Skip to content
LumenaryAcademy
Frontier 90 min

Two-dimensional and quantum materials

Graphene's Dirac cones computed from a tight-binding model, the transition-metal dichalcogenides, twisted bilayers and moiré physics, topological insulators in outline with a Chern number computed, and the materials inside quantum computers — 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

One atom thick

In 2004 Andre Geim, Konstantin Novoselov and colleagues at the University of Manchester peeled graphite apart with adhesive tape until they had flakes one atom thick, found them on a silicon wafer by the faint colour they gave under an optical microscope, and showed that their conductivity could be changed by a gate voltage, like a transistor channel. Graphene had been a theorist's model since Philip Wallace's 1947 band calculation; few expected it to survive as a free-standing crystal. The 2010 Nobel Prize in Physics followed, and so did a new field: hundreds of layered crystals can be thinned to a single layer, each a material that is all surface, whose electrons can be tuned by a gate and whose layers can be stacked like paper in any order and at any angle.

Quantum materials is a broader and vaguer label for solids whose useful properties come from distinctly quantum effects acting collectively: topology, strong correlations between electrons, long-lived quantum coherence. Two-dimensional crystals are one of its main playgrounds, and the qubits of quantum computers are among its main customers. This lesson computes the band structure that started it all, then follows the ideas to moiré superlattices, topological invariants and the materials problems of quantum hardware. It builds on the tight-binding chain of Electronic structure and bands, on Semiconductor materials and on Superconductors.

DefinitionsH

Graphene's bands from a tight-binding model

Graphene's carbon atoms sit on a honeycomb: a triangular Bravais lattice with a basis of two atoms, A and B, 1.42 Å apart. Each atom uses three of its four valence electrons in σ bonds to its neighbours; the fourth sits in a pzp_z orbital perpendicular to the sheet, and those orbitals make the bands near the Fermi level.

a₁a₂δhoneycomb: two atoms per cellΓMKK′Brillouin zone
Left: the honeycomb lattice. Filled and hollow circles are the two sublattices, A and B; the thin arrows from one A atom are the three nearest-neighbour vectors δ, and the thick arrows the lattice vectors a₁ and a₂, which span a unit cell holding one A and one B atom. Right: the hexagonal Brillouin zone with its high-symmetry points Γ, M, K and K′, and the path Γ–M–K–Γ along which the bands are computed.

Definition— Nearest-neighbour tight-binding model of graphene

With one pzp_z orbital per atom and a hopping energy −t-t between nearest neighbours (t≈2.8t \approx 2.8 eV), a Bloch state ψk=cA ϕA,k+cB ϕB,k\psi_\mathbf{k} = c_A\,\phi_{A,\mathbf{k}} + c_B\,\phi_{B,\mathbf{k}} obeys the 2×22 \times 2 eigenproblem

(Δ/2−tf(k)−tf(k)∗−Δ/2)(cAcB)=E(cAcB),f(k)=∑j=13eik⋅δj,\begin{pmatrix} \Delta/2 & -t f(\mathbf k) \\ -t f(\mathbf k)^* & -\Delta/2 \end{pmatrix} \begin{pmatrix} c_A \\ c_B \end{pmatrix} = E \begin{pmatrix} c_A \\ c_B \end{pmatrix}, \qquad f(\mathbf k) = \sum_{j=1}^{3} e^{i\mathbf k \cdot \boldsymbol\delta_j},

where the δj\boldsymbol\delta_j are the three vectors from an A atom to its B neighbours, and Δ\Delta is an energy difference between the two sublattices (zero in pristine graphene). The bands are E±(k)=±Δ2/4+t2∣f(k)∣2E_\pm(\mathbf k) = \pm\sqrt{\Delta^2/4 + t^2 |f(\mathbf k)|^2}.

Theorem— Dirac cones

With Δ=0\Delta = 0, the two bands touch at the corners K and K′ of the Brillouin zone, and near a corner, at k=K+q\mathbf k = \mathbf K + \mathbf q, E±≈±ℏvF∣q∣E_\pm \approx \pm \hbar v_F |\mathbf q| with ℏvF=32t aCC\hbar v_F = \tfrac{3}{2} t\, a_\text{CC}, where aCCa_\text{CC} is the carbon–carbon distance.

