Superlubricity of graphene nanoribbons on gold surfaces

See allHide authors and affiliations

Science  26 Feb 2016:
Vol. 351, Issue 6276, pp. 957-961
DOI: 10.1126/science.aad3569

A golden opportunity for graphene

Reducing friction can limit wear and improve the energy efficiency of mechanical devices. Graphene is a promising lubricant because the friction between sheets is minuscule under certain circumstances. Kawai et al. show that the same ultra-low frictional properties extend to other surfaces. They find ultralow friction when dragging graphene nanoribbons across a gold surface using an atomic force microscope. This discovery sets up the potential for developing nanographene frictionless coatings.

Science, this issue p. 957


The state of vanishing friction known as superlubricity has important applications for energy saving and increasing the lifetime of devices. Superlubricity, as detected with atomic force microscopy, appears when sliding large graphite flakes or gold nanoclusters across surfaces, for example. However, the origin of the behavior is poorly understood because of the lack of a controllable nanocontact. We demonstrated the superlubricity of graphene nanoribbons when sliding on gold with a joint experimental and computational approach. The atomically well-defined contact allows us to trace the origin of superlubricity, unraveling the role played by ribbon size and elasticity, as well as by surface reconstruction. Our results pave the way to the scale-up of superlubricity and thus to the realization of frictionless coatings.

Graphene offers distinctive properties as a solid lubricant (1) and has the potential to be used as an ultrathin coating material on surfaces, almost suppressing energy consumption in mechanical components. The interpretation of such superlubric behavior is based on the premise that (25) (i) the high lateral stiffness of graphene makes a commensurable contact with most solid surfaces nearly impossible, and (ii) combined with the weak interaction with most materials, incommensurability leads to a state of ultralow friction when graphene slides over a different material. To substantiate this hypothesis and establish a connection with the tribological properties observed on macro- and mesoscales, it is highly desirable to measure the mechanical response of a graphene flake down to the nanometer level. In such measurements, one has to ensure that both of the contacting surfaces are atomically well defined, that their common interface is free from contaminants, and that the ultralow forces accompanying the sliding motion can be distinguished from the background noise. Whereas clean atomically flat surfaces as substrates can reliably be obtained in ultrahigh vacuum (UHV), atomically defined graphene systems as sliding objects are rarely prepared. Carbon nanotubes have exceptional superlubric properties up to a length scale of a few centimeters (6), but their curvature makes them difficult to manipulate in a controlled way. Nevertheless, the problem can be overcome by using graphene nanoribbons (GNRs), which recently have been synthetized on a metal substrate by means of an on-surface chemical reaction (7). Their structure is well defined by the precursor molecule, as confirmed by high-resolution scanning tunneling microscopy (STM) and atomic force microscopy (AFM). For this reason, GNRs are an appropriate candidate for our goal. Apart from that, GNRs are also very promising in a series of applications [e.g., nano-electromechanical systems (8), nanofillers (9), transistors (10), and other electronic and spintronic devices (11)] for which assessing their mechanical stability is pivotal.

We investigated the frictional, adhesive, and elastic properties of GNRs by means of lateral manipulation on an Au(111) substrate, using dynamic AFM in UHV at a low temperature (4.8 K). The ends of selected GNRs were anchored to the probing tip and dragged back and forth in a controlled way while the friction force was recorded. An accompanying computational experiment allowed us to relate the origin of the measured superlubricity to the molecular dynamics occurring at the interface.

