Spin magnetic proximity effect in graphene superlattices

Abstract

The magnetic proximity effect induces spin splitting in graphene through interfacial exchange coupling, enabling control of spin-resolved band structure, most clearly revealed near charge neutrality where low carrier density enhances spin-dependent transport signatures. Here, we use cobalt contacts to induce magnetic proximity in graphene and probe spin-resolved bands with pure spin currents, observing a gate-tunable inversion of the nonlocal spin signal near the charge neutrality point. Similar inversions occur at satellite neutrality points in graphene-boron nitride aligned superlattices, demonstrating that proximity-induced spin splitting governs spin transport across both primary and reconstructed bands. In a bilayer graphene superlattice device, where a bandgap enhances energy-selective spin filtering, we observe spin polarizations approaching 50% and nonlocal spin resistances exceeding 300 Ω, nearly two orders of magnitude larger than those away from the charge neutrality point. Such electrically controlled spin polarization via proximity interactions at low carrier densities opens opportunities for low-power spintronic devices.

This study demonstrates gate-controlled inversion of large spin signals near charge neutrality points in graphene and its superlattices, probed via pure spin currents, which arises from magnetic proximity-induced spin splitting.

Introduction

Graphene’s low-energy band structure features massless Dirac fermions, with conduction and valence bands intersecting at the charge neutrality point (CNP) in a linear, gapless, and spin-degenerate configuration^1–3. This dispersion supports a variety of quantum transport phenomena including the Dirac fluid⁴, giant magnetoresistance⁵, and anomalous diamagnetism⁶ and establishes graphene as a solid-state platform for exploring relativistic physics in two dimensions. Yet, directly accessing spin-dependent features near the Dirac point remains experimentally challenging. In pristine graphene, spin–orbit coupling (SOC) opens a gap of only tens of μeV^7–10- well below the energy broadening induced by disorder and thermal fluctuations^11–13. While SOC effects have recently been resolved in ballistic quantum point contacts in bilayer graphene¹⁴, conventional charge transport measurements remain largely insensitive to these subtle features.

Magnetic proximity effect (MPE) offers an alternative route to lift spin degeneracy and engineer spin-resolved band structure in graphene^15–24. It has been demonstrated that MPE can be induced by magnetic insulators^{15,16,19–23}, where exchange splitting ranges from a few meV to several tens of meV^15,16,19,22. In these systems, MPE has mainly been indirectly probed through anomalous Hall signals^15,16,22,23 or nonlocal voltages arising from the Zeeman spin Hall effect^19–21. In parallel, first-principles calculations predict that cobalt (Co), even when separated from graphene by up to three layers of hexagonal boron nitride (hBN), can induce exchange splitting approaching 10 meV via spin-selective hybridization with Co’s d-orbitals^25,26. This interaction strongly modifies the spin-dependent band structure near the CNP, enabling selective access to spin-polarized states and offering new pathways for device-level control of spin transport. A few studies have investigated such metal-induced magnetic proximity effects in graphene using spin transport measurements^17,18, but their direct and gate-tunable manifestation in spin-resolved transport near charge neutrality - where the effects of exchange splitting are expected to be most pronounced - remains largely unexplored.

In contrast to charge transport, which averages over spin channels, nonlocal spin transport can selectively probe spin-split bands. In a spin valve geometry, when only one contact induces proximity magnetism, electrostatic gating of the charge carrier density across the CNP reverses the dominant spin species at that contact, leading to signal inversion as its polarization switches relative to the non-proximitized electrode. This inversion across the CNP would offer a direct experimental signature of band-selective spin occupation. Extending this strategy to bilayer graphene (BLG) could amplify its impact, given that the presence of a bandgap in BLG alters the low-energy density of states, enabling energy-selective access to spin-polarized bands. The spin splitting of this gap would result in transport dominated by a single spin species - effectively mimicking half-metallicity - and enable spin polarizations approaching 100%²⁷. This engineered band structure promises enhanced spin signals, far exceeding the few-ohm values typical of graphene spin valves²⁸, and is promising for spintronics devices that require large, gate-tunable spin densities²⁹.