Proof

Take δ1=a(0,1)\boldsymbol\delta_1 = a(0, 1), δ2,3=a(±32,−12)\boldsymbol\delta_{2,3} = a(\pm\tfrac{\sqrt3}{2}, -\tfrac12) with a=aCCa = a_\text{CC}, and the zone corner K=(4π33 a,0)\mathbf K = \bigl(\tfrac{4\pi}{3\sqrt3\,a}, 0\bigr). Then K⋅δj=0,2π3,−2π3\mathbf K \cdot \boldsymbol\delta_j = 0, \tfrac{2\pi}{3}, -\tfrac{2\pi}{3}, so the three phases ωj=eiK⋅δj\omega_j = e^{i\mathbf K\cdot\boldsymbol\delta_j} are the three cube roots of unity and f(K)=0f(\mathbf K) = 0: both bands are at E=0E = 0. To first order in q\mathbf q, f(K+q)≈i q⋅∑jωjδjf(\mathbf K + \mathbf q) \approx i\,\mathbf q \cdot \sum_j \omega_j \boldsymbol\delta_j, and a short calculation gives ∑jωjδj=3a2(i,1)\sum_j \omega_j \boldsymbol\delta_j = \tfrac{3a}{2}(i, 1), so f≈3a2(−qx+iqy)f \approx \tfrac{3a}{2}(-q_x + i q_y) and ∣f∣=3a2∣q∣|f| = \tfrac{3a}{2}|\mathbf q|. Hence E±=±t∣f∣≈±32ta∣q∣E_\pm = \pm t|f| \approx \pm \tfrac{3}{2} t a |\mathbf q|, a cone. With Δ≠0\Delta \neq 0 the same expansion gives E±≈±Δ2/4+(ℏvFq)2E_\pm \approx \pm\sqrt{\Delta^2/4 + (\hbar v_F q)^2}, a gap of Δ\Delta.

∎

The program diagonalises the Hamiltonian along the path Γ–M–K–Γ and draws the two bands as a text plot, then checks the cone's slope against the theorem.

python
import numpy as np

t = 2.8            # nearest-neighbour hopping, eV (a typical fitted value for graphene)
acc = 1.42         # carbon-carbon distance, angstrom
# Vectors from an A atom to its three B neighbours; the lattice vectors are differences of them.
delta = acc * np.array([[0.0, 1.0], [np.sqrt(3) / 2, -0.5], [-np.sqrt(3) / 2, -0.5]])
a1, a2 = delta[0] - delta[1], delta[0] - delta[2]

def bands(k, gap=0.0):
    """Eigenvalues of the 2x2 tight-binding Hamiltonian at wavevector k (1/angstrom).
    gap: energy +gap/2 on A sites and -gap/2 on B sites (as when the two sublattices differ)."""
    f = np.exp(1j * delta @ k).sum()
    H = np.array([[gap / 2, -t * f], [-t * np.conj(f), -gap / 2]])
    e = np.linalg.eigvalsh(H)
    return np.where(np.abs(e) < 1e-9, 0.0, e)

# Reciprocal lattice and the high-symmetry points of the hexagonal Brillouin zone.
A = np.array([a1, a2])
B = 2 * np.pi * np.linalg.inv(A).T          # rows b1, b2 with a_i . b_j = 2 pi delta_ij
G, M, K = np.zeros(2), B[0] / 2, (2 * B[0] + B[1]) / 3
path = [("G", G), ("M", M), ("K", K), ("G", G)]

print("bands along G-M-K-G: each * is a band, one character per 0.25 eV, | marks E = 0")
for i, ((n1, p1), (n2, p2)) in enumerate(zip(path, path[1:])):
    steps = np.linspace(0, 1, 7) if i == len(path) - 2 else np.linspace(0, 1, 7)[:-1]
    for s in steps:
        lo, hi = bands(p1 + s * (p2 - p1))
        row = [" "] * 73
        row[36] = "|"
        for e in (lo, hi):
            row[int(round(36 + e / 0.25))] = "*"
        label = n1 if s == 0 else (n2 if s == 1 else " ")
        print(f"{label} {lo:6.2f} {hi:6.2f} " + "".join(row))

print(f"\nat G: {bands(G)}   at M: {bands(M)}   at K: {bands(K)}")
q = 1e-3                                       # step away from K, 1/angstrom
slope = bands(K + np.array([q, 0.0]))[1] / q   # eV * angstrom: E = hbar v_F |q|
hbar = 6.582e-16                               # eV s
print(f"cone slope {slope:.3f} eV*A (3 t a_cc / 2 = {1.5 * t * acc:.3f}); v_F = {slope * 1e-10 / hbar:.2e} m/s")
print(f"with a 0.06 eV difference between the sublattices, the gap at K is {np.diff(bands(K, gap=0.06))[0]:.3f} eV")
Runs entirely in your browser, in an isolated worker. Nothing is sent to a server.

