Magic-Angle Twisted Bilayer Graphene

• Magic-angle twisted bilayer graphene is a moiré system formed by stacking two graphene sheets with a small twist near 1.1°, resulting in nearly flat electronic bands.
• The flat bands enhance electron correlations, giving rise to superconductivity, interaction-induced insulating states, magnetism, nematicity, and quantized Hall effects.
• Experimental control through twist angle, pressure, substrate alignment, and gating is key to tuning MATBG's rich phase diagram and emergent many-body phenomena.

Magic-angle twisted bilayer graphene (MATBG) is a graphene-based moiré system created by stacking two atomically thin graphene sheets with a small relative twist. Near the largest magic angle, $\theta_0 \approx 1.1^\circ$, the lowest conduction and highest valence moiré minibands become extremely narrow, kinetic energy is suppressed, and electron-electron interactions dominate. In this regime, MATBG hosts superconductivity, interaction-induced insulating states, magnetism, electronic nematicity, linear-in-$T$ low-temperature resistivity, and quantized anomalous Hall states. Direct momentum-space spectroscopy further shows that the low-energy bands, the remote bands, and their evolution with twist angle are strongly reshaped by lattice relaxation and interlayer-spacing modulation, rather than by twist angle alone (Andrei et al., 2020, Li et al., 2024).

1. Moiré geometry and the magic-angle mechanism

MATBG is governed by the long-wavelength moiré superlattice generated by the relative rotation of two honeycomb Dirac systems. The moiré period is

$L \sim \frac{a_0}{2\sin(\theta/2)},$

and the characteristic momentum scale is

$k_\theta \sim 2K\sin(\theta/2).$

At large twist angles the layers are only weakly coupled and the system behaves much like two nearly independent graphene sheets. As $\theta$ is reduced, the two layer Dirac cones move closer in momentum space, moiré hybridization becomes strong, and the coupling produces saddle points and van Hove singularities. In the continuum description, the low-energy Hamiltonian is written as two graphene-layer Dirac Hamiltonians coupled by a moiré tunneling term $T(\mathbf r)$, and the ratio of interlayer tunneling to the kinetic-energy scale, often summarized by $w/(\hbar v_F k_\theta)$, controls the band flattening (Andrei et al., 2020).

The original Bistritzer-MacDonald mechanism identifies a sequence of magic angles at which the Fermi velocity vanishes. The largest of these, near $\theta_0 \approx 1.1^\circ$, is the one most closely associated with correlated phenomena. The filling factor is naturally defined on the moiré unit cell, with $v=-4$ for empty flat bands, $v=+4$ for full flat bands, and $T$0 at charge neutrality. This filling-based description is central because very small changes in carrier density can cause large changes in behavior. Hydrostatic pressure provides a second axis of control by increasing interlayer coupling and shifting the magic condition upward in angle; in first-principles calculations this motivates the use of a pressured $T$1 structure as a computational proxy for the flat-band regime (Andrei et al., 2020, Chen et al., 2020).

2. Evolution of flat and remote bands and the role of lattice relaxation

NanoARPES and AFM measurements on the same regions of the same samples have directly mapped how the electronic structure evolves from $T$2 to $T$3. Near $T$4, the low-energy dispersion contains an isolated flat band near the Fermi level, separated by a gap from remote bands at higher binding energy. By following peak positions in energy-distribution curves, the flat-band bandwidth was extracted as $T$5 meV. As the twist angle increases, the flat band evolves into a much more dispersive, M-shaped band, and the bandwidth grows to $T$6 meV at $T$7. The measured dispersions agree well with Bistritzer-MacDonald-type calculations, reinforcing the continuum-model picture that the flat band emerges from moiré hybridization of the two graphene Dirac cones (Li et al., 2024).