In this work, we systematically probe magnetic proximity effects in graphene using nonlocal spin transport across three distinct devices. We begin with monolayer graphene, where spin signal inversion near the CNP reveals asymmetric spin injection into proximity-split bands - establishing a well-controlled and direct signature of exchange-induced band reconstruction. Extending this approach to aligned monolayer graphene, we observe similar inversion near moiré-induced satellite neutrality points, confirming that the same spin-resolved mechanism operates across reconstructed minibands. Finally, in doubly aligned bilayer graphene, where a bandgap enables energy-selective spin filtering, we detect non-local spin resistances exceeding 300 Ω and effective spin polarizations approaching 50%. These signatures persist to room temperature (RT), positioning spin transport as a sensitive probe of magnetic band reconstruction and demonstrating the viability of engineered graphene superlattices for high-efficiency spintronic devices.

Results and discussion

We begin by investigating spin transport in monolayer graphene misaligned with both top and bottom hBN layers, using a device fabricated with multiple one-dimensional (1D) cobalt edge contacts (Fig. 1a). The contact resistance R_c as a function of back gate voltage V_BG is characterized using a three-probe configuration across four representative contacts labeled a, b, c, and d. As shown in Fig. 1b, all contacts exhibit gate dependence with a pronounced electron-hole asymmetry, peaking at V_BG < 0 V. Among them, contacts c and d show similar resistance–width products R_cw, whereas contacts a and b display distinctly different values - indicative of different interfacial coupling strengths to graphene. Based on the relative magnitudes of R_cw, we define the a–b contact pair as asymmetric and the c–d contact pair as symmetric. While a symmetric configuration can in principle be formed by two contacts with either similarly high or low resistances, in this device it refers to the latter, as the one-dimensional edge contacts typically exhibit low-resistance. To investigate how this contact asymmetry influences spin transport, we perform nonlocal spin-valve measurements using the standard four-terminal geometry as depicted in Fig. 1a (see Methods). Figure 1c shows the magnetic-field dependence of the nonlocal spin signal. The symmetric injector–detector pair (c–d) exhibits a conventional spin-valve response, whereas the asymmetric pair (a–b) shows a clear inversion in spin signal polarity³⁰. A similar inversion is observed in Hanle measurements as well (see Supplementary Information). This contrasting behavior highlights the critical role of contact-induced exchange in enabling spin signal reversal and requires a detailed investigation of gate-dependent spin band inversion.

Fig. 1. Device structure and spin transport characteristics under symmetric and asymmetric electrodes in misaligned monolayer graphene at 20 K.

a Schematic of the non-aligned hBN/graphene/hBN device with 1D transparent contact. Gray and yellow slabs denote the Co and Au contacts, respectively. Inner Co contacts are used as injector (left) and detector (right) electrodes. A charge current is applied between the spin injector and left Au electrode, while a nonlocal voltage is measured between the spin detector and right Au electrode. Inset: the layer structure of the encapsulated graphene channel. b Dependence of the contact resistance–width product R_cw on back gate voltage V_BG for four different contacts labeled (a–d). c Nonlocal spin signal R_NL as a function of in-plane magnetic field B_∥ for symmetric (c, d, red) and asymmetric electrode (a, b, green) configurations at V_BG = 14 V and 12 V, respectively. The red curve shows a conventional spin-valve response, with the parallel configuration corresponding to the high-resistance state and the antiparallel configuration corresponding to the low-resistance state.

To probe the mechanism behind this inversion, we perform systematic nonlocal spin valve measurements on both measurement configurations. Figure 2a top and bottom panels display colormaps of the differential nonlocal resistance δR_NL = R_NL^trace − R_NL^retrace, plotted as a function of back gate voltage V_BG and in-plane magnetic field B_∥. Vertical color bands near B_∥ ≈ ±30 mT indicate antiparallel magnetization alignment (AP) between injector and detector electrodes, due to the magnetization reversal of said contacts, whereas the rest of the range corresponds to parallel alignment (P). For the symmetric contact pair, the spin signal exhibits the expected behavior: a consistent sign and gate dependence that is approximately symmetric about the CNP (located at V_BG ≈ −4 V)³¹. In stark contrast, the asymmetric configuration reveals a pronounced polarity reversal in the spin signal across the CNP. The opposite color contrast on either side of charge neutrality indicates that the sign of the nonlocal spin signal flips—a direct experimental signature of spin-dependent band occupation driven by magnetic proximity effect.

