Single Crystalline Metal Films as Substrates for Graphene Growth
Abstract
Single crystalline metal films deposited on YSZ-buffered Si(111) wafers were investigated with respect to their suitability as substrates for epitaxial graphene. Graphene was grown by CVD of ethylene on Ru(0001), Ir(111), and Ni(111) films in UHV. For analysis a variety of surface science methods were used. By an initial annealing step the surface quality of the films was strongly improved. The temperature treatments of the metal films caused a pattern of slip lines, formed by thermal stress in the films, which, however, did not affect the graphene quality and even prevented wrinkle formation. Graphene was successfully grown on all three types of metal films in a quality comparable to graphene grown on bulk single crystals of the same metals. In the case of the Ni(111) films the originally obtained domain structure of rotational graphene phases could be transformed into a single domain by annealing. This healing process is based on the control of the equilibrium between graphene and dissolved carbon in the film. For the system graphene/Ni(111) the metal, after graphene growth, could be removed from underneath the epitaxial graphene layer by a pure gas phase reaction, using the reaction of CO with Ni to give gaseous Ni(CO)4.
1 Introduction
Experiments on the extraordinary properties of graphene have mostly been performed with samples exfoliated from graphite.1, 2 However, to make use of these properties for electronic applications upscalable preparation methods have to be found, a condition not fulfilled by the exfoliation technique. An alternative method that has been intensively studied is the synthesis of graphene by chemical vapor deposition (CVD) of hydrocarbons on transition metal surfaces.3-5 It has been shown that high-quality graphene can be prepared in this way, but to be suitable for electronic devices, a metal-based method has to fulfill rather extreme conditions: (i) The graphene has to be basically defect-free to provide high charge carrier mobilities; (ii) the graphene layer should have a controlled azimuthal orientation, e.g., if nanoribbons with defined edge terminations are to be made; (iii) when the graphene film, after growth by CVD, has to be detached from the metal and transferred to an insulating support, this has to happen without creating defects; (iv) and finally, the method has to be implemented into fabrication processes in an economic way to fulfill the scalability condition. In combined form, these goals have not yet been reached. Attempts to avoid the metal substrates and directly grow graphene by CVD on insulators or semiconductors, such as SiC, SiO2, AlN, h-BN, mica, Si or Ge were partially successful. However, in terms of grain size, number of layers, defect density, and azimuthal orientation the graphene flakes grown on these substrates, possibly except for the cases H-terminated Ge(110) and SiC, did not yet reach the quality grown on metals.6-9
Most CVD experiments on metal surfaces have been performed with bulk single crystals or with polycrystalline metal foils. Bulk single crystals provide a defined surface orientation and thus epitaxial control, so that defined lattice orientations can be obtained and grain boundaries may be avoided.3, 4 However, methods based on bulk crystals are not economically viable for the majority of metals of interest. In contrast, polycrystalline metal foils represent sacrificial supports that can simply be etched after graphene growth, fulfilling the low cost condition,10-14 but they do not provide control of the graphene orientation. An obvious idea is to combine the advantages of these two substrate types by using thin single crystalline metals films, and there have been several attempts following this notion.15-24
We here report about a project in which we have investigated a special, high-quality type of single crystalline metal films as supports for graphene growth. The films consist of single-crystalline, 100 to 150 nm thick layers of various metals on 4 inch Si(111) wafers, thus fulfilling the conditions of scalability. To prevent silicide formation, the metal films are separated from the Si support by 40 to 120 nm thick buffer layers of yttria-stabilized zirconia (YSZ). The metal films display low mosaic spreads (table S1 in the Supporting Information) that correspond to a crystalline quality comparable to standard metal bulk single crystals. The preparation of these films has been described in detail previously.25
The growth of graphene was studied with films of Ir(111), Ru(0001), and Ni(111) that represent different classes with respect to the interactions with the graphene layer. The graphene-metal bonding type ranges from weak, physisorption-like to almost chemisorption-like, which in turn affects the epitaxial growth. Ir(111) belongs to the weakly, Ru(0001) to the strongly interacting class,3, 4 and in both cases the lattice mismatch of the hexagonal metal with the graphene leads to characteristic moiré structures.26, 27 Ni(111) also belongs to the strongly interacting class3, 4 but appears structurally simpler as the lattice constant is almost identical to that of graphene. On the other hand, nickel is also more complicated, as nickel carbide can form in a competing reaction, and significant amounts of carbon can dissolve in the bulk.28, 29 The different chemical properties of the three metals also play a role for possible detachment methods. Films of Cu, the most common metal substrate for graphene growth, were not investigated, mainly because the quality of the Cu films on YSZ/Si(111) has so far been insufficient. Moreover, graphene flakes on Cu, although often large, are not azimuthally well-oriented,30 so that the advantage of the single crystalline films would not become operative.
