Edge-spin-derived magnetism in few-layer MoS2 nanomeshes

Magnetism arising from edge spins is highly interesting, particularly in 2D atomically thin materials in which the influence of edges becomes more significant. Among such materials, molybdenum disulfide (MoS2; one of the transition metal dichalcogenide (TMD) family) is attracting significant attention. The causes for magnetism observed in the TMD family, including in MoS2, have been discussed by considering various aspects, such as pure zigzag atomic-structure edges, grain boundaries, and vacancies. Here, we report the observation of ferromagnetism (FM) in few-layer MoS2 nanomeshes (NMs; honeycomb-like array of hexagonal nanopores with low-contamination and low-defect pore edges), which have been created by a specific non-lithographic method. We confirm robust FM arising from pore edges in oxygen(O)-terminated MoS2-NMs at room temperature, while it disappears in hydrogen(H)-terminated samples. The observed high-sensitivity of FM to NM structures and critical annealing temperatures suggest a possibility that the Mo-atom dangling bond in pore edge is a dominant factor for the FM.

TMD family), 3 and to few-layer hexagonal boron-nitride (hBN; an insulator (with a wide band gap of 5.92 eV) composed of a hexagonal network with an equal number of alternating B and N atoms in sp 2 hybridization that are bonded covalently). 4 We revealed that the O-terminated few-layer zigzag BP-NMs produce evident room-temperature FM with a magnitude approximately 100 times/area larger than that reported for the H-GNMs at room temperature. 3 The O-terminated zigzag-hBN-NM also exhibited similar room-temperature FM with a much smaller amplitude. 4   Here, atom-thin MoS2 is attracting significant attention because of its direct bandgap of 1.5−1.8 eV in the 2H semiconducting phase. 5 We reported an atomically thin MoS2 lateral Schottky junction created only by electron beam irradiation that caused (2H-semiconductor)-to-(1T-metal) phase transition, 6 and its photoresponse. 7 Various phenomena and applications of atomically thin TMD family, and also their van der Waals (vdW) heterostructure with other 2D atomically thin layers have been reported. Magnetism derived from edge spins has been also reported in TMD family (e.g., MoS2, MoSe2, WS2). [8][9][10][11][12][13][14][15][16] The causes for magnetism have been discussed by considering various aspects, such as zigzag edges, 8-10 grain boundaries, 11 vacancies, 12 ion irradiation, 17 and Mo-doping. 18 However, no consensus has been reached regarding these. Each layer of TMD family typically has a thickness of 6-7 Å, which consists of a hexagonally packed layer of metal atoms sandwiched between two layers of chalcogen atoms ( Fig. 1(a)(b)). 5 Therefore, two different kinds of dangling bonds at zigzag edges exist over the two atom layers within the AB stacking structure; e.g., one zigzag edge of a NR has a Mo dangling bond (right red circle of Fig. 1(a)), while the other has a S dangling bond (left blue circle of Fig. 1(a)). In this sense, the termination of these different edge dangling bonds by foreign atoms is analogous to the case of the edge dangling bonds of hBN-NR, in which only the edge O-N bond yielded FM, as mentioned above. Thus, it is interesting to clarify which dangling bonds in the zigzag edges of MoS2 are a dominant factor for FM. Raman spectrum of the bulk flake (red curve) and nanomesh right after pore formation without annealing (black-dot curve) and right after 700 Cand 800 C-annealing (green and blue dot curves, respectively). Inset: Optical micro scope image of a measured MoS2 flake. White spot means the part irradiated by laser beam. (d) FESEM top-view images of a MoS2-NM formed on the sample in (a) with difference observation scales. The mean pore diameter () and interpore distance (w) are 80nm and 10nm, respectively. (e) Typical XPS spectra of a ferromagnetic MoS2-NM ( Fig.  3(a)-sample) for binding energy of orbitals of Ni2p, Co2p, and Fe2p. Diameter of the irradiated X ray is 100 m. The upper limit for impurity detection of the XPS is 0.05 at%.