The bands span ±8.40 eV (±3t\pm 3t) at Γ, pass ±2.80 eV (±t\pm t) at M, and meet at exactly zero at K. The numerical slope, 5.962 eV Å, matches 32taCC=5.964\tfrac32 t a_\text{CC} = 5.964 eV Å, giving a Fermi velocity of 9.06×1059.06 \times 10^5 m/s, about 1/330 of the speed of light, close to measured values near 1×1061 \times 10^6 m/s. With one pzp_z electron per atom the lower band is exactly full, so the Fermi level sits at the cone's tip: graphene is a semimetal whose electrons near EFE_\text{F} behave like massless particles described by a two-dimensional Dirac equation, with the A/B sublattice playing the role of spin ("pseudospin").

ComputationC

Density of states and the electric-field effect

The cone has a striking consequence for the density of states, g(E)g(E). In two dimensions, a band with E∝∣q∣E \propto |\mathbf q| gives g(E)∝∣E∣g(E) \propto |E|, vanishing at the Dirac point; with four-fold degeneracy (two spins, two valleys K and K′), the carrier density for a Fermi energy EFE_\text{F} above the Dirac point is n=EF2/(π(ℏvF)2)n = E_\text{F}^2/\bigl(\pi(\hbar v_F)^2\bigr). So few states lie near E=0E = 0 that a gate can move the Fermi level by a large fraction of an electronvolt, switching the carriers from holes to electrons. The program samples the whole Brillouin zone to get g(E)g(E), then converts carrier densities into Fermi energies and gate voltages.

python
import numpy as np

t, acc = 2.8, 1.42                                       # eV, angstrom
delta = acc * np.array([[0.0, 1.0], [np.sqrt(3) / 2, -0.5], [-np.sqrt(3) / 2, -0.5]])
a1, a2 = delta[0] - delta[1], delta[0] - delta[2]
B = 2 * np.pi * np.linalg.inv(np.array([a1, a2])).T     # reciprocal lattice vectors (rows)

# Sample the whole Brillouin zone on an n x n grid and histogram the upper band |t f(k)|.
n = 600
s = (np.arange(n) + 0.5) / n
k = (s[:, None, None] * B[0] + s[None, :, None] * B[1]).reshape(-1, 2)
E = t * np.abs(np.exp(1j * k @ delta.T).sum(axis=1))
width = 0.5
hist, edges = np.histogram(E, bins=np.arange(0, 8.5 + width, width))
dos = hist / len(E) / width                             # states per eV per cell, per spin, upper band
print("density of states of the upper band (per eV per cell per spin)")
for lo, d in zip(edges[:-1], dos):
    print(f"{lo:4.2f}-{lo + width:4.2f} eV  {d:5.3f} " + "#" * int(round(d * 100)))

# Near the Dirac point the DOS is linear, so the carrier density goes as E_F^2.
hbar_vf = 1.5 * t * acc                                 # eV * angstrom
print("\ncarrier density n (per cm^2) -> Fermi energy above the Dirac point")
for dens in (1e11, 1e12, 1e13):
    n_per_A2 = dens * 1e-16                             # 1 cm^2 = 1e16 angstrom^2
    EF = hbar_vf * np.sqrt(np.pi * n_per_A2)            # n = E_F^2 / (pi (hbar v_F)^2), 4-fold degeneracy
    print(f"  n = {dens:.0e}: E_F = {1000 * EF:5.0f} meV")
eps0, e = 8.854e-12, 1.602e-19
for d_nm, eps_r in ((300, 3.9), (10, 3.9)):
    C = eps_r * eps0 / (d_nm * 1e-9)                    # F/m^2, a gate through SiO2
    print(f"gate on {d_nm} nm of SiO2: 1e12 cm^-2 needs {1e16 * e / C:5.1f} V")
Runs entirely in your browser, in an isolated worker. Nothing is sent to a server.

