Realization of Kagome Kondo lattice

Abstract

The Kondo lattice, describing a grid of the local magnetic moments coupling to itinerant electrons, is a fertile ground of strongly correlated states in condensed matter physics. While the Kagome lattice has long been predicted to host Kondo physics with exotic magnetism and nontrivial topology, no experimental realization has been achieved to the best of our knowledge. Here, we report the discovery of CsCr₆Sb₆, a van der Waals-like Kagome Kondo lattice featuring extremely flat, isolated bands at the Fermi level that composed entirely of Cr-3d electrons. We observe heavy fermions with the effective mass over 100 times greater than those of its vanadium counterpart. We also observe Kondo insulating behavior in an ultra-low carrier density of 10¹⁹cm^-3 and dimensionality-induced Kondo breakdown. Counterintuitively, mechanical exfoliation of the frustrated bulk reveals hidden A-type antiferromagnetism with even-odd layer-dependent anomalous Hall effect. The realization of Kondo physics in Kagome lattice opens avenues for exploring diverse quantum criticalities in a strongly-correlated frustrated system.

Accessing strong correlation effects in Kagome materials remains challenging. Here, the authors realize a Kagome Kondo lattice in CsCr₆Sb₆ exhibiting flat, isolated Kagome bands at the Fermi level.

Introduction

Kagome lattice, characterized by symmetry-protected Dirac points, van Hove singularities and flat bands, provides an exciting platform to intertwine nontrivial topology, frustrated magnetism and strongly correlated electrons^1–16. However, accessing strong correlation effects in Kagome materials remains challenging, as their flat bands are often distant from E_F or disrupted by structural distortions that deviate from the ideal Kagome structure^12–16. The canonical Kondo phase diagram (Fig. 1a) is typically realized in f-electron compounds, where strong correlations emerge from the interaction between two essential ingredients: itinerant electrons and local magnetic moments^17–28. Recent observations of strange metallicity in Ni₃In and pressurized CsCr₃Sb₅ suggest that bringing portions of the Kagome flat bands closer to E_F could effectively enhance electronic correlations^29–31, yet the absence of local moments in these systems precludes the development of Kondo physics. In contrast, Nb₃Cl₈ exhibits local magnetic moments when its Kagome flat band approaches the Fermi level, but the absence of itinerant electrons leads to a Mott insulating ground state³². Therefore, simultaneously bringing a dispersive band and a Kagome flat band to E_F is a promising route to realize a Kagome Kondo lattice.

Fig. 1. Crystal and electronic structures of CsCr₆Sb₆.

a Illustration of the competing phases of heavy fermion liquid and magnetic ordered state. b Optical image of a CsCr₆Sb₆ single crystal. c Crystal structure obtained from single crystal X-ray diffraction. d Double Kagome structure viewed along the c-axis, with the bottom Kagome layer shown in varying transparency. e Atomic-resolution image of CsCr₆Sb₆ viewed along the a-axis with the crystal structure superimposed.

In this article, we report the realization of a Kagome Kondo lattice in the van der Waals-like crystal CsCr₆Sb₆, which hosts extremely flat, isolated bands at the Fermi level, giving rise to heavy fermions with ultra-low carrier density and strong correlations in a non-f-electron system. Remarkably, as the material is thinned down to a few layers, its bulk magnetic frustration gives way to an emergent A-type antiferromagnetic order and layer-parity-dependent anomalous Hall response, a feature absent in conventional 2D magnets. These distinct phenomena arise from the delicate interplay between Kondo screening and RKKY interaction. Our findings establish CsCr₆Sb₆ as a rare platform where frustration geometry, strong correlations, and tunable dimensionality converge to yield exotic quantum states.

Results and discussion

A Kagome Kondo lattice

Figure 1b shows the optical image of the Kagome material CsCr₆Sb₆, whose crystal structure is depicted in Fig. 1c (detailed single crystal characterization see Supplementary Fig. 1). Unlike other 166-Kagome materials such as RMn₆Sn₆¹¹ and RV₆Sn₆³³ (R = rare earth elements) that exhibit a three-dimensional (3D) nature, CsCr₆Sb₆ is van der Waals like, featuring a space group of R-3m and lattice parameters a = 5.546(2) Å and c = 34.52(1) Å. Figure 1d illustrates the building blocks of the double Kagome (DK) layer. Successive three DK layers are shifted through [2/3, 1/3, 1/3] to form an ABC stacking lattice, each segregated by Cs atoms. Figure 1e presents an atomic-resolution scanning transmission electron microscopy (STEM) image with the superimposed crystal structure along the a-axis. CsCr₆Sb₆ single crystals can be mechanically exfoliated into a single DK layer (~1.1 nm, namely 1/3 of the unit cell, see Supplementary Fig. 2). Additionally, we note that both single crystals and thin flakes of CsCr₆Sb₆ are robust against moisture, greatly facilitating future characterization and manipulation.

