Thanks to their unique transport properties, carbon nanotubes (CNT), graphene sheets and nanoribbons are promising components for future nanoelectronics1. A close attention is paid to the connections of these graphitic structures to external metallic leads where the injection and the collection of charges are controlled2,3,4. Indeed, bad interfaces might ruin the transport properties of such carbon-based devices. To this respect, the chemical nature of the contacting leads is of major importance; it affects the electronic properties5 and, depending on the reactivity, the geometry of the contact6. The impact of these two aspects on the transport properties is entangled and, for mesoscopic structures, it is challenging to address them separately. Exploring the evolution of these parameters for contacts shrunk to the limit of individual atoms might solve this issue.

Scanning tunnelling microscopy (STM) contact experiments with fullerene molecules have revealed the decisive impact of atomic-scale modifications on transport properties7,8,9,10. Recently, some of us demonstrated that the conductance of a Cu–C60–Cu junction varies by more than an order of magnitude as a function of the number of Cu atoms in direct contact with the C60 molecule11. For small contact sections (that is, a single-atom contact) the conductance is limited by charge injection to the molecule, and depends essentially on the electronic and geometrical properties of the metal-C atomic contact, which acts as a conductance ‘bottleneck’.

Here we probe the charge injection efficiency at the interface between a C60 molecule and single metallic atoms of different chemical nature. We demonstrate that information on the reactivity between the fullerene and each specific atom can be extracted from our experiments, which match the properties observed at the mesoscopic scale. First-principles transport simulations with a novel scheme to describe the single-C60 junction reproduce the experimental findings and reveal how the chemical valence of the contacting atom determines the conductance of the junction12,13,14. Our study further demonstrates that atomic-scale junctions may be used to explore the properties of mesoscale graphitic contacts.


Metal adatoms as chemically controlled electrodes

To use different metal adatoms (M1=Cu1, Au1, Pd1, Fe1, Ti1, Al1) deposited on a Cu(111) surface as chemically controlled electrodes for molecular contacts, the apex of our STM tip was functionalized with a C60 molecule previously evaporated on the surface15. Figure 1a shows metallic adatoms imaged with a C60 tip at a sample voltage corresponding to the second lowest unoccupied molecular orbital (LUMO+1) of the molecular tip. The image reveals that this particular C60-functionalized tip is oriented with a bond between a pentagon and a hexagon (5:6 bond) towards the substrate16. Depending on the chemical nature of the adatom, a varying ‘apparent height’ is exhibited in the STM images as highlighted in the close-up views in Fig. 1b–d.

Figure 1: Individual metal adatoms contacted with a C60 -functionalized STM tip.
figure 1

(a) STM image (7.0 × 6.2 nm2) of different metal adatoms on Cu(111) acquired with a C60 tip at a sample voltage V=1.7 V. (bd) Close-up views (1.4 × 1.4 nm2) of Cu1, Au1 and Fe1 images with a C60 tip for tunnelling conditions corresponding to the initial parameter of the traces in f. (e) Sketch of the C60 tip where z=0 corresponds to contact with the flat surface. (f) Experimental conductance traces G(z) in units of the conductance quantum G0=2e2/h. Black crosses mark the contact points defined as the intersection of the contact and transition regimes such as indicated by the dashed grey lines in panel f for the bare surface data.

Figure 1e–f show contact experiments obtained with a C60-terminated STM tip and the different adatoms. As expected, the conductances vary exponentially with the tip approach in the tunnelling regime (large z). At shorter distances, a change in the slopes is observed, which marks a transition towards a contact regime. A contact distance zc and a contact conductance Gc (black crosses in Fig. 1f) are determined for each conductance trace following the method detailed in ref. 11. We emphasize that the contact measurements performed on Au1, Cu1 and Fe1 were acquired with the same C60 tip and can be directly compared. The traces obtained on Pd1, Ti1 and Al1 pertain to other sets of measurements where they were compared with Cu1. Taking Cu1 as a reference, it is therefore possible to compare data from different experiments with different adatom species. Using this method, the impact of different molecular orientations and of different tip-side interfaces between C60 and metal on the experimental data is strongly reduced. Figure 2 and Table 1 summarizes the relative distances and conductances at contact.