Our measurements originate from the unintentional manipulation of GNRs aligned along the [–1,0,1] direction of the Au(111) substrate, when the GNRs were imaged by STM using a gold tip. The GNRs were always displaced along their longitudinal axis, even with a relatively large separation, indicating high diffusivity (figs. S1 to S4). To measure the static friction force (Fstat), we switched to AFM, using the same tip. After imaging a sample area covered by GNRs (Fig. 1A), we acquired a two-dimensional (2D) frequency shift map while the tip was scanned laterally along a GNR (x direction) at different constant z distances (Fig. 1B and fig. S5). Following the method of Ternes et al. (12), we reduced the tip-GNR distance stepwise during scanning until we observed an abrupt decrease in the frequency shift (Δf) at the distance defined as z = 0 (Fig. 1C). We found that the GNR was displaced by a distance d = 2.2 nm (Fig. 1A). We also observed that the Δf(x) profile was repeated after the same distance. Langewisch et al. reported a similar observation in their manipulation experiments on perylenetetracarboxylic dianhydride molecules (13). We found that d varied with both the GNR length and the adsorption site. Furthermore, jumps with smaller d values were rarely observed, and the GNRs were never dragged continuously, meaning that the junction formed between the tip and the GNRs is weak. To quantify Fstat, we first estimated the energy landscape experienced by the tip by integrating two times the Δf(z) sections extracted from Fig. 1B. Then we differentiated the 2D potential map along the x direction and multiplied the result by the factor −2kc/f = −0.15 N m−1 Hz−1, where kc = 1800 N m–1 is the spring constant, and f = 24.7 kHz is the resonance frequency of the free tuning fork. The manipulation occurred when Fstat ≈ −105 pN (Fig. 1D). Fstat is exceptionally low, considering that the linear size of the GNR is well above that of the single atoms and conventional molecules that are typically manipulated by AFM (12, 14, 15). This result is a strong confirmation of the superlubric properties of graphene on the nanoscale, as observed in previous friction measurements taken on graphene flakes of undefined size (1618). Another signature of superlubricity is the decrease of the friction force per unit of contact area with increasing size of the contact (1921). To look for this, in our quasi-1D system, we repeated the measurements on GNRs of different lengths (Fig. 1E). In spite of the spread in the measured data (due to the surface reconstruction, discussed below) the force per unit length was found to decrease with increasing GNR length.

Fig. 1 Static friction force measurement.

(A) STM topographies of GNRs on Au(111) before and after a tip-induced lateral manipulation (the green arrow indicates the sliding direction). (B) 2D Δf map along the longitudinal axis of the manipulated GNR. (C) Distance dependence of Δf before, during, and after the GNR displacement. (D) Calculated lateral force. The cross symbol corresponding to the red arrow in (B) shows the position at which the GNR starts moving and the corresponding value of the static friction force Fstat. (E) The absolute value of Fstat as a function of the GNR length (black) and Fstat per unit length (red). Dots correspond to single measurements, whereas bars connect the largest and the smallest values measured while manipulating the same ribbon on different surface regions. Measurement parameters: tunneling current I = 2 pA, bias voltage V = −200 mV (A); oscillation amplitude A = 34 pm [(B) to (D)].

We estimated the diffusion barrier (ΔE) for the GNR by assuming a simple sinusoidal interaction as ΔEFstat a/π ≈ 40 meV, where a = 0.41 nm is the lattice constant of the Au(111) substrate (15). This low value means that a single isolated GNR would diffuse spontaneously at room temperature, making measurements challenging. We also observed a rotation of short GNRs (2 nm), although we always scanned the tip exactly along the GNR axis (fig. S6). This behavior is predicted theoretically for graphene flakes dragged on graphite (22). We even observed a vertical motion of the shortest GNRs (1 nm) before the start of lateral manipulation (fig. S7). We could not perform reliable static friction force measurements on GNRs longer than 22 nm, because other GNRs were often found nearby, and the measured forces were considerably affected by the interaction with those neighbors. Nevertheless, superlubricity allowed us to manipulate GNRs up to 55 nm long (figs. S8 and S9).

Although our measurements allow a precise estimation of the static friction force, they do not provide any details on the complex dynamics of the sliding motion of the GNRs. To gain more insight, we applied the procedure that some of us previously introduced for polymer chains (23) and succeeded in attaching a short edge of a GNR to an Au-coated tip. We then oscillated the GNRs along the [−1,0,1] direction of the Au(111) surface with the tip kept at a constant distance (z) from the substrate. We repeated the measurements several times at increasing values of z (Fig. 2A). The corresponding variations in Δf are shown in Fig. 2, C to F, for increasing values of z. We found that the frequency shift oscillated with a periodicity of 0.28 nm, except when the GNR was driven backward with a tip-surface distance of z = 5 nm. The amplitude of the Δf oscillations was not constant along x but modulated on distances of a few nanometers, varying by a factor of 2. We observed curves with roughly half-periodicity at a small scanning distance of z = 1 to 2 nm (figs. S10 and S11). We also imaged the sample at the end of the process to ensure that we manipulated only the target ribbon (Fig. 2B and fig. S12).