The remote bands provide a separate diagnostic of microscopic interlayer tunneling. In NanoARPES they split into two pockets, $T$8 and $T$9, arising from hybridization between the top- and bottom-layer Dirac cones. Comparison of the measured separation $L \sim \frac{a_0}{2\sin(\theta/2)},$0 with theory yields $L \sim \frac{a_0}{2\sin(\theta/2)},$1 meV and $L \sim \frac{a_0}{2\sin(\theta/2)},$2, where $L \sim \frac{a_0}{2\sin(\theta/2)},$3. A further twist-angle-dependent effect is the transfer of spectral weight between the two pockets: at $L \sim \frac{a_0}{2\sin(\theta/2)},$4 the inner pocket $L \sim \frac{a_0}{2\sin(\theta/2)},$5 is stronger, whereas at $L \sim \frac{a_0}{2\sin(\theta/2)},$6 and above the outer pocket $L \sim \frac{a_0}{2\sin(\theta/2)},$7 becomes dominant. Simple continuum calculations with fixed interlayer spacing reproduce the dispersions but not this intensity switching; including variation of the average interlayer spacing $L \sim \frac{a_0}{2\sin(\theta/2)},$8 does reproduce it. In the ARPES intensity expression, the phase factor $L \sim \frac{a_0}{2\sin(\theta/2)},$9 controls interference between the two layers and is identified as the dominant driver of the spectral-weight transfer. Changing $k_\theta \sim 2K\sin(\theta/2).$0 from $k_\theta \sim 2K\sin(\theta/2).$1 Å to $k_\theta \sim 2K\sin(\theta/2).$2 Å can switch the relative strength of $k_\theta \sim 2K\sin(\theta/2).$3 and $k_\theta \sim 2K\sin(\theta/2).$4, and fitting the intensity ratio gives an average interlayer spacing of about $k_\theta \sim 2K\sin(\theta/2).$5 Å near the magic angle (Li et al., 2024).

These measurements show that the lattice is not a rigid moiré overlay. The extracted $k_\theta \sim 2K\sin(\theta/2).$6 decreases near the magic angle, indicating stronger lattice relaxation: AA-stacked regions shrink and become more isolated, AB-stacked regions expand, and the spacing approaches the AB value of about $k_\theta \sim 2K\sin(\theta/2).$7 Å. Calculated local density of states further shows that near the magic angle the flat-band states become more localized in AA regions, whereas remote-band states are more delocalized and resemble AA-like and AB-like dispersions. The study explicitly connects this to the heavy-fermion description of MATBG, in which localized flat-band states hybridize with more itinerant remote-band states, and also to the idea that lattice corrugation can act like a pseudo-magnetic confinement mechanism (Li et al., 2024).

3. Interaction scales and spectroscopic many-body structure

Near the magic angle, the flat-band bandwidth is comparable to, or smaller than, the relevant Coulomb scale. A commonly used estimate is

$k_\theta \sim 2K\sin(\theta/2).$8

where $k_\theta \sim 2K\sin(\theta/2).$9 is the moiré period. In the spectroscopic data at $\theta$0, the flat-band bandwidth is smaller than this interaction energy, placing MATBG in the regime where correlations are strongly enhanced and kinetic energy is suppressed by band flattening (Li et al., 2024).

STM/STS established early direct evidence for strong many-body correlations. At weak doping, the local density of states shows two sharp peaks identified as van Hove singularities from the nearly flat conduction and valence bands, along with weaker step-like features attributed to remote bands. When one flat band is partially filled, however, the spectra become strongly distorted and broadened over an energy range of about $\theta$1–$\theta$2 meV, and the effect also strongly modifies the opposite flat-band peak. The data were not explained by a non-interacting continuum model or by self-consistent Hartree-Fock, whereas a phenomenological extended Hubbard model treated by cluster exact diagonalization reproduced the qualitative broadening and splitting at partial filling. In this description, the relevant on-site interaction $\theta$3 is comparable to or larger than the band width, and the system is more naturally viewed as Hubbard-like than weakly perturbed band-like (Xie et al., 2019).

High-resolution STM then resolved a filling-dependent cascade of spectroscopic transitions at each integer moiré filling. The spectra exhibit a cascade of 14 fine features around $\theta$4, organized in four approximately evenly spaced intervals between $\theta$5 and $\theta$6, and another four between $\theta$7 and $\theta$8. The interpretation is that Coulomb interactions split the fourfold-degenerate flat bands into Hubbard sub-bands. The extracted interaction scales are $\theta$9 meV and $T(\mathbf r)$0 meV, while a perpendicular magnetic field of $T(\mathbf r)$1 T strongly suppresses the chemical-potential cusps and zero-bias density-of-states dips associated with the cascade (Wong et al., 2019).

