Confinement of superconductivity into quasi-one-dimensional channels by a super-Moiré crystal

Abstract

The precise spatial engineering of quantum phases stands as a major challenge in condensed matter physics, with profound implications for fundamental science and quantum technologies. While moiré heterostructures have enabled remarkable tunability on the correlated states, the deterministic spatial manipulation of emergent quantum phases like superconductivity and magnetism remains an open challenge. Here, we investigate hexagonal boron nitride (hBN)-aligned twisted bilayer graphene (TBG), in which fabrication-induced uniaxial heterostrain gives rise to a periodic quasi-one-dimensional (1D) super-moiré supercell. This supercell imprints a striking chain-like architecture onto the electronic states. Using atomic-scale spectroscopy and point-contact Andreev reflection, we demonstrate that intrinsic superconductivity arising from TBG emerges exclusively within these nanoscale channels, a phenomenon unprecedented in other quantum materials. Furthermore, we show that the superconducting state is governed by the local inversion symmetry, which can be deliberately tuned via the moiré registry with the hBN substrate. This work establishes a structural approach to atomically pattern quantum phases, bridging the gap between nanoscale symmetry control and macroscopic quantum functionality. Our results position super-moiré lattices as a programmable platform for quantum phase design, opening routes to explore exotic dimensionality crossovers in correlated electron systems.

Spatially patterning emergent quantum phases remains challenging. Here, atomic-scale spectroscopy and Andreev reflection show that a strain-induced super-moiré in hBN-aligned twisted bilayer graphene confines superconductivity to quasi-one-dimensional channels governed by local inversion symmetry.

Introduction

The pursuit of reduced-dimensional superconductivity is driven by its potential to uncover exotic physics absent in bulk materials and to advance technological applications^1,2. Among low-dimensional systems, one-dimensional (1D) and quasi-one-dimensional (quasi-1D) superconductors are of particular interest. In these systems, electrons are confined to a linear world where enhanced quantum fluctuations challenge conventional superconductivity. Theoretically, they are predicted to host unconventional pairing, including triplet and odd-parity states, as well as an increased propensity for topological superconductivity, a key ingredient for fault-tolerant quantum computing^3–7. However, the experimental realization of tunable quasi-1D superconductivity remains challenging due to the scarcity of reliable atomic-scale fabrication techniques. Conventional approaches, such as nanowire growth and lithographic patterning, often suffer from structural disorder and limited tunability^8–11, impeding precise control over emergent superconducting phases.

Recent progress in van der Waals heterostructures and moiré physics has introduced a powerful platform for exploring correlated phases in reduced dimensions (2D layers)^12–14. While magic-angle twisted graphene hosts diverse strongly correlated states, including Chern insulators^15–24, orbital magnetism^25–28, and unconventional superconductivity^29–34, they typically lack intrinsic dimensional confinement or the capability for programmable geometric design. Here, we overcome this limitation by employing dual moiré potentials in a heterostructure of twisted bilayer graphene (TBG) aligned with hexagonal boron nitride (hBN). By coupling two competing moiré patterns, and assisted by uniaxial strain, we create a periodic super-moiré that atomically “sculpts” correlated electrons into chain-like architectures. This enables us to have precise control over their spatial modulation and emergent quantum properties. Through spatially resolved dI/dV mapping combined with point contact spectroscopy (PCS), we confirm the formation of a superconducting state within these periodic quasi-1D channels, which remains robust over extended distances. Our work establishes a pathway to isolate and stabilize superconductivity within nanoscale channels by deliberately controlling the local twist angle and inversion symmetry. Furthermore, our dual moiré coupling approach redefines the role of moiré engineering, transitioning it from a tool for band structure tuning into a versatile platform for dimensionality-driven quantum design.

