Chun-I
Lu
ab,
Chih-Heng
Huang
ac,
Kui-Hon
Ou Yang
d,
Kristan Bryan
Simbulan
b,
Kai-Shin
Li
e,
Feng
Li
f,
Junjie
Qi
f,
Matteo
Jugovac
g,
Iulia
Cojocariu
g,
Vitaliy
Feyer
g,
Christian
Tusche
g,
Minn-Tsong
Lin
hi,
Tzu-Hung
Chuang
a,
Yann-Wen
Lan
*b and
Der-Hsin
Wei
*a
aScientific Research Division, National Synchrotron Radiation Research Center, Hsinchu, Taiwan. E-mail: dhw@nsrrc.org.tw
bDepartment of Physics, National Taiwan Normal University, Taipei, Taiwan. E-mail: ywlan@ntnu.edu.tw
cInternational PhD Program for Science, National Sun Yat-Sen University, Kaohsiung, Taiwan
dGraduate Institute of Applied Physics, National Taiwan University, Taipei, Taiwan
eNational Nano Device Laboratories, National Applied Research Laboratories, Hsinchu, Taiwan
fState Key Laboratory for Advanced Metals and Materials, School of Materials Science and Engineering, University of Science and Technology Beijing, Beijing, People's Republic of China
gForschungszentrum Jülich, Peter Grünberg Institut (PGI-6), 52425, Jülich, Germany
hDepartment of Physics, National Taiwan University, Taipei, Taiwan
iInstitute of Atomic and Molecular Sciences, Academia Sinica, Taipei, Taiwan
First published on 29th April 2020
Magnetic anisotropy (MA) is a material preference that involves magnetization aligned along a specific direction and provides a basis for spintronic devices. Here we report the first observation of strong MA in a cobalt–molybdenum disulfide (Co/MoS2) heterojunction. Element-specific magnetic images recorded with an X-ray photoemission electron microscope (PEEM) reveal that ultrathin Co films, of thickness 5 monolayers (ML) and above, form micrometer (μm)-sized domains on monolayer MoS2 flakes of size tens of μm. Image analysis shows that the magnetization of these Co domains is oriented not randomly but in directions apparently correlated with the crystal structure of the underlying MoS2. Evidence from micro-area X-ray photoelectron spectra (μ-XPS) further indicates that a small amount of charge is donated from cobalt to sulfur upon direct contact between Co and MoS2. As the ferromagnetic behavior found for Co/MoS2 is in sharp contrast with that reported earlier for non-reactive Fe/MoS2, we suggest that orbital hybridization at the interface is what makes Co/MoS2 different. Our report provides micro-magnetic and micro-spectral evidence that consolidates the knowledge required to build functional heterojunctions based on two-dimensional (2D) materials.
New conceptsThe new concept in this work is the observation of spontaneously induced magnetic anisotropy in monolayer MoS2 due to charge donation at the interface of the Co/MoS2 heterojunction. In this way, monolayer transition metal dichalcogenide (TMD) materials can be magnetized by simply attaching a suitable material, instead of doping them, to avoid damaging the atomic structure. Furthermore, the magnetic properties of TMDs could be fine-tuned by choosing different materials, which involves different orbital hybridization at the interface. |
MoS2 can also serve as a spacer in a spin-valve device to exploit its semiconducting nature and its stable spin polarization in the out-of-plane direction.20,21 Magnetoresistance (MR) of 0.73% at 20 K and 0.23% at 240 K has been demonstrated in a NiFe/MoS2/NiFe structure.22 MR denotes the change of electrical resistance when two ferromagnetic (FM) electrodes sandwiching a MoS2 layer switch their magnetization alignment from parallel to anti-parallel. The confirmation of a spin-valve effect in a MoS2-based heterostructure is encouraging, but the large discrepancy between measurement (MR less than 1%)22 and prediction (MR as large as 300%)23 indicates that we have yet to identify all players relevant to the spin-dependent transport in TMD-based spin valves.
Direct investigation of the fundamental magnetic properties of FM–TMD heterojunctions is believed to be informative but remains scattered. A recent experimental study of a Fe/MoS2 heterojunction found that deposited Fe aggregates into nanoparticles with no sign of magnetic coupling to MoS2. Neither is there charge transfer between Fe and MoS2, according to measurements of their X-ray photoemission spectra (XPS).24,25 In contrast, Co/MoS2 was suggested to be different. According to a first-principles calculation reported by Garandel et al., the energetically favored Co–S bonding at the Co/MoS2 interface would lead to a spin imbalance on the MoS2 side.26 We thus expect Co/MoS2 to exhibit magnetic properties different from those of Fe/MoS2.
