Introduction

Heterostructures are promising to host emergent phenomena and device functions not present in constituent parts1,2,3,4,5,6,7,8,9,10. One well-known example is the integration of two insulating complex oxides leading to a conducting two-dimensional electron gas at the interface2, with surprising coexistence of superconductivity and ferromagnetism3. The recently explored van der Waals (vdW) magnets have pushed the research frontier to 2D magnetism where exotic magnetic ground states and quantum phases can emerge11,12,13,14,15,16,17,18. Magnetic vdW heterostructures provide a new toolbox to explore magnetic proximity and related effects4,5,6,7,8,9,10. A largely unexplored arena is to combine two different magnetic orders and investigate the magnetic proximity at the interface, which could allow modulation of magnetic interactions and establish exotic magnetic properties. It is also of fundamental significance to effectively control the exchange interactions and magnetic anisotropy, with the latter being crucial to stabilize the long-range magnetic orders.

The studies of layered semiconducting chromium trihalides have shown exotic magnetic behaviors and rich tunability by stimuli14,15,16,17,18. Typically, the few-layer (FL) CrI3 is an antiferromagnet with Ising-like perpendicular magnetic anisotropy (PMA)19,20, as schematically depicted in Fig. 1a. The interlayer antiferromagnetic coupling is ascribed to the exchange interactions between Cr mediated by ligand atoms17,20. In contrast, FL CrCl3 is an easy-plane interlayer antiferromagnet, where spins prefer to lie in the layers20,21. In particular, the single-ion anisotropy from spin-orbit coupling (SOC) of Cr and the anisotropic exchange from SOC of Cl nearly cancel out each other20,21. Therefore, CrCl3 is located close to the boundary between PMA and in-plane anisotropy, suggesting that its magnetic properties may be particularly susceptible to external perturbations. The combined heterostructure of CrI3 and CrCl3 is possibly a fertile system to realize rich magnetic phases and manipulate them. Such a heterostructure has not yet been explored.

Fig. 1: CrI3/CrCl3 heterostructures and MOKE measurements.
figure 1

a Schematics of the magnetic ground states in bilayer (2L) CrI3 and few-layer (FL) CrCl3 before (left) and after (right) forming heterostructure. Only four layers of CrCl3 are shown for simplicity. b Optical micrograph of a 2L CrI3/FL CrCl3 heterostructure. c Atomic force microscopy of the heterostructure in the same position as in b. The height profile (along the yellow dotted line in the image) at the edge of CrI3 indicates the thickness of a bilayer. d MOKE signal (after subtracting a polynomial background, as done for all MOKE curves in the main text) of the 2L CrI3 region and the 2L CrI3/FL CrCl3 heterostructure region as a function of perpendicular magnetic field. Two curves of each region represent forward and backward sweeps of the field, respectively. The data are taken at the spots marked by red in b. Insets depict magnetic ground states of 2L CrI3 and the CrI3/CrCl3 heterostructure (showing only the interfacial CrCl3 layer, highlighted by red dashed rectangles). e MOKE signal of another monolayer (1L) CrI3/FL CrCl3 heterostructure, compared with that measured in the 1L CrI3 (from the same CrI3 flake as in the heterostructure region).

Full size image

Here, we fabricate CrI3/CrCl3 heterostructures and demonstrate interfacial ferromagnetism between the two antiferromagnets. Figure 1a left panel schematically depicts the expected spin configurations for the magnetic ground states in bilayer (2L) CrI3 and FL CrCl3 with PMA and in-plane anisotropy19,20,21, respectively. Due to the strong intralayer ferromagnetic coupling in chromium trihalides17, we can denote all spins in a given layer by a macroscopic spin (out-of-plane: ↑, ↓; in-plane: ←, →). The optical micrograph of a representative 2L CrI3/FL CrCl3 heterostructure is shown in Fig. 1b. The 2L CrI3 is partially stacked on top of FL CrCl3, allowing the comparison between regions of 2L CrI3, FL CrCl3 and heterostructure. Atomic force microscopy (AFM) confirms flake thickness of 1.6 nm for 2L CrI3 (Fig. 1c) and 9.5 nm for FL CrCl3, respectively.