Results

Super-moiré structure and spatially modulated electronic states

Figure 1a illustrates the structure of our hBN-aligned TBG device, which comprises two rotationally misaligned graphene layers stacked on an hBN substrate. The interlayer rotation angles (θ_G-G and θ_G-hBN) generate two distinct moiré patterns with periodicities L_G-G and L_G-hBN. In our sample, θ_G-G is approximately $1^{\circ }$, and θ_G-hBN is nearly $0^{\circ }$ (aligned with the hBN lattice), resulting in two moiré patterns with closely matched periods of ~14.5 nm. When coupled with a G/hBN moiré potential of similar period, the G/G moiré pattern undergoes reconstruction, giving rise to super-moiré structure^35–39. In our device, however, the TBG experiences a weak uniaxial strain during fabrication, while the underlying hBN remains essentially rigid. This heterostrain distorts both the G/G and G/hBN moiré patterns (as shown in the right panel of Fig. 1a) and ultimately drives the formation of a unique super-moiré structure.

Fig. 1. Super-moiré structure in hBN-aligned TBG.

a Left panel, schematic of strained hBN-anligned twisted bilayer graphene (TBG) heterostructure. Right panel, Sketch of a distorted moiré superlattice. b STM topography of a 360 × 360 nm² region (V_Bias = −100 mV, I = 100 pA; scale bar: 100 nm). Lattice relaxation drives AA-stacked regions into 1D quadruple (4×1) supercells. The yellow and red dashed diamonds depict one moiré unit cell and 4 × 1 supercell, respectively. Inset: FFT of the topography showing G/G moiré Bragg peaks (blue circles) and G/hBN peaks (purple diamonds). c Zoom-in of the area in the brown box in b (V_Bias = −100 mV, I = 100 pA; scale bar: 30 nm). Yellow dashed lines mark domain walls between supercells. Green, red, blue, and brown circles indicate the four inequivalent AA regions in one supercell. d Local twist-angle map of the region shown in c. The white dashed rectangle outlines one 4×1 supercell. Scale bar, 30 nm. e Histogram of twist angles from (d), showing two populations corresponding to inner (AA₂ & AA₃) and outer (AA₁ & AA₄) AA regions. f STM topography highlighting the G/G moiré pattern (V_Bias = −100 mV, I = 50 pA; scale bar: 20 nm). Red arrows mark lattice vectors L₁ and L₂. The super-moiré has the same period as the moiré along L₁, but quadrupled along L₂. g STM topography of the same area highlighting the G/hBN moiré (V_Bias = −500 mV, I = 50 pA; scale bar: 20 nm). Green arrows indicate lattice vectors L₁’ and L₂’. h Schematic diagram demonstrating the formation of the 4×1 super-moiré structure. Vectors L₁ and L₁’ are aligned and equal in length, whereas L₂ and L₂’ are misaligned. The relation: 4L₂ − 4L₂’ ≈ L₁’ (where L₁’ ≈ L₁) dictates the commensurate reconstruction.

Figure 1b shows a representative STM topography image of the resulting super-moiré structure. Notably, the AA stacking regions (bright spots) reorganize into a periodic array of 1D chains. These chains form a superlattice with a unit cell that is quadrupled relative to the original moiré period (outlined by the diamonds in Fig. 1b). We denote this as a 4 × 1 supercell. Adjacent chains are separated by broad ‘domain walls’ approximately 16.5 nm in width. This structure contrasts sharply with the uniform, C₆-symmetric distribution of AA regions in original TBG. A close-up STM topographic image (Fig. 1c) further confirms that the original moiré pattern’s translational symmetry is broken and replaced by a new periodicity, which is four times the original moiré length.

The domain walls create inequivalent structural environments for the four AA regions within each supercell, which we label as AA₁, AA₂, AA₃, and AA₄ (marked by green, red, blue, and brown circles in Fig. 1c, respectively). To quantify these structural differences across the 1D supercell, we extracted the local twist angle point by point from the STM topography, as detailed in Supplementary Note II. The resulting map (Fig. 1d) reveals a 1D periodic modulation of the twist angle, with the same period as the supercell. Specifically, the AA₁ and AA₄ regions adjacent to the domain walls show reduced twist angles (light blue area). In contrast, the AA₂ and AA₃ regions near the supercell center maintain a higher twist angle (red area). The statistical distribution of the local twist angles across the region in Fig. 1c is plotted in Fig. 1e. The angles are clustered predominantly around two distinct values, i.e., $0.89^{\circ }$ and $0.98^{\circ }$, directly confirming the modulated local structural environments.

