Atomic-Scale Interfacial Magnetism in Fe/Graphene Heterojunction

Successful spin injection into graphene makes it a competitive contender in the race to become a key material for quantum computation, or the spin-operation-based data processing and sensing. Engineering ferromagnetic metal (FM)/graphene heterojunctions is one of the most promising avenues to realise it, however, their interface magnetism remains an open question up to this day. In any proposed FM/graphene spintronic devices, the best opportunity for spin transport could only be achieved where no magnetic dead layer exists at the FM/graphene interface. Here we present a comprehensive study of the epitaxial Fe/graphene interface by means of X-ray magnetic circular dichroism (XMCD) and density functional theory (DFT) calculations. The experiment has been performed using a specially designed FM1/FM2/graphene structure that to a large extent restores the realistic case of the proposed graphene-based transistors. We have quantitatively observed a reduced but still sizable magnetic moments of the epitaxial Fe ML on graphene, which is well resembled by simulations and can be attributed to the strong hybridization between the Fe 3dz2 and the C 2pz orbitals and the sp-orbital-like behavior of the Fe 3d electrons due to the presence of graphene.

Scientific RepoRts | 5:11911 | DOi: 10.1038/srep11911 In any proposed graphene-based transistors, the best opportunity for spin transport could only be achieved when no magnetic dead layer exists at the FM/graphene interface. Previous studies on various FM/semiconductor (FM/SC) heterojunctions revealed the possibility that the magnetic ordering near a region of the surface or interface of FM/SC may be modified due to interdiffusion, termination and hybridization; and controversial reports make this issue rather complex [15][16][17][18][19][20][21] . Calculations for transition-metal/nanotube hybrid structures exhibit substantial magnetism 17 . For Fe-, Co-, and Ni-doped carbon nanotubes, the interactions are found ferromagnetic for Fe and Co while nonmagnetic for Ni 18 . A ~1.2 nm magnetic dead layer of Co was observed on a topological insulator surface 19 . Whether a deposited FM on graphene is magnetically ordered at the FM/graphene interface is a must-addressed issue before an efficient graphene-based transistor can be developed. In this Letter, we present a comprehensive XMCD study of the ML epitaxial Fe/graphene interface, combined with DFT calculations. The experiments have been performed using a specially designed FM 1 /FM 2 /SC structure that to a large extent simulates the realistic FM/graphene interface of the proposed graphene-based transistors [4][5][6] and at the same time allows a direct determination of the interface magnetism of FM/graphene. Fundamentally all the intriguing spintronic phenomena observed in the FM/graphene heterojunctions strongly depend upon the interfacial hybridization and magnetic exchange interaction [8][9][10][11][12][13][14] . A direct demonstration of the magnetic and electronic state of the FM/graphene interface down to ML scale remains a nontrivial task, even today, partially due to the inaccessibility of the buried layer between the topmost atoms and substrate. On the one hand, for samples comprising of several nanometers thick FM atop the SC substrate, it is always hard to separate the contributions of the interface and the bulk magnetization. On the other hand, the low coverage of FM (in the form of atoms and clusters) will be paramagnetic or ferromagnetic with extremely low Curie temperatures (T c ) 22,23 , and consequently no longer be representative of a realistic device. Moreover, while the FM atoms reduce to a minute amount, many global detection techniques become invalid, as the experimental conditions like vacuum, sensitivity, cryogen etc. must be simultaneously satisfied at a high level. To overcome these obstacles, we employed a unique FM 1 /FM 2 /SC structure (in this letter, FM 1 = 30 MLs Ni, FM 2 = 1 ML Fe, and SC = graphene). The thick FM 1 layer provides the FM 2 ML with a source of exchange interaction and these two together restore the interfacial behavior of the thick ferromagnetic FM 2 /SC. Combined with the unique elemental selectivity of XMCD, such structure allows direct observation of the magnetization of FM 2 at the FM 2 / SC interface.