Figure 2: Mechanical and conductive trends with different adatom electrodes.
figure 2

(a) Experimental (black crosses) contact distances zc and (b) contact conductances Gc for each of the considered metallic adatoms M1, obtained with different C60 tips. As a common reference, the data are (a) compared with or (b) normalized by the value obtained on Cu1. In panel a, the experimental contact distances are compared with calculated binding energies between an atom M1 and a single C60 molecule. In panel b, the experimental data are compared with calculated conductances (normalized by Cu1) for different electrode separations and binding sites (see text). The scattering in the theoretical data (blue triangles) corresponds to calculations for different lateral adatom positions and distances with respect to the C60 tip.

Table 1 Absolute experimental contact distances and contact conductances .

Mechanical aspects of contact formation

First we focus on the mechanical aspects of the different junctions. Figure 2a shows that the contact takes place at different tip-sample distances zc depending on the chemical nature of the metallic adatom. For instance, the contact point occurs systematically at larger zc values with Ti1 than with any of the other species. We find that the experimental variation in zc correlates very well with binding energies Eb (blue triangles) between an atom M1 and a single C60 molecule calculated using density functional theory (DFT), see Methods section. The substrate has little impact on these binding energy trends, as discussed in Supplementary Fig. 1.

This suggests that the shape of the conductance trace around the point of contact is a measure of the attractive chemical forces between the adatom and the C60 tip. Indeed, when the C60 tip approaches the adatom, the attractive force between them increases giving rise to an elastic response17. The larger the attractive force between the molecule and the adatom, the ‘sooner’ a contact is established (that is, larger zc). Our DFT-generalized gradient approximation (GGA) calculations18,19 for the various junctions (Fig. 3a) also support that the variation observed in Fig. 2a is not due to the natural height variations of the different adatoms on Cu(111) (Supplementary Fig. 2). These findings also agree with measurements of the sticking behaviour of mesoscopic metallic electrodes on single-wall CNT6, supporting a hierarchy Eb(Pd)>Eb(Fe)>Eb(Al)>Eb(Au) for the binding energy. Theoretical works have revealed similar hierarchies for metallic adatoms20, clusters of metallic atoms5, and metal surfaces21 interacting with graphene sheets. While the reactivity between sp2 carbon atoms and metallic electrodes varies with the dimensionality (0D, 1D or 2D) and the curvature of the graphitic structure, it is noticeable that our atomic-scale experiment reproduces accurately all these trends.

Figure 3: Analysis of calculated junction properties.
figure 3

(a) Model of the C60–M1 junctions considered in the calculations. (b) PDOS onto adatom basis (full lines: adatoms positioned in hollow sites on the molecular symmetry axis, L=18.5 Å) compared with the density of states for the same species in its bulk environment (dashed lines, Supplementary Fig. 5). For Fe and Ti, the two spin components are shown with opposite sign. The datasets are offset for clarity. (c) Comparison of adatom PDOS at the Fermi level (L=18.5 Å) with the calculated junction conductance Gc(M1) at different electrode separations, normalized with respect to Cu1.