The histogram rises linearly from zero — 0.012, 0.036, 0.063 per eV in the first three bins — and peaks between 2.5 and 3.0 eV, at the van Hove singularity from the saddle point at M (E=tE = t). Near the Dirac point, 101210^{12} carriers per cm² put the Fermi level 106 meV from the cone's tip, and 101310^{13} put it 334 meV away, which a 300 nm silicon dioxide gate (the Manchester group's original substrate) reaches at about 14 V per 101210^{12} cm⁻². This tunability, a carrier mobility that exceeds 10⁵ cm²/V s in clean devices encapsulated in hexagonal boron nitride, and graphene's stiffness, strength and thermal conductivity are what made it famous.

graphenegapped honeycombtopological insulatorbulk gap, metallic surfacegapE_Fbulkbulk
Schematic band diagrams near the Fermi level (dashed). Left: graphene's Dirac cone, with no gap. Centre: a gap opened by making the two sublattices inequivalent (as in boron nitride, or graphene aligned on it). Right: a topological insulator, whose bulk has a gap (shaded bands) crossed by metallic surface states that form a single Dirac cone.
Worked examplesH

Transition-metal dichalcogenides and van der Waals stacks

The transition-metal dichalcogenides (TMDs), MX₂ with M = Mo or W and X = S, Se or Te, are layered crystals in which a sheet of metal atoms is sandwiched between two sheets of chalcogen atoms, the layers held together by van der Waals forces. Bulk MoS₂ is an indirect-gap semiconductor (about 1.2–1.3 eV); in 2010, groups at Columbia University and at Berkeley showed independently that a monolayer becomes a direct-gap semiconductor, at about 1.9 eV, and lights up in photoluminescence. In a monolayer, inversion symmetry is broken and strong spin–orbit coupling splits the bands differently in the K and K′ valleys, so circularly polarised light can address one valley or the other ("valleytronics"); and because there is so little material to screen the Coulomb attraction, electrons and holes bind into excitons with binding energies of hundreds of meV, compared with about 4 meV in bulk GaAs.

Two-dimensional crystals can be stacked without matching their lattices, because van der Waals bonding does not require it. Geim and Grigorieva called the result van der Waals heterostructures: graphene encapsulated in hexagonal boron nitride (an atomically flat insulator with a 6 eV gap) for clean transport, TMD heterobilayers whose electrons and holes sit in different layers, and tunnelling and light-emitting devices built layer by layer. The research workhorse is still mechanical exfoliation and transfer with polymer stamps, done one device at a time.

For electronics, monolayer TMDs are candidates for the transistor channel after silicon's scaling limits, because a channel 0.65 nm thick keeps good gate control at very short gate lengths where silicon's mobility collapses. Semiconductor manufacturers and research institutes in Belgium, Taiwan, the United States, Korea, China and elsewhere grow MoS₂ and WSe₂ on 200–300 mm wafers by metal–organic chemical vapour deposition and report research transistors, and in 2025 a team at Fudan University reported a simple 32-bit RISC-V processor built from about 5,900 MoS₂ transistors. As of 2026 the open problems are materials problems: low-resistance contacts to an atomically thin semiconductor, p-type devices to complement n-type MoS₂, gate dielectrics that do not damage or dope the channel, and defect densities in wafer-scale films. Industry roadmaps place 2D channels, if at all, in the 2030s. Lumenary EE's transistor scaling covers the device side.

ComputationC

Twisted bilayers and moiré physics

Lay one graphene sheet on another with a small twist θ\theta and a moiré pattern appears, a superlattice of period L=a/(2sin⁡θ2)L = a/(2\sin\tfrac{\theta}{2}) with a=2.46a = 2.46 Å, in which the stacking varies smoothly from regions where atoms sit on top of each other to regions where they interleave. Electrons tunnelling between the layers feel this landscape. In 2011 Rafi Bistritzer and Allan MacDonald predicted that at particular magic angles, the first near 1.1°, the bands near the Fermi level become almost perfectly flat. Their criterion compares the interlayer tunnelling energy ww with the kinetic energy ℏvFkθ\hbar v_F k_\theta across the moiré Brillouin zone, where kθ=2∣K∣sin⁡θ2k_\theta = 2|\mathbf K|\sin\tfrac{\theta}{2} is the separation of the two layers' Dirac points; the first flat band appears when α=w/(ℏvFkθ)\alpha = w/(\hbar v_F k_\theta) is close to 1/31/\sqrt3.

