Is graphene chemically inert?

G. Carraro, L. Savio, L. Vattuone


1 Introduction

Ever since it became clear that graphene is not the inert material that it was thought to be, the challenge has become how to control and tune its chemical activity, at least up to a certain extent. A significant amount of studies have developed on the interaction of graphene layers with different kinds of reactants (mainly gases, but also liquids). When investigating such interaction, one of the key parameters is the strength of the bond forming between the graphene layers and the adsorbate, the optimal value of which depends on the application envisaged for the material. For example, for the use of graphene as an active element in electronic devices it is necessary to open a gap in the band structure, but also that the Fermi level lies inside the gap; this is not straightforward, since charge donation may be present and shift the Fermi level out of range. Indeed, adsorption of several molecular species causes a band gap opening but only for a few of them (e.g. H2, tetracyanoquinodimethane −TCNQ− and its fluorinated derivative F4-TCNQ, tetracyanoethylene −TCNE− and fullerenes) the condition to keep the Fermi level within the gap is observed. The topic is quite complex since chemical modifications affect the effective mass of the charge carriers, thus reducing their mobility. The stability of the gap under ambient conditions, where of course also other species are present, is another quite important issue. Therefore, in practical applications a compromise between several factors must be found and, in general, strong chemical bonds are required to ensure the stability of the gap opening and of its value induced by adsorption. On the contrary, if the scope is to use graphene as a gas-sensing element, relatively weak molecular bonds are required to ensure the possibility to reset the sensor as easily as possible.

It is therefore clear that the interaction of graphene with other molecular species is of utmost interest for a variety of applications; for this reason, we shall focus here on the fundamental factors affecting the adsorption of simple molecules at graphene.

The first demonstration of the possible use of graphene as a very sensitive gas detector (2007) rapidly followed the discovery of graphene (2004). Schedin et al. demonstrated that adsorption of several simple molecules on graphene caused a change in the resistivity. The effect was explained considering that molecules such as NO2 and H2O act as electron acceptors and cause p-type doping while NH3 and CO act as donors and induce n-type doping. p-type doping causes a negative shift of the Fermi level, which falls below the vertex of the Dirac cone (the so-called K point in the Brillouin zone), while n-type doping induces the opposite effect. The change in resistivity, and hence the sensitivity to the gas, is higher for molecules with larger charge transfer (i.e. NO2 and NH3) and increases linearly with their concentration in the gas phase. However, the initial resistivity of bare graphene could not be recovered after removal of the gas and annealing to 150 °C (or, alternatively, exposure to UV Light) was needed to reset the sensor to the initial state. Such behaviour is indicative of a relatively high adsorption energy of the molecules and represents a significant drawback because it does not allow to operate the sensor at room temperature in a continuous way. It is also not compatible with the adsorption energy values predicted by Density Functional Theory calculations: 67 meV for NO2, 31 meV for NH3 and only 14 meV for CO. If molecules adsorbed so weakly at the graphene surface, their lifetime at the surface should be so short that a prompt recover of the initial resistivity should occur upon removal of the gas pressure already at room temperature. This contradiction stimulated further experimental studies and it was even found that the sensing activity is suppressed after removing the contaminants introduced using nanolithography.

The signal given by a certain concentration of molecules at a given temperature is determined indeed by the number of molecules adsorbed at the surface and by the charge transfer they induce. The charge transfer depends on the nature of the molecule, on its orientation with respect to the surface and on its adsorption site. The amount of molecules that can adsorb at a given pressure and temperature increases directly with their adsorption energy, which also determines whether the sensor can be re-set at a given temperature just by removing the gas or whether annealing or exposure to light is needed.

From a more fundamental point of view, it is important to establish whether gas molecules adsorb at regular or at defected graphene sites and whether physisorption or chemisorption takes place. In the former case, no chemical bond is formed between the molecule and the surface and the adsorption energy is ruled by van der Waals interactions: it is typically lower than 0.5 eV/molecule for light species and it increases with molecular weight. On the contrary, chemisorption implies the formation of a true chemical bond between the molecule and the surface, so that the adsorption energy is higher.

In this frame, we have investigated adsorption of simple molecules on supported graphene samples by using the standard tools of Surface Science. In particular, we have conducted our experiments under controlled Ultra High Vacuum conditions (UHV, i.e. for pressures lower than 10–9 mbar). UHV guarantees a negligible contamination of the surface over the time required to perform the experiment (a few hours) and allows the use of electron-based spectroscopies such as X-ray photoemission Spectroscopy (XPS) and High Resolution Electron Energy Loss Spectroscopy (HREELS) for chemical and vibrational analysis, respectively. In case of physisorption, the internal vibrational modes of the adsorbed molecule are almost unaffected by the interaction with the surface and their frequencies are close to those observed in the gas phase. When chemisorption occurs, on the contrary, the molecules are significantly deformed due to the interaction with the surface and the frequencies of some normal modes are significantly modified. In case of very strong interaction even dissociation of the molecule can occur thus providing radicals which adsorb at the surface and can react with other species. The latter step is especially relevant in heterogeneous catalysis, but it is not expected to occur on surfaces used for sensing purposes.