Fig. 2. Spin and charge transport in misaligned monolayer graphene at 20 K.

a Gate and field-dependent nonlocal spin signal δR_NL measured using asymmetric contact (top panel) and symmetric contact (bottom panel). δR_NL is defined as $\delta R_{\mathrm{NL}}=R_{\mathrm{NL}}\left(\mathrm{trace}\right)-R_{\mathrm{NL}}\left(\mathrm{retrace}\right)$, where a magnetic field sweep from negative to positive is referred to as a trace, and a sweep in the opposite direction as a retrace. b Resistivity ρ and ΔR_NL as a function of V_BG. Black dashed line indicates the CNP. Error bars represent the standard error of the mean of ΔR_NL obtained from magnetic field sweeps. Inset: spin-dependent charge carrier density (n_↑ and n_↓) as a function of energy in monolayer graphene with MPE.

We analyse the gate dependence of the spin signal by extracting the difference in the nonlocal resistance between the parallel and antiparallel states, ΔR_NL = R_NL^P − R_NL^AP, plotted alongside the longitudinal resistivity ρ in Fig. 2b. While ρ peaks sharply at the CNP, ΔR_NL shows antisymmetric extrema on the electron and hole sides, with a clear sign reversal centered at the CNP. This inversion is a hallmark of asymmetric spin injection under magnetic proximity: the lower-resistance contact induces an interfacial exchange field that spin-splits the graphene bands, while the higher-resistance contact remains weakly coupled and effectively does not induce MPE. In the modified band structure (schematically shown in the inset of Fig. 2b), both spin subbands cross the Fermi level and therefore contribute to transport. The resulting spin polarization originates from the unequal occupation and density of states of the two spin-split bands at the Fermi energy. As the Fermi level crosses the CNP, the dominant spin species in the proximitized region switches, reversing the sign of the detected spin accumulation when it becomes misaligned with the detector’s polarization. The result establishes a direct connection between spin signal polarity and the spin-resolved band structure induced by magnetic proximity. Consequently, a significant enhancement in polarization is observed near the CNP (see Supplementary Information). It is important to note that while first-principles calculations predict energy-dependent reversals in the induced spin polarization in Co/graphene systems, these occur over a broad (>1 eV) energy range²⁵. In contrast, the polarity reversal we observe takes place within a narrow Fermi energy window (<0.05 eV), confirming that it arises from spin band splitting near the Dirac point rather than a sign change in the interfacial spin polarization of Co/graphene itself.

To evaluate the generality of proximity-driven effect beyond the intrinsic Dirac point, we study Device 2, consisting of an aligned monolayer graphene/hBN channel with 1D asymmetric cobalt contacts. Figure 3a (top) presents ρ vs V_BG, revealing a primary peak at V_BG = −6.6 V and a secondary peak at V_BG = 21 V, corresponding to the main and electron-side moiré Dirac points (sDPs), respectively. The corresponding spin signal ΔR_NL, shown in Fig. 3a (bottom), exhibits again a distinct sign reversal near the electron-side sDP, mirroring the behavior observed at the main CNP in Fig. 2. We note that a clear modulation in the spin signal is seen near V_BG = 10 V, which we interpret to reflect the presence of MPE near van Hove singularities in the reconstructed miniband density of states³². Additional spin signal reversals are less resolved close to the main CNP and the hole-side sDP. Spin valve measurements taken on either side of the electron sDP (Fig. 3b) further confirm a polarity flip in the spin signal, demonstrating proximity-modified miniband structures in designer graphene systems.

Fig. 3. Spin and charge transport in aligned monolayer graphene heterostructures at 20 K.

a Gate-dependent ρ and ΔR_NL. Black and gray dashed lines indicate the CNP and sDP, respectively. b R_NL as a function of B_∥ at two gate voltages on either side of sDP, at V_BG = 30 V (red) and V_BG = 13 V (green).