python
import numpy as np

a = 2.46                     # graphene lattice constant, angstrom
hbar_vf = 1.5 * 2.8 * 1.42   # eV * angstrom, from the tight-binding cone
w = 0.110                    # interlayer tunnelling energy, eV (a typical fitted value)
K = 4 * np.pi / (3 * a)      # distance from G to K, 1/angstrom

print("twist    moire period   atoms per moire cell   alpha = w / (hbar v_F k_theta)")
for deg in (5.0, 3.0, 2.0, 1.5, 1.1, 1.05, 0.8, 0.5):
    th = np.radians(deg)
    L = a / (2 * np.sin(th / 2))                 # moire period
    atoms = 4 * (L / a) ** 2                     # two layers x two atoms x (L/a)^2 cells
    k_theta = 2 * K * np.sin(th / 2)             # separation of the two layers' Dirac points
    alpha = w / (hbar_vf * k_theta)
    print(f"{deg:5.2f} deg  {L / 10:7.1f} nm     {atoms:12,.0f}            {alpha:5.3f}")

# The first magic angle of the Bistritzer-MacDonald model is where alpha is about 1/sqrt(3).
th_magic = 2 * np.arcsin(w * np.sqrt(3) / (hbar_vf * 2 * K))
print(f"\nalpha = 1/sqrt(3) at a twist of {np.degrees(th_magic):.2f} degrees")

# Kinetic versus interaction energy at 1.1 degrees: the flat band's width is a few meV,
# while the Coulomb energy between electrons one moire period apart (screened by hBN, eps ~ 4) is
L = a / (2 * np.sin(np.radians(1.1) / 2)) * 1e-10
coulomb = 1.602e-19 / (4 * np.pi * 8.854e-12 * 4 * L)
print(f"Coulomb energy of two electrons one moire period apart: {1000 * coulomb:.0f} meV")
Runs entirely in your browser, in an isolated worker. Nothing is sent to a server.

At 1.1° the moiré period is 12.8 nm and each moiré cell holds about 10,850 carbon atoms, far beyond what DFT can treat directly, which is why continuum models like Bistritzer and MacDonald's are the standard tool. The criterion gives a first magic angle of 1.07° for these parameters. The physics that follows is in the last line: a flat band has a bandwidth of only a few meV, while two electrons one moiré period apart repel each other with about 28 meV. When interactions beat kinetic energy, electrons stop behaving as independent particles.

In 2018 Pablo Jarillo-Herrero's group at MIT, with hBN crystals from Kenji Watanabe and Takashi Taniguchi at Japan's National Institute for Materials Science (the source of the clean hBN used in a very large share of the field's devices), reported that magic-angle twisted bilayer graphene becomes a correlated insulator at half filling of its flat bands and a superconductor nearby, all tunable by a gate. Since then the moiré family has grown: moiré TMD bilayers that act as tunable simulators of the Hubbard model; superconductivity in twisted bilayer WSe₂; and in 2023 the fractional quantum anomalous Hall effect — fractionally quantised Hall conductance with no applied magnetic field — in twisted bilayer MoTe₂, reported by groups at the University of Washington and Shanghai Jiao Tong University, followed in 2024 by an MIT group in rhombohedral five-layer graphene aligned with hBN.

ComputationC

Topological insulators, in outline

Band theory classifies insulators by their gap. Topology adds a finer classification: two insulators with the same gap can differ in a property of their occupied Bloch states over the whole Brillouin zone that cannot change without closing the gap. The prototype is the Chern number.

Definition— Berry curvature and Chern number

For a band with Bloch eigenvectors ∣uk⟩|u_\mathbf k\rangle over a two-dimensional Brillouin zone, the Berry connection is A(k)=i⟨uk∣∇kuk⟩\mathbf A(\mathbf k) = i\langle u_\mathbf k|\nabla_\mathbf k u_\mathbf k\rangle and the Berry curvature Ω(k)=∂kxAy−∂kyAx\Omega(\mathbf k) = \partial_{k_x}A_y - \partial_{k_y}A_x. The Chern number of an isolated band is C=12π∫BZΩ(k) d2kC = \dfrac{1}{2\pi}\displaystyle\int_\text{BZ}\Omega(\mathbf k)\,d^2k, which is always an integer. A two-dimensional insulator whose filled bands have total Chern number CC has Hall conductance σxy=C e2/h\sigma_{xy} = C\,e^2/h and ∣C∣|C| chiral conducting channels along each edge.