A momentum-resolved extension of this picture was provided by the quantum twisting microscope. Away from the magic angle, at $T(\mathbf r)$2, the bands largely follow the noninteracting Bistritzer-MacDonald model. At the magic angle, $T(\mathbf r)$3, the bands are instead transformed by interactions: over most of the mini-Brillouin zone they are almost completely flattened and separated by a large interaction-induced gap of about $T(\mathbf r)$4 meV, but near the $T(\mathbf r)$5 point they remain dispersive and gapless. Upon doping, the $T(\mathbf r)$6-point feature stretches strongly, giving an effective bandwidth of about $T(\mathbf r)$7 meV at $T(\mathbf r)$8 and $T(\mathbf r)$9 meV at $w/(\hbar v_F k_\theta)$0. The same measurements identify Mott-like cascades of the heavy particles, Dirac revivals of the light particles, and a persistent low-energy excitation at approximately $w/(\hbar v_F k_\theta)$1 meV tied to the heavy sector (Xiao et al., 25 Jun 2025).

4. Correlated phases and transport phenomenology

The experimentally established phase diagram of MATBG includes superconductivity, interaction-induced insulating states, magnetism, electronic nematicity, strange-metal transport, and topological anomalous Hall behavior. In high-quality hBN-encapsulated devices, low-temperature transport revealed superconducting domes adjacent to insulating phases. Interaction-induced insulating states appear at integer moiré fillings, especially around $w/(\hbar v_F k_\theta)$2, where transport shows resistive gaps or insulating behavior and local probes detect dips in the density of states and peaks in inverse compressibility. MATBG can also host orbital ferromagnetism and Chern insulating states; one state at $w/(\hbar v_F k_\theta)$3 exhibited Hall resistance quantized to within $w/(\hbar v_F k_\theta)$4 of $w/(\hbar v_F k_\theta)$5, and a sequence of Chern insulators with $w/(\hbar v_F k_\theta)$6 was reported at $w/(\hbar v_F k_\theta)$7 (Andrei et al., 2020).

Electronic nematicity appears as charge stripes in STM-based spectroscopy and as resistivity anisotropy in transport. Another central transport signature is approximately linear-in-temperature low-temperature resistivity over a broad filling range within the flat bands. This strange-metal behavior is often written as

$w/(\hbar v_F k_\theta)$8

and the review literature emphasizes its similarity to Planckian dissipation, while also noting that phonon scattering remains an alternative explanation that has not been fully excluded (Andrei et al., 2020).

Screened devices isolate this metallic state more clearly. In a superconducting device with $w/(\hbar v_F k_\theta)$9 and screening-layer separation $\theta_0 \approx 1.1^\circ$0, screening suppresses the correlated insulating state near $\theta_0 \approx 1.1^\circ$1, exposing an uninterrupted metallic ground state. In the center of the flat band the resistivity obeys

$\theta_0 \approx 1.1^\circ$2

down to $\theta_0 \approx 1.1^\circ$3 mK and over a broad doping window roughly

$\theta_0 \approx 1.1^\circ$4

The same regime exhibits linear magnetoresistivity,

$\theta_0 \approx 1.1^\circ$5

with reported slopes $\theta_0 \approx 1.1^\circ$6. By contrast, near charge neutrality, near full filling, and in devices twisted away from the magic angle, transport reverts to the Fermi-liquid forms $\theta_0 \approx 1.1^\circ$7 and $\theta_0 \approx 1.1^\circ$8. Suppressing superconductivity with a small perpendicular magnetic field reveals the strange metal beneath it, with a reported critical field $\theta_0 \approx 1.1^\circ$9 in one case (Jaoui et al., 2021).

5. Pressure, substrate alignment, bosonic coupling, and magnetic field control

MATBG is unusually sensitive to external control parameters. Nearby metallic gates alter screening and can favor superconductivity over insulating states. Alignment of one graphene layer to hBN breaks $v=-4$0 symmetry, opens a gap at the Dirac points, modifies the correlated phase diagram, and can assist orbital magnetism and topological states. Pressure increases interlayer coupling $v=-4$1 and shifts the magic angle upward, providing a route to flat-band physics at larger structural twist angles (Andrei et al., 2020).