In the following, we present a résumé of results on the surface quality of the films, on the quality of the epitaxial graphene grown on these films, and, in the case of Ni(111), on a new detachment method. Individual results have been published in detailed previous reports,15-17 but also new data will be shown.
2 Experimental
After preparation of the metal films by the established method,25 the wafers were taken out of the preparation chamber and cut into ∼10 mm x 10 mm pieces. These samples were then introduced into one of two ultra-high vacuum (UHV) chambers for further preparation and to perform the graphene growth experiments.
One of the chambers (base pressure ≤ 1.3 × 10−10 mbar) was equipped with a home-built scanning tunneling microscope (STM),31 an Auger electron spectrometer (AES), and a four grid low-energy electron diffraction (LEED) optics. The second chamber was a SPECS FE-LEEM P90 system (base pressure 2 × 10−10 mbar). In this apparatus low-energy electron microscopy (LEEM) and photoemission electron microscopy (PEEM) experiments were performed. A LOT LSB610 Hg arc lamp served as light source for PEEM. Sample temperatures were measured with an IR pyrometer. To remove the initial contamination layers from the samples and improve their surface quality the samples were first cleaned by Ar+ ion sputtering (10 min; 1 keV; ∼8 μA) and by annealing for 1 – 2 h at 800 – 900°C. The Ru(0001) films were also annealed at 1050°C. For further cleaning the samples were repeatedly sputtered for 5 min with Ar+ ions and annealed at 700 - 800°C for 2 min. Graphene was grown by decomposition of ethylene at temperatures between 600 and 850°C (675 - 950°C for Ru) and at pressures in the range between 10−9 and 10−7 mbar. Typical ethylene dosages for full monolayers were 20 – 40 L (1 L = 1.33 × 10−6 mbar · s) corresponding, in the range of applied pressures, to deposition times between 10 min and 6 h.
In a third UHV chamber X-ray photoelectron spectroscopy (XPS) measurements were performed, using a VSW TA10 X-ray source providing non-monochromatic Mg Kα radiation and a VSW HA100 hemispherical analyzer. For Raman spectroscopy a HORIBA Jobin Yvon HR800 UV spectrometer equipped with a HeNe laser (λLaser = 633 nm) was used.
3 Results and Discussion
3.1 Morphology of the Metal Films
To analyze the morphology of the pristine metal films, STM data were taken (before performing extensive further preparation steps). Figure 1a shows an STM image of an Ir(111) film that had been annealed at 400°C in 2.7 × 10−7 mbar of O2 to oxidize contaminants (a clean surface was confirmed by AES). The annealing temperature was considerably lower than the film growth temperature range of 600 - 700°C to preserve the original morphology. The main features in the STM image are monoatomic steps. Their density is relatively high, corresponding to an average terrace width of only a few hundred angstroms. The black features are holes in the metal film, with typical diameters of about 400 – 900 Å and densities of 7 – 20 μm−2. The depths of the holes could not be determined by STM because of the finite tip diameter. They most likely result from an incomplete coalescence of the grains during the original growth of the films. A Ni(111) film was at first only annealed at 400°C in UHV to remove a contamination layer. The film then showed similar holes (Figure 1b), and the overall height modulation was even stronger than for Ir(111). Similar holes have also been reported for the Rh(111)/YSZ/Si(111) film system.32 The as-grown, cleaned Ru(0001) films (5 min Ar+ sputtering, followed by annealing at ∼550°C in UHV) also displayed holes, but of considerably larger diameters, and the holes were more elongated, forming trench-like structures (Figure 1c).