The integer is the same kind of object as the Euler characteristic of a surface — a global count that a smooth deformation cannot change. In 1988 Duncan Haldane showed that graphene's honeycomb could have C≠0C \neq 0 with no net magnetic field: add a complex hopping t2e±iϕt_2 e^{\pm i\phi} between second neighbours, which breaks time-reversal symmetry. The block computes the Chern number with the lattice method of Fukui, Hatsugai and Suzuki, which multiplies the overlaps of neighbouring eigenvectors around each small plaquette of a discretised zone; the phases sum exactly to 2πC2\pi C, even on a coarse grid, as long as the gap stays open.

python
import numpy as np

# Haldane's model: graphene's honeycomb (hopping t1) plus a sublattice energy difference M and a complex
# second-neighbour hopping t2*exp(i*phi), which breaks time-reversal symmetry without a net magnetic field.
delta = np.array([[0.0, 1.0], [np.sqrt(3) / 2, -0.5], [-np.sqrt(3) / 2, -0.5]])   # C-C distance = 1
nnn = np.array([delta[1] - delta[2], delta[2] - delta[0], delta[0] - delta[1]])  # second neighbours
B = 2 * np.pi * np.linalg.inv(np.array([delta[0] - delta[1], delta[0] - delta[2]])).T

def hamiltonian(k, t1, t2, phi, M):
    """2x2 Bloch Hamiltonians for an array of wavevectors k with shape (..., 2)."""
    f = t1 * np.exp(1j * k @ delta.T).sum(-1)
    dz = M - 2 * t2 * np.sin(phi) * np.sin(k @ nnn.T).sum(-1)
    d0 = 2 * t2 * np.cos(phi) * np.cos(k @ nnn.T).sum(-1)
    H = np.empty(k.shape[:-1] + (2, 2), dtype=complex)
    H[..., 0, 0], H[..., 1, 1] = d0 + dz, d0 - dz
    H[..., 0, 1], H[..., 1, 0] = f, np.conj(f)
    return H

def chern_and_gap(t1, t2, phi, M, n=36):
    """Chern number of the lower band by the Fukui-Hatsugai-Suzuki lattice method, and the minimum gap."""
    s = np.arange(n) / n
    k = s[:, None, None] * B[0] + s[None, :, None] * B[1]          # (n, n, 2) grid over the zone
    E, V = np.linalg.eigh(hamiltonian(k, t1, t2, phi, M))
    u = V[..., :, 0]                                               # lower-band eigenvectors
    link = lambda a, b: np.sum(np.conj(a) * b, axis=-1)
    u1, u2 = np.roll(u, -1, axis=0), np.roll(u, -1, axis=1)        # neighbours (periodic zone)
    u12 = np.roll(u1, -1, axis=1)
    F = np.angle(link(u, u1) * link(u1, u12) * link(u12, u2) * link(u2, u))   # Berry flux per plaquette
    return F.sum() / (2 * np.pi), (E[..., 1] - E[..., 0]).min()

t1, t2 = 1.0, 0.1
print(f"critical mass 3*sqrt(3)*t2 = {3 * np.sqrt(3) * t2:.3f} (for phi = 90 degrees)\n")
print(" phi (deg)    M      Chern number   smallest gap")
for phi_deg, M in ((0, 0.2), (0, 0.6), (90, 0.0), (90, 0.3), (90, 0.5), (90, 0.7), (-90, 0.3), (45, 0.2)):
    C, gap = chern_and_gap(t1, t2, np.radians(phi_deg), M)
    C = 0.0 if abs(C) < 5e-4 else C                                # print -0.000 as 0.000
    print(f"  {phi_deg:5d}     {M:4.2f}      {C:+6.3f}        {gap:6.3f}")
Runs entirely in your browser, in an isolated worker. Nothing is sent to a server.

With ϕ=0\phi = 0 (no time-reversal breaking) a sublattice mass opens an ordinary gap and C=0C = 0. With ϕ=90°\phi = 90° the lower band has C=+1C = +1 for every mass below the critical value 33 t2=0.5203\sqrt3\,t_2 = 0.520, and C=0C = 0 above it; reversing ϕ\phi gives C=−1C = -1. Near the transition, at M=0.50M = 0.50, the gap has shrunk to 0.039: the only way to change the integer is to close the gap, which happens at M=0.520M = 0.520. In a real sample, the boundary between a C=1C = 1 region and the vacuum (C=0C = 0) must therefore host states inside the gap — the edge channels — and they are robust because no local imperfection can change a global integer.