The emergence of these 1D super-moiré structures originates from the interplay between the G/G moiré and G/hBN moiré patterns. Figures 1f and 1g present STM topography of the G/G moiré and G/hBN moiré patterns at the same area acquired by different bias voltages. We extracted the basis vectors of the G/G moiré lattice (L₁ and L₂) and the G/BN moiré lattice (L₁’ and L₂’). While L₁ and L₁’ are nearly identical in both length and direction, L₂ and L₂’ differ by a finite angle (Fig. 1h). Furthermore, they satisfy the relation: 4L₂ - 4L₂’ ≈ L₁’ (L₁). This indicates that although the two moiré patterns are misaligned along L₂, the accumulated mismatch is compensated every four periods, thereby generating a new superlattice with a quadrupled periodicity along this direction. This model is general and applies to different super-moiré configurations in our samples (such as 3×1 and 2×1 patterns), all of which are analytically verified (see Supplementary Fig. 3). All observed structures conform to a universal scaling law: n(L₂ - L₂’) ≈ L₁’ (L₁), where n×1 super-moiré periodicity originates from an n-period mismatch along L₂.

Notably, a simple superposition of two C₆-symmetric moirés cannot produce this 1D structure. The critical ingredient is the mismatches between the L₂ and L₂’ basis vectors, especially when L₁ and L₁’ are well aligned. We attribute this distinct L₂ and L₂’ mismatch to uniaxial strain introduced during the sample fabrication. Such mismatch leads to an energy minimization via periodic reconstruction, where the system relaxes the accumulated strain and lattice mismatch every n periods along L₂, resulting in a modulated 1D strain field that pins the observed n×1 superlattice. In essence, the uniaxial strain provides the initial L₂/L₂’ mismatch, while the moiré-moiré interaction transforms it into a periodic potential that sculpts the electronic landscape. Through a detailed geometric analysis of the supercell (see Supplementary Note III), we find that by appropriately tuning the magnitude and direction of the uniaxial strain as well as the twist angles, the experimentally observed 1D structure can be reproduced, which further supports the scenario proposed above.

Spatially modulated correlation gap feature at 4.4 K

The 1D lattice reconstruction in the super-moiré structure modulates the effective local twist angle, resulting in distinct local electronic states across the inequivalent AA regions. To probe this spatial electronic modulation, Fig. 2a shows dI/dV spectra at the charge neutral point (CNP) along L₂ direction (the white dashed line in Fig. 1c), which spans two supercells. We find that although the characteristic flat bands of TBG persist in all AA regions within a supercell, they exhibit notable differences in the energy gap between the flat bands, their bandwidth, and other spectral features. Moreover, a comparison of the spectra across two supercells reveals a precise spatial repetition of the modulated electronic structure, confirming that the electronic states inherit the translational symmetry of the super-moiré lattice. Consistent with these experimental observations, by combining the theoretically modeled 1D structural reconstruction with the continuum model, we also reproduce the 4×1 electronic modulation (see Supplementary Fig. 5), further confirming both the structural tunability and the connection between the lattice reconstruction and the resulting electronic states.