Chemical trends of the adatom on contact conductance

We next turn to the discussion of the contact conductance values. The experimental data in Fig. 2b (black crosses) reveal that Gc is on average about 1.6 times higher for Au1 and Pd1 than for Cu1 contacts. The largest average ratio was observed for Al1. Although we took the greatest care to locate the molecular tip over each adatom with the same relative position before the tip approach, sub-angström variations of this parameter are unavoidable. Such small variations can lead to different contact geometries and, consequently, to different contact conductances9. This explains the scattering in the experimental data in Fig. 2b. In our simulations we account for this aspect by considering different lateral positions of the adatoms with respect to the C60 tip11. We also consider different C60-adatom distances around the point of contact (Fig. 3a). The result of these simulations is displayed as blue triangles in Fig. 2b. An overall good agreement is obtained between experiment and theory. The conductance ratios are quantitatively reproduced for the Au1, Fe1 and Pd1 species. Ti1 and Al1 are also found to be the most conductive species, but contrary to the experiment theory assigns a larger conductance to Ti1 than to Al1. Our conductance ratios also agree with simulations for metal-graphene and metal-CNT interfaces5,22. Importantly, these results confirm that the efficiency of charge injection to a C60 molecule can vary substantially for different metal adatoms, independently of any geometrical considerations.

To rationalize the observed hierarchy, it is useful to consider the projected density of states (PDOS) on the different metallic adatoms as shown in Fig. 3b and Supplementary Fig. 3. For Cu1, Au1 and Pd1 a d-orbital adatom resonance is present significantly below the Fermi level EF. Around EF, these three species as well as Al1 exhibit no particular spectral features corroborating that their sp electronic states of the free atoms hybridize strongly with Cu(111). The situation is very different for Fe1 and Ti1 which both exhibit a spin-polarized electronic structure with d-orbital resonances located around EF. For Fe1 (Ti1), the PDOS at EF is dominated by the minority (majority) spin channel, rather analogous to the case of these metals adsorbed on graphene20. The transmission spin polarization23 is predicted to be as high as 88% in the case of Ti (Supplementary Fig. 4). Figure 3c compares the calculated junction conductances with the PDOS(EF) of the adatom at relatively large electrode separation (L=18.5 Å, Fig. 3a). To a first approximation, it is observed that the conductance is related to the PDOS.

This interpretation in terms of PDOS on the adatoms also offers an explanation for the overestimated conductance of Ti1 in the simulations. Close to resonant transport conditions—that is, when the adatom d-orbitals are nearly aligned with EF—the conductance depends sensitively on the resonance position and width. An inadequate description of the strong Coulomb repulsion between the 3d electrons localized on the adatom with standard DFT methods (and consequently their hybridization with the substrate) can therefore have a large impact on the calculated conductance. Although theory still provides a qualitative agreement with experiment in these cases (Fe1 and Ti1), a quantitative comparison needs to be taken with caution.

Figure 3b also compares PDOS of the various metal adatoms with the corresponding density of states in the bulk (Supplementary Fig. 5). Interestingly, despite of very different environments, their spectral features are rather similar. The main difference is obtained for Pd, where the d-band in bulk reaches EF while it is located well below for Pd1 on Cu(111). This general correspondence supports the notion that our single adatoms can be considered representative for meso- and macroscopic electrodes. Therefore noble metals Cu1 and Au1 as well as Pd1 (closed d-shell elements) are generally less favourable for charge injection to C60 than the open d-shell elements Fe1 and Ti1 or the open p-shell element Al1. These observations suggest that the ability of a given metal to inject charges to sp2 carbon can, to some extent, be intuited from its chemical valence.

Adatom influence on C60 orbitals

Finally, the electronic properties of the C60 are affected by the hybridization with the different adatoms11,24 with consequences for the junction conductance. Figure 4 shows the PDOS onto the C60 basis at relatively small electrode separation (L=17.2 Å, Fig. 3a). It reveals the characteristic spectral features of the C60 tip (essentially the highest occupied molecular orbital around −1.2 eV, the LUMO around 0.4 eV and the LUMO+1 around 1.2 eV), but also subtle differences depending on the adatom species. Compared with the PDOS for the isolated C60 tip (dashed black lines), it is observed that the spectrum is only weakly perturbed by the interaction with the adatoms. For Pd1, Fe1 and Ti1, additional features appear coinciding with the d-resonances shown in Fig. 3b. The stronger impact of these adatoms on the C60 orbitals is consistent with their significant reactivities reported in Fig. 2a.

