Atomic-scale control of graphene magnetism by using hydrogen atoms

See allHide authors and affiliations

Science  22 Apr 2016:
Vol. 352, Issue 6284, pp. 437-441
DOI: 10.1126/science.aad8038

Hydrogen atom makes graphene magnetic

Graphene has many extraordinary mechanical and electronic properties, but it's not magnetic. To make it so, the simplest strategy is to modify its electronic structure to create unpaired electrons. Researchers can do that by, for example, removing individual carbon atoms or adsorbing hydrogen onto graphene. This has to be done in a very controlled way because of a peculiarity of the graphene's crystal lattice, which consists of two sublattices. Gonzales-Herrero et al. deposited a single hydrogen atom on top of graphene and used scanning tunneling microscopy to detect magnetism on the sublattice lacking the deposited atom (see the Perspective by Hollen and Gupta).

Science, this issue p. 437; see also p. 415


Isolated hydrogen atoms absorbed on graphene are predicted to induce magnetic moments. Here we demonstrate that the adsorption of a single hydrogen atom on graphene induces a magnetic moment characterized by a ~20–millielectron volt spin-split state at the Fermi energy. Our scanning tunneling microscopy (STM) experiments, complemented by first-principles calculations, show that such a spin-polarized state is essentially localized on the carbon sublattice opposite to the one where the hydrogen atom is chemisorbed. This atomically modulated spin texture, which extends several nanometers away from the hydrogen atom, drives the direct coupling between the magnetic moments at unusually long distances. By using the STM tip to manipulate hydrogen atoms with atomic precision, it is possible to tailor the magnetism of selected graphene regions.

Adding magnetism to the long list of graphene’s capabilities has been pursued since this material was first isolated (1). From a theoretical point of view, magnetic moments in graphene can be induced by removing a single pz orbital from the π-graphene system; this removal creates a single π-state at the Fermi energy (EF) around the missing orbital. The double occupation of this state by two electrons with different spins is forbidden by the electrostatic Coulomb repulsion; namely, once an electron occupies the state, a second one with opposite spin needs to “pay” an extra energy U. This leaves a single electron occupying the state and therefore a net magnetic moment (26). The strength of U, which determines the spin splitting, depends on the spatial localization of the state, because this defines the proximity of the electrons (Fig. 1A). In contrast to magnetic moments of a strongly localized atomic character that are commonly found in magnetic materials, these induced moments are predicted to extend over several nanometers, suggesting a strong direct coupling between them at unusually long distances. The coupling rules between the induced magnetic moments are also expected to be simple. Because of the bipartite atomic structure of graphene—which consists of two equivalent triangular sublattices, labeled A and B—and according to Lieb's theorem (7), the ground state of the system possesses a total spin given by S = 1/2 × |NANB|, where NA and NB are the number of pz orbitals removed from each sublattice (4, 8, 9). Thus, to generate a net magnetic moment in a particular graphene region, a different number of pz orbitals from each sublattice needs to be locally removed.

Fig. 1 Spin-split state induced by atomic H on graphene.

(A) Illustration of the origin of the spin-split state in terms of its spatial extension (r) given by the square of its wave function ψ2, and the corresponding electrostatic Coulomb repulsion U. Arrows indicate the energy position of spin-up (n) and spin-down (n) levels. For a fully polarized one-electron state, the majority level spin is filled and the minority one is empty; therefore, n= 1 and n= 0, and the energy splitting (EE) is given by U [section 3 in (27)]. (B) STM topography of a single H atom chemisorbed on neutral graphene (0.2 V, 0.1 nA, 7 × 7 nm2). (C) dI/dV spectrum measured on the H atom, showing a fully polarized peak at EF, andmeasured on bare graphene far from the H atom (a.u., arbitrary units). The spectra were acquired at a nominal junction impedance of 2 gigaohms (–100 mV, 50 pA). (D and E) DFT-simulated STM image (D) and DOS (E) of an H atom chemisorbed on neutral graphene. (F) dI/dV spectra and DFT calculation of the DOS induced by a single H atom on n- (left) and p-type (right) doped graphene. The minimum of the dI/dV spectra, acquired on bare graphene (black curves), determines the position of ED. The spectra were acquired at a nominal junction impedance of 8 gigaohms (–400 mV, 50 pA). STM data were acquired and processed using the WSxM software (36). Calculations for all simulated images were performed at the same energy as in the corresponding experimental conditions. All experimental data were acquired at 5 K.