Fig. 2. Spatially periodic quasi-1D gap features (4.4 K).

a Spatially resolved dI/dV spectra measured along the white dashed arrow in Fig. 1c at the charge neutrality point (CNP), showing a spatial periodicity quadruple that of the original moiré lattice. The green, red, and blue dashed lines mark the positions of AA₁, AA₂, AA₃ regions, respectively. b Gate-dependent dI/dV spectra acquired at the three inequivalent AA regions (V_Bias = −80 mV, I = 500 pA). Black dashed lines mark the occupied isospin flavors at integer fillings, signaling flavor polarization. A pseudogap-like feature between ν = −2 and −3 emerges exclusively in the AA₂ region (purple arrow). c Gate-dependent dI/dV intensity plots for the AA₂ (left) and AA₃ (right) regions, (V_Bias = −10 mV, I = 500 pA). The color scale represents the differential conductance. d Representative dI/dV spectra at selected gate voltages from c, highlighting the contrasting behavior at E_F (black dashed line) between the gapped AA₂ and metallic AA₃ regions. e Direct comparison of dI/dV spectra at ν = −2.24, providing direct evidence of the gap opening in AA₂. f STM topography of 100 × 100 nm² region containing two supercells (scale bar, 20 nm). The green dashed lines outline the AA₂ region within each supercell, also referenced in (g) and (h). g dI/dV intensity map at E_F (V_Bias = 0 mV) and ν = −2.24. The purple stripes (demarcated by green dashed lines) indicate regions of strongly suppressed density of states, forming quasi-1D chains. h Spatial map of the extracted gap size (2Δ) at ν = −2.24, directly confirming the formation of periodic, quasi-1D gapped channels.

As mentioned earlier, the average θ_G-G in this region is around 1°, placing it within the “magic angle” regime⁴⁰. At this angle, Coulomb interactions become dominant over the electron kinetic energy, leading to various strongly correlated phases. Previous STM studies of magic-angle twisted bilayer graphene (MATBG) have demonstrated that gate-dependent dI/dV spectra exhibit characteristic signatures of correlated electronic states as the flat bands pass through the Fermi level, including cascade transitions at integer fillings^41–43 and interaction-induced gaps^23,24. As shown in Fig. 2b, our gate-dependent dI/dV spectrums measured at 4.4 K from each AA region (see Supplementary Fig. 9 for AA₄ data) closely resemble those of typical MATBG, mainly the characteristic cascade of electronic transitions (marked by black dashed lines) at integer fillings, a hallmark of strong correlation effect. Notably, a gap feature is observed at E_F within the doping range of -3 <ν < -2 in the AA₂ region (purple arrow in Fig. 2b), which is absent in all other AA regions. Temperature-dependent measurements show that this gap closes at 8 K, while the cascade feature persists (see Supplementary Note IV), confirming its origin from electronic correlations. In typical hBN-unaligned MATBG, a similar gap in this filling range is often identified as a pseudogap-like state, a gapped phase that persists even when superconductivity is suppressed^31,44.

The absence of the pseudogap-like state in AA₁ and AA₄ regions could be attributed to their local twist angles deviating from the ‘magic angle’. However, a more intriguing puzzle arises when comparing the AA₂ and AA₃ regions. Despite possessing comparable local twist angles and heterostrain (see Supplementary Table 1), they exhibit dramatically different electronic ground states. Zoomed-in gate-dependent dI/dV spectra (Fig. 2c, d) reveals a pronounced symmetric gap at E_F in AA₂ region within the doping range of ν = −2 to −2.5, while AA₃ remains persistently metallic without the gap feature. This is further evidenced by a comparison of the dI/dV curves at ν = −2.24 (Fig. 2e). The stark difference in the ground state between these two structurally similar regions indicates the necessity of invoking additional underlying mechanisms, which will be discussed in subsequent sections.

To directly visualize the real-space distribution of this strongly correlated state, we performed a large-area dI/dV map at ν = -2.24 across a region encompassing two supercells, with the corresponding STM topography shown in Fig. 2f. Since gapped regions exhibit a strong suppression of the DOS at E_F relative to metallic regions (Fig. 2e), we identified them using the dI/dV intensity map at E = 0 meV (Fig. 2g). As outlined by green dashed lines, the gapped regions form two parallel chains with a width of 13.6 nm and an interchain spacing of 57.7 nm, corresponding to approximately one and four times the moiré period, respectively. By comparing with the corresponding STM topography in Fig. 2f, we find that these chains fully encompass the AA₂ regions within each supercell and extend partially into adjacent AB stacking areas.