Results and discussion

We employ magneto-optical Kerr effect microscopy (MOKE) under the polar configuration as the primary measurement due to its high sensitivity to the magnetic moments perpendicular to the sample surface22. Figure 1d shows the MOKE signal (θK) of the 2L CrI3 and the 2L CrI3/FL CrCl3 heterostructure as a function of the perpendicular magnetic field. In the 2L CrI3 region, θK stays close to zero at low field, corresponding to the antiferromagnetic states ↓ ↑ or ↑↓ with zero net magnetization. Beyond critical field ±0.76 T, θK abruptly jumps to ferromagnetic states with finite magnetization. This is consistent with the reported spin-flip transitions in 2L CrI323.

In the 2L CrI3/FL CrCl3 heterostructure, the antiferromagnetic-to-ferromagnetic spin-flip transition of 2L CrI3 is still present and its critical field decreases from ±0.76 T in 2L CrI3 region to ±0.57 T in the heterostructure region. Remarkably, a significant square hysteresis loop is observed with coercive field ~±0.1 T, indicating a magnetic transition between two different phases with non-zero net magnetization, in sharp contrast to the antiferromagnetic ground states in 2L CrI3 (↓↑ or ↑↓). Such a ferromagnetic-like loop is absent in either 2L CrI3 or FL CrCl3, suggesting its origin from interfacial magnetic interaction. This phenomenon should not be due to the charge transfer/doping-induced antiferromagnetic-to-ferromagnetic transition reported in 2L CrI314,15, which does not exhibit such a coexistence of antiferromagnetic-type and ferromagnetic-type transitions. Recent works have shown that a twist of two chromium trihalides layers may lead to noncollinear antiferromagnetic-ferromagnetic domains and thus finite MOKE signals9,10,18,24,25,26,27. However, this scenario is also less likely to be relevant, as such domains are predicted to emerge for sufficiently large moiré periodicity. Due to large lattice constant mismatch18, the CrI3/CrCl3 heterostructure can hardly form large moiré periodicity even at zero twist angle. Furthermore, the magnetization behaviors (including the field and temperature dependence) measured in twisted CrI39,10,25,26,27 exhibit qualitative difference from what we observe in our samples. We further rule out several other possible origins (see detailed discussions in Supplementary Text 1). We propose that at least three spin layers (two layers of CrI3 plus one neighboring layer of CrCl3) are responsible for the observed transitions. The neighboring CrCl3 layer is acting as the third spin layer with out-of-plane magnetic order after being stacked in proximity with CrI3, as schematically shown in Fig. 1a and denoted in the red dashed rectangles in Fig. 1d, e. Note that a perpendicular magnetic field induces canting of the planar CrCl3 spins, giving rise to a continuously varying MOKE background28 (Supplementary Fig. 1), which is typically subtracted and eliminated from our MOKE signal. Therefore, only perpendicular spin-flip transitions are discussed in this work.

Similar to trilayer CrI3, in principle several potential antiferromagnetic configurations can be considered: ↑↓↓ (−1), ↓↑↑ (+1), ↓↑↓ (−1), ↑↓↑ (+1) with the first two and the third spins referring to the 2L CrI3 and the neighboring CrCl3 layer respectively and the numbers in brackets denoting the net magnetic moments. To figure out the coupling type for the neighboring CrI3 and CrCl3 layers, we study the 1L CrI3/FL CrCl3 heterostructure where the magnetic behavior can directly verify the interlayer coupling type. Figure 1e shows θK of the 1L CrI3 and the 1L CrI3/FL CrCl3 heterostructure, fabricated from the same CrI3 flake. Interestingly, both show a ferromagnetic behavior with a single hysteresis loop. This observation indicates that the neighboring CrI3 and CrCl3 layer is ferromagnetically coupled, in contrast to the interlayer antiferromagnetic coupling in FL CrI319. Careful inspection on the hysteresis loop in Fig. 1d shows more transition steps, possibly due to the switching of magnetic domains23,29. Note that due to thin-film optical interference effect23,30. it is not possible to associate the magnitude or sign of the MOKE signal to the magnetization of different samples (e.g., the doubling of the MOKE signal in the 1L CrI3/FL CrCl3 heterostructure relative to that in the 1L CrI3 in Fig. 1e does not imply that the probed magnetization doubles). In the following text, we mainly focus on other features (e.g., emergent hysteresis loop, transition/coercive fields, critical temperatures). To exclude any unique causes related to the stacking sequence, we also studied reversely stacked heterostructures with FL CrCl3 on top of 2L CrI3 and observed similar hysteresis loops (Supplementary Figs. 2 and 3). The larger coercive field of the hysteresis loop observed in the reversed stack may be due to sample differences or twist angle dependence and is out of the scope of this work.