Figure 4: Influence of adatoms on the C60 orbitals.
figure 4

PDOS onto the C60 basis at relatively small electrode separation (L=17.2 Å. Fig. 3a). Adatoms in hollow sites on the molecular symmetry axis (thin lines) as well as adatoms shifted one hollow site away (thick lines) are considered. For Fe and Ti, the two spin components are shown with opposite sign. The datasets are offset for clarity and compared with PDOS for the isolated C60 tip (dashed black lines).

Figure 4 also reveals that the molecular resonances shift to lower energies as the contact is established, suggesting that these adatoms transfer additional electrons to C60 (for the full evolution with electrode separation L, see Supplementary Fig. 6). As a consequence, the LUMO becomes more resonant with the Fermi level and conductance is enhanced.11 These shifts are weak with Pd1 (little charge transfer) but strong with Al1 and Ti1 (significant charge transfer). The differences among the adatoms may therefore affect the conductance hierarchy for larger contacts (more than one atom); for example, the conductance would be expected to increase more rapidly with the number of atoms in the cluster for Al1 and Ti1 than for Pd1.


We have presented a systematic approach to probe the impact of electrode material on structure and electron transport at metal-sp2 carbon interfaces. DFT simulations confirm our observations and rationalize the observed properties. We find that some experimental parameters (the contact distance zc) directly reflect the strength of the metal-C bonds, but also that this strength does not necessarily reflect its conductive properties. In fact, we conclude that the ability of a metal atom to efficiently inject charges into a sp2 carbon atom is intimately linked to its chemical valence. For transition metals, it appears that elements in the centre of the d-block present a higher ability for charge injection than atoms from the d-block extrema due to a higher density of states at the Fermi level. Charge transfer between the metal adatom and the contacted C atoms of the C60 molecule modifies somewhat the position of the molecular resonances that also affects the conductance of the atomic-scale contact. Our single-C60 junctions are thus good models to explore the properties of electric contacts to sp2 carbon materials.


STM setup

The experiments were performed with an Omicron low-temperature STM operated at ≈4.5 K in ultrahigh vacuum (below 10−10 mbar). Cu(111) samples and etched W tips were prepared by Ar+ bombardment and annealing. W tips were indented into the sample surface to cover them with Cu. Approximately 0.2 monolayer of C60 molecules were deposited on a sample kept at room temperature. Au, Pd, Fe, Ti or Al atoms were deposited on the C60-covered Cu(111) surface kept at low temperature (4.5 K). Cu atoms were deposited using the method described in ref. 25.

Identification of the different metallic species

In a first step, each of the metallic species (except Cu1) was evaporated individually (that is, different experimental sets) on a pristine Cu(111) sample maintained at low temperature. As referenced in the paper, Cu1 was systematically deposited by contacting the Cu(111) surface with the STM tip. For Cu1, Au1 and Fe1, it was possible to register constant height differential conductance (dI/dV) spectra over a large energy range (Fig. 5a,b). The dI/dV spectra were acquired using lock-in detection with a modulation frequency of 740 Hz and a root-mean-square modulation amplitude of 10 mV. These data reveal resonances characteristic of the different species. The dI/dV spectra of Cu1/Cu(111) are even referenced in the literature26. Using these signatures, it was easy to discriminate these three adatoms from each other on the surface.The dI/dV spectra of Pd1, Ti1 and Al1 could not be measured because of instabilities at high voltages. This is why these elements were measured separately and were only compared with Cu1 (which can be easily identified). It is therefore impossible to take one species for another.

Figure 5: Experimental constant-height spectra of deposited adatoms recorded with a metallic STM tip.
figure 5

(a) Cu1, Au1 and (b) Fe1 deposited on Cu(111). Note that the spectrum in b was acquired with a lower current set point to limit instabilities at high voltages. These spectra are characteristic of the different species and are used to identify them. (c) Ti1 deposited on Au(111). The same features as those reported in ref. 27 confirm that we evaporated Ti1. The structure around ≈−0.5 V corresponds to the localization of the Au(111) surface state at the adatom.