Kane and Mele showed in 2005 that spin–orbit coupling makes graphene two copies of Haldane's model with opposite Chern numbers for the two spins, a quantum spin Hall insulator protected by time-reversal symmetry; in graphene the gap is far too small to observe, but in 2007 Laurens Molenkamp's group in Würzburg found the effect in HgTe quantum wells, as Bernevig, Hughes and Zhang had predicted. The three-dimensional versions followed: Bi₂Se₃ and Bi₂Te₃, whose surfaces host a single Dirac cone, seen by photoemission from 2009. In 2013 a team led by Qi-Kun Xue at Tsinghua University observed the quantum anomalous Hall effect — Haldane's idea made real — in thin films of magnetically doped (Bi,Sb)₂Te₃, though only below about 30 mK at first; intrinsic magnetic topological insulators such as MnBi₂Te₄ raised the temperatures. As of 2026 the clearest practical prospect is in metrology, where quantum anomalous Hall devices are being explored as resistance standards that need no magnet.

ExplanationH

Materials for quantum computing

Every qubit platform is limited by materials, usually by tiny populations of defects that absorb energy or scramble phase. Lumenary EE's quantum hardware covers the devices; Lumenary Chemistry's quantum computing for chemistry covers what they might compute.

Superconducting circuits. Transmon qubits are aluminium or niobium circuits on silicon or sapphire with Al/AlOₓ/Al Josephson junctions, operated near 10 mK. Their energy relaxation is dominated by two-level systems — atomic-scale defects, mostly in the amorphous oxides on metal surfaces and at the metal–substrate and substrate–air interfaces — that resonantly absorb microwave photons. Materials work has moved the field: in 2021 a Princeton group showed that tantalum films, whose native oxide is thinner and better behaved, gave transmons with coherence times above 0.3 ms; in 2025 the same groups reported tantalum qubits on high-resistivity silicon with lifetimes up to about 1.7 ms. Surface treatments, cleaner substrates, encapsulation layers that prevent oxide growth, and junction fabrication with fewer defects are where coherence is now gained. A second materials problem is quasiparticles and ionising radiation: cosmic rays and natural radioactivity deposit energy in the substrate that briefly breaks Cooper pairs across many qubits at once, a correlated error that error correction handles badly; phonon traps and shielding are being engineered against it.

Defects as qubits. The nitrogen–vacancy centre in diamond (a nitrogen atom next to a missing carbon) has an electron spin that can be initialised and read out with green light at room temperature, with a ground-state splitting of 2.87 GHz; its coherence improves in isotopically purified ¹²C diamond, since ¹³C nuclei carry spins that cause noise. NV centres are excellent sensors of magnetic fields at the nanoscale and were used in 2015 for a loophole-free Bell test at Delft. Silicon carbide hosts divacancies and silicon vacancies with similar properties in a material with wafer-scale industrial production. Silicon itself, isotopically enriched to more than 99.9 per cent ²⁸Si, which has no nuclear spin, gives electron and nuclear spin qubits in devices close to those of standard chip manufacture, and optically active centres such as the T centre are being developed for networking. For all of these, the materials questions are how to place a single defect where you want it, keep its environment quiet and keep many of them identical.

Topological qubits, carefully. Majorana zero modes in superconductor–semiconductor nanowires would store quantum information non-locally and be protected from local noise. The experimental record has been difficult: a 2018 Nature paper reporting quantised Majorana conductance was retracted in 2021 after data problems were found, and Microsoft's February 2025 Nature paper on interferometric parity measurement in InAs–Al devices was published with an editorial note stating that its results did not represent evidence for Majorana zero modes; Microsoft's accompanying announcement of a topological processor went further than the peer-reviewed paper. Follow-up data in 2025 continued the debate. As of late 2026, a topological qubit has not been demonstrated to the satisfaction of most of the field. The materials challenge — atomically clean semiconductor–superconductor interfaces with very little disorder — is real and is being worked on.

ExplanationH

What is real, what is promising, and what is hype

Graphene's story is a useful calibration. It was announced in the late 2000s as a material that would transform electronics, batteries, water filtration and aircraft. Twenty years on, graphene and few-layer graphene are in real products — heat-spreading films in phones, conductive additives in battery electrodes and coatings, strengthening additives in polymers and concrete, sensors — sold by companies in China, Europe, North America and elsewhere, and the European Union's ten-year Graphene Flagship has moved many research results towards industry. It is not in your processor, and much of the "graphene" sold in bulk is in fact multilayer flakes of varying quality, which is why standards bodies developed measurement standards for graphene powders in the early 2020s. The other 2D materials are earlier on the same road.