Results and discussions
XMCD measurements. The XAS and XMCD of the Fe and Ni L 2,3 absorption edges were performed on beamline I10 at the Diamond Light Source, UK. Circularly polarized X-rays with 100% degree of polarization were used in normal incidence with respect to the sample plane and parallel with the applied magnetic field, as shown in Fig. 1, in order to minimize the nonmagnetic asymmetries and saturation effects. The XAS spectra were obtained using both total electron yield (TEY) and total florescence yield (TFY) detection simultaneously. The XMCD was taken as the difference of the XAS spectra, i.e., σ − − σ + , obtained by flipping the X-ray helicity at a fixed magnetic field of 3 T, under which the sample is fully magnetized with little paramagnetic contribution. The X-ray absorption (XAS) and XMCD of the Fe and Ni L 2,3 absorption edges were performed and typical spectra obtained at 5-300 K are presented in Fig. 2a,b, respectively. The XAS spectra of the interface Fe and the stabilizing layer Ni for both left-and right-circularly polarized X-rays show a white line at each spin-orbit split core level without prominent splitting, suggesting that the sample has been well protected from oxidation. The spin (m spin ) and orbital magnetic moment (m orb ) were calculated by applying the sum rules 24  where the effective number of 3d-band holes, n h , was assumed as 3.7 accounting for the charge transfer from Fe to the atop stabilizing Ni layer (n h = 3.4 in bulk) 25 . In order to exclude non-magnetic parts of the XAS spectra, a sigmoidal function is used to fit the threshold 26,27 . As can be seen from Fig. 2a, both integrations (the dashed lines) of XMCD and summed XAS spectra become flat within the selected range, proving a proper background offset. The calculated m spin , m orb , and total magnetic moment (m total ) of the interface Fe at temperatures from 5 K to 300 K were plotted in Fig. 3 (left). The m total of Fe displays a decreasing trend with the increasing temperature from 1.23 μ B /atom at 5 K to 1.12 μ B / atom at 300 K, i.e., ~9% reduction, pointing to a T C close to bulk-like Fe. At the lowest temperature (5 K), we obtained m spin = (1.06 ± 0.1) μ B /atom, which is ~50% reduced from the bulk-like Fe (whose m spin = 2.2 μ B /atom) and m orb = (0.18 ± 0.02) μ B /atom, corresponding to an enhancement of ~200% compared to m orb = 0.085 μ B /atom of that in the bulk 28 . The enhancement of m orb of Fe can be attributed to the symmetry breaking of the ultrathin films, in which the electrons are rather localized around the nucleus, leading to an orbital degeneracy lifting as that reported in Fe/GaAs 20 and Fe/InAs 21 . A further factor of influence is the modified chemical environment of Fe due to the presence of graphene, as intensive spin and charge transfer can occur at the FM/graphene interface 25,29-31 . The stacking of the topmost FM with respect to graphene is another noteworthy question 32 . Within the context of the occupation sites, the best agreement between experiment   The calculated m spin , m orb , and m total of the Ni stabilizing layer (see Fig. 3 (right)) are in good agreement with the previous reports of metallic Ni 33 . The sum-rules derived m total of Ni exhibits a trend of slight decrease with the increasing temperature from (0.63 ± 0.06) μ B /atom at 5 K to (0.52 ± 0.06) μ B /atom at 300 K, pointing to a T C close to the bulk-like Ni 34 . Unlike the interfacial Fe, whose m orb shows a significant enhancement compared to that of bulk-like Fe, the Ni stabilizing layer shows a nearly quenched m orb . This is expected for bulk-like Ni, where the effect of the crystal field plays a dominant role. As an additional note, the graphene is likely to carry induced magnetic moment at the Fe/graphene interface 25,[29][30][31] . However this magnetization is not sizable enough to quantitatively demonstrate with XMCD owing to the small crossection of C atoms.  Figure S4 presents the initial (upper row) and the relaxed (lower row) geometries of Fe fcc(111), bcc(100) and bcc(110) stacking on graphene, respectively. The calculations suggest that the Fe prefers to follow the fcc(111)-like structure of the graphene substrate. Significant deformations were observed for both Fe bcc(110) and (100) on graphene due to the large lattice mismatch. The averaged Fe-Fe bond length after relaxation changes from 2.73 to 2.35 Å for Fe bcc(110), and from 2.91 to 2.38 Å for Fe bcc(100) ML, respectively, whilst that for Fe fcc(111) remains unchanged from 2.47 Å.