In this work, flakes of few-layer MoS2 are mechanically exfoliated on SiO2/Si substrate from bulk MoS2 (Smart Element Co.) in a glove box using the (non-magnetic) Scotch tape method and confirmed using optical and atomic force microscopies ( Fig. 2(a)(b)) and Raman spectroscopy ( Fig.   2(c)). Both the in-plane E 1 2g (382 cm -1 ) and the out-of-plane A1g (487 cm -1 ) vibration modes are confirmed for bulk sample in Fig. 2(c). A thickness of 3 nm is also confirmed in Fig. 2(b). These suggest that the bulk MoS2 flake is at least five-layer thick. Similar flakes are collected for magnetization measurement so as to satisfy the total area of 1cm 2 . Following the abovementioned non-lithographic method (Supplementary file), 1-4 few-layer MoS2-NMs are fabricated and confirmed by field-emission SEM (FESEM) (Fig. 2(d)) and atomic force microscopy. The absence of magnetic impurities (e.g., Ni, Co, and Fe) is also confirmed by X-ray Photoelectron Spectroscopy ( Fig. 2(e)).
All the NM samples are annealed at high temperatures for 30 min under high vacuum. Raman spectrum are shown in Fig. 2(c). The peaks become broad and full width of half maximum (FWHM) increases from bulk (e.g. A1g 7 cm -1 ) to pore formation without annealing (A1g 20 cm -1 ). After the 700 C annealing, the peak becomes sharp again and FWHM decreases to A1g 12 cm -1 , while it becomes broader and FWHM increases to A1g 15 cm -1 after the 800 C annealing. As an interpretation, these results mean that the quality of sample degrades by forming NM structure by Ar ion etching, while it recovers by the 700 C annealing. On the other hand, higher T annealing (i.e., the 800 C annealing) reduces the sample quality again. Another interpretation should be associated with one dimensional (1D) structure (i.e., NR) of the interpore region of NM, because, in the case of graphene NM, larger D-peak height corresponded to the appearance of arm chair edge in 1D GNR (i.e., the interpore region of graphene NM). Three peak heights do not reach to those of bulk sample because the area of remained interpore regions in NM is very small. Moreover, it suggest absence of defects because there are no peaks other than E 1 2g and A1g peaks, unlike reported by ref. 19. After the annealing, the NMs are exposed to air for the O-termination of pore edges, while they are annealed at a high temperature in a H atmosphere for 30 min for the H-termination.    Fig. 3(a) at H = 10 Oe. Magnetization monotonically decreases with increasing temperature up to 300 K, which is the upper limit temperature of our SQUID. Hence, the Currie temperature of this sample is estimated to be above 300 K. It is also interesting to note that the magnetization value slightly increases below 10 K with decreasing temperature.   Fig. 4(c) exhibits Ms as a function of the annealing temperature. It implies that the annealing temperature has an optimum value to yield a large Ms (i.e., 700 °C). We have defined Tan as the highest annealing temperature at which the NM structure is maintained even after annealing. The NM structure disappears due to pore-edge atomic reconstruction by annealing at temperatures slightly above Tan (20 °C), because the interpore regions are NRs with widths as low as 10-20 nm (Supplementary file). The amplitudes of FMs observed in H-terminated GNMs 1 and O-terminated BP-NMs 3 actually showed the largest values, when the samples were annealed at this Tan. In contrast, the present MoS2-NM structure disappears after annealing at 820 °C and thus Tan should be 800 °C, whereas the largest Ms is obtained for annealing at 700 °C in Fig. 4(c). This is in contrast to the cases for GNMs and BP-NMs, while it is qualitatively similar to those of hBN-NMs, in which Tan was 600 °C but the annealing temperature for the largest Ms was 500 °C. 4 As mentioned above, magnetism in atomically thin MoS2 layer has been discussed by considering various aspects, such as pure zigzag edges, 8 grain boundaries (e.g., composed of pentagon-heptagon pairs), 11 and vacancies (e.g., arising from 2H/1T-phase combination, 12 with impurity-doping).
However, this does not correspond to the present case, because the samples without annealing, which should have the largest amount of as-grown defects or boundaries, did not show FM.
Moreover, if irregular boundaries and defects are the cause of FM, the exact structure dependence (Figs. 4(a) and (b)) should not appear. Raman spectrum (Fig. 2(c)) support absence of such defects in NMs after the 700 °C annealing. Therefore, we interpret that the contribution of the possible pure zigzag pore edges might be dominant for the present FMs.
We discuss about the relationship between the observed FM and the zigzag edges at pores. Li et al. 8 calculated that pure zigzag-edged MoS2 NRs without termination by foreign atoms provided a stable ferromagnetic ground state with an energy difference ΔE between their spin-unpolarized and spin-polarized total energies. Both ΔE and the total magnetic moment increased with increasing NR width and thickness (i.e., number of layers). For example, the largest-width MoS2 NR had the highest ΔE (57.30 meV) and a total magnetic moment as large as 0.879 μB. Moreover, the unpaired spin mainly concentrated on the edge Mo and S atoms, while the inner Mo atoms less contributed to the unpaired spin. Densities of the unpaired spins in Mo atom dangling bonds in zigzag edges (e.g., right red circle of Fig. 1(a)) were much larger than those at S atoms (e.g., left blue circle of Fig. 1(a)).