We next study the temperature dependence of the magnetism in the heterostructure. Figure 2a, b shows the temperature dependence of θK in 2L CrI3 and 2L CrI3/FL CrCl3 heterostructure. The extracted H1, H1*, H2* and ∆θ1, ∆θ1*, ∆θ2* as a function of temperature are shown in Fig. 2c, d, respectively. The antiferromagnetic spin-flip transitions (at H1 and H1*) in both 2L CrI3 and the heterostructure disappear at temperatures larger than TC ~ 40 K and is consistent with previous measurements in 2L CrI331. The ferromagnetic-like hysteresis loop (at H2*) observed only in the heterostructure region survives up to a higher temperature TC* ~ 48 K. Another experiment on 1L CrI3/FL CrCl3 heterostructure shows critical temperatures of TC ~ 33 K and TC* ~ 37 K for 1L CrI3 region and heterostructure region, respectively (Supplementary Fig. 4). In 2D magnets, the critical temperature is determined by the spin-wave excitation gap, which is dictated by the anisotropies present in the system12,23,32,33. Our density functional theory results suggest an increase in the effective single-ion anisotropy of CrI3 when brought in proximity to CrCl3. On the other hand, thanks to the induced ferromagnetic coupling, the CrCl3 layer now sees an effective anisotropy field that depends both on the interfacial ferromagnetic coupling as well as the anisotropy of CrI3, which is expected to enlarge the spin-wave gaps for both the materials. This is consistent with the observed increase of TC for both 1L CrI3/FL CrCl3 and 2L CrI3/FL CrCl3 systems.

Fig. 2: Temperature dependence of the magnetism of 2L CrI3 and 2L CrI3/FL CrCl3 heterostructure.
figure 2

MOKE signal in the 2L CrI3 region (a) and the 2L CrI3/FL CrCl3 heterostructure region (b) as a function of perpendicular magnetic field at different temperatures. Critical fields H1, H1*, H2* and magnitudes in the change of MOKE signal ∆θ1, ∆θ1*, ∆θ2* of magnetic transitions are labeled. c Temperature dependence of the critical fields of magnetic transitions. PM paramagnetic. The critical fields are extracted from the peak of derivative dθK/dH and the error bars are the peak widths. d Temperature dependence of the magnitudes in the change of MOKE signal at magnetic transitions. Solid curves are fitted by a power-law equation11. The critical temperatures TC and TC* are indicated. The error bars are the uncertainties in extracting the transition magnitudes.

Full size image

To better understand the observations, we explore the magnetic ground states of the CrI3/CrCl3 bilayer using first-principles calculations (Supplementary Text 2). We find that the perpendicular ferromagnetic state (↑↑) is more favorable than three other magnetic configurations: perpendicular antiferromagnetic state (↑↓), the states that one layer is out-of-plane polarized while the other in-plane polarized (↑→ or →↑). Further consideration on magnetic dipole-dipole interaction and different commensurate twist angles (0° and 30°) does not undermine the favorable perpendicular ferromagnetic state. The interlayer exchange energy Jinter in CrI3/CrCl3 can be approximated by the energy difference between perpendicular antiferromagnetic and ferromagnetic configurations17. We estimate Jinter ≈ −77 (−64) μJ/m2 for 0° (30°)-twisted CrI3/CrCl3, compared to the reported interlayer exchange ~80 μJ/m2 in 2L CrI332,34,35. Such interfacial exchange coupling in the heterostructure wins over the in-plane anisotropy of CrCl3 and results in the out-of-plane magnetic order in the CrCl3 layer next to CrI3, in agreement with our observations.

We next turn to explore the electrical tunability of the observed interfacial magnetism. A unique aspect of the CrI3/CrCl3 heterostructure, when compared with previously explored monolayer and/or homobilayer systems14,15,36, is the absence of structural inversion symmetry. The Neumann’s principle37 states that the spin-charge coupling is dictated by the symmetries of the system. We thus expect to observe spin-charge coupling phenomena for the interlayer magnetic order. In particular, breaking of structural inversion allows for direct electric-field modification of the magnetic anisotropy and the interlayer exchange interactions via terms of the form (see detailed discussions in Supplementary Text 3):