We had a particularly critical look at the Ti1 and Al1 cases. Although the evaporation of Cu1, Au1, Pd1 and Fe1 is well known and relatively easy to control, Ti1 and Al1 are less documented. Ti1 was deposited by T. Jamneala et al.27 on Au(111). In this case, dI/dV data reveal a small Kondo feature at the Fermi level as well as a characteristic feature around 150 mV (most likely a 3d-orbital resonance). We were able to reproduce these features (Fig. 5c), which validates the evaporation of Ti1. Unfortunately, Al1 reveals no features in dI/dV spectroscopy and this strategy can thus not be used for the identification. To investigate if the evaporated material really corresponds to individual atoms (and not clusters of atoms), we used atom manipulation techniques to form dimers Al2 and trimers Al3 (Fig. 6). The fuzzy image in Fig. 6e is a strong indication of a dimer, similar to the Cu2/Cu(111) case26. While we cannot rule out a possible contamination of the deposited adatoms, we believe the manipulation sequence is a strong support for Al1.

Figure 6: Experimental STM images of a manipulation sequence on Cu(111).
figure 6

(a) Three individual Al1 are arranged into (b) Al1 and a dimer Al2, and finally into (c) a trimer Al3. Panels (e,f) show image close ups of the cluster to the upper right. The scale bar, 1 nm.

C60–M1 binding energy calculations

The binding energy Eb between a C60 molecule and a single metallic atom M1 is defined as Eb(C60M1)=Etot(C60)+Etot(M1)−Etot(C60M1), where Etot(i) is the total energy of system i from a spin-polarized calculation. A positive binding energy thus corresponds to a stable system. DFT calculations were carried out with the Vasp code28 using a 515-eV planewave cutoff, the generalized gradient approximation (GGA) of Perdew-Burke-Ernzerhof (PBE) for the exchange-correlation functional29, a tetragonal supercell of 14 × 14 × 20 Å3, a Gaussian smearing of 1 meV for the occupancies, and relaxations until residual forces were smaller than 0.02 eV Å−1. Constraints were imposed to fix the metal atom to different binding sites on the C60 cage. The results for the binding energies Eb, spin magnetic moment μ and characteristic bond lengths are given in Tables 2 and 3. These trends are mostly in agreement with those reported in ref. 23. We checked the role of applying a dipole correction for electrostatic interactions between neighbouring cells along the dimer axis, but this only affects the binding energies by a few percent for our simulation cells (Supplementary Table 1).

Table 2 Calculated binding energies and spin magnetic moments of C60-M1 clusters.
Table 3 Calculated bond lengths of C60-M1 clusters.

Electronic structure of model junctions

The electronic structure for the various junction geometries (Fig. 3a) was calculated with the Siesta18 pseudopotential DFT method with the GGA-PBE exchange-correlation functional29 as described in refs 11 and 15. The junctions are modelled by structures comprizing a 13-layer slab Cu(111) in a 4 × 4 representation, the adatom as well as the C60 molecule. A standard single-zeta plus polarization basis was employed for bulk Cu (0.15-eV energy shift) and a long-ranged (0.02-eV energy shift), double-zeta plus polarization basis for C60 and the adatoms. The pseudopotentials were constructed according to the parameters specified in Supplementary Table 2. Real-space grid integrations were carried out using a 200 Ry energy cutoff. The 3D Brillouin zone was sampled with a 2 × 2 × 1 Monkhorst-Pack k-mesh. The lattice constant for the Cu crystal was set to 3.70 Å. As a function of varying the electrode separation L (Fig. 3a), the adatoms and underlying surface layer were relaxed to 0.02 eV Å−1. The adatom heights h and radial distances r to the molecular symmetry axis are reported in Supplementary Fig. 7. PDOS were calculated using Inelastica30 with a 13 × 13 k||-mesh of Gauss–Kronrod points and a broadening of η=0.1 eV in the bulk part of the semi-infinite electrodes.

