing, which happens even on a shelf and is fastest when warm and fully charged, and cycle ageing. Post-mortem studies sort the damage into a few modes: loss of lithium inventory (lithium trapped somewhere it can no longer cycle), loss of active material, and rising resistance.

  • SEI growth consumes lithium continuously. Reactants must cross the existing film, so growth is diffusion-limited and the capacity lost grows roughly as t\sqrt{t} — the mathematics of the oxide scales in Fatigue, creep and corrosion. A cell that has lost 4 per cent in its first year will lose about 4 per cent more by the fourth.
  • Lithium plating. Charge graphite too fast or too cold and lithium deposits on its surface as metal; some is stranded, and some grows filaments that can pierce the separator. This is why fast charging slows near full charge and in winter.
  • Cathode cracking. Nickel-rich oxides change their lattice parameters strongly and anisotropically near full charge. In the usual polycrystalline particles — secondary spheres a few micrometres across made of many small grains — this opens cracks along grain boundaries, which electrolyte then fills. Manufacturers have moved to single-crystal particles, which have no internal grain boundaries to crack, and to surface coatings and dopants.
  • Surface reconstruction and dissolution. Delithiated nickel-rich surfaces lose oxygen and rearrange into a resistive rock-salt layer; manganese and nickel dissolve, migrate to the negative electrode and damage its SEI ("cross-talk").
  • Silicon cracks and re-forms SEI on every cycle unless its particles are small and its binder compliant.

The most serious failure is thermal runaway: an internal short, overcharge or external heat starts exothermic reactions (SEI breakdown, electrolyte attacking the electrodes, oxygen from a charged layered oxide burning the organic electrolyte) that heat the cell faster than it can shed heat. Materials change the onset — LFP releases far less oxygen than nickel-rich oxides, ceramic-coated separators keep their shape when hot — and pack design and battery-management electronics do the rest.

ComputationC

Solid electrolytes

A solid electrolyte replaces the flammable liquid with an ion-conducting solid, promising safety and, if it allows a lithium-metal negative electrode, the energy gain the stack model showed. The families, each with its trade-offs:

  • Sulfides: thiophosphates such as Li₁₀GeP₂S₁₂ (2011), whose room-temperature conductivity matched the liquid electrolyte, and the cheaper argyrodites Li₆PS₅X (X = Cl, Br, I). They are soft enough to press into dense layers at room temperature, but react with moisture to release toxic hydrogen sulfide (so they are made in dry rooms) and have a narrow electrochemical window.
  • Oxides: the garnet Li₇La₃Zr₂O₁₂ (2007, doped with Al, Ga or Ta to keep its fast-conducting cubic form), stable against lithium metal but hard, brittle and sintered at high temperature; NASICON-type phosphates; and the thin-film LiPON used in microbatteries.
  • Halides such as Li₃YCl₆ and Li₃InCl₆ (from about 2018), stable at high voltage, and polymers such as PEO with lithium salts, flexible and processable but conducting well only when warm.

Bulk conductivity used to be the obstacle. The block takes representative, rounded room-temperature conductivities and activation energies, extrapolates them with the Arrhenius law, and computes the area-specific resistance (ASR =L/σ= L/\sigma, in Ω cm²) of a 30 µm layer and the voltage it costs at 1C for 4 mAh/cm² electrodes. A liquid sits in a porous separator, which conducts about eight times worse than the bulk liquid (its MacMullin number).

python
import math

k = 8.617e-5   # eV/K
# Representative, rounded room-temperature conductivities (S/cm) and activation energies (eV);
# each family spans a range in the literature, depending on composition and processing.
electrolytes = {
    "liquid, 1 M LiPF6 in carbonates":   (1.0e-2, 0.15),
    "sulfide, Li10GeP2S12 (LGPS)":        (1.2e-2, 0.25),
    "sulfide, argyrodite Li6PS5Cl":       (2.0e-3, 0.35),
    "oxide, garnet Li7La3Zr2O12 (doped)": (8.0e-4, 0.35),
    "thin film, LiPON":                   (2.0e-6, 0.55),
}

def sigma(s25, ea, celsius):
    """Arrhenius extrapolation from 25 C."""
    T = celsius + 273.15
    return s25 * math.exp(-ea / k * (1 / T - 1 / 298.15))

L = 30e-4         # a 30 um electrolyte layer, in cm
macmullin = 8.0   # a liquid in a porous separator conducts ~8 times worse than the bulk liquid
i_1c = 4e-3       # A/cm^2: 1C for 4 mAh/cm^2 electrodes
print(f"{'30 um layer':36s} {'sigma -20 C':>11s} {'ASR at -20/25/60 C (ohm cm2)':>30s} {'drop at 1C, 25 C':>17s}")
for name, (s25, ea) in electrolytes.items():
    f = macmullin if name.startswith("liquid") else 1.0
    asr = [L * f / sigma(s25, ea, t) for t in (-20, 25, 60)]
    print(f"{name:36s} {sigma(s25, ea, -20):11.1e}   {asr[0]:8.3g} {asr[1]:8.3g} {asr[2]:8.3g}   {1000 * i_1c * asr[1]:10.0f} mV")

print("\nthickest layer that keeps the resistance below 10 ohm cm2 at 25 C:")
for name, (s25, ea) in electrolytes.items():
    if not name.startswith("liquid"):
        print(f"  {name:36s} {1e4 * 10 * s25:8.1f} um")
Runs entirely in your browser, in an isolated worker. Nothing is sent to a server.