To further quantify the spatial variation, we extracted the local gap size (2Δ) in Fig. 2h, defined as the energy separation between the two coherence peaks flanking the gap. The regions with finite gap sizes (ranging from 6 to 8 meV) are predominantly localized near the AA₂ regions of each supercell. The results provide direct evidence for the formation of quasi-1D pseudogap-like chains in real space. Similar structures were consistently observed across multiple quadruple super-moiré areas (see Supplementary Figs. 10–12), confirming that this phenomenon is reproducible and intrinsically linked to the underlying structural modulation.

Correlated gap to superconductivity transition at 0.4 K

The electronic state of the AA₂ region at 4.4 K is highly consistent with conventional MATBG, including the presence of a pseudogap-like state between ν = -2 ~ -3. This suggests that superconductivity could emerge at lower temperatures, with its spatial distribution dictated by the super-moiré structure. To monitor the transition from pseudogap-like feature to superconductivity, we cooled the sample to 0.4 K and acquired gate-dependent dI/dV spectrum in the AA₂ region (Fig. 3a). At this temperature, a pronounced gap feature emerges between ν = −2 ~ −3, spanning a similar doping range as that at 4.4 K (a direct comparison is shown in Supplementary Fig. 13). A zoomed-in view of the spectra (Fig. 3b) highlights the doping evolution of this gap, which is flanked by two symmetrically positioned coherence peaks, consistent with a superconducting gap previously observed in MATBG³¹. We modeled the tunneling spectrum in Fig. 3c using a Dynes function for the quasiparticle density of states, considering both s-wave nodeless and nodal superconducting scenarios (green and orange curves, respectively). Both fits yield a superconducting gap of 2Δ’ = 4.0 meV, in good agreement with previous STM studies of TBG³¹.

Fig. 3. PCS measurements to identify gap states in the AA₂ region.

a Gate-dependent dI/dV density plot for the AA₂ region at 0.4 K (V_Bias = -40 mV, I = 500 pA). The red dashed rectangle labels the superconducting phase. b Zoomed in dI/dV density plot with V_g from -8 to −17 V (upper panel, V_Bias = −20 mV, I = 500 pA) and the corresponding line traces (lower panel). c Dynes-function fits to the experimental tunneling spectrum (blue curve) measured at ν = -2.50 in the AA₂ region at 0.4 K using the model quasiparticle DOS for a nodeless s-wave superconductor (green curve) and a nodal superconductor (orange curve). d A schematic drawing illustrates Andreev reflection occurring at the interface between a normal metallic STM tip and an SC sample. e Normalized setpoint-dependent dI/dV line cuts at the superconducting phase (ν = -2.25) ranging from I = 1 nA to 320 nA which indicate the occurrence of Andreev reflection. f Magnetic-field dependent PCS dI/dV spectra at the superconducting phase (ν = -2.25). Curves are normalized for clarity. g Temperature dependent PCS dI/dV spectra at the superconducting phase (ν = -2.50).

Temperature- and magnetic field-dependent evolution could provide important evidence for identifying a superconducting gap^45,46. Here, we present the evolution of the gap spectra in the AA₂ region as a function of temperature and magnetic field (see Supplementary Fig. 8). However, given the presence of a pseudogap state previously reported in this system, we further employed point-contact spectroscopy (PCS), an established method for identifying superconductivity in graphene-based moiré systems^31,47, to clearly distinguish the superconducting gap from the pseudogap. In this technique (Fig. 3d), the metallic tip is gradually brought into contact with the sample by increasing the tunneling current setpoint. Upon contact, electrons from the tip can then enter the superconductor, form Cooper pairs, and reflect the holes, a process known as Andreev reflection, which enhances the dI/dV spectrum intensity within the superconducting gap^48,49. Figure 3e shows a series of dI/dV spectra acquired with increasing current setpoints, illustrating the transition from the tunneling to the contact regime. As the tunneling current increases from the nA range to the hundreds of nA range, the dI/dV intensity at E_F rises. The gap feature transits to a peak, confirming the emergence of an Andreev reflection signature. Furthermore, from the magnetic field and temperature (Figs. 3f, g) dependence of the Andreev reflection signature (peak feature), we extract a critical field $B_{\text{c}}\approx 100\text{mT}$ and a critical temperature $T_{\text{c}}\approx 1.0\text{K}$, consistent with previous reports on MATBG^29–32. These results support the superconducting states in the AA₂ region.