The strong height modulations of the as-grown films and the deep holes would certainly be detrimental for the quality of the graphene grown on these films. It was therefore attempted to improve the film surface morphologies prior to graphene growth by annealing at considerably higher temperatures in UHV than during preparation. Figure 1d–f shows the results. In all three cases the holes, and even the massive trenches in the Ru(0001) case, vanished completely. Also the height modulations were strongly reduced, in particular for the relatively rough Ni(111) and Ru(0001) films, and quite flat surfaces were obtained. For the annealed Ni(111) films (Figure 1e), the height modulation after annealing was only 10 atomic layers, corresponding to approximately 20 Å, on a length scale of 3000 Å. Previous observations with Ni(111) films on other supports, of new holes formed by annealing or even of a complete loss of the metal were not confirmed.23, 33, 34 The Ir(111) films (Figure 1d) were even flatter than the Ni(111) films; the Ru(0001) films (Figure 1f) remained somewhat rougher and showed quite a high density of screw dislocations.
However, in particular in the case of the Ni(111) films, the annealing led to a characteristic triangular pattern of straight lines crossing each other at 60°/120° angles (Figure 1e). In less clear form the pattern was also visible before the annealing (Figure 1b). The Ir(111) films also showed straight lines but at a lower density; on the Ru(0001) films no lines were observed. [Similar straight steps have also been reported for Rh(111) films on YSZ/Si(111) supports.]32, 35 The lines run along the close-packed directions of the films. They are formed by atomic steps of exactly the same height as ordinary monoatomic steps on bulk single crystals. In contrast, atomic steps on well-annealed, close-packed surfaces of bulk single crystals do not show preferential directions, and the enclosed terraces have a round shape. Such "ordinary" steps can also be seen on all three film materials (Figure 1d–f), but for Ni(111) and Ir(111) they are crossed by the additional straight steps.
LEEM measurements during annealing provided insight into the origin of the straight step pattern. Figure 2 shows data from a Ni(111) film partially covered by graphene [the islands at the left and right edges of the field of view (FOV)]; the center area shows a stripe of uncovered nickel. The first image (Figure 2a) was taken at 486.5°C. The following images (Figures 2b and c) were consecutively recorded at 0.6 s intervals during cooling of the sample (the full data set is shown as movie in the Supporting Information). The dark lines on the uncovered area are monoatomic steps, and the first image exclusively shows steps of the "ordinary" type as they would be observed on a Ni(111) bulk crystal. In the next image (Figure 2b) three straight steps have appeared (orange arrows), oriented at 60°/120° angles with respect to each other and crossing the existing "ordinary" steps. In the next image (Figure 2c) the sharp intersections between the two step types have started to round off (orange cycles), i.e., the straight and the "ordinary" steps merged. Obviously, at this elevated temperature the metal atoms are sufficiently mobile that the energetically unfavorable sharp intersections could be replaced by smooth connections between the two step types. This can explain why, during annealing at high, constant temperatures, the films did not show the straight steps but only the "ordinary" steps. When the temperature was lowered to below approximately 400°C, the straight steps, formed during the temperature change, did no longer disappear because the surface mobility of the metal atoms became too low. The LEEM image taken from the same location after cooling to room temperature (Figure 2d) accordingly showed a high density of straight steps. Under these imaging conditions one could also see that the graphene islands displayed the same triangular pattern as the uncovered Ni, and by closer inspections one finds that the straight steps crossed the borders between the empty and the graphene-covered areas. The dark phase in Figure 2d is Ni carbide that was formed by segregation of dissolved carbon at temperatures below 400°C (see also Figure 6).
glide planes.
The model shown in Figure 2e explains these observations. The line pattern is a result of the much higher thermal expansion coefficient of the metal than of the silicon substrate (Figure S1a). When the temperature was reduced after annealing tensile stress built up in the metal layer that was released by slipping along glide planes in the metal film. For the fcc metals Ni and Ir the main glide planes are of the 
family.36 Only the planes tilted with respect to the (111) surface participate in the glide processes as schematically shown in Figure 2e. Slipping along these planes allows the film to release stress parallel to the film plane. Each slipping event by one lattice constant leaves the bulk lattice intact but creates a monoatomic step at the surface. Because of the orientations of the glide planes in the film the newly formed surface steps are straight along close-packed directions, they cross existing surface steps, and they cross each other at 60°/120° angles, all as observed by LEEM. That only monoatomic steps were formed can be explained by the adhesion of the metal film to the substrate creating a restoring force that prevented larger shifts in one step. Slip lines were also observed when the temperature was increased which can be understood analogously.