Figure 2a presents the angle-resolved photoemission spectroscopy (ARPES) band structures along high-symmetry directions at 15 K. Notably, a dispersionless flat band is observed along the Γ-M and Γ-K high symmetry directions in momentum space, positioned right beneath E_F. Enlarged second derivate of band structures can be found in Supplementary Fig. 3. To further highlight the flat band feature, we plot the energy distribution curves (EDCs) within a narrower energy window in Fig. 2c. By tracing the EDCs peaks, as indicated by the red tips, the detailed band structure can be discerned. We find that the observed flat band feature is not a single band but consists of two bands around the Γ point. One band crosses the E_F, forming a small pocket around Brillouin zone center (as show in Fig. 2b), while the other is a dispersionless flat band situated around 40 meV below E_F, extending towards the K and M points. Around the K point, the flat band crosses E_F, forming a triangular pocket, while near the M point, the flat band just touches E_F, creating a prominent spot feature on the Fermi surface.

Fig. 2. Extreme flat bands of CsCr₆Sb₆.

a Band structures measured along the Γ-K and Γ-M high symmetry directions in the momentum space at 15 K using He I lamp (21.2 eV). b Fermi surface mapping of CsCr₆Sb₆ measured at 15 K. It is obtained by integrating the spectral intensity within 10 meV with respect to the E_F. The blue solid curves schematically depict the shape of Fermi surface. c Energy distribution curve (EDCs) of the photoemission spectra along the Γ- K and Γ-M high symmetry directions. The EDCs at K and M points are highlighted. The red tips mark the EDC peaks to trace the band dispersions. d Density functional theory (DFT) calculation of CsCr₆Sb₆ without considering electron correlation. Color variation represents the spectral weight percentage of Cr relative to the summation of Cr and Sb content. e Dynamical mean field theory (DMFT) result of a two-orbital model of CsCr₆Sb₆ with U = 1.8 eV.

Another prominent observation is the presence of a band gap between the flat bands and valence bands, spanning from 50 meV to 0.5 eV. These findings make CsCr₆Sb₆ stand out among other Kagome materials, where previously reported flat bands are either mixed with other bands, or positioned far from E_F.

Figure 2d presents a local-density-approximation (LDA) calculation of CsCr₆Sb₆ without considering electron correlation, which reveals a Kagome flat band located exactly at E_F. However, we notice that the calculated band dispersions do not align well with our ARPES measurements, particularly around the K point, indicating significant correlation within the system. We determined the renormalization factor by the ratio of experimental bandwidth (~40 meV) to that of calculated conduction band minimum relative to the E_F, which is significantly large, ~10, when comparing the ARPES experiments with theoretical calculations. To address this discrepancy, we employ dynamical mean field theory (DMFT) to renormalize the band structure, which successfully captures the features of the flat band at E_F, as shown in Fig. 2e. Notably, the orbital projection in Fig. 2d (detailed in Supplementary Fig. 4) shows that all bands near E_F are all primarily contributed by Cr’s 3d electrons, confirming that the observed flat bands are pure Kagome bands. These flat bands suppress the kinetic energy of electrons, effectively localizing the magnetic moments of Cr, while residual pockets around the Γ point support mobile electrons.

We further carried out scanning tunneling microscopy (STM) and spectroscopy (STS) measurements to prove Kondo hybridization by decreasing the temperature. As shown in Supplementary Fig. 5, the STS spectra at 77 K show a metallic-like behavior with a finite density of states at the Fermi level. However, at 2 K, a clear gap opens around E_F, with the gap size estimated to be ~10–15 meV. This gap is consistent with the formation of a Kondo insulating state, where the hybridization between the localized Cr-3d electrons (flat band) and the itinerant electrons (dispersive band) leads to the opening of an energy gap. Therefore, CsCr₆Sb₆ fulfill the requirements of a Kagome Kondo lattice.