Two habits protect against hype. First, ask for the number that matters in the application, measured in a device under realistic conditions: a mobility in a transistor with a gate dielectric and contacts, not in a suspended flake at 2 K; a coherence time for a qubit in a processor, not a champion device. Second, wait for independent replication of extraordinary claims. In July 2023 a claim of room-temperature, ambient-pressure superconductivity in a copper-doped lead apatite ("LK-99") was tested by laboratories on several continents within weeks; it was not reproduced, and the observed effects were traced to impurity phases such as copper sulfide. That was science working quickly and in public, and Superconductors tells related stories.

Common misconceptionsH

Common misconceptions

Misconception

"Graphene's electrons travel at the speed of light." They behave like massless particles, with a linear dispersion like light's, but their velocity is about 10610^6 m/s, roughly 1/300 of the speed of light, set by the hopping energy and the bond length.

Misconception

"A topological insulator is an insulator coated with a metal." Its metallic surface is not a separate layer; it is required by the bulk's topological invariant, and appears at any boundary with an ordinary insulator or vacuum. It cannot be removed without closing the bulk gap or breaking the protecting symmetry.

Misconception

"Any monolayer is a better semiconductor than bulk." Thinning changes the band structure (MoS₂ goes from indirect to direct gap), but monolayers are also more sensitive to their substrate, contacts, adsorbates and defects, and their room-temperature mobilities are lower than silicon's. Their advantage is electrostatic control at extreme thinness, not mobility.

Misconception

"Quantum materials are a new class of substance." The term is a shorthand for solids whose useful properties need quantum mechanics beyond the independent-electron band picture: topology, strong correlations, coherence. Many are old materials, such as bismuth telluride, a thermoelectric since the 1950s, looked at with new questions.

Exercises

Exercises

1. Two-dimensional and quantum materials

Warm-upHuman

Two graphene layers (a=2.46a = 2.46 Å) are twisted by 0.5°. What is the moiré period L=a/(2sin⁡θ2)L = a/(2\sin\tfrac{\theta}{2}), in nanometres, to the nearest integer?

2. Two-dimensional and quantum materials

CoreHuman

A graphene device is gated to an electron density of 4×10124 \times 10^{12} cm⁻². Using ℏvF=5.96\hbar v_F = 5.96 eV Å and four-fold degeneracy (spin and valley), so that n=EF2/(π(ℏvF)2)n = E_\text{F}^2/(\pi(\hbar v_F)^2), how far above the Dirac point is the Fermi level, in meV, to the nearest integer?

3. Two-dimensional and quantum materials

CoreHuman

Which change to graphene's nearest-neighbour model opens a gap at the K points?

Choose one

4. Two-dimensional and quantum materials

CoreHuman

Which statements about Chern numbers are correct? Select all that apply.

Choose all that apply

5. Two-dimensional and quantum materials

CoreHuman

Why did switching from niobium or aluminium to tantalum films, and later to high-resistivity silicon substrates, lengthen the lifetimes of superconducting transmon qubits?

Choose one

6. Two-dimensional and quantum materials

CoreComputational

Write graphene_bands(kx, ky, t=2.8, acc=1.42) returning the two band energies (E_minus, E_plus) of nearest-neighbour tight-binding graphene at wavevector (kx,ky)(k_x, k_y) in Å⁻¹, using the neighbour vectors aCC(0,1)a_\text{CC}(0, 1) and aCC(±32,−12)a_\text{CC}(\pm\tfrac{\sqrt3}{2}, -\tfrac12) and E±=±t∣f(k)∣E_\pm = \pm t|f(\mathbf k)|. The tests check Γ, K, M, the slope of the cone and random points against direct diagonalisation.

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

7. Two-dimensional and quantum materials

StretchComputational

Write a general chern_number(h, n=40) for two-band models whose Brillouin zone is the square [0,2π)2[0, 2\pi)^2, using the Fukui–Hatsugai–Suzuki method: take the lower-band eigenvector at every grid point, form the four link overlaps around each plaquette, sum the phases of their products, and divide by 2π2\pi. The tests use the Qi–Wu–Zhang model H=sin⁡kx σx+sin⁡ky σy+(m+cos⁡kx+cos⁡ky) σzH = \sin k_x\,\sigma_x + \sin k_y\,\sigma_y + (m + \cos k_x + \cos k_y)\,\sigma_z, which is topological for 0<∣m∣<20 \lt |m| \lt 2 and trivial for ∣m∣>2|m| \gt 2.

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

Further reading and resources