At room temperature, a 30 µm sulfide separator costs 1 to 6 mV at 1C and a garnet 15 mV, against 10 mV for the liquid in its separator: in the bulk, the best solids are as good as or better than the liquid. The cold column shows the cost of a higher activation energy (a garnet's resistance rises to 42 Ω cm² at −20 °C), and LiPON, at 2 µS/cm, only works as a film a fraction of a micrometre thick, which is how thin-film microbatteries use it.

So why are solid-state cells not everywhere? Because the resistance that matters is at the interfaces, and solids do not flow. Electrode particles swell and shrink on every cycle and lose contact with a rigid electrolyte unless the stack is held under pressure; sulfides decompose at the high-voltage cathode unless it is coated; and lithium metal, which in a liquid grows dendrites, in a ceramic grows along grain boundaries and pores and short-circuits the cell above a critical current density that depends on microstructure, pressure and interface quality — fracture mechanics from Fracture and toughness applied to a ceramic being split by soft lithium. Manufacturing thin, dense, defect-free layers at scale is its own challenge.

As of late 2026, "semi-solid" cells, which keep some liquid, have been in production vehicles in China since 2024. All-solid-state cells, mostly sulfide-based and some "anode-free" (plating lithium onto a bare current collector on the first charge), are on pilot lines in Japan, Korea, China, the United States and Europe, with small-volume vehicle use announced for about 2027–2028, at a cost per kWh still above conventional cells. Screening for new solid electrolytes is one of the more productive uses of the methods in AI-assisted materials discovery.

Worked examplesH

Sodium-ion and other chemistries

Sodium-ion cells swap lithium for sodium, which is about a thousand times more abundant in the Earth's crust and is obtained from ordinary salt. Na⁺ is larger than Li⁺ and does not form a stable compound with graphite, so the negative electrode is hard carbon, a disordered carbon made from biomass or resins, with nanopores and curved layers that store sodium. Positive electrodes come in three families: layered sodium transition-metal oxides; polyanion compounds such as Na₃V₂(PO₄)₂F₃; and Prussian white, Na₂Fe[Fe(CN)₆], with large open channels and only cheap iron. Sodium-ion cells store less energy than lithium-ion (roughly LFP-like at best), but work better in the cold, can be shipped at zero volts, need no lithium, cobalt or nickel, and can use aluminium instead of copper on the negative side, because sodium does not alloy with aluminium. CATL showed its first sodium-ion cells in 2021; production began in China in 2023, and second-generation cells announced in 2025 at about 175 Wh/kg were shown in a passenger car in early 2026, with sales planned for later that year; BYD and others are building sodium-ion factories too. The economics move with the price of lithium carbonate, which fell steeply from its 2022 peak.

Other chemistries, in outline:

  • Lithium–sulfur: cheap sulfur stores 1672 mAh/g, but dissolved polysulfides shuttle between the electrodes and the electrode swells by about 80 per cent; cycle life is the barrier.
  • Lithium–air: a very high theoretical energy, but clogging discharge products and electrolyte side reactions keep it a research topic.
  • Aqueous zinc, iron–air and flow batteries (vanadium and others): cheap, water-based chemistries for multi-hour to multi-day grid storage, where energy per kilogram hardly matters; vanadium flow batteries are established in China, and iron–air systems reached early commercial deployment in the United States in the mid-2020s.
ExplanationH

Critical minerals and recycling

A nickel-rich electric-vehicle battery of 60 kWh contains of the order of 5–8 kg of lithium (as metal), tens of kilograms of nickel, a few to ten kilograms of cobalt depending on the grade, and 50–70 kg of graphite; LFP replaces the nickel and cobalt with iron and phosphate. Supply is concentrated, according to the USGS Mineral Commodity Summaries and the IEA: the Democratic Republic of the Congo mines around 70 per cent of the world's cobalt, Indonesia more than half of its nickel, Australia, Chile, China and Argentina most of its lithium, and China most of its graphite; China also refines the majority of the world's lithium chemicals and cobalt and almost all battery-grade graphite. China placed export licences on graphite in 2023, and every large economy now has a critical-minerals policy. Lumenary Materials' Sustainable materials takes up the general case; cobalt mining's documented social problems, including artisanal mining and child labour, are one reason for the shift towards nickel-rich, cobalt-free and LFP cathodes.