To further test this model the density of slip lines was determined as a function of temperature for the Ni(111) case. The model predicts that the difference between the thermal expansion coefficients of the metal and the silicon should be balanced by a defined number of slip lines formed in a given temperature interval. Figure S1b shows that this was in fact the case. That the density of slip lines was lower for the Ir(111) films (Figure 1d) is explained by the lower thermal expansion coefficient of Ir, which is much closer to that of Si (Figure S1a). The fact that no slip lines were observed on the Ru(0001) films (Figure 1f) also fits into the picture as the main glide plane of the hcp metal Ru is the (0001) plane that cannot release thermal stress as it is parallel to the film plane.37
The formation of steps underneath the graphene layer during cooling to room temperature apparently did not create defects in the graphene layers grown on these films. The plastic deformation even seemed to be favorable since in this way the metal film could follow the thermal expansion of the Si support which is much closer to that of graphene. As a consequence, we have never observed wrinkles in the graphene layers grown on the metal/YSZ/Si substrates, which is often a problematic effect for graphene on bulk metal substrates caused by the strongly different thermal expansion coefficients.38
3.2 Graphene Growth on the Metal Films
On the Ru(0001) films highly ordered graphene monolayers were successfully grown by CVD of ethylene at temperatures between 675 and 950°C and at ethylene pressures in the range of 10−9 – 10−7 mbar. These parameters correspond to the typical growth parameters reported for bulk single crystals.39-42 STM and LEED data are shown in Figure 3. The STM image in Figure 3a shows a hexagonal superstructure with a periodicity of (31.4 ± 0.8) Å, in good agreement with the (11.5 × 11.5) structure observed for bulk single crystals. (The actual superlattice has a twice as large, (23 × 23), period, but the height modulation resolved by STM corresponds to half of this period.43, 44) The superstructure results from the moiré effect caused by the superposition of the graphene and the Ru lattices that have slightly different lattice constants. The atomic structure within the moiré unit cell was also resolved by STM (Figure 3b). In these higher-resolved images one can also identify the typical three contrast levels in the unit cell (bright, medium, and dark) that are caused by the different positions of the carbon atoms with respect to the underlying Ru atoms.26, 43) Moiré superlattices amplify misorientations between two lattices,45 e.g., in this case, a rotational disorder would be amplified by a factor of ∼10. That Figure 3a shows a perfectly ordered moiré structure on a large scale (Figure 3a) thus also proves a perfect alignment on the atomic scale. The LEED pattern showed sharp diffraction spots of the moiré superlattice and an exact alignment of these spots with respect to the spots from the Ru lattice (Figure 3c). The graphene was thus also well-ordered on the greater length scale of the LEED experiment. That on Ru(0001) well-ordered graphene could be obtained in a relatively straightforward way can be explained by the fact that Ru belongs to the strongly interacting class of metals, so that there is a considerable driving force for an oriented binding of the graphene to the metal. With the Ru(0001) films the same degree of epitaxial control as for Ru bulk single crystals was achieved.
Iridium belongs to the weakly interacting class of metals, and accordingly graphene often forms rotational domains on Ir(111).4, 15, 46 In order to exclusively obtain lattice-aligned graphene more efforts were needed. From experiments with bulk Ir(111) it is known that at high growth temperatures, 1257°C at an ethylene pressure of 5 × 10−6 mbar, exclusively the lattice-aligned structure can be prepared.47 However, these conditions were not accessible with the Ir(111) films; at approximately 1100°C the films started to dewet from the support (Figure S2).15 It was therefore attempted to find other conditions under which lattice-aligned graphene could be grown. Figure 4a shows results of experiments performed over a wide range of temperatures in which also the ethylene pressure was varied. Green triangles represent cases in which exclusively lattice-aligned graphene was obtained; red dots represent cases of mixtures of rotated phases. One can see that within most of the tested parameter space rotated graphene was obtained. However, at temperatures around 800°C and at ethylene pressures ≤ 1.3 × 10−8 mbar exclusively lattice-aligned graphene was formed. (The two green data points at 650 and 700°C we regard as outliers). By lowering the ethylene pressure and accordingly the graphene growth rate it was thus possible to reduce the temperature to limits allowed by the stability of the films.