As outlined above, several factors can boost the reactivity of the graphene layer: doping, strain and interaction with the substrate. We report here about the recent activity performed at the joint CNR-IMEM and Physics Department group at the University of Genova about the role of the graphene-substrate interaction and of doping and defects.

2 Adsorption of carbon monoxide at graphene/Ni (111)

As a first point in our investigation of the interaction of simple gas molecules with supported graphene layers, we wanted to unravel the role of the support on the chemical reactivity of graphene. To this aim we have chosen to compare the adsorption of carbon monoxide (CO) on a single layer of graphene (G) grown on Ni (111) and on polycrystalline copper. The choice of G/Ni (111) is motivated by the lower graphene-substrate distance, which maximizes the potential role of the support, while graphene on polycrystalline Cu (G/Cupoly) is representative of a weakly interacting system and is widely used in applications. The choice of a poisoning molecule like CO as probe molecules is motivated by the weak charge transfer it induces: the sensitivity of sensors to its presence is thus still limited and improvements in sensitivity would be strongly welcome.

Figure 1 shows the comparison between HREEL spectra recorded after dosing CO both on G/Cupoly (left panel) and on G/Ni (111) (right panel). While no signature of CO adsorption is present on either sample at room temperature (RT), a vibrational loss at 256 meV is evident for G/Ni (111) at T =87 K. Such peak, corresponding to the internal CO stretch vibration, has a frequency lower than the one reported for gas phase and physisorbed CO (265 meV) and it is thus a clear fingerprint of CO chemisorption at mono-coordinated (i.e. on-top) sites. The loss starts to decrease in intensity upon annealing above 150 K and completely vanishes above 225 K. This temperature range for desorption under UHV conditions is indicative of an adsorption energy between 0.35 and 0.58 eV/molecule, much higher than expected for physisorption and indeed compatible with weak molecular chemisorption. The actual value depends on the amount of molecules present at the surface and usually decreases while approaching saturation.

Although relatively weak, the adsorption energy is much larger than the physisorption energy and thus guarantees that for nearly atmospheric pressure a significant equilibrium coverage of CO can be attained: at room temperature an equilibrium coverage of the order of 0.1 ML (i.e. one molecule of CO every ten surface atoms) is expected at the surface for a gas pressure of 10 mbar of CO. This allows to imagine a possible role of graphene on Ni as an active support for catalytic applications: graphene would not be the catalyst itself, but it would just stabilize a high enough coverage of molecules to supply reactants to catalytic nanoparticles deposited on it. The use of graphene/Ni as a sensing element is on the contrary not trivial because of the metallic nature of the substrate and also because of the disappearance of the Dirac cone for such a reactive substrate.

We still need to answer the key question whether CO molecules adsorb at defect or at regular graphene sites. To do that, we dosed 1 L of CO on the G/Ni (111) sample at 87 K and we monitored the surface morphology by LT-STM. The outcome is reported in fig. 2, in which the clean graphene layer and the CO-covered sample are compared. While panel a) and the corresponding inset show the symmetry of the clean G layer, panel b) shows that after CO exposure the surface is completely covered by the adsorbate. CO-admolecules organize in short rows along the $\langle \bar{2}11\rangle$ direction at a minimum distance of two lattice spacings in both directions, as determined by the line scans (panel c) and as schematized in the drawing of panel d).

For sake of clarity we have to mention that the CO coverage is not uniform over the whole G sample and that some patches not covered by CO (not shown here) can indeed be found upon extended STM inspection. This is explained by the fact that different domains can form for G/Ni (111), depending on the relative position of the C atoms of the G layer with respect to the topmost layer of atoms of the Ni substrate. Such domains are usually addressed as top-fcc, top-bridge and top-hcp, in a clearly self-explaining nomenclature. Ni2C can also form as alternative product in the graphene growth process. All these phases can coexist on the surface and their relative amount depends on the preparation conditions and on post-growth treatment. Careful analysis of the XPS spectra enabled us to correlate the reactivity of the graphene layer with the different amount of each graphene phase and of nickel carbide. We proved that the top-fcc phase is the most reactive one, while the presence of Ni2C and/or of carbon dissolved in the subsurface region has a poisoning action with respect to the adsorption of carbon monoxide.

3 The effect of nitrogen doping on chemical reactivity Theoretical studies suggest that a significant increase of the adsorption energy, which correlates with changes of the electronic and chemical properties, can be obtained by doping with donors and acceptors atoms.

In order to address the role of doping, we introduced heteroatoms in the graphene lattice by bombarding a single layer of G/Ni (111) with 110 eV N2+ ions before exposing it to CO. It is already proved in the literature that such process leads to the formation of a doped G layer in which N atoms substitute some of the C atoms; due to the different external electronic configuration on N with respect to C and hence to the ability of N to form only three 3 bonds instead of four, N can assume a graphitic, pyridinic or pyrrolic configuration, thus creating defect sites in the perfect hexagonal G lattice.