It was found that the Fe atoms deposited on graphene prefer to follow fcc(111) stacking and there exist three inequivalent positions of the fcc Fe allocations on graphene as presented in Fig. 4a, namely, top (blue), hollow (red) and bridge (green), among which the top configuration is most energetically favorable with an averaged equilibrium distance of 1.90 Å. The calculated magnetic moments of Fe for the three different configurations as well as those derived from the experimental measurements were gathered in Table 1. Consistent with the experiment, the theoretical simulation also reveals a ~50% reduction in the magnetic moment of Fe. Although Fe top was found the most stable geometry on graphene according to the calculation, the observed numeric results from XMCD, i.e., m spin = (1.06 ± 0.1) μ B /atom, is more likely a mixture of Fe top (m spin = 1.23 μ B /atom) and Fe bridge (m spin = 0.67 μ B /atom), given the small energy difference (∆E = 43 meV) of their calculated total energies.
The DFT derived spin-resolved band structures for a freestanding Fe ML (left column) and the ML Fe top /graphene (right column), respectively, together with their corresponding partial density of states (DOSs) are presented in Fig. 4b. Remarkable exchange splitting (2.67 eV) for the spin-up and spin-down electrons of the freestanding Fe were observed from the partial DOSs and were traced back to those of the Fe 3d z2 orbitals (see the white shed areas), which is responsible for the large m spin (2.76 μ B /atom). In contract, the Fe 3d z2 orbitals in ML Fe top /graphene become strongly delocalized in the reciprocal space and tend to show an sp-orbital-like partial band character. The exchange splitting sharply reduces from 2.67 eV to 0.94 eV due to the presence of graphene, pointing to a weakened exchange interaction between the 3d electrons of the Fe in Fe top /graphene and consequently a reduced spin polarization near the Fermi level (E F ). Further examinations of the symmetry matching and spatial overlap suggest that the strong Fe-C hybridization predominantly originates from that between the C1 p z orbitals and the Fe 3d z2 orbitals, although the interactions between the C2 p z orbitals and the Fe 3d xz and 3d yz orbitals cannot be neglected (here, C1 stands for the C atom located underneath the Fe atom in Fe ML, while C2 is located at the hollow site, as shown in Fig. 4a). Figure 5 presents the differential electronic and spin density map of the Fe top /graphene. The 3D charge accumulation and depletion was obtained by subtracting the charge density of Fe top /graphene from that of the freestanding Fe ML and graphene as shown in Fig. 5a. The Bader charge analysis suggests that by average 0.13 e is transferred from Fe to C atoms per unit cell. This substantial charge transfer mainly occurs at the interfacial region of the Fe top /graphene and predominately via the C1 p z orbitals. The Fe-C hybridization, in turn, slightly polarizes the C atoms in Fe top /graphene in opposite direction to the Fe moments (m spin = − 0.02 μ B /atom for the C1 and − 0.01 μ B /atom for the C2 atoms) as can be seen from  Fig. 5b, and consequently the Dirac point of intact graphene is destroyed. Further details of the induced magnetization of the C atoms can be found in the Supplementary Materials. Moreover, the spin and charge transfer of the metastable Fe bridge /graphene (not shown) are qualitatively same but quantitatively different as that of Fe top /graphene (not shown) and m spin of − 0.008 μ B /atom are induced for both the C1 and C2 atoms.