Imaging quasi-1D superconducting channels

Having affirmed superconductivity in the AA₂ region at low temperatures, we next investigated its spatial modulation by the super-moiré lattice. Gate-dependent dI/dV spectra acquired at 0.4 K across three characteristic regions (Fig. 4a; data for the AA₄ region are provided in Supplementary Fig. 9) reveals a superconducting gap between ν = -2 ~ -3 only in the AA₂ region (green arrow). This spatial confinement is further corroborated by the dI/dV intensity at E_F as a function of V_g (Fig. 4b), which shows strong suppression (green shading) only within this same doping range for AA₂. These results demonstrate that superconductivity is strictly confined to the AA₂ regions, forming an array of quasi-1D superconducting channels defined by the underlying super-moiré lattice.

Fig. 4. Visualization of quasi-1D superconducting channels at 0.4 K.

a Gate-dependent dI/dV spectra acquired at 0.4 K for the AA₁, AA₂, and AA₃ regions (V_Bias = -50 mV, I = 500 pA). A superconducting gap, marked by the green arrow, emerges exclusively in the AA₂ region within the range of ν = -2 to −3. b dI/dV intensity at E_F as a function of V_g. A pronounced suppression, signature of the superconducting gap, is observed only within the AA₂ region. c STM topography of one supercell, with the four distinct AA regions labeled. d Spatially resolved dI/dV spectra at ν = -2.25 along trajectory in (c). The green dashed rectangle highlights the spatially localized superconducting gap at the AA₂ region. e Line profile of the dI/dV at E_F, derived from (d), directly mapping the spatial confinement of superconductivity. f Corresponding point-contact spectroscopy (PCS) measurements along the same path. The blue dashed rectangle indicates the emergence of a zero-bias conductance peak, a hallmark of Andreev reflection occurring in the superconducting AA₂ region. g Large-area dI/dV map (120 × 60 nm²) at E_F and ν = −2.5. The chain-like patterns of suppressed conductance are co-located with the AA₂ regions, confirming the formation of extended quasi-1D superconducting channels. Scale bar, 20 nm. h Structural origin of the difference in superconducting state. Left panel: STM topography showing AA₂ (red circle, inversion-symmetric) and AA₃ (blue circle, symmetry-broken) regions relative to the G/hBN moiré pattern. Scale bar, 5 nm. Right panel: Corresponding atomic schematics. The AA₃ site, at the moiré corner, couples asymmetrically to the hBN substrate, breaking the symmetry.

To directly visualize this spatial confinement, we performed combined STS and PCS measurements along a linecut crossing all four characteristic regions at ν = −2.25 (white dashed line in Fig. 4c). The results highlight two distinct signatures of superconductivity localized at AA₂: a pronounced suppression at E_F in the STS (tunneling regime, green dashed box, Fig. 4d) and a clear Andreev reflection peak in the PCS (contact regime, blue dashed box, Fig. 4f). The dI/dV intensity at E_F (green region in Fig. 4e) provides a direct spatial map of the superconducting region, confirming its strict confinement to AA₂. Additionally, the resulting superconducting domain exhibits atomically sharp boundaries, demonstrating that the super-moiré potential ultimately pins the interface of the superconducting phase.