Fig. 2 Frequency shift versus pulling height.

(A) Schematic drawing of the lateral manipulation procedure. (B) STM topographies before and after a GNR has been displaced on the Au(111) surface in the direction of the yellow arrows. The length of the GNR is 6.28 nm, corresponding to seven connected monomers. (C to F) Frequency shifts accompanying the lateral motion at different heights (z = 2, 3, 4, and 5 nm). Oscillation amplitude A = 38 pm.

Figure 3 and fig. S13 show the variations of the normal force Fz and lateral force Fx, calculated by molecular dynamics (MD) simulations, as the tip drives the GNR parallel to the unreconstructed Au(111) surface at a low separation (z = 2 nm).Two characteristic lengths of 0.06 and 0.11 nm correspond to the lateral shift between three stable configurations (Fig. 3C). We estimated Δf as recorded in the AFM measurements by multiplying the force derivative [Fz(x, zz)–Fz(x, z)]/Δz by the conversion factor 0.15 N m–1z = 0.05 nm). The obtained profile (Fig. 3B) maintains the same periodicities of the manipulation curves, allowing comparison between simulations and experiments, although the relative heights of the peaks are different.

Fig. 3 Simulated sliding behavior.

(A) Lateral force Fx(x,z) (blue curve) and normal force Fz(x,z) (orange curve) while pulling the 6.28-nm GNR along its longitudinal axis at a distance z = 2 nm from an unreconstructed Au(111) surface. The force Fz has been also calculated at z = 2.05 nm (green curve), which allows us to estimate the frequency shift variation Δf(x) shown in (B). (C) Sketch of a generic row of C atoms in the GNR (black), showing that the atoms sit most of the time in three nonequivalent configurations marked as 1, 2, and 3 (also marked in other panels); these configurations give rise to a periodicity of ∼0.06 nm for short jumps from 1 to 2 or ∼0.11 nm for long jumps from 2 to 3 or 3 to 1 (Au atoms are shown in orange). (D) Tip trajectories on reconstructed Au(111) for the scan of panels and GNR configurations corresponding to minimum and maximum friction (the simulation cell size is 25.7 × 7.0 nm). Dashed lines represent the boundaries between the HCP and FCC regions, and rectangles represent the attached portion of the GNR during the scan (with red, green, and blue colors corresponding to increasing lateral force). In the background, dots indicate C atoms and surface, middle, and inner layers. (E to G) Frequency shift Δf(x) along the scan lines in (D). The corresponding lateral force profiles are shown in fig. S13.

The regular profile (Fig. 3B) is considerably modified by the herringbone reconstruction, which deforms the top Au(111) layer and makes it slightly wavy (0.02-nm corrugation). The commensurability degree between the GNR and the substrate is modulated correspondingly, and the same modulation appears in the friction (or frequency shift) profiles. We studied via our simulations the effect of the reconstruction starting at three different locations on the surface (Fig. 3, D to G). The friction force is reduced if the whole GNR lies on the face-centered cubic (FCC) or hexagonal close-packed (HCP) regions (red arrows), is still small when the GNR crosses the boundary between the FCC and HCP regions (green arrow), but increases and reaches a maximum value when the free edge of the GNR or the point of detachment from the substrate sits over the boundary between the FCC and HCP regions (blue arrows). The GNR short edge binds more strongly to the substrate than to the inner atoms, and this effect is more pronounced in the boundary regions between HCP and FCC, where the substrate structure becomes more commensurate with that of the GNR. We established further support for this behavior by means of additional measurements taken on different GNRs (fig. S10); it is also supported by theoretical studies on the pinning role of the nanostructure edges (24). The role of the herringbone reconstruction is confirmed by a similar experiment that we attempted on an unreconstructed Ag(111) surface. In this case, the GNRs merged and formed a moiré pattern with the substrate (fig. S14). Manipulation with force values similar to those used on Au(111) was not possible in this case.