On the other hand, the H-terminated zigzag MoS2 NRs (the edge S and Mo atoms are saturated with one and two H atoms, respectively) showed weaker magnetism.
These theories qualitatively and partially agree with our observation as follows. As mentioned above, interpore regions correspond to NRs in our NM structure and the observed FM slightly increases with increasing W (i.e., NR width; Fig. 4(a)) and d (Fig. 4(b)). Moreover, the FM amplitude decreases in H-terminated MoS2-NMs (Fig.3(b)). Here, Tan ( 800 °C) and the annealing temperature to obtain the largest Ms (700 °C) are different, as mentioned for Fig. 4(c). This can be interpreted that the annealing temperature to form the most stable pore edges with Mo dangling bonds is 700 °C, which results in the densities of the unpaired spins at Mo dangling bonds and the observed largest Ms (ref.8). As discussed above, annealing at Tan results in the formation of assumed zigzag pore edges through edge atomic reconstruction in our NM structure. This is because NMs disappear by annealing over Tan and zigzag edge is the most stable from the chemical and thermal viewpoints after the edge atomic reconstruction caused by annealing at Tan. In both GNM and BP-NM, this Tan and annealing temperature to obtain the largest FM are the same, because these NMs consist of a single-kind atom (i.e., carbon and black phosphorus, respectively).
In contrast, few-layer hBN-NMs exhibited a relationship with the annealing temperature similar to that of the present few-layer MoS2-NMs. As a common structure, MoS2 and hBN consist of two-different kind atoms (i.e., Mo and S, and B and N, respectively). Consequently, in hBN-NM, there were two different kinds of edge dangling bonds (i.e., B and N atoms) and only the edge O-N bond theoretically yielded FM due to the evident spin splitting in its px and pz orbitals. 4 Thus, we interpreted that the pore edges of hBN-NM became N dangling-bond rich after annealing at the temperature for the largest Ms (500 °C), while the pore edges became disorder-dominant with non-zigzag by after annealing at Tc (600 °C) and FM disappeared. This is because N dangling bonds had a binding energy larger than that for B dangling bonds and they could become sufficiently stable after annealing at temperature 500 °C through the edge-atomic reconstruction. After annealing at much higher temperature (600 °C), even these pore edges with N dangling bond were destroyed, while interpore bulk-parts of hBN-NMs still remained due to the strong B-N bonds.
This mechanism should be analogous to the present MoS2-NMs. As calculated by Li et al., 8

edge
Mo dangling bonds exhibit an unpaired spin density higher than that of S dangling bonds. Hence, it is speculated that the pore edges of MoS2-NM become Mo dangling bonds rich after the annealing at temperature for the largest Ms (700 °C), while the pore edges become disorder-dominant with non-zigzag after annealing at Tc (800 °C) but keeping the interpore bulk-parts of NM structure and FM disappears. This is also because Mo dangling bonds have a binding energy (232 eV for 3d3/2 and 229 eV for 3d5/2) larger than that of S dangling bonds (164 eV for 2p1/2 and 162 eV for 2p3/2). 6 In the O-terminated hBN-NMs, the evident spin splitting in px and pz orbitals of the edge O-N bond led to the FM. In the present case, Mo dangling bonds have unsaturated bonds in core 3d3/2 and 3d5/2 orbitals at least. O-termination of these unsaturated bonds should lead to the significant spin splitting and the observed FM. With this viewpoint, the theoretical confirmation of the mechanism of ferromagnetic spin coupling at the edge O-Mo bonds is indispensable.
The ferromagnetic spin coupling at the edge O-Mo bonds has a weaker vdW-interlayer interaction as explained by Li et al. 8 because of the AB stacking of the puckered honeycomb lattice.
Thus, the Ms value monotonically increases with increasing d as shown in Fig. 4(b). Moreover, since the oxidation of pore edges is easily obtained only by exposing MoS2-NMs to air, all dangling bonds of pore edges in each NM can be fully O-terminated and can easily provide a robust FM even at room temperature. Consequently, the present finding of edge-spin-derived room-temperature FM must enable the use of atomically thin (or few-layer) MoS2 in unique magnetic and spintronic applications in addition to well-established electronic and photonic applications.