We next examine how magnetic proximity effects manifest in a gapped band structure by studying Device 3 - a BLG heterostructure aligned to both top and bottom hBN layers. The top bilayer hBN spacer acts as a tunnel barrier between graphene and cobalt electrodes, with all crystals isolated and assembled in a glovebox to ensure atomically clean, residue-free interfaces—a critical step, as such contamination has been identified as a dominant source of spin dephasing in graphene^13,33. This doubly aligned configuration generates a super-moiré potential that reconstructs the bilayer graphene band structure into minibands and opens a bandgap at the CNP^34–36. Here, injector and detector contacts are similarly asymmetric (see Supplementary Information). Figure 4a schematically illustrates the expected spin polarization in the presence of a proximity-induced exchange field: spin-split conduction and valence bands produce a spin-dependent density of states, defining an energy window where transport is governed by a single spin species. The presence of the enlarged band gap and the reduced overlap between spin channels help suppress the influence of charge fluctuations, allowing the Fermi level to more easily access the highly spin-polarized regime. As a result, this spin filtering effect is anticipated to be stronger than in monolayer graphene.

Fig. 4. Spin and charge transport in doubly-aligned bilayer graphene at 2.5 K.

a Spin-dependent charge carrier density (n_↑ and n_↓) as a function of energy in bilayer graphene with MPE. b Schematic of hBN/graphene/hBN device with tunneling contact. Upper-right inset: a moiré pattern formed by hBN and graphene. Lower-right inset: the schematics of the heterostructure. c ρ (top panel) and ΔR_NL and dρ/dV_BG (bottom panel) as a function of V_BG. The twist angles (θ_top and θ_bottom) between graphene and the top and bottom hBN layers are approximately 0.6° and 0.8°, respectively. The extracted moiré wavelengths λ are 13.9 nm and 10.7 nm on the electron side, and 13.4 nm and 10.8 nm on the hole side. Error bars represent the standard error of the mean of ΔR_NL obtained from magnetic field sweeps. d Hanle and (e), nonlocal spin valve measurements performed at V_BG = 0.6 V and 0.25 V, with ΔR_NL values of 220 Ω and -133 Ω, respectively.

Figure 4c shows the gate dependence of the nonlocal spin signal ΔR_NL alongside the longitudinal resistivity. The resistivity peak defines the position of the CNP, with a corresponding maximum in ΔR_NL on either side. Crucially, the spin signal undergoes a full sign inversion as V_BG sweeps through the CNP. A second inversion appears near the hole-side satellite Dirac point, consistent with miniband formation in the moiré-reconstructed spectrum. The bottom panel of Fig. 4c compares ΔR_NL to dρ/dV_BG, which serves as a spectroscopic proxy for the density of states. Their matching features at the main and secondary CNPs confirm that the spin signal reversal is rooted in underlying band structure modifications. Interestingly, the sign of the spin signal near the CNP in Device 3 is opposite to that observed in Devices 1 and 2, suggesting a reversal of the dominant spin polarization under proximity. We note that Device 3 employs 2D traditional cobalt contacts²⁸, which increase the proximitized graphene area compared to the 1D edge contacts used previously—a factor likely contributing to the observed contrast. Such behavior is consistent with first-principles predictions for graphene proximitized by cobalt across a bilayer hBN spacer, where the exchange field can invert depending on interface configuration and stacking²⁶. Finally, unlike Device 2, which exhibits a single alignment signature, Device 3 displays multiple resistivity peaks and spin signal inversions across both electron and hole sides, consistent with a super-moiré potential from dual alignment to top and bottom hBN³⁵. This confirms that magnetic proximity effects can be tuned through both alignments, offering a pathway to engineer the number and position of spin signal inversions by controlling the stacking angles.