Many theoretical proposals have been put forward on this subject, involving zigzag edges, graphene clusters, grain boundaries, and atomic defects (2, 4, 5, 811). Experimentally, the removal of pz orbitals from the π system has been achieved by randomly creating atomic vacancies or adsorbing adatoms (1216). However, removing those pz orbitals in a controlled manner has turned out to be challenging. In this work, we relied on the simplest (albeit demanding) experimental approach to remove a single pz orbital from the graphene network by means of the adsorption of a single H atom. Atomic H chemisorbs on graphene on top of carbon atoms, changing the initial sp2 hybridization of carbon to essentially sp3 (17, 18) and effectively removing the corresponding pz orbital (4, 19, 20). In this sense, chemisorbed H atoms are equivalent to carbon vacancies (4, 12, 14) but with the advantage that, unlike vacancies, they leave the graphene atomic lattice with no unsaturated dangling bonds, preserving the threefold symmetry. Our experiments, supported by ab initio calculations, provide a comprehensive picture of the origin, coupling, and manipulation of the magnetism induced by H atoms on graphene layers.

We deposited atomic H on graphene grown on a SiC(000-1) substrate (21). In this system, the rotational disorder of the graphene layers electronically decouples the π bands, leading to a stacking of essentially isolated graphene sheets (2224). Scanning tunneling microscopy (STM) visualizes single H atoms as a bright protrusion (apparent height, ~2.5 Å) surrounded by a complex threefold √3×√3 patternthat is rotated 30° degrees with respect to the graphene lattice. (25, 26) [Fig. 1B and section 1 in (27)]. The resolution that we achieved allowed us to identify the adsorbate as a single H atom and to determine the atomic site (and thus the corresponding atomic sublattice) where each H atom was chemisorbed by means of comparison with density functional theory (DFT)–simulated STM images [Fig. 1D and section 1 in (27)]. As depicted in Fig. 1A, graphene magnetic moments induced by H adsorption should be reflected in the appearance of a spin-polarized state at EF, which, according to DFT calculations, should be characterized by two narrow peaks in the density of states (DOS) (Fig. 1E) (4).

Differential conductance spectra (dI/dV; I, current; V, sample voltage) probe the energy-resolved local DOS under the STM tip position and thus are ideal for investigating this question. Figure 1C shows two dI/dV spectra, measured at 5 K, that are representative of our findings. The dI/dV spectra measured on clean graphene, located far enough away from defects, have the characteristic featureless V shape of graphene, with a minimum at EF indicating the position of the Dirac point ED. The dI/dV spectra measured on top of single H atoms have two narrow peaks, one below and one above EF, separated in energy by a splitting of ~20 meV. We attribute these two features to the spin-polarized state, in which the Coulomb repulsion is large enough to fully separate the two spin components. The observed charge neutrality (the splitting is essentially symmetric around EF) and the well-defined peak splitting indicate the complete spin polarization of the state. DFT calculations show that the magnetic moment associated with the unpaired electron that is left over in the graphene system after H adsorption would be 1 bohr magneton (fig. S7). Our interpretation of the experiment is fully supported by DFT calculations, as can be seen in Fig. 1E, which shows the expected DOS for a single H atom in a graphene super cell containing 218 carbon atoms. The theoretical energy splitting depends on the size of the graphene super cell (5). Our calculations show that the splitting decays with the size of the graphene super cell, suggesting a small but finite splitting for the isolated H [section 9 in (27)], which is in agreement with the experimental observations.