This work began with the deposition of ultrathin films of Co onto flakes of SiO2-supported monolayer MoS2,6 followed by an examination of the magnetic domain configuration with a photoemission electron microscope (PEEM)27 and chemical states with μ-XPS.28 Element-specific images revealed that, with Co covering both MoS2 and SiO2, the magnetic domains appeared only at the area of Co/MoS2. Furthermore, the direction of magnetization of Co domains seems to correlate with the underlying MoS2 lattice structure. μ-XPS measurements28 on Co/MoS2 disclose further that a small charge transfer, induced by the formation of a covalent bond between S and Co atoms, occurs after the deposition of Co. Our observation of magnetic domains with preferred directions of magnetization is solid proof of magnetic anisotropy at a Co/MoS2 (monolayer) heterojunction. The charge transfer identified with μ-XPS indicates that a high spin-injection efficiency at this particular interface might be possible.
Fig. 1 Monolayer of MoS2. (a) OM image of CVD-grown MoS2 islands on SiO2. The dashed square indicates a triangular MoS2 sample, which is also featured in Fig. 3. (b) The Raman spectrum of the sample shows two dominant Raman modes at 383 and 404 cm−1 of separation 21 cm−1. (c) The PL spectrum of the sample indicates that the photoluminescence peak energy is equal to 1.83 eV. These two spectra indicate that the MoS2 islands are monolayers. |
A schematic band structure of a typical FM 3d-transition metal is drawn in Fig. 2(a), in which a net magnetic moment originates from the asymmetric nature of the 3d band. According to the selection rules, an incident energetic photon beam with either right or left circular polarization (RCP or LCP) can excite electrons that fulfill the transition conditions: Δs = 0 and Δm = ±1. The probabilities of each pair of corresponding transitions from the occupied 2p core-level up to the empty 3d band are, however, unequal for the majority and minority spin channels because each channel has a different number of empty states available for a transition. A schematic diagram of the experimental setup appears in Fig. 2(b), which shows an incident circularly polarized beam illuminating the sample – a Co ultrathin film on MoS2 – at an angle of 25° of grazing incidence. The spatial distribution of the photo-emitted electrons under the X-ray magnetic-circular-dichroism (XMCD) effect30,31 is resolved with the PEEM, allowing the observation of magnetic domains in the Co layer. Fig. 2(c) shows an XMCD image recorded from a Co film (9 ML) deposited on monolayer MoS2. In that image, two regions of distinct contrast are labeled as regions A and B. From there, we extracted two micro-area spectra according to their intensity variation as a function of photon energy. As the difference of these two spectra in Fig. 2(d) shows a typical XMCD signature – opposite enhancement at Co L3 and L2 resonances – the contrast seen in Fig. 2(c) has indeed a magnetic origin.
Fig. 3(a)–(c) show XMCD images of Co films at thicknesses 5, 7, and 9 ML. After recording the first XMCD image from a Co film at thickness 5 ML, the images in succeeding sets were acquired after each subsequent deposition. After comparing the series of images, we concluded that the additional Co deposition would only enhance the magnetic contrast of the existing domains. The XMCD image can be conceived as an inner product of the sample magnetization () and the beam polarization (), i.e. I ∝ ·.33 Under the conditions of fixed incident angle and beam polarization, the XMCD image is effectively a measure of the spatial distribution of magnetization directions with respect to the polarization of the photon beam. In our study, the thickness of Co was assumed to be uniform on MoS2,34,35 and likewise the magnitude of its magnetization. The fact that we observed only several distinct contrasts in Fig. 3 indicates that the localized magnetizations are somehow oriented into selected directions. To explain the directions of magnetization clearly, we plotted the contrast levels of domain images into histograms. The greyscale shown at the bottom of the histogram indicates the range of contrast levels taken from images, whereas the intensity of the histogram represents the frequency of finding a specific contrast level from the image pixels. In the hypothetical case of out-of-plane magnetization, only two contrast levels would be possible, inward-pointing normal and outward-pointing normal. As all histograms in Fig. 3 display more than two contrast levels within a finite greyscale, we conclude that the magnetizations of Co domains not only lie either in-plane or canted but also have preferred orientations.