Heavy fermions with ultra-low charge carrier density

Figure 3a displays the temperature-dependent resistivity of a bulk sample. The large resistivity of CsCr₆Sb₆ is intrinsic, as confirmed by measurements on three independent sample batches (Supplementary Fig. 6b). The high resistivity arises from the dominance of Kagome-derived flat bands near the Fermi level, which significantly suppress the kinetic energy of electrons. This contrasts with other Kagome metals, such as CsV₃Sb₅, where additional dispersive bands from non-Kagome lattices contribute to their metallicity. It exhibits metallic behavior at high temperatures and reaches a minimum. Upon further cooling, the resistance increases, showing a canonical -logT divergence. This resistance minimum arises from enhanced scattering between local moments and itinerant electrons³⁴, as described by

$$\rho =a{T}^5+c{\rho }_0-c{\rho }_1\log T$$ 1

Fig. 3. Frustrated magnetism and heavy fermion with ultra-low carrier density.

a Resistance of bulk CsCr₆Sb₆ single crystal. A kink at 72 K indicates the onset of frustrated magnetic state. The red line is a -log T plot of resistivity, which gradually deviate from linear curve below T_K. Inset shows the temperature dependence of the Hall coefficient, which shows a drastic reduction at T_F. Three regions are divided by T_F and T_K. b Heat capacity (C_p) of CsCr₆Sb₆ shows a subtle kink at 72 K. Inset is the T dependent C_p/T curve around 72 K, revealing a broad hump at the same temperature. c Linear fitting of C_p/T versus T², the linear residual gives an electron contribution of 75 mJ mol⁻¹ K⁻². d Correlation strength of several correlated materials, which places CsCr₆Sb₆ within the realm of f-electron heavy fermions^17,39–45. HTSC and HF are short for high temperature superconductors and heavy fermion materials, respectively. e Magnetic susceptibility χ of CsCr₆Sb₆ measured along and perpendicular to the c-axis. The dχ/dT curve shown in the inset peaks at 72 K, consistent with the observed T_F shown in (a). f Contour plot of MR, divided into three regions same to (a). Region II and III are dominated by moments and singlets scattering, respectively. Dashed line in region III marks the boundary of magnetic destruction. Magnetic field is along c-axis.

Below T_K = 20 K, the resistance deviates from the -log T dependence and continue to increase with a mild slope, marking the formation of Kondo singlets and indicating the emergence of heavy fermions due to Kondo resonance.

To demonstrate this, we compare the heat capacity of CsCr₆Sb₆ with its isostructural counterpart CsV₆Sb₆ as shown in Fig. 3b, c. The extracted Sommerfeld coefficient γ_Cr-166 = 75 mJ·mol⁻¹·K⁻² is three times larger than that of CsV₆Sb₆. Note that CsCr₆Sb₆ exhibits insulating behavior at low temperatures with an ultra-low carrier density of 10¹⁹cm⁻³ (as R_H shown in the inset of Fig. 3a, Hall measurements data and the carrier density extracted from Supplementary Figs. 6 and 7), in contrast to the metallic CsV₆Sb₆ with a carrier density of 10²¹cm⁻³. This value is even lower than the rare low-carrier-density Kondo systems^35–37, such as CeNi₂As₂³⁸. Calculation using the free electron gas model reveals a large effective mass for CsCr₆Sb₆ of m*/m₀ = 60, which is two orders of magnitude larger than that of CsV₆Sb₆. This value surpassing most d-electron systems on record and reaching the territory of f-electron heavy fermion materials^17,39–45 (Fig. 3d). Instead of a heavy fermion metal, the insufficient Kondo screening results in insulating-like behavior due to incoherent scattering, similar to observations in low-carrier-density f-electron systems CeNi₂As₂ and NaYbSe₂⁴⁶.

Notably, the resistivity exhibits an inconspicuous kink at T_F = 72 K, which is more clearly visible in Supplementary Fig. 6a. This kink is related to a magnetic interaction, as indicated by the magnetization drop shown in Fig. 3e. Even though, no long-range magnetic order emerges due to the frustration inherent in the Kagome lattice, as evidenced by the broad hump in magnetization and the absence of the λ-type anomaly in heat capacity shown in Fig. 3b. Curie-Weiss fitting of the magnetic susceptibility reveal a magnetic moment of 1.76 μ_B per Cr (Supplementary Fig. 8). The fitted Neel temperature is much higher than T_F, suggesting the strong frustration. The origin of these short-range magnetic interactions remains unclear, potentially arising from the incomplete localization of wavefunction.