An independent proof that the split dI/dV peaks are induced by magnetism can be obtained by changing the occupation of the split states (n, n) by means of graphene doping. This is based on the idea originally proposed in (28), according to which the transition from a magnetic state to a nonmagnetic one can be realized by tuning the energy position of the impurity level with respect to the Fermi level [section 3A in (27) gives a detailed description of this system in terms of the Anderson impurity model]. In graphene, the impurity level (zero-energy mode) should be dictated by the position of ED [E= ED + U(n – 1/2); E = ED+ U(n1/2)] (29), which can be tuned by doping the graphene layers. For a large enough electron (hole) doping, the occupation of both the n and n levels can be tuned to 1 (0), in which case the energy levels will be degenerate, leading to a single nonmagnetic state close to ED (E = EED). In Fig. 1F, we show how graphene doping affects the splitting of the H-induced magnetic state. Our dI/dV spectra demonstrate that both n- and p-type graphene doping cause the splitting of the H-induced graphene state to vanish; only one sharp peak appears at ED, which we ascribe to a transition to the nonmagnetic state. This interpretation is fully supported by our DFT calculations for doped graphene layers [Fig. 1F and section 3B in (27)]. If the split peaks appearing in neutral graphene after H adsorption had an origin that was not associated with a magnetic moment (i.e., a single-electron origin), we would observe a rigid shift of the peak position with doping, and the doublet structure would remain unmodified [section 3 in (27)]. Our results are consistent with the case of sp3 defects in graphene, for which the possibility to control graphene magnetic moments by molecular doping has been reported (15).

We explored, with atomic precision, the spatial extension of the spin-polarized electronic state induced by H atoms in undoped graphene. The relatively modest 20-meV energy splitting observed in our experiments suggests a large spatial extension of the magnetic state (Fig. 1A). Figure 2A shows a conductance map plotted with respect to distance and energy [dI/dV(x, E)], with the distance x measured along the 6-nm line that crosses the H atom in Fig. 2B. The state extends several nanometers away from the H atom, confirming that it is a quasi-localized graphene state (3). It presents strong atomic-scale modulations of the peak intensities, with maxima (bright yellow features in the conductance map) corresponding to carbon atoms in the sublattice opposite to the one on which H is chemisorbed. On carbon sites in the same sublattice as H chemisorption, the peaks vanish almost completely (fig. S11). Because our results show a complete spin polarization of the state, the spatial evolution of the height of the dI/dV occupied peak provides the spatial distribution of the local magnetic moment induced by H chemisorption [DFT calculations in Fig. 2, C and D, and section 4 in (27)]. This is further illustrated in Fig. 2E, which shows that the magnetic moment is essentially induced on the carbon atoms in the graphene sublattice opposite to the locus of H chemisorption.

Fig. 2 Spatial extension of the spin-polarized electronic state induced by H atoms in undoped graphene.

(A) Conductance map [dI/dV(x,E)] along the dashed line in (B). The spectra were acquired at a nominal junction impedance of 3 gigaohms (100 mV, 33 pA). (B) STM topography of a single H atom on graphene (0.2 V, 0.1 nA, 7 × 5 nm2). (C) Comparison between DFT calculations for the local magnetic moment and the height of the occupied projected DOS (PDOS) peak, calculated on different carbon atoms [section 4 in (27)]. (D) Calculated magnetic moments induced by H chemisorption (the lengths of the arrows signify the relative magnitudes of the magnetic moments). (E) Schematic of the graphene structure along the dashed line in (B). Green (purple) balls indicate the positions of carbon atoms belonging to the same (opposite) sublattice with respect to the locus of H chemisorption. The dotted line shows the evolution of the height of the measured occupied peak, and the arrows show the relative magnetic moment contribution of each carbon atom. All experimental data were acquired at temperature T = 5 K.

The large extension of the local magnetic moments associated with H chemisorption suggests that long-range magnetic interactions mediated by direct exchange should take place. This is different from substrate-mediated interactions such as the Ruderman-Kittel-Kasuya-Yosida interaction, because in this case the coupling results from the direct overlap of magnetized graphene states. In addition, the critical sublattice dependence that we observed for the spin-polarized peak implies that the magnetic coupling should be radically different depending on whether H atoms are adsorbed on the same or different sublattices. Consistent with this expectation, our DFT calculations reveal that two H atoms chemisorbed on the same sublattice (AA dimer) show ferromagnetic coupling with a total spin S = 1, whereas for H atoms on different sublattices (AB dimer), the solution is nonmagnetic. This result is reproduced for all possible H-H arrangements up to the largest distances (~1.5 nm) achievable with our super cell sizes (Fig. 3A). For a separation of 1.5 nm, the computed exchange energy for AA dimers is ~35 meV [section 5 in (27)]. Furthermore, the total energy of all H dimers that we studied is lower than that of two isolated H atoms (Fig. 3A), confirming the observed tendency of H to form dimers on graphene surfaces at high enough H concentrations (3032).