To directly verify the sign inversion, we perform Hanle spin precession and spin valve measurements at two gate voltages flanking the primary CNP (Fig. 4d, e). Both measurements show clear and opposite polarity in ΔR_NL, confirming the gate-controlled reversal of the spin signal and validating proximity-driven spin splitting in this gapped bilayer regime. Remarkably, due to the sizeable energy gap, the magnitude of ΔR_NL reaches over 300 Ω near the primary CNP- over 100 times larger than the few-ohm values observed at high doping, outside of CNPs (Fig. 4c). At the hole-side satellite peak, although the energy gap is smaller than that at the primary CNP, it remains sufficiently large to yield a ΔR_NL exceeding 70 Ω, representing a more than 20-fold enhancement, while on the electron side, where the energy gap is further reduced, no comparably large ΔR_NL is observed. These results highlight the crucial role of the energy gap and confirm that strong proximity-driven spin polarization extends across reconstructed minibands. We estimate the effective polarizations at the hole- and electron-side extrema of the main CNP to be 13.6% and 10%, respectively, assuming symmetric injector and detector contributions. To isolate the effect of the proximity-modified injector, we estimate the detector polarization separately (P_d ≈ 4%) using an independent geometry with a symmetric contact pair (see Supplementary Information), allowing us to extract injector polarizations of ~ 47% and ~ 25% - demonstrating a highly spin-polarized current injected via proximity-magnetized contacts³⁷. It is important to note that the complete spin signal reversal from maximum positive to negative values occurs over a narrow ~ 300 mV gate range, enabled by the optimized hBN/graphite back-gate configuration that provides precise electrostatic control.

Next, we examine the behavior of magnetic proximity effects at room temperature in the same device. As shown in Fig. 5a, the spin signal ΔR_NL exhibits a clear polarity inversion near the primary CNP, closely aligned with features in the longitudinal resistivity. The spin signal again reverses polarity near the CNP, reaching up to 15 Ω on both sides - confirming robust proximity-induced spin polarization at RT. Figure 5b shows spin valve measurements on either side of this inversion, validating the signal. It is important to note that two additional, smaller sign changes appear near V_BG ≈ −6.25 V and +3.75 V, without corresponding signatures in ρ. These may reflect satellite Dirac points that remain unresolved in charge transport but are detectable via spin transport, highlighting the enhanced spectroscopic sensitivity of spin-based measurements to subtle features in the moiré-reconstructed band structure.

Fig. 5. Spin and charge transport in doubly-aligned bilayer graphene at 300 K.

a ρ (top panel) and ΔR_NL and dρ/dV_BG (bottom panel) as a function of V_BG. b Nonlocal spin valve measurements performed at V_BG = 0.8 V and −2.18 V, with ΔR_NL values of 16 Ω and -14 Ω, respectively.

These additional features arise from the moiré reconstruction imposed by graphene–hBN alignment. The twist angle determines the position of satellite neutrality points and the magnitude of the associated band gaps, and since the proximity-induced spin signal is directly linked to the underlying band structure, variations in alignment naturally shift the location and amplitude of the corresponding spin-signal extrema. Beyond moiré effects, first-principles calculations have shown that proximity-induced exchange in Co/hBN/graphene heterostructures originates from spin-dependent hybridization between graphene p_z orbitals and Co d orbitals²⁵, and that both the magnitude and sign of the exchange splitting depend sensitively on hBN thickness²⁶ and atomic registry at the interface³⁸. These studies predict exchange splittings on the order of several meV and even sign reversals under different stacking configurations. While our devices employ polycrystalline Co electrodes and therefore sample multiple local registries, the measured spin polarization reflects the net proximity-induced exchange. In this sense, the existing theoretical framework^25,26,38, captures the essential mechanism, and our results provide experimental validation of proximity-driven spin-band reconstruction in gate-tunable graphene superlattices.

Our findings establish the magnetic proximity effect in graphene as a robust, gate-tunable mechanism for spin signal inversion, observed not only at the intrinsic Dirac point but also at moiré-induced satellite neutrality points arising from crystallographic alignment to hBN. In bilayer graphene, the presence of the gap enables this effect to produce nonlocal spin signals exceeding 300 Ω. Given that such high spin polarizations could be demonstrated at sufficiently high current densities and reduced contact areas, this opens a potential pathway toward nonlocal spin-transfer torque applications, advancing the feasibility of all-spin logic architectures²⁹. More broadly, the sensitivity of spin transport to fine features in the electronic spectrum underscores its value as a spectroscopic probe of spin-resolved band structures. Extending this approach to moiré-engineered quantum systems such as twisted graphene heterostructures could offer unprecedented insights into spin-dependent electronic structures and their emergent correlated states³⁹.