First-principles transport simulations

The electronic structure from Siesta was used to calculate the transport properties for the TranSiesta19 setup using Inelastica30. The zero-bias conductance G is determined by the electron transmission function T(E) evaluated at the Fermi energy EF:

where G0=2e2/h is the conductance quantum. We calculate the electron transmission in two different approaches as described below.

The typical approach is to calculate T(EF) per unit cell for a system which is periodically repeated in the plane perpendicular to the electron transport. This periodic (P) approach is consistent with the DFT treatment that relies on this periodicity. As a consequence of Bloch’s theorem, the electron transmission is averaged over k-points in the 1st Brillouin zone (with weights wk=1/Nk when using a linearly spaced mesh with Nk points):

Here T(E,k) is the electron transmission resolved in terms of electron momentum k

and where the retarded Green’s function G in the device region is

with S(k) and H(k) being the overlap and Hamiltonian matrices, respectively. The electrode-coupling rates are related to the self-energies via

The experiment, however, concerns a single C60-functionalized STM tip in contact with a single adatom on Cu(111). Our alternative computational approach, more in line with this situation, corresponds to partitioning the system into periodic leads (where quantities are k-sampled) and a ‘non-periodic’ device region (which does not depend on k). This is similar to a real-space DFT approach for STM simulations31,32 and methods used for phonon transport through nanoconstrictions33,34. For our simulation cells, the natural choice is to consider the adatom and C60 as the device, since the electronic coupling between these atoms and their periodic repetitions is significantly smaller than for the Cu atoms in the two semi-infinite electrodes. The transmission TNP(E) is then given by

where now the device Green’s function G is thought to concern a single junction

with only the self-energies being sampled over k. In practice, since the device region may not be completely decoupled from its periodic repetitions, we sample TNP(E) over a coarse 3 × 3-k-mesh for the device part (H and S) and check that there are only small variations on this mesh.

Figure 7 compares the conductance ratios for the two computational schemes as well as for different sampling of k-space (6 × 6 linearly spaced points versus 13 × 13 Gauss–Kronrod points). We find that the latter ‘non-periodic’ approach, implemented on this occasion in Inelastica30, gives a slightly better overall agreement with the experimental conductance ratios (in particular for the Al1 case). The figure also highlights the convergence in k-space as essentially the same conductance ratios are obtained for the two periodic sets of calculations (blue diamonds versus red triangles). In the Gauss–Kronrod scheme, a broadening of η=0.1 eV was used in the bulk electrode.

Figure 7: Comparison of conductance ratios with different computational approaches.
figure 7

Calculations for periodic arrays of C60 junctions with the transmission T(EF) sampled on a k-mesh of 6 × 6 linearly spaced points (blue diamonds) or with 13 × 13 Gauss–Kronrod points (red triangles) in 1st Brillouin zone (1BZ). Calculations for a ‘non-periodic’ molecular junction with electrode self-energies ΣL/R sampled on a k-mesh of 13 × 13 Gauss–Kronrod points in 1BZ (turquoise triangles). The device region consists of the adatom and C60. The experimental ratios (black crosses) are shown for comparison.

The absolute conductances for both schemes are shown in Supplementary Figs 8 and 9. Further, for the spin-polarized species (Fe and Ti) the transmission per spin channel as well as the transmission spin polarization is shown in Supplementary Fig. 4. We also checked that the calculated conductance ratios for Al1 shows no significant dependence on the C60-orientation (Supplementary Fig. 10).

Kondo physics

We note that eventual Kondo physics35,36, beyond the DFT methods employed here, is not expected to impact our results. No Kondo resonances were experimentally resolved in dI/dV for any of the considered species on Cu(111). Therefore, even if a narrow Kondo feature would exists for species like Fe1 and Ti1, it has a negligible effect on the measured conductance at V=−50 mV.

Additional information

How to cite this article: Frederiksen, T. et al. Chemical control of electrical contact to sp2 carbon atoms. Nat. Commun. 5:3659 doi: 10.1038/ncomms4659 (2014).