In our experimental conditions we produced n-G/Ni (111) with a significant doping (around 11%), in which all the N atoms were in pyridinic or pyrrolic configuration, according to XPS analysis. When probing this layer towards CO adsorption, we found the appearance of an additional CO-related peak in the HREEL spectrum (see fig. 1). Such moiety is characterised by a CO stretch frequency of 238 meV, which is usually considered as a fingerprint of doubly coordinated chemisorbed CO, and it is still present at 190 K, desorbing completely only around 240 K. On this basis it is possible to estimate an adsorption energy of 0.50–0.85 eV/molecule, depending on coverage, i.e. even higher than the one of 0.34–0.54 eV/molecule estimated for mono-coordinated (i.e. on-top) CO at regular G sites. This result unambiguously proves the positive effect of N doping on graphene chemical reactivity, at least for G/Ni (111).

4 The role of vacancies

Many theoretical studies suggest an increased heat of adsorption (up to 2.33 eV/molecule) for CO at vacancies of free-standing graphene. In order to test this prediction, we have introduced vacancies in the G/Ni (111) and G/Cupoly layers by low-energy (150 eV) sputtering with Ne+ ions impinging at normal incidence. The samples were then exposed to CO at room temperature. Vibrational spectra (top spectra in fig. 1) clearly indicate that no CO adsorption occurs on defected G/Cupoly, while on defected G/Ni (111) CO-related vibrational losses at 236 and 253 meV are present. The frequencies are however different with respect to those observed for pristine G/Ni (111) and closer to those reported for CO on Ni (111). In addition, the desorption temperature of CO is quite similar to the one reported for CO on bare Ni (111). Therefore, the experimental data indicate that CO has not adsorbed at the graphene vacancy but rather that it has intercalated below graphene and adsorbed at the underlying Ni substrate. This interpretation is further supported by XPS analysis of the same system, which suggests that graphene layer partly detaches from the substrate upon ion bombardment, thus favouring intercalation. Notably this phenomenon is not observed for G/Cupoly because the adsorption energy of CO is much lower on Cu than on Ni, so CO does not adsorb on Cu under the investigated experimental conditions.

This experiment proves that theoretical results for free-standing graphene are not adequate to extrapolate the behaviour on supported layers: the substrate always plays an important role. Indeed, more recent DFT calculations simulating the behaviour of a single vacancy at graphene supported on Pt and on Cu show that the dangling bonds of the C atoms adjacent the vacancy saturate towards the metal substrate. In the case of the more reactive Pt surface, they are no longer available for the following reaction with water. Due to the reactivity on the Ni substrate, a similar scenario is envisaged also for the CO/G*/Ni (111) system.

Artificially created vacancies could be in principle also active sites for catalysis and for sensing. For both applications it is however essential for adsorption and desorption to occur reversibly. We have thus checked whether CO intercalation through vacancies is a reversible process by repeating several adsorption-desorption cycles. The experiment indicates that the amount of adsorbed CO decreases at each cycle, which suggests a modification of the vacancy sites upon interaction with the gas. In order to understand what is happening, we have studied the process by LT-STM. Figure 3 shows the morphology of the G/Ni (111) surface after sputtering (A), after exposure to CO (B) and after annealing to 400 K (C). Sputtering creates lines of defect on the surface, the aspect of which is modified upon CO exposure. CO desorption induced by annealing causes a partial healing of the vacancies.

This result implies that CO molecules have reacted and released C atoms which repair the vacancy: at the next adsorption cycle the number of possible holes for intercalation is thus lower, accounting for the reduction in the amount of adsorbed CO.

We can speculate that a Boudouard reaction (catalysed by the Ni substrate) between two intercalated CO molecules has taken place leading to CO2 and C: the former species desorbs and the latter has remained to fix the vacancy. This explanation is consistent with the scenario of chemistry under cover: graphene is not directly involved in the chemical reaction but just forces the reactants adsorbed at the substrate underneath to meet and react. The presence of the cover can favour the reaction both due to the constraining effect of the graphene layer and to the lowering of the adsorption energy at the catalyst, i.e. at the support under graphene. The lowering of the adsorption energy leads to a decrease of the activation barrier for the reaction thus resulting in a higher turnover frequency than observed for the same reaction without cover. Interestingly the mechanism is not peculiar to graphene as also hexagonal boron nitride layers have been shown to promote reactivity in a similar manner.

4 Conclusions and perspectives

Our recent studies have shown that the substrate plays a key role in tuning the reactivity of the graphene layer. The strongly interacting G/Ni (111) system seems more promising for its possible role in catalysis than for sensing applications: indeed the substantial increase in the adsorption energy at the surface enables to use graphene as an active support for reactants while reactions can occur or under cover or by suitable nanoparticles deposited above graphene.

The recent theoretical suggestion about the effect of the local curvature of graphene on its reactivity suggests also a possible use for hydrogen storage. Until now hydrogen adsorption on graphene has been obtained only by using an atomic hydrogen source but the possibility to control the strain of supported graphene might allow to obtain dissociative adsorption at the strained region and to release molecular hydrogen when the strain is removed.