$${E}_{{{\mbox{elec}}}}\left({{{{{{\bf{m}}}}}}}_{i},{\sigma }_{i}\right)=\left({\sigma }_{1}-{\sigma }_{2}\right)\left({\beta }_{1}{m}_{z1}^{2}+{\beta }_{2}{m}_{z2}^{2}+{\beta }_{3}{{{{{{\bf{m}}}}}}}_{1}\cdot {{{{{{\bf{m}}}}}}}_{2}\right),$$
(1)

where \({{{{{{\bf{m}}}}}}}_{i}\), \({\sigma }_{i}\) are the magnetization and charges of the respective layers, \(({\sigma }_{1}-{\sigma }_{2})\sim\) electric field and \({\beta }_{{{{{\mathrm{1,2,3}}}}}}\) parameterizes the strength of respective interactions. Microscopically, the electric-field control of interfacial magnetic interactions could arise from electric-field-induced changes in the orbital occupancy in conjunction with spin-orbit interactions. Such a mechanism has attracted significant interest for constructing low-dissipation spintronic memory and logic devices38,39.

To check the electric-field tuning of the observed interfacial magnetism, we fabricated a dual-gated 1L CrI3/FL CrCl3 device, as shown in Fig. 3a, b. This structure allows us to study the magnetization of the CrI3/CrCl3 heterostructure (as well as that of the 1L CrI3 region in the same device) under the top-gate voltage Vtg and back-gate voltage Vbg. The two voltages are converted to electrostatic doping density n and displacement field D (Methods). Figure 3c shows the coercive field (Hc) in 1L CrI3/FL CrCl3 heterostructure increases from ~700 Oe to ~1000 Oe when the D is tuned from −1.4 V nm−1 to 1 V nm−1, indicating the enhancement of the magnetic anisotropy of the interfacial ferromagnetism in the heterostructure. The full mappings in Fig. 3d, e present the extracted Hc as a function of both n and D in 1L CrI3 and the 1L CrI3/FL CrCl3 heterostructure, respectively. A quite weak modulation of Hc is observed in 1L CrI3, suggesting that the magnetism of 1L CrI3 can hardly be tuned under the range of gating voltages of this work. Separate experiments on FL CrCl3 demonstrate that the magnetism of CrCl3 also can hardly be tuned by electrostatic gating (Supplementary Fig. 5d). However, significant tunability of the Hc is observed by the D applied to the heterostructure. Such a dramatic tunability in the CrI3/CrCl3 is in agreement with the electric field control of interfacial magnetic interactions allowed by the structural symmetry breaking, predicted in the above theoretical analysis. The intriguing electrical tunability allowed by symmetry breaking is also observed in a heterostructure containing a bilayer CrI3 (Supplementary Text 4).

Fig. 3: Electrical control of the magnetism in 1L CrI3/FL CrCl3 heterostructure.
figure 3

a, b Optical micrograph and schematic structure of a dual-gated 1L CrI3/FL CrCl3 device. Three few-layer graphene (FLG) flakes are used as back/top gates and the contact to the stack. c Normalized MOKE signal as a function of perpendicular magnetic field in 1L CrI3/FL CrCl3 heterostructure under different displacement fields D = −1.4, −0.2, 1.0 V nm−1. Coercive field Hc as a function of electrostatic doping density n and displacement field D of 1L CrI3 (d) and 1L CrI3/FL CrCl3 heterostructure (e), respectively. The black dots in e correspond to the MOKE curves in c. The data are taken at the spots marked by red in a.

Full size image

In summary, we studied the interfacial magnetism in CrI3/CrCl3 heterostructures and demonstrated the interfacial ferromagnetic coupling between neighboring CrI3 and CrCl3 layers. The demonstrated ability to engineer magnetoelectric phenomena by breaking symmetries via vdW heterostructures provides opportunities for vdW spintronics. The compatible hybrids of 2D magnets with other quantum materials, such as unconventional superconductors, ferroelectrics or topological materials are predicted to demonstrate exotic topological phases and many-body interactions6,12,13, as well as to design new spintronic devices and therefore are highly desirable for further study.

Methods

Crystal growth

Single crystal CrI3 was synthesized using the chemical vapor transport (CVT) method40. The Cr powder and iodine pieces were mixed with a stoichiometric ratio and loaded into a quartz tube (inner diameter, 10 mm; length, 180 mm). The quartz tube was sealed under vacuum and then transferred to a double temperature zones furnace. The temperatures of the hot and cold ends of the furnace were set at 650 °C and 550 °C, respectively. The growth with such a temperature gradient lasted for 7 days. Finally, the furnace was shut down, and the quartz tube naturally cooled down to room temperature. The black plate-like CrI3 crystals can be found at the cold end of the quartz tube.