Our next task was to examine the relevance of the MoS2 crystalline structure for the domain formation. Fig. 4(a) and (b) display the domain images of a Co film (9 ML) grown on a single crystalline monolayer MoS2 flake and a polycrystalline monolayer MoS2 film, respectively.34 By controlling the degree of crystallinity in monolayer MoS2, we were able to adjust the lateral dimensions of the magnetic domains from tens of micrometers to sub-micrometers. In addition, as Fig. 4(b) shows a broader spread of greyscale, the Co domains on a polycrystalline surface are believed to be less aligned. Another observation worthy of mention concerns the domain boundaries on triangular MoS2 flakes. As displayed in Fig. 4(c), the paths of the domain boundary are not arbitrary but follow a particular crystalline axis of MoS2. Considering that CVD-grown triangular MoS2 flakes are typically treated as a single crystalline grain, and that the edge of the grain is in either an armchair or zig-zag configuration as illustrated in Fig. 4(d),36 it seems that well-defined crystallinity in the MoS2 layer would not only promote the magnetization alignment in a Co layer but also affect how the domains are divided. We emphasize here that the same behavior was found repeatedly on other flakes of MoS2 covered with Co.
Finally, we examined the electronic structure of Co/MoS2 (flake) with μ-XPS. Fig. 5 shows the μ-XPS recorded from the Co(4 ML)/SiO2 and the Co(4 ML)/MoS2 regions, respectively. As the Co layer was prepared in situ, under UHV conditions, the Co 2p core levels acquired in the two separate areas contain contributions mainly from Co(0) (2p3/2 = 778.3 eV),37 which corresponds to the metallic state of Co. Moreover, apart from a broad spectral feature corresponding to the Auger emission of S atoms in MoS2, there are noticeable differences between the two spectra displayed in Fig. 5. A fit of the Co 2p spectrum of Co/MoS2 (flake) produces two additional spectral features, namely Co(II) (2p3/2 = 781.5 eV) and Co(III) (2p3/2 = 779.7 eV).37,38 As Co(III) has an emission energy similar to that of Co oxidization found in Co/SiO2,39 we suggest that both Co(II) and Co(III) found on Co/MoS2 (flake) are the result of charge transfer from Co to MoS2. The spectra shown in Fig. 5 are inconsistent with the concept of chemical doping, which is generally accompanied by chemical shifts in XPS.40 Instead, the absence of an energy shift of all major emission peaks – Co(0), Mo 3d, and S 2p (Fig. S6, ESI†) – implies a small net extent of charge transfer.
Fig. 5 μ-XPS evidence of charge transfer in Co. Co 2p μ-XPS recorded for Co/SiO2 (upper) and Co/MoS2 (lower). |
Based on our experimental observations, we confirm that an ultrathin Co film deposited on monolayer MoS2 is able to form micrometer-sized ferromagnetic domains. Furthermore, the magnetization and the boundaries of domains have preferred directions or paths that are parallel to either the zig-zag or armchair directions of the MoS2 crystal structure. The presence of magnetic order and magnetic anisotropy in Co/MoS2 (flake) proves the possibility of using two-dimensional materials of monolayer thickness to evoke the anisotropy of a magnetic layer deposited thereon.
The crystal structure of Co offers the most straightforward explanation of the observed MA, because an ultrathin metallic film under a lattice strain can display properties different from those found in its pristine structure. An effective example to show what a lattice strain can do to the magnetic properties of a material is Co/GaAs(110); the Co layer stabilized on the body-center-cubic (BCC) structure displays an α-Fe-like behavior.41 However, as the low energy electron diffraction (LEED) results on Co/MoS2 (bulk crystal) failed to produce a sharp pattern, we suspect that Co/MoS2 (flake) might not be epitaxial in nature; in fact, the Co layer grown on the MoS2 surface seems amorphous (Fig. S7, ESI†). Although we lack the appropriate tool to determine the crystal structure of Co/MoS2 (flake), we know that the shape anisotropy can play no major role because of the non-comparable thickness between the Co layer and the area of the MoS2 flake (several nm vs. tens of μm2).