A LEED pattern of graphene grown at 700°C and 2.7 × 10−8 mbar ethylene (i.e. in the parameter range of mainly rotated domains) is shown in Figure 4b. One can see an almost continuous ring with a radius corresponding to the graphene lattice constant, indicating an almost random spread of azimuthally rotated graphene (Figure 4b). Discrete spots on top of the ring showed that structures rotated by 0°, 19°, and 30° were preferentially formed that represent known graphene phases on bulk Ir(111).46
Under optimized conditions (800°C and 6.7 × 10−9 mbar ethylene), the LEED pattern only showed one graphene structure on the Ir(111) films. Figure 4c shows such a case, with sharp diffraction spots and a parallel alignment of the graphene (and moiré) spots with respect to the substrate spots. The spot positions correspond to an approximate (9.07 × 9.07) structure, in reasonable agreement with the lattice-aligned, incommensurate (9.32 × 9.32) superstructure known from bulk Ir(111).27 Atomically resolved STM images confirmed this known structure (Figure 4d). Therefore, using optimized conditions, well-ordered aligned graphene could also be obtained on the Ir/YSZ/Si(111) system.
For the Ni(111) films LEED (not shown here, see Ref. [16]) and STM data (Figure 5) showed that under well-controlled conditions [680°C and 2 × 10−9 mbar ethylene, followed by annealing at 750°C for 45 min (see Figure 8)] a (1 × 1) graphene structure could be grown. A (1 × 1) structure is expected from the almost identical lattice constants of graphene and Ni(111), and has been reported for bulk Ni(111) single crystals 48 as well as for thin Ni films.16, 17, 23, 49 Figure 5a shows two terraces covered by the (1 × 1) structure and a coherently overgrown atomic step. That steps are overgrown without defects is an important factor for a macroscopically coherent graphene layer and has been reported for epitaxial graphene on other metals before.5 However, some of the STM images showed point defects as shown in Figure 5b. Yet, at other tunneling voltages they were completely invisible (Figure S3), and their concentrations were generally much higher than the defect concentrations derived from Raman spectra (after delamination of the graphene layers). These are strong indications that most of these STM features were defects in the Ni substrate rather than in the graphene layer.16
However, this structural simplicity of the Ni(111) case is superimposed by complications, namely by the high solubility of carbon in nickel 29, 50, 51 and by the competing growth of a Ni2C surface carbide.52, 53 Also the easy formation of rotated graphene domains on Ni(111) is a well-known difficulty with this substrate.49, 54 Reasons suggested for the formation of rotated graphene domains are the transformation of Ni2C into graphene,55 the presence of oxygen,18 growth temperatures above 650°C 54, and a high concentration of dissolved carbon at the beginning of the growth.56 (It has also been reported that the transformation of Ni2C resulted in aligned graphene.53)
We confirm that these complications known from bulk Ni(111) are also present for the Ni(111) films. In some experiments striking non-reproducibilities were found. For instance, under nominally identical CVD conditions, 690 – 700°C, 2.7 × 10−8 mbar C2H4, and ethylene doses of 27 L, in some experiments a graphene layer was formed, in other experiments only the carbide phase. The latter was in contrast to expectation as the carbide is only stable below 460°C 57 and should not have formed under these conditions.
To analyze these discrepancies AES measurements were performed during CVD. To measure the formation of surface carbon during the dosing the peak-to-peak intensity IPP of the C KLL transition was recorded (Figure 6). However, in the first run (500°C, ethylene pressure 4 × 10−9 mbar) (Figure 6a) no carbon signal was detected during the entire dosing time of 140 min (corresponding to an exposure of 25 L). The ethylene valve was then closed, the chamber was pumped down to UHV, and the sample cooled down. Immediately a carbon signal appeared which then remained constant during further cooling. The peak shape of the C KLL signal (Figure 6a inset) indicated that nickel carbide had formed.52, 57 The nickel carbide layer was then removed from the surface by one cycle of Ar+ sputtering. Then ethylene was dosed in a second run (500°C, 8 × 10−9 mbar C2H4) (Figure 6b). At first, again no carbon signal was detected. However, after about 40 min (∼15 L), a carbon signal appeared that continuously increased during further dosing. The peak shape of the C KLL transition (Figure 6b inset) showed that this time graphene had formed.