Next, we examine the impact of local moment screening on magnetotransport properties. Figure 3f shows the contour plot of magnetoresistance (MR) of the bulk sample at different temperatures. At T_F, a crossover from positive to negative MR is manifested, and its amplitude reaches a maximum at T_K. These two characteristic temperatures divide the behavior of MR into three regions. In region II, the negative MR can be attributed to the suppression of local moment scattering by external magnetic fields. In the III (below T_K), the partial Kondo screening cancels out the negative MR caused by local moment scattering. The MR returns to negative at high magnetic fields, which can be attribute to the destruction of heavy fermions and the resurgence of the local moment scattering⁴⁷. Magnetic destruction occurs when the Zeeman energy exceeds T_K, so the destruction field increases with cooling. The temperature-dependent MR behavior and magnetic destruction of heavy fermions observed in CsCr₆Sb₆, combined with the prominent enhancement of effective mass, verify the formation of Kondo resonance at T_K.

Dimensionality induced Kondo breakdown

The van der Waals nature of CsCr₆Sb₆ enables the exploration of dimensional reduction effects on the interplay of various emergent quantum phenomena, which is a long-standing aspiration in study of Kondo lattice physics but remains a significant challenge due to the 3D atomic and electronic structures of conventional heavy fermion compounds^45,48,49. Recent results in exfoliated CeSiI suggest a small step towards this goal, despite the nearly unchanged Kondo temperature and coercive magnetic fields. Here, we demonstrate the successful tuning of T_K in CsCr₆Sb₆ by exfoliating the bulk sample approaching its 2D limit.

Note that T_F is defined as the temperature at which the MR reverses sign (Supplementary Fig. 9), while T_K is at the maximum negative MR. Figure 4a, b shows MR at 9 T and dR/dT of CsCr₆Sb₆ in bulk and 36 L sample, respectively. The high similarity of MR and dR/dT in both bulk and thin flakes indicates the two characteristic temperatures are both reflected in the resistance behavior. Based on this, we depict the dimensionality effect on the Kondo exchange (J_K) in CsCr₆Sb₆ by fabricating devices with thickness varies from bulk to 2 L (Fig. 4d). The dR/dT data are renormalized and shifted for clarity. We observe that T_K can be feasibly tuned as the sample thickness is reduced to tens of nanometers, exhibiting a monotonic decrease. These results suggest that dimensional reduction diminishes the J_K, which is consistent with the general recognition of weakened Coulomb screening approaching the 2D limit. The hopping of conduction electrons is thus suppressed, exacerbating the insufficient Kondo screening. Meanwhile the quantum fluctuations raised by dimensional reduction may also destabilize the singlets.

Fig. 4. Highly tunable T_K and a hidden AFM state.

a, b MR at 9 T and dR/dT of the bulk and the 36 L sample, respectively. Both bulk and thin flakes exhibit a kink and a minimum in dR/dT coinciding with T_F and T_K. c Temperature dependent resistivity of 2 L sample, with residual resistivity of 700 μΩ·cm. Inset show the linear dependence of the resistivity at low temperature. d dR/dT of samples with varying thickness from bulk to 2 DK layers (2 L, hereafter). The curves are normalized and vertically offset for clarity. T_k and T_K are indicated by blue and red triangles, respectively. e Hall resistance R_xy of 3 and 36 L samples. A hysteresis and anomalous Hall resistance appear in the 3 L sample with a coercive field of 3.8 T. f temperature dependent MR of the 3 L sample, where the critical temperature is defined as the temperature at which the hysteresis vanishes. g R_xy of samples with thickness of 4, 5 and 20 L. red and blue lines represent up and down sweep of the magnetic field. Even-layer samples exhibit AFM states, while odd-layer ones are in a ferromagnetic (FM) state. Magnetic fields are all along c-axis.

In the 36 L sample, MR measurement indicates that the magnetic destruction field is 2.8 T at 1.8 K (Supplementary Fig. 10d), significantly lower than the 8.7 T observed in the bulk (Fig. 3f), verifying the reduced energy scale of J_K. By further reducing the thickness to 5 L, the minimum of dR/dT vanishes, indicating a complete suppression of J_K and the realization of a dimensionality-induced Kondo breakdown. Meanwhile, T_F persists and continues to shift toward lower temperatures. Interestingly, in the region of complete Kondo breakdown, the 2 L sample exhibits strange metal behavior with a residual resistivity exceeding 700 μΩ·cm (Fig. 4c), highlighting a unique dimensional effect as the system approaches the atomic limit.