Methods

Crystal growth and sample preparation

Devices 1 and 2 are stacked via a dry transfer method⁴⁰, using flakes of graphene and hBN exfoliated onto separate Si/SiO₂ substrates. Following stacking, the heterostructures are annealed in a H₂/Ar atmosphere at high temperature (> 300 °C), for 4 h. Transport channels are defined by exposing a polymer hard mask in electron beam lithography, followed by reactive ion etching, which exposes a narrow ledge of graphene at the channel edges (~ 10 nm). A pattern of nanoscale electrodes is exposed in a second lithography step, after which metal is deposited (Co 60 nm/Au 20 nm) via electron beam evaporation. The magnetic nanowire electrodes make electrical contact at the exposed graphene edges, enabling the injection and detection of spin polarized current.

The third device was fabricated in a nitrogen-filled glovebox. A polymer-based polycarbonate/polydimethylsiloxane (PC/PDMS) stamp was used to sequentially pick up the top hBN, graphene, and bottom hBN flakes from Si/SiO₂ substrates, with crystallographic alignment. The full stack was finally transferred onto a pre-exfoliated graphite flake. The resulting heterostructure was annealed at 340 °C for 6 hours under vacuum conditions. The graphene surface was never exposed to ambient conditions or any polymer during the fabrication. After forming the Co (35 nm)/Au (5 nm) contacts, a final step annealing process was performed to further remove fabrication residues.

hBN single crystals were synthesized using the temperature gradient method under high-pressure and high-temperature conditions, as described in Ref. ⁴¹.

Measurement

Cryogenic measurements were conducted using cryogenic systems with a base temperature of 2.5 K for Device 3 and 18 K for Devices 1 and 2. For charge transport measurements, a Keithley 2450 Sourcemeter was used to apply both the source-drain bias and the gate voltage and the dropping voltage was recorded with Keithley 2182 A. Spin transport signals were recorded either using a Keithley 2182 A nanovoltmeter or SR830 lock-in amplifier using low frequency ac currents.

For nonlocal spin valve measurements (Figs. 1a and 4b), a charge current I is applied between a gray (Co) electrode acting as the spin injector and a yellow (Au) electrode, while a nonlocal voltage V_NL is measured between another Co electrode serving as the spin detector and a second Au electrode. The nonlocal resistance is then defined as R_NL = V_NL/I. All spin-transport measurements were performed using excitation currents within the linear regime (see Supplementary Information). The Co electrodes possess a predominantly in-plane magnetization due to shape anisotropy, and their different widths ensure distinct coercive fields. By sweeping an in-plane magnetic field along their easy axis, the parallel and antiparallel magnetization configurations of the injector and detector can be switched controllably. The resulting changes in the nonlocal spin signal give a measure of the spin polarization of contacts.

For clarity in comparing charge and spin signals, a small horizontal offset was applied to the charge data to align the Dirac point with the spin measurements in Figs. 4c and 5a; this accounts for slight gating hysteresis and time-dependent doping, as the spin data were acquired over significantly longer durations than the rapid charge sweeps.

Author contributions

A.A. and I.J.V-M. designed and coordinated the work. Y.L., D.B., V.H.G.M. and J.C.T-F. fabricated and measured the samples. Y.J. supported the device fabrication. K.W. and T.T grew hBN crystals. Y.L., D.B., J.C.T-F., I.J.V-M. and A.A. analyzed the results and wrote the manuscript.

Peer review

Peer review information

Nature Communications thanks Qingjun Tong and the other anonymous reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Data availability

The data supporting the findings of this study are available within the paper and its Supplementary Information. The datasets generated during and/or analysed during this study are available from the corresponding author upon reasonable request.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-026-71915-w.