A distinct spectroscopic signature of the superconducting regime is the appearance of replica flat bands in hBN-unaligned MATBG. Micro-ARPES found higher-binding-energy copies of the flat band in superconducting devices with twist angles near $v=-4$2–$v=-4$3, uniformly spaced by $v=-4$4 meV. These replicas are absent in hBN-aligned MATBG and in off-magic-angle TBG such as $v=-4$5 and $v=-4$6. Frozen-phonon calculations identify the most likely origin as strong inter-valley electron-phonon coupling to the in-plane transverse optical phonon near the graphene $v=-4$7 point, and the replica intensities follow a Poisson-like distribution with $v=-4$8. The authors explicitly note that this does not prove that phonons are the sole or primary pairing glue, but it does establish strong bosonic dressing in superconducting MATBG (Chen et al., 2023).

Pressure and electric field can also stabilize or suppress magnetism. First-principles calculations for pressured MATBG at $v=-4$9 find that the flat-band regime appears around $v=+4$0, which corresponds to about $v=+4$1 GPa. At this point a ferromagnetic solution is lower in energy than the nonmagnetic solution for doping levels ranging from 2 electrons to 4 holes per moiré unit cell, with the strongest stability near 2 holes. The spin density is localized at AA stacking sites, the zero-field zero-doping magnetic moment is $v=+4$2 per moiré unit cell, and an out-of-plane electric field drives a transition from the ferromagnetic to the nonmagnetic state between $v=+4$3 and $v=+4$4. In the one-electron-doped ferromagnetic case studied there, the flat bands have trivial band topology, with $v=+4$5 and $v=+4$6 (Chen et al., 2020).

Magnetic fields produce both single-particle and strongly interacting reconstructions. In a tight-binding Peierls treatment at $v=+4$7, orthogonal magnetic fields yield dispersive Landau levels at $v=+4$8 T and much flatter levels at $v=+4$9 T, while strong in-plane fields shift and distort the low-energy bands through minimal coupling. In ultraclean MATBG at $T$00, the perpendicular-field Hofstadter regime hosts symmetry-broken Chern insulator states and fractional states. The observed SBCI cascade includes $T$01, $T$02, and $T$03, while the fractional sequence includes $T$04, $T$05, $T$06, and $T$07. These fractional states disappear at high magnetic field, above about $T$08–$T$09 T, in a field-driven transition from composite-fermion phases to a dissipative Fermi liquid. The corresponding magnetic subbands are argued to be better understood as in-field fractional Chern insulators than as ideal lowest-Landau-level states (Bigeard et al., 2023, He et al., 2024).

6. Fabrication, homogeneity, and gate-defined device architectures

Because correlated phases are extremely sensitive to twist-angle disorder, strain, trapped contamination, and hBN alignment, fabrication quality is a central part of MATBG physics. An optimized dry-transfer process based on deterministic graphene anchoring and aggressive bubble suppression addresses this directly. In a set of 34 fully fabricated and measured devices, 13 devices ($T$10) contained at least one region between two contacts with $T$11, and the yield increased to $T$12 within $T$13. Typical homogeneous regions were about $T$14 with $T$15, and the best case reached $T$16. Optical microscopy and AFM gave average bubble densities of $T$17 microbubble and $T$18 nanobubbles per $T$19, while some devices were bubble-free over regions larger than $T$20 (Diez-Merida et al., 2024).

This level of control enables local electrostatic definition of superconducting circuits within a single MATBG flake. A split-gate-defined superconducting channel was realized in a MATBG mesa with a graphite back gate, split gates separated by a $T$21 nm gap, and a $T$22 nm channel gate. At $T$23 mK, progressive gate tuning drove a smooth transition from superconductivity to highly resistive transport. For $T$24, the channel resistance rose to $T$25, pinching off the superconducting channel; electrostatic analysis implied an estimated minimum normal-conducting channel width of about $T$26 nm, much smaller than the lithographic $T$27 nm width. The same work showed that the supercurrent can be guided through the constriction, through the full device, or blocked entirely (Zheng et al., 2023).