The model shown in Figure 6c explains these observations. In the beginning of the experiment, the as-grown Ni film was virtually free of carbon. Due to its high solubility in nickel, the carbon formed by dissociation of the ethylene molecules during the CVD process at first completely dissolved in the metal film (center sketch).53, 58 The surface remained free of carbon (as detected by AES) as long as saturation in the bulk was not yet reached. In the first run, saturation had not been reached at the end of the dosing, but when the sample was cooled the carbon solubility dropped with decreasing temperature, and the excess carbon segregated to the surface. This led to the carbide phase rather than to graphene, an effect explained by the kinetics (Figure S4). At the typical fast cooling rates the temperature quickly dropped to the stability range of Ni2C, so that the carbide phase formed. A similar observation of Ni2C segregation during cooling, but underneath already existing, rotated graphene, has been reported before.53 In the second CVD run the bulk of the film was already partially saturated with carbon, so that during dosing the saturation point was reached and surface carbon built up at a temperature (500°C) in the stability regime of graphene. Hence a graphene layer grew on the surface. This model can also explain the apparent non-reproducibility in many Ni experiments that can be traced back to the different pretreatments of the Ni samples, namely to the varying amounts of dissolved carbon from previous experiments. In this respect, the thin Ni films possess a strong advantage over bulk single crystals. Bulk Ni crystals practically represent infinite reservoirs for the dissolution and segregation of carbon atoms, making these processes extremely hard to control. By contrast, the amount of dissolved carbon in the small volumes of the Ni films is finite, of the order of half a graphene layer, and could be used as a controllable parameter (s. below).
To get information about the quality of the graphene layer on a larger scale than usually investigated by STM, movies were recorded by LEEM and PEEM during growth of graphene on Ni(111) films. Figure 7a–c shows three LEEM images from such a movie, taken during CVD at approximately 650°C. (The full movie is available in the Supporting Information.) The first image shows the uncovered Ni(111) film, with thin dark lines most likely representing monoatomic steps, and thicker lines representing step bunches (Figure 7a). The straight steps seen in Figure 2 were absent because the temperature was constant and higher than 400°C. After an induction period, in which the Ni film became saturated with dissolved carbon, monolayer graphene (the bright area at the left side of Figure 7b) grew into the FOV of the experiment. The data show that the graphene layer grew across atomic steps and step bunches in a coherent fashion, following the well-known "carpet mode".59 In the experiment, the ethylene valve was closed after Figure 7b and the chamber was pumped down. The position of the graphene island edge at this time instant is indicated by the red line in Figure 7c. However, the dropping ethylene pressure was still high enough that the graphene layer could grow by some further distance (Figure 7c). Inspection of Figure 7c shows that the decreasing growth rate at the end of the dosing affected the quality of the graphene. The graphene area on the left side of the red line displays some internal contrast, whereas the area on the right side, grown at lower and finally vanishing speed, appears homogenous. Using PEEM, the contrast variations were resolved much more clearly (Figure 7d, showing a larger FOV from the same area). Here the bright areas (graphene) clearly display two grey levels, indicating different domains; the black area is the uncovered metal. By taking local diffraction images with the LEEM apparatus it was shown that the contrast variations within the graphene layer were caused by different rotational domains (light grey: lattice aligned graphene; medium grey: rotated graphene). Bi- and multilayer graphene was excluded by analysis of the I(V) curve of the (0,0) spot measured with LEEM. Hence, even at relatively low ethylene pressures (in the experiment: 4.0 × 10−9 mbar up to Figure 7b), rotational domains were formed. The resulting high density of grain boundaries is, of course, counterproductive for applications which require high charge carrier mobilities. By further reducing the pressure, like in Figure 7c, the growth clearly became more homogeneous. Nevertheless, even then the desired, lattice-aligned graphene, with its thermodynamically stable (1 × 1) structure, was not obtained (as concluded from the contrast in PEEM). However, the low growth speed of graphene prevented the formation of new, rotated grains. (Growth rates determined from the edge velocities in LEEM, at 10−9 mbar of ethylene and 600 - 700°C, were between 5 and 7 nm/s for aligned and 20 to 60 nm/s for rotated graphene, in agreement with previous reports.54)