Single crystal CrCl3 was grown by the CVT method. The commercial CrCl3 polycrystal powder (99.9%) was sealed in a silica tube with a length of 200 mm and an inner diameter of 14 mm. The tube was pumped down to 0.01 Pa and sealed under vacuum, and then placed in a two-zone horizontal tube furnace. The two growth zones were raised slowly to 973 K and 823 K for 2 days, and then held there for another 7 days. After that, the furnace was shut down and cooled down naturally. Shiny, plate-like crystals with lateral dimensions of up to several millimeters can be obtained from the growth.

Device fabrication

FL CrI3, CrCl3 and hexagonal boron nitride (hBN) flakes are exfoliated onto the silicon wafer covered by 285-nm thermal oxide layer. Flakes with proper thickness are selected by optical contrast23 and later confirmed by AFM and MOKE measurements. CrI3 flakes used in this work have 1~2 layers and the FL CrCl3 flakes are around 5~10 nm (0.6 nm for each layer) thick. Heterostructures of CrI3 and CrCl3 are fabricated by the dry-transfer method and encapsulated between two hBN flakes with a typical thickness of ~10 nm. Specifically, a stamp made of a thin polycarbonate and polydimethylsiloxane is then employed to pick up the flakes in sequence under an optical microscope. In the end, the finished stack is deposited onto the target substrate with polycarbonate on top which is removed by chloroform afterwards. The whole process is performed inside a glovebox to avoid material degradation. The exposure time to air is kept below ten minutes before transferring the fabricated sample into the measurement chamber and pumping down.

For the dual-gated heterostructure device and magnetic tunneling junction device, FL graphene flakes are exfoliated and integrated into the stack following the above processes. The target substrate is pre-patterned with electrodes fabricated by standard e-beam lithography, Au/Ti deposition and lift-off processes. The stack is carefully aligned and transferred onto the target pattern to make contact between graphene flakes and electrodes.

MOKE microscopy

The polarization of a linearly polarized light reflected from a magnetic material will be rotated by a Kerr angle θK, which is proportional to the magnetization of the material. In this work, the incident light is normal to the sample plane and MOKE is in the polar geometry, meaning that the magnetic vector being probed is perpendicular to the sample surface and parallel to the incident light. A balanced photodetector and lock-in method are used to obtain the MOKE signal. A laser is used here with wavelength of 633 nm and power of 5 µW. The sample is placed in a helium-flow optical cryostat with the temperature down to 6 K and magnetic field (perpendicular to sample surface) up to 5 T. The laser is focused onto the sample surface by an objective with the spot diameter of 0.5 µm.

Electrical control of the dual-gated device

Top-gate and back-gate voltages can be applied to the FL graphene gates in the heterostructure device, while the graphene contact to the heterostructure is grounded. The dual-gate structure allows independent control of the doping density and displacement field applied on the heterostructure. The doping density n and displacement field D are extracted by the simple parallel plate capacitor model. For simplicity, the CrI3/CrCl3 heterostructure is regarded as one channel, on which the doping density and electric field are applied. The quantum capacitance of CrI3 and CrCl3 is much larger than that of graphene due to the nearly flat bands of these two magnetic semiconductors15. Therefore, only geometric capacitances \({C}_{{{\mbox{bg}}}}\) and \({C}_{{{\mbox{tg}}}}\) are considered. The doping density and displacement field can be written as \(n={C}_{{{\mbox{bg}}}}\cdot {V}_{{{\mbox{bg}}}}+{C}_{{{\mbox{tg}}}}\cdot {V}_{{{\mbox{tg}}}}\) and \(D=({D}_{{{\mbox{bg}}}}+{D}_{{{\mbox{tg}}}})/2=({\varepsilon }_{{{\mbox{bg}}}}\cdot {V}_{{{\mbox{bg}}}}/{d}_{{{\mbox{bg}}}}-{\varepsilon }_{{{\mbox{tg}}}}\cdot {V}_{{{\mbox{tg}}}}/{d}_{{{\mbox{tg}}}})/2\), respectively. The relative dielectric constant of hBN14 is εbg = εtg = 3. For the device in Fig. 3, the thicknesses of bottom hBN and top hBN are obtained by AFM measurement to be dbg = 19.6 nm and dtg = 14.9 nm, respectively.