Multi-gated MATBG also supports Josephson junctions, edge tunneling spectroscopy, and Coulomb-blockaded islands defined solely by local gates. In one geometry, a narrow top gate over a globally back-gated device produced a Josephson weak link with $T$28 K and $T$29 nA. The magnetic interference pattern was governed by the Pearl regime of ultrathin superconductors, characterized by

$T$30

rather than by conventional bulk electrodynamics. Intrinsic MATBG band insulators were used as tunnel barriers for edge tunneling spectroscopy, yielding fitted superconducting gaps $T$31 and $T$32 in two devices. A double-barrier geometry further produced single-electron-transistor behavior with Coulomb diamonds and extracted capacitances $T$33 aF, $T$34 aF, and $T$35 aF (Rodan-Legrain et al., 2020).

7. Theoretical frameworks and unresolved questions

Theoretical descriptions of MATBG span continuum models, atomistic approaches, projected interacting Hamiltonians, and heavy-fermion constructions. Within the continuum family, the BM model, a generalized BM model with nonlocal tunneling, and fully relaxed models yield markedly different predictions for layer-projected Hall conductivity and in-plane orbital magnetization. In the generalized BM model with $T$36, the layer Hall response is amplified by about an order of magnitude relative to BM and flips sign on the electron-doped side; self-consistent Hartree fields then renormalize the response in a model-dependent way. This makes layer Hall counterflow a sensitive proposed probe of structural relaxation, nonlocal tunneling, and Hartree band renormalization (Zhu et al., 2023).

Weak-coupling theory emphasizes the dominance of Hartree electrostatics near charge neutrality. In a self-consistent Hartree plus RPA treatment, the chemical potential shifts by about $T$37 meV across the flat-band filling range, the compressibility remains positive, and flavor-paramagnetic states are stable over a range near charge neutrality. The same analysis argues that mean-field theory strongly overestimates the energy difference between flavor-polarized and paramagnetic states: estimates of order $T$38 meV per moiré cell are reduced to less than $T$39 meV per moiré cell when dynamically screened correlations are included (Zhu et al., 2024).

Nonperturbative and emergent-lattice descriptions instead stress flavor polarization and hybridization. Exact diagonalization projected onto the continuum-model flat bands finds that for $T$40, where superconductivity is strongest, the ground state is spin ferromagnetic, and near $T$41 the system forms Chern-insulator ground states with spontaneous spin, valley, and sublattice polarization. The same work emphasizes that remote-band self-energy is necessary for a reliable description of flat-band correlations (Potasz et al., 2021). In the topological heavy-fermion framework, MATBG is represented as an emergent Anderson lattice with localized $T$42-orbitals carrying the AA-centered flat-band weight, itinerant $T$43-orbitals carrying the dispersive sector, and a hybridization $T$44 that produces the flat-band topology. Realistic estimates place the system in an intermediate regime with $T$45 meV and $T$46 meV, and perturbative RG below about $T$47 eV increases both $T$48 and $T$49 by about $T$50, leaves $T$51 nearly unchanged, and lowers $T$52, thereby pushing the low-energy theory toward the projected-limit or chiral side (Huang et al., 23 Jul 2025). Related large-$T$53 slave-boson work frames MATBG as a topological mixed-valence problem with SU(8) structure and reports a Kondo-scale crossover around $T$54 K in one representative case (Li et al., 2023).

A separate interacting tight-binding program shows that integrating out high-energy modes can strongly widen the active bands. At $T$55, the bare flat-band bandwidth of $T$56 meV increases to $T$57 meV for $T$58 eV and to $T$59 meV for $T$60 eV, while the wavefunctions move toward the chiral, particle-hole-symmetric, sublattice-polarized limit (Sánchez et al., 15 Jan 2025). In an exactly solvable flat-band interacting Hamiltonian built from the chiral-limit Bistritzer-MacDonald model, the many-body ground-state manifold is the linear span of ferromagnetic Slater determinants, and no other ground states occur (Stubbs et al., 25 Mar 2025).

The remaining open questions are correspondingly sharp. The review literature identifies four in particular: what precisely drives superconductivity in MATBG; what microscopic mechanism causes the linear-$T$61 resistivity; how much of the phase diagram is controlled by electron-electron interactions versus electron-phonon coupling; and how best to model the flat bands while preserving topology, remote-band effects, and interaction renormalization. It also stresses that more reliable control over average twist angle and twist-angle disorder is essential for further progress (Andrei et al., 2020).