Recycling routes, from rough to refined:

  • Pyrometallurgy smelts cells: cobalt, nickel and copper go to an alloy, lithium and manganese mostly to the slag. Robust with mixed feed, energy-hungry, and poor for LFP, which contains little valuable metal.
  • Hydrometallurgy shreds cells into "black mass", leaches it in acid and separates battery-grade salts by precipitation and solvent extraction. It recovers lithium well and dominates new capacity, at the cost of reagents and wastewater treatment.
  • Direct recycling keeps the cathode crystal, restores its lithium ("relithiation") and reuses it, as studied at Argonne's ReCell Center. It saves the most energy but needs sorted feed of known chemistry, and the recovered cathode may be a generation out of date.

The EU Battery Regulation of 2023 set the first binding framework: recycling efficiencies (65 per cent of the mass of lithium-based batteries by the end of 2025), material recovery targets (lithium 50 per cent by the end of 2027 and 80 per cent by 2031; cobalt, nickel and copper 90 then 95 per cent), minimum recycled content in new industrial and vehicle batteries from August 2031 (16 per cent cobalt, 6 per cent lithium, 6 per cent nickel, rising in 2036), carbon-footprint declarations and a digital battery passport. In September 2026 the Commission published an assessment of whether to revise the lithium-recovery targets, so check the current text. China, the largest recycler by far, has its own standards for recycling enterprises, and the United States supports recycling through its critical-minerals programmes. Recycling cannot yet supply a growing market — batteries made today return in ten to fifteen years — but it is the long-term mine.

Misconception

"Batteries are not recyclable." Lithium-ion batteries are recycled at industrial scale in Asia, Europe and North America, with metal recoveries above 90 per cent for cobalt, nickel and copper in modern hydrometallurgical plants. What is difficult is collection, safe transport of damaged cells, economics for low-value chemistries such as LFP, and recovering graphite and electrolyte, which most processes still lose.

Common misconceptionsH

Common misconceptions

Misconception

"Higher capacity per gram means a better electrode." Volume, swelling, voltage, coulombic efficiency and cost matter as much. Silicon's tenfold gravimetric advantage shrinks to about three times per lithiated volume, and its cracking and SEI growth limit how much can be used.

Misconception

"Solid-state batteries cannot catch fire, and they are about to replace lithium-ion." Removing the liquid removes the main fuel, but lithium metal is reactive, sulfide electrolytes release hydrogen sulfide with moisture, and a charged cathode still stores energy. As of 2026 all-solid-state cells are on pilot lines; liquid and semi-solid lithium-ion and sodium-ion cells will carry most of the market for years.

Exercises

Exercises

1. Battery materials

Warm-upHuman

Lithium titanate, Li₄Ti₅O₁₂ (molar mass 459.1 g/mol), takes up three lithium ions per formula unit to become Li₇Ti₅O₁₂, reducing three Ti⁴⁺ to Ti³⁺. What is its theoretical specific capacity in mAh/g, to the nearest integer? (F=96 485F = 96\,485 C/mol; 1 mAh = 3.6 C.)

2. Battery materials

CoreHuman

A cell must keep 80 % of its capacity after 1000 full cycles, and all fade comes from charge lost on each cycle at a constant coulombic efficiency. What is the largest allowed loss per cycle, in parts per million of the cell's charge, to the nearest integer?

3. Battery materials

CoreHuman

Why is the open-circuit voltage of a LiFePO₄ electrode almost constant over most of its discharge?

Choose one

4. Battery materials

CoreHuman

A solid-state cell has a 50 µm garnet electrolyte layer with an ionic conductivity of 0.5 mS/cm. Ignoring the interfaces, what voltage, in mV, does the layer's ionic resistance cost at a current density of 4 mA/cm²?

5. Battery materials

CoreHuman

LiFePO₄ looked impractical in 1997 and became the leading chemistry for grid storage. Which statements about its materials engineering are correct? Select all that apply.

Choose all that apply

6. Battery materials

CoreComputational

Calendar ageing from SEI growth is often modelled as a capacity loss proportional to the square root of time, Qloss=atQ_\text{loss} = a\sqrt{t}. Write fit_sqrt_fade(days, loss), which returns the least-squares value of aa for a line through the origin in t\sqrt{t}, and days_to_loss(a, target), which returns how many days the model takes to reach a target loss. (Losses are in per cent.)

Runs entirely in your browser, in an isolated worker. Nothing is sent to a server.

7. Battery materials

StretchComputational

For the symmetric regular-solution lattice gas of the lesson, the two-phase region runs from x1x_1 to 1−x11 - x_1, where x1<1/2x_1 \lt 1/2 solves kBTln⁡x1−x=Ω(2x−1)k_\text{B}T \ln\dfrac{x}{1-x} = \Omega(2x - 1). Write miscibility_edge(omega, T) that returns x1x_1 (by bisection), or None when Ω≤2kBT\Omega \le 2k_\text{B}T and there is no gap. The tests reproduce the lesson's numbers (0.010 at 25 °C and 0.162 at 300 °C for Ω=0.12\Omega = 0.12 eV).

Runs entirely in your browser, in an isolated worker. Nothing is sent to a server.

Further reading and resources