Readers might curious about the absence of magnetic domains in the part Co/SiO2. The uniformity of the Co layer could cause such a difference. However, with our Co deposition known to vary by at most a few percent within a diameter of 6 mm on the specimen and the confirmation of having the same Co L-edge intensity measured from Co/MoS2 and Co/SiO2 (Fig. S11, ESI†), we believe that the thickness of the Co layer is not responsible for the observed difference in domain configurations. Also, the AFM image (Fig. S12, ESI†) of a Pd protected Co film on the substrate shows that nanoclusters form on both MoS2 and SiO2. The absence of magnetic domains of Co/SiO2 means that the orientations of spin are in a high symmetry phase whose orientations have equal probabilities of pointing in every direction. This could be the reason why XMCD-PEEM cannot observe the magnetic domain, although Co is a ferromagnetic material. In contrast, the presence of magnetic domains is a signature of symmetry lowering in which spins are grouped and oriented into selected directions.
Regarding the XPS evidence of charge transfer from Co to MoS2, we find it consistent with the theoretical work of Garandel et al. that indicates that Co atoms at the interface bond covalently with the topmost S atoms of MoS2.26 According to that reported work,26 the charge transfer at Co/MoS2 would result in a spin-polarized metallic interface and facilitate efficient spin injection. Another reason motivating us to look into the electronic structures of Co/MoS2 and Co/SiO2 is the distinct behavior recently reported in Fe/MoS2 (Fig. S8 (ESI†), and ref. 24); neither magnetic domains nor charge transfer between Fe and MoS2 was found.
The MoS2 samples grown on the SiO2(90 nm)/Si(001) substrates were loaded into TLS endstation 05B2 and heated at 150 °C for 6 h under ultra-high vacuum (UHV) conditions for outgassing. The Co layers were grown with a commercial EFM3; their thickness was defined in units of monolayer (ML). The rate of Co deposition was calibrated with the medium-energy electron-diffraction (MEED) oscillation recorded during the deposition of Co onto a Cu(001) single crystal. Note that it is necessary to check the structural stability of a 2D layer in a heterojunction. For that, we examine the micro-area XPS spectra taken from Co/MoS2/SiO2 (Fig. S6, ESI†). Compared to previous publications with ones acquired from CVD growth of MoS2, ref. 29, and Co-doped MoS2, ref. 40, the molybdenum Mo 3d and sulfur S 2p spectra reported in this work are in great resemblance to the ones acquired from a single layer of MoS2. We conclude that the chemical bond of monolayer MoS2 is preserved after Co deposition. Our argument above can be strengthened further by the nature of the strong covalent bonds within MoS2, whose melting point is 2375 °C. As the Co deposition (whose melting point is 1495 °C) is accomplished via an EFM3 evaporator at a slow deposition rate (## ML per minute), we believe the way we deposited Co is not likely to damage the atomic structure of MoS2.
After the Co deposition, the sample was transferred into the PEEM chamber27 under UHV conditions for XMCD imaging. We enhanced the domain images (or XMCD images) by the absorption asymmetry, IA = (IL3 − IL2)/(IL3 + IL2); that is, each XMCD image (IA) was actually an asymmetric superposition of two images, IL2 and IL3, recorded at L2 and L3, respectively. As this image processing can eliminate geometric inhomogeneity, the contrast in the IA image served to recognize the magnetic domains of varied magnetization directions.32,33 The exact magnetization direction of the domains could not be determined as it was not possible to rotate the sample in situ or to apply an external magnetic field in the PEEM experimental chamber. The system is hence sensitive only to whether the magnetic domains are parallel (brighter) or antiparallel (darker) with respect to the incident light. The XMCD images were leveled by subtracting the polynomial background, which is supposedly a consequence of a shift of the beam spot caused by a change in the photon energy. The micro-area X-ray absorption spectra (μ-XAS) recorded on A and B domains (displayed in Fig. 2(d)) were extracted from a stack of images recorded at a step of 0.2 eV from hν = 750 eV to 810 eV, across the Co L2,3 edges.
The μ-XPS measurements were performed at beamline 1.2L NanoESCA at the Elettra Synchrotron (Trieste, Italy).28 Before the deposition of cobalt, the MoS2/SiO2 samples were outgassed at 250 °C in a UHV preparation chamber for 2 h. Co 2p spectra were recorded using an s-polarized photon energy of 950 eV. Because of the decreased beam spot size on the sample (<20 μm) it was possible to record μ-XPS selectively from the Co/MoS2 flakes and Co/SiO2 regions.
Footnote |
† Electronic supplementary information (ESI) available. See DOI: 10.1039/d0nh00108b |
This journal is © The Royal Society of Chemistry 2020 |