Hidden magnetic ordered states

In a Kondo lattice, the Ruderman–Kittel–Kasuya–Yosida (RKKY) interaction is generally suppressed by Kondo screening. A key distinction between Kondo lattices and impurity Kondo effect is the realization of long-range magnetic ordered states by tuning various parameters. As we have established the dimensionality tuned J_K, we further investigate the magnetic response of this Kagome Kondo lattice, particularly near the Kondo breakdown, through magneto-transport measurements.

As expected, we observe the emergence of a long-range antiferromagnetic ordered state below ~20 L, which is absent in the bulk. For comparison, we plot the transverse resistance R_xy of 3 L and 36 L thin flakes at 5 K in Fig. 4e. Distinct from the 36 L and bulk samples, R_xy of 3 L sample shows prominent jumps at 2.5 T and display hysteresis upon sweeping the magnetic fields. Notably, the anomalous Hall resistance observed at zero field indicates the formation of a ferromagnetic ground state with the easy axis oriented out-of-plane (Supplementary Fig. 11). The temperature dependent MR of the 3 L sample shows the characteristic butterfly hysteresis of ferromagnetism (Fig. 4f).

The odd and even DK layers provide further insights into the magnetic structure. As shown in Fig. 4g, the even-layer samples exhibit antiferromagnetic hysteresis, while the odd-layer samples display ferromagnetic loops. This even-odd-layer-dependent behavior is a hallmark of A-type antiferromagnetism. Consequently, the magnetic unit cell of CsCr₆Sb₆ should contain at least six DK layers. Meanwhile, the magnetic ordering within each DK and its thickness evolution remains unclear thus far. Moreover, the R_xy of few-layers CsCr₆Sb₆ flakes show a lot of steps at lower temperatures (additional temperature-dependent R_xy data are presented in Supplementary Figs. 12 and 13), suggesting more complex magnetic states or domain structures compared to those observed in MnBi₂Te₄⁵⁰ and MnBi₄Te₇⁵¹, warranting further investigation.

This finding is unique from the perspective of 2D magnetism. Recently search for 2D magnetic materials have focused on van der Waals compounds with high Curie or Néel temperatures and strong magnetocrystalline anisotropy to maintain magnetic order approaching their monolayer limit^52–54. CsCr₆Sb₆ is the only known example of emergent magnetism upon thinning, to the best of our knowledge. This peculiar thickness-dependent magnetism can now be fully understood within a unified framework of Kondo phase diagram.

As shown in Fig. 5, the blue circles and pink diamonds are T_F and T_K, and the orange squares represent the onset of magnetic ordering at T_N, determined as the temperature at which the magnetization-induced jumps in MR or R_xy vanish. In the bulk, the strong geometric frustration of the Kagome lattice prevents the formation of long-range antiferromagnetic order, which may promote the emergence of the heavy fermions; otherwise, the ground states of the bulk would enter the AFM phase. In the 20 L sample, the emergence of ordered states cannot totally suppress T_K. The Kondo screening occurring in an AFM state extends beyond the Doniach phase diagram, where typically only one phase dominants. The moisture stability of few-layer CsCr₆Sb₆ further underscores its potential as a leading candidate for next-generation spintronic devices.

Fig. 5. Magnetic phase diagram.

Blue circle and pink diamonds are T_F and T_K, respectively. Squares represent the AFM magnetic transition T_N, which is determined as the temperature at which the magnetization-induced jumps in MR or R_xy vanish. Error bars are defined by the temperature steps in MR measurements. The thickness of the samples tunes J_K driving the system from a frustrated state in the bulk into a long-range AFM ordered state with dimension reduction down to few layers.

In conclusion, Kagome Kondo lattice has been experimentally realized in CsCr₆Sb₆. The discovery of heavy fermion behavior, Kondo singlets, Kondo breakdown, frustrated magnetic states, and a hidden AFM ordering establishes CsCr₆Sb₆ as an ideal archetype for investigating various strongly-correlated phenomena. Emergent phenomena including quantum critical point, non-Fermi liquid behavior, heavy fermion superconductivity and high-temperature fractional quantum Hall states are waiting to be explored.