Having established the formation of quasi-1D superconductivity from local probes, we now verify its periodic structure over a larger scale. Figure 4g shows a large-area zero-bias dI/dV intensity map at ν = −2.5 encompassing two supercells (corresponding topography in Supplementary Fig. 15). Regions with strongly suppressed dI/dV intensity form well-defined chain-like patterns, corresponding to the superconducting channels. These channels are spatially co-located with the pseudogap regions observed at 4.4 K. (See Supplementary Fig. 14 for a direct comparison). Additional zero-bias dI/dV mapping at different doping levels (see Supplementary Fig. 15) confirms that these quasi-1D superconducting channels persist robustly throughout the superconducting doping regime. Together, these findings demonstrate that the super-moiré structure generates stable, quasi-1D superconducting channels in real space. Notably, using the measured perpendicular critical field of B_c ∼ 0.1 T, we estimate a superconducting coherence length of ξ ∼ 57 nm. However, this estimate assumes a homogeneous two-dimensional superconductor. In our strongly anisotropic system, the much narrower ∼15 nm channel width may instead result from super-moiré-induced electronic interfaces that suppress the transverse propagation of superconducting correlations while preserving stronger coherence along the channel direction.

We now turn to the factors governing the emergence of superconductivity. Although AA₂ and AA₃ regions share similar local twist angles and heterostrain, they exhibit dramatically different superconducting behaviors. The line cut in Fig. 2a reveals that the DOS of the valence flat band exhibits no significant difference between the superconducting AA₂ region and the non-superconducting AA₃ region. This suggests that superconductivity is not directly correlated with the DOS and cannot be simply attributed to electron-phonon coupling. Instead, an additional structural factor must play a decisive role.

Figure 4h presents an STM topography encompassing both AA₂ (red dashed circle) and AA₃ (blue dashed circle) regions, where both G/G and G/hBN moiré patterns are visible. A key structural difference is that the superconducting AA₂ region in G/G moiré lies in the ridge region of the G/hBN moiré, whereas in the non-superconducting AA₃ region (blue circle), it resides in the corner of the G/hBN moiré. Since the underlying hBN lattice consists of two inequivalent elements, it breaks the inversion symmetry of the atop graphene lattice in the AA₃ region at the atomic scale (right panel in Fig. 4h). In contrast, the superconducting region (AA₂) is located in the ridge of the G/hBN moiré; thus, the inversion symmetry is preserved due to the irregular stacking configuration with hBN. Examination of multiple regions confirms that the preservation of local inversion symmetry, i.e., C₂ symmetry, universally correlates with the emergence of superconducting AA regions (see Supplementary Note VIII).

The existence of superconductivity in hBN-aligned MATBG has been debated, with transport measurements reporting conflicting results: both its presence^19,50 and absence^25,26. Such discrepancies likely originate from the complex microstructures of dual-moiré systems³⁸, which are averaged in transport measurements. A previous STM study reported no superconductivity in hBN-aligned MATBG³¹, where the AA region of the G/G moiré was aligned exactly with the inversion symmetry-broken honeycomb of the G/hBN moiré. In contrast, our samples exhibit a variety of super-moiré structures where both symmetry-preserving and symmetry-broken regions coexist, leading to the simultaneous presence of superconducting and non-superconducting behaviors. These results strongly correlate the presence of superconductivity with preserved inversion symmetry in hBN-aligned MATBG. Although further validation is needed, this insight opens a new perspective for understanding unconventional superconductivity in moiré systems.

In summary, by coupling two distinct moiré patterns, we have constructed a super-moiré structure that periodically modulates the correlated electronic properties of magic-angle twisted bilayer graphene in real space. Local variations in twist angle, heterostrain, and inversion symmetry across the supercell strongly modify the strongly correlated states, ultimately confining superconductivity to well-defined spatial channels. Using large-area spectroscopic mapping, we have demonstrated the formation of periodic quasi-1D superconducting channels that remain stable over extended distances. Recent advances in mechanically controlled deformation, such as AFM-tip-induced bending of graphene on hBN⁵¹, offer a feasible route to continuously tune the local strain, twist angle, and moiré distortion. This approach could enable the deterministic design of super-moiré configurations and provide a pathway for engineering a broader range of spatially structured correlated phases. Together, these possibilities highlight super-moiré engineering as a versatile strategy for spatially controlling correlated states and exploring the coexistence of multiple quantum phases in moiré systems.