As an additional note, the effect of the topmost Ni was also considered. The bulk-like Ni atop the Fe ML serves as a stabilizing layer, in order to make the ML Fe in such configuration to be representative of the interface Fe of a bulk-like Fe on graphene. The two FM layers are expected to ferromagnetically couple with each other, provided an atomically clean interface is prepared, e.g., using UHV growth techniques. The calculation suggests that the stabilizing Ni layer has a limited influence on the underneath Fe by offering only a small charge transfer, i.e. 0.017 e/atom. In the fully relaxed structure, the Ni tends to pull the Fe slightly away from the graphene substrate, resulting in a higher m spin of 1.53 μ B /atom for Fe. The inclusion of the topmost 7 MLs Ni has not modified the favorable stacking of the ML Fe in conjunction with the graphene. Furthermore, the calculated average m spin of the 7 ML Ni atop Fe is 0.59 μ B / atom, well reproducing the experimental value of m spin = (0.59 ± 0.06) μ B /atom obtained from XMCD.

Conclusions
To summarize, we have demonstrated an unambiguous ferromagnetic epitaxial Fe/graphene interface by XMCD. This was aided by the inclusion of a Ni stabilizing layer atop the Fe film, which facilitated the study of the temperature dependence of the Fe/graphene interface. We observed strong dichroic XAS spectra of Fe, suggesting a well-maintained ferromagnetic order of the Fe top /graphene interface up to room temperatures. We obtained a reduced but still sizable magnetic moment of the ML Fe on graphene, i.e., (1.23 ± 0.1) μ B /atom, which is well resembled by the DFT calculations. The observed suppression of the Fe magnetization is attributed to the strong hybridization between the Fe 3d z2 and the C 2p z orbitals, leading to greatly reduced exchange splitting, and the sp-orbital-like behavior of the Fe 3d electrons due to the presence of graphene. Our study restores a realistic scenario of the proposed graphene-based transistors and addresses the open question of the magnetic and electronic nature of the Fe/graphene interface. Although possibilities such as the discontinuous growth of the ML Fe and the contact of Fe with the SiO 2 substrate cannot be fully excluded from the experiment, our study has demonstrated a valuable model of exploring the interfacial magnetism of FM on graphene and the results are encouraging based on the contemporary sample preparation technique. Future work to explore the tuning of the spin polarized band structure of both the FM and graphene via the interface engineering will be of great interest and have strong implications for both fundamental physics and the emerging spintronics technology.

Methods
The single layer graphene used in this study was prepared by chemical vapor deposition (CVD) on Cu foil and then transferred on top of 300 nm SiO 2 /Si substrate 35,36 . After 1 hour of annealing at 200 °C, 1 ML Fe was grown on the graphene by molecular beam epitaxy (MBE) using an e-beam evaporator, whose rate was monitored by a quartz microbalance and calibrated by ex-situ atomic force microscope (AFM). 30 ML Ni was then deposited onto the Fe and the Ni/Fe/graphene was finally capped by 15 ML Cr for easy transport to the synchrotron facility. Raman scattering measurements were performed on the FM/ graphene and an as-grown area of the sample without FM deposition. From both part the character spectra of graphene were obtained, suggesting that the structure of graphene was well maintained after the transfer 36 , annealing, and the FM deposition processes. Further details of the Raman spectroscopy experiment can be found in the supplementary materials.
The first-principle calculations were performed using the projector augmented wave methods implemented in the Vienna ab initio simulation package (VASP) 37,38 . The electron exchange and correlation interactions were described by the local density approximation (LDA) 39 . The plane wave kinetic energy cutoff was set to be 520 eV. A Monkhorst-Pack mesh 40 of 31 × 31 × 1 k-points was used to sample the two-dimensional Brillouin zone for the thin film electronic structure calculations. The total energies were obtained using the tetrahedron method with the Blöchl corrections. In each unit cell, all atomic positions are fully relaxed with the conjugate gradient procedure until the residual forces vanished within 0.02 eV/Å. The vacuum space of 15 Å was set to separate the interactions between neighboring slabs of the unit cells. Test calculations have been performed for the bulk bcc Fe (a 0 = 2.87 Å) and the freestanding fcc monolayer Fe (a 0 = 2.55 Å) using this method and the results compare well with the pioneering reports 25,[29][30][31][41][42][43] , proving the validity of the selected computational method 44,45 .