Configuration 2 of the GNR (Fig. 3C) becomes unstable as the transitions from 1 to 2 and from 2 to 3 are suppressed. This leads to the half-periodicity that we sometimes observed while scanning at very close separations. Just before a “slip” from 1 to 3 or 3 to 1 occurs, the GNR becomes almost insensitive to the substrate, except for its short edge, which is still attached to the substrate (fig. S15). Because a C atom at this edge lies in a potential well U0 a few milli–electron volts deep, and it is essentially driven by the spring k = 1.5 N m–1 that connects the GNR to the tip apex [C-C bonds have an estimated stiffness of a few hundred newtons per meter (25)], we can apply a well-known result of the Prandtl-Tomlinson model for atomic-scale friction and estimate the characteristic parameter η = 4π2 U0/(ka2) (26). The resulting value of η is well below 1 and indicates a continuous transition between the two equilibrium states (this is valid strictly at the particular instant that we have considered). When the GNR is pinned in configuration 1 or 3 and is pulled by the spring at the same time, all C atoms in contact with the substrate oppose a certain resistance, but the overall value of the friction force remains very small (a few hundred piconewtons). Thus, our MD simulations are fully consistent with the commonly accepted interpretation of the superlubricity of graphene. Because of its exceptional lateral stiffness, this material is not prone to stretch and adapt to the substrate lattice while sliding. Combined with the weak interaction between graphene and the substrate, the resulting incommensurability leads to the almost frictionless sliding of the GNR.

We plotted Δf(x) curves at increasing separations z from the surface and also reversed the direction of motion (Fig. 4). Configuration 1 becomes unstable when z > 2 nm, and only the 3-to-3 transitions remain (corresponding to the more frequently measured periodicity of 0.28 nm). Lastly, we noticed that the forward and backward scan traces can be either in phase or in antiphase. MD simulations allow us to attribute this effect to the different bending of the suspended portion of the GNR in the two directions (Fig. 4). In the substrate regions with a large friction force, the bending of this portion can be much larger when scanning backward, thus leading to a delay in the slip events (fig. S16). If z = 5 nm (i.e., when the GNR is close to complete detachment), the agreement between the model and experimental results becomes weak; this is presumably due to the fact that the H atoms passivating the GNR edges, which are neglected in the MD simulations, start to play an important role at this point.

Fig. 4 Simulated frequency shifts at different pulling heights.

(A to D) Frequency shifts for forward and backward scans at z = 2, 3, 4, and 5 nm. Note that the half-periodicity disappears if z > 2 nm. The diagrams on the right show the different bending of the detached portion of the GNR in the forward (black) and backward (red) scans. The corresponding friction force loops are shown in fig. S16.

The pinning and releasing processes that occur in a sliding contact, as described here, are pivotal in the development of friction between two solid surfaces in reciprocal sliding (27). The GNR-Au(111) contact is almost superlubric, with static and kinetic friction force values in the range of 100 pN. The detailed dynamics of the sliding motion are nevertheless influenced by local surface properties, such as the variable degree of commensurability caused by the surface reconstruction. These details are clearly observable when the tip-surface separation z is small but tend to disappear as z increases and the bending (elastic) properties of the suspended piece of GNR become important. Our findings will aid in understanding and improving AFM-based nanomanipulation techniques and will motivate the design of novel nano-functionalized interfaces for friction control.

Supplementary Materials

Materials and Methods

Supplementary Text

Figs. S1 to S16

References (2844)


Acknowledgments: This work was supported in part by the Japan Science and Technology Agency PRESTO program, under the project “Molecular technology and creation of new function;” the National Center of Competence in Research Nanoscale Science program; grant CRSII2 136287/1 from the Swiss National Science Foundation; the Swiss Nanoscience Institute; COST (European Cooperation in Science and Technology) Action MP1303; the European Commission, under the Graphene Flagship (award no. CNECT-ICT-604391); the U.S. Office of Naval Research Basic Research Challenge program; and Comunidad de Madrid, under the MAD2D-CM (S2013/MIT-3007) project. The Partnership for Advanced Computing in Europe (project 2012071262) and the Empa high-performance computing facility Hypatia are acknowledged for computational resources. All data are tabulated in the main text and the supplementary materials.

Stay Connected to Science

Navigate This Article