Fig. 3 Sublattice dependence of the magnetic coupling between neighboring H atoms.

(A) Calculated total energy, relative to twice the adsorption energy of a single H atom, and magnetic state of a pair of H atoms adsorbed on the same (AA dimer) and different (AB dimer) sublattices, plotted as a function of the H-H distance. (B) STM image showing two different pairs of H atoms, with one pair in an AA (purple circle) and the other pair in an AB (green circle) configuration (0.2 V, 0.1 nA, 7.8 × 6.6 nm2). (C) Calculated STM image of the AB dimer and (D) the AA dimer, with the corresponding diagrams for H atoms (blue balls) on graphene (purple and green balls). (E) Experimental dI/dV spectra and (F) calculated DOS for the AA dimer, AB dimer, and clean graphene. The spectra were acquired at a nominal junction impedance of 8 gigaohms (–400 mV, 50 pA). All experimental data were acquired at 5 K.

To test this scenario experimentally, we explored the local electronic structure of many different H dimers with high-resolution scanning tunneling spectroscopy (STS). The STM image in Fig. 3B shows two H dimers in AA (purple circle) and AB (green circle) configurations; the corresponding calculated STM images are shown in Fig. 3, C and D. The dI/dV spectra acquired on the AB dimer (green line in Fig. 3E) show a featureless local DOS that is indistinguishable from that measured on bare graphene (black), as would be expected for a nonmagnetic configuration. In contrast, dI/dV spectra measured on the AA dimer (purple) show the split state in the vicinity of EF, as expected for a ferromagnetic coupling between the H atoms. As shown in Fig. 3F, our calculated DOS reproduce these observations, confirming the ferromagnetic (nonmagnetic) nature of the AA (AB) dimer. For all H dimers measured in this study, AA dimers presented a fully split state close to EF, which was absent in AB dimers (fig. S12). Our STS data show that this sublattice-dependent magnetic coupling persists for very long distances, even for H dimers separated by more than 1 nm [section 5 in (27)].

We further demonstrated the capability of inducing magnetic moments on selected graphene regions by using STM to perform atomic manipulations (3335). We proved that individual H atoms can be removed, laterally moved, and even deposited on graphene surfaces with atomic precision to ultimately tailor their local magnetic state [section 6 in (27)]. Figure 4 shows two representative examples of these manipulation experiments, in which the local graphene magnetism was selectively switched on and off. The graphene region in Fig. 4A shows two H atoms in an AB dimer configuration. Our STS data measured on those H atoms (Fig. 4C) show that this AB dimer configuration does not induce any magnetism onthe graphene layer, in good agreement with the coupling rules previously discussed. Figure 4B shows the same graphene region as in Fig. 4A, after the controlled extraction of one H atom by carefully approaching it with the STM tip. As shown in Fig. 4D, a spin-split state immediately emerges on the graphene layer after the H removal, confirming the creation of a local magnetic moment in graphene. The insets show the corresponding DFT calculations of the resulting magnetic moment for each situation. We next performed a lateral manipulation on the H dimer, shown in the central region of Fig. 4E. Initially, the dimer was in an AA configuration, with both H atoms chemisorbed on the same carbon sublattice. The STS spectrum for that configuration (Fig. 4G) shows the presence of a spin-split state, as expected for ferromagnetic coupling. To switch off the graphene magnetic moments induced by this H dimer, we turned it into a nonmagnetic AB dimer configuration by laterally moving one of its H atoms to the opposite sublattice. Figure 4F shows the same graphene region after the H manipulation (the AB dimer in the upper part of the image serves as reference). The STS spectrum measured on the constructed AB dimer shows the disappearance of the polarized peaks, indicating that local graphene magnetism was effectively switched off.