Methods

Sample fabrication

The devices were fabricated using a conventional “tear-and-stack” approach. Monolayer graphene (MLG) and few-layer hexagonal boron nitride (hBN) flakes were first exfoliated onto SiO₂/Si substrates and identified by optical and atomic force microscopy. Using a polypropylene carbonate (PPC)/polydimethylsiloxane (PDMS) stamp, we sequentially picked up hBN, half of the MLG, and then the remaining half after rotating the stage by approximately 1°, thereby forming the desired TBG structure. All pick-up steps were carried out at 40–50 °C. The assembled heterostructure was then released onto a SiO₂/Si substrate (SiO₂ thickness: 285 nm) at 105 °C, followed by removal of the PPC layer via high-vacuum annealing at 350 °C. To enable scanning tunneling microscopy (STM) measurements, the heterostructures were patterned into gate-tunable devices using electron-beam lithography and Au/Ti evaporation. Finally, the samples were annealed in an ultrahigh-vacuum chamber at 400 °C overnight to remove surface residues prior to STM characterization.

STM/STS measurements

STM/STS measurements were conducted at 4.4 K under ultrahigh vacuum conditions (<1 × 10⁻¹⁰ mbar). The STM tips were calibrated on Cu(111) before use. Topographic images were acquired in constant-current mode, while differential conductance (dI/dV) spectra and maps were recorded using a standard lock-in technique with a 1 mV AC modulation at 713 Hz added to the DC bias. Additional measurements at 0.4 K were obtained using a ³He cryostat. Magnetic fields were applied perpendicular to the sample surface using a superconducting magnet.

Spatially resolved STS maps were acquired as follows. With the feedback loop open, the STM tip was positioned in constant-current mode and then stabilized. The feedback loop was subsequently closed, and the sample bias was swept in constant-height mode to record the dI/dV spectrum.

Author contributions

J.M. and Y.J. conceived the research. Y.H., Y.W., and S.L. performed STM experiments with the assistance of Z.W. and L.C. Y.H., X.W., and J.S. prepared the samples. Y.H. and Z.T. analyzed the experimental data and plotted figures. Y.H. wrote the manuscript with guidance of J.M. and Y.J. K.W. and T.T. provided the hBN crystals. J.K. performed theoretical calculations and analyses. Y.H., Y.W., S.L., C.S., J.K., Y.Y., H.-J.G., Y.J., and J.M. participated in discussions of the results and contributed to the manuscript preparation.

Peer review

Peer review information

Nature Communications thanks the anonymous reviewers for their contribution to the peer review of this work. A peer review file is available.

Funding

This work was supported by the Fundings from the National Natural Science Foundation of China (Grants No. 12474477, No. 12550405, No. 61888102, No. 12574159, No. 12074276, No. 12321004, and No. W2511003); the Beijing Outstanding Young Scientist Program; the National Key R&D Program of China (2024YFA1207700); Fundamental Research Funds for the Central Universities; Scientific Research Innovation Capability Support Project for Young Faculty (SRICSPYF-ZY2025071); the robotic AI-Scientist platform of Chinese Academy of Sciences; the EMEXT Element Strategy Initiative to Form Core Research Center through Grant No. JPMXP0112101001; the Strategic Priority Research Program of Chinese Academy of Sciences (No. XDB1710000); CAS Project for Young Scientists in Basic Research (YSBR-003); the Double First-Class Initiative Fund of ShanghaiTech University, and the start-up grant of ShanghaiTech University.

Data availability

Source data supporting the findings of this study are provided with this paper. Source data are provided with this paper.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s41467-026-77046-6.