Methods

Crystal synthesis and characterization

Single crystalline CsCr₆Sb₆ was synthesized using self-flux method. The mixture of Cs ingot (99.9%, Alfa), Cr grains (99.9%, Alfa) and Sb powder (99.999%, Alfa) were weighted with the mole ratio of 10:3:30, loaded into an alumina crucible and sealed in a quartz ampoule. The ampoule was heated to 1223 K and soaked there for 24 h. Then it was cooled down to 923 K at a rate of 2 K/h followed by furnace cooling. Finally, the collected melt was immersed in water and the van-der-Waals-like hexagonal crystals were harvested with typical size of 3 mm × 3 mm × 50 μm. Single crystal diffraction data was collected using a Rigaku XtaLAB Synergy R diffractometer (Supplementary Fig. 1a, resolved structure in Supplementary Table 1). X-ray diffraction (XRD) was performed on a Rigaku Smart Lab diffractometer with Cu K_α radiation (λ = 1.5418 Å) (Supplementary Fig. 1b). The magnetic properties were measured by a magnetism property measurement system (MPMS, Quantum Design). The high-angle annular dark-field (HAADF) images were captured using a JEOL ARM-200F scanning transmission electron microscope (STEM) operating at 200 Kv (Supplementary Fig. 1d). For structural analysis, the sample was prepared using the focused ion beam (FIB) method.

High resolution ARPES measurements

High resolution angle-resolved photoemission measurements were performed using a lab-based ARPES system equipped with the 21.2 eV He I lamp and a hemispherical electron energy analyzer DA30L (Scienta-Omicron). The energy resolution was set at 20 meV. All the samples were cleaved in situ at 15 K and measured in ultrahigh vacuum with a base pressure better than 5 ×10⁻¹¹ mbar. The Fermi level is referenced by measuring clean polycrystalline gold that is electrically connected to the sample.

Fabrication and characterization of few-layer CsCr₆Sb₆ devices

We obtain few-layer CsCr₆Sb₆ samples using an Al₂O₃-assisted exfoliation technique as described in ref. ⁵³ Firstly, the bulk samples are exfoliated to expose a fresh surface using scotch tapes. Then, 60–90 nm Al₂O₃ is deposited on the surface. Next, the stack was picked up by a thermal-released tape. Through this process, few layer samples are exfoliated benefit from the tight adhesion between sample surface and Al₂O₃. Subsequently, the stamp of Al₂O₃/CsCr₆Sb₆ is released onto a piece of PDMS (polydimethylsiloxane) with the CsCr₆Sb₆ side in contact with the PDMS surface. Supplementary Fig. 2a shows a typical optical image of few-layer CsCr₆Sb₆ samples on PDMS. Sample thickness is determined using the atomic force microscope and the optical contrast. Here, the optical contrast is transmittance defined as $G_{\mathrm{sample}}^{\mathrm{T}}/G_{\mathrm{substrate}}^{\mathrm{T}}$, where $G_{\mathrm{sample}}^{\mathrm{T}}$ and $G_{\mathrm{substrate}}^{\mathrm{T}}$ are the intensity of the transmission through the sample and substrate, respectively, in blue channel of the image captured with a charge-coupled device (CCD) camera. The transmittance of various numbers of layers follows the Beer-Lambert Law (Supplementary Fig. 2b) which enables us to determine the layer number quickly and precisely. After that, selected samples are finally stamped assembly onto a silicon substrate coated with 300 nm SiO₂. The electrical transport devices are fabricated through a standard electron beam lithography (EBL) process and a metal deposition with 5/60 nm Cr/Au as the electrodes.

Author contributions

T.P.Y. and X.L.C. conceptualized the idea. B.Q.S. carried out the majority of sample synthesis, device fabrication, measurements, and data processing, with assistance from H.L., S.J.T., H.C.L. and J.-G.G. J.C. and Q.H.Z. performed the electron diffraction. Y.Y.X., L.Z. and X.J.Z. carried out the ARPES measurements. X.Z., Q.H.Z. and X.T.L. performed STM measurements. W.-J.L. and S.-L.Y. performed the theoretical calculations. B.Q.S. and T.P.Y. analyzed the data and wrote the paper with inputs from all authors.

Peer review

Peer review information

Nature Communications thanks Jungseek Hwang, Dalmau Reig-i-Plessis and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Data availability

Most relevant data generated in this study (Figs. 2–4) are provided in the Source Data file. All other data that support the findings of this study are available from the corresponding authors upon request. 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 10.1038/s41467-025-60785-3.