Fig. 4 Manipulation of graphene local magnetic moments by STM.

(A) STM image of an H dimer in an AB configuration. (B) STM image after the removal of one H atom. (C) dI/dV spectra measured on the AB dimer in (A) and (D) the single H atom in (B). The spectra were acquired at a nominal junction impedance of 4 gigaohms (200 mV, 50 pA). The insets present the corresponding DFT calculations for H atoms (blue balls) on graphene (purple and green balls), with blue arrows being the magnetic moments induced on graphene. (E) STM image of an H dimer in an AA configuration. (F) STM image after laterally moving one H atom. (G) dI/dV spectra measured on the AA dimer in (E) and (H) the AB dimer in (F). The spectra were acquired at a nominal junction impedance of 4 gigaohms (200 mV, 50 pA). An additional dI/dV spectrum that is better resolved in the vicinity of EF measured on the AA dimer in (E) is shown in fig. S20. Insets show the corresponding DFT calculations. (I to L) STM images showing the same graphene region during different steps of a manipulation experiment involving a large number of H atoms. The point defect outlined with a gray circle is used as a reference. Tunneling parameters were 0.2 V, 0.1 nA, and 6.5 × 4.0 nm2 for (A) and (B); 0.2 V, 0.1 nA, and 9.5 × 5.5 nm2 for (E) and (F); and 0.4 V, 0.03 nA, and 28 × 28 nm2 for (I) to (L). All experimental data were acquired at T = 5 K.

Last, we explored the possibility of selectively tuning the collective magnetic moment in a graphene region by inducing an imbalance between H atoms on opposing sublattices A and B. For this purpose, we systematically manipulated a large number of H atoms [section 6 in (27)]. In Fig. 4, I to L, we present an example in which we first removed all H atoms from a graphene region by using the STM tip (Fig. 4I). Then, we selectively deposited 14 H atoms on this same region to reach a configuration with seven H atoms chemisorbed on each graphene sublattice (Fig. 4J). Our experimental findings and existing calculations (4, 7) indicate that a very low (if any) net magnetic moment should be expected on this region, because of these equal sublattice populations. Next, by selectively removing all the H atoms chemisorbed on sublattice B, we created a ferromagnetic configuration with the seven remaining H atoms on sublattice A (Fig. 4K). As the final step, we combined several manipulation processes to reverse the situation and construct an H arrangement with all seven H atoms chemisorbed on sublattice B (Fig. 4L). The degree of complexity shown in our manipulation experiments demonstrates the high reproducibility of the procedure, which paves the way to the realization of atomically controlled experiments in graphene magnetism, an area that has thus far been restricted to a purely theoretical framework.

Supplementary Materials

Materials and Methods

Figs. S1 to S20

Movie S1

References (3759)

References and Notes

  1. Materials and methods are available as supplementary materials on Science Online.
  2. Acknowledgments: We thank V. Cherkez (Institut NEEL, CNRS, and Université Grenoble Alpes) for his help with the fabrication of samples of graphene grown on SiC, and we thank D. Wong (University of California–Berkeley) for his careful reading of the manuscript. This work was supported by Spain’s Ministerio de Economía y Competitividad under grant nos. MAT2013-41636-P, CSD2010-00024, PCIN-2015-030, FIS2013-47328, and FIS2012-37549-C05-03; the European Union structural funds and the Comunidad de Madrid MAD2D-CM program under grant nos. S2013/MIT-3007 and P2013/MIT-2850; the Generalitat Valenciana under grant no. PROMETEO/2012/011; THE CNRS PICS (Projets Internationaux de Coopération Scientifique) program under grant no. 6182; and the European Union FP7 (7th Framework Programme for Research and Technological Development) Graphene Flagship (grant 604391) and FLAG-ERA programs. The authors acknowledge the computer resources and assistance provided by the Centro de Computación Científica of the Universidad Autónoma de Madrid.
View Abstract

Stay Connected to Science

Navigate This Article