We found, however, that the special properties of the thin films could be used to significantly improve the quality of the graphene layer grown on Ni(111). Figure 8 shows an experiment in which a graphene-covered Ni(111) film was annealed at selected temperatures, and the changes were monitored by PEEM.17 The first PEEM image (Figure 8a) shows the state of the sample as obtained from the CVD process. In this case 1.2 layers of graphene had grown, i.e., some bilayer had already formed. One can discriminate three contrast levels caused by the different work functions of the graphene phases. Light grey represents the lattice-aligned monolayer with its (1 × 1) structure, medium grey the rotated monolayer, and dark grey the bilayer (for better visibility the domains are also shown in color-coded form in Figure 8e). LEED confirmed the presence of rotated graphene domains in the beginning (inset). When the temperature was increased from 617°C to 700°C a surprising dynamics set in (Figure 8b–d, the full movie is shown in Ref. [17]). The bilayer areas started to dissolve (light blue in the color-coded Figure 8e). At the same time small holes in the monolayer areas formed (black in the original data, green in the color-coded presentation). The holes started to travel through the graphene layer in such a way that they moved through the areas of rotated graphene (dark blue). The areas through which the holes had moved were transformed into lattice-aligned graphene (orange). After a further increase of the temperature to 730°C (not shown) and a subsequent reduction to 700°C these processes had completely transformed the original multi-domain layer into a single domain of lattice-aligned monolayer graphene (Figure 8d). Only a few of the small holes remained. Subsequent cooling to 300°C (Figure 8f) did not change this situation, and in particular no carbon resegregated to form bilayer islands. The (1 × 1) LEED pattern recorded afterwards confirmed the exclusive presence of the lattice-aligned graphene (inset). It is obviously possible to heal the original defect-rich graphene on the Ni(111) film by an annealing program, a surprising result that certainly will be important for the future development of controlled synthesis strategies of perfect graphene monolayers on Ni(111).
It was shown that the healing process is based on the equilibrium between surface carbon and bulk-dissolved carbon.17 Controlling this equilibrium by the temperature was only possible because of the small volume of the film and thus the finite amount of dissolved carbon. The driving force for the "travelling holes" was the greater thermodynamic stability of the (1 × 1) structure compared to the rotated structures. Of course, the kinetics may also play a role. However, it has previously been reported that graphene starts to dissolve in a Ni film [grown on a W(110) substrate] at T ≥ 657°C because then the rate becomes sufficiently fast,60 indicating that the healing experiments, which were performed at higher temperatures (Figure 8), were not limited by the kinetics. That carbon did not resegregate upon cooling to form the bilayer can be explained by the stability of the (1 × 1) structure that prevents formation of a second layer underneath (that bilayers only segregate under rotated domains has been reported before.54). It has also been suggested that the near-surface range of the Ni has become depleted of carbon.56
3.3 Detachment of Graphene From the Metal Films
We have also performed extensive experiments trying to find methods to detach the graphene layers from the metal films and transfer them to insulating supports, an essential step if the graphene is to be used for electronic devices.61 Before the detachment processes none of the Ru, Ir, and Ni samples showed Raman signals of graphene (consistent with the exclusive presence of lattice-aligned graphene monolayers 59, 62). For the Ru(0001) films wet chemical etching methods were tried, using solutions of ceric ammonium nitrate (like in Ref. [24]) and of sodium hypochlorite, but in all cases the graphene became substantially damaged. Ir is even less chemically reactive than Ru, so that we tried etching the Ir(111) films electrochemically by high AC voltages, but no conditions were found under which the graphene was not oxidized at the same time. For the Ir(111) films the known bubbling transfer method 63, 64 worked, although the Raman spectra taken thereafter showed a relatively high D band.
The reaction, first described in 1890 by Ludwig Mond65 and currently used on an industrial scale to refine nickel,66 takes place at 75°C and CO pressures of ∼1 bar in the presence of sulfide catalysts. The idea of using this reaction for etching was that the conditions are very mild and that only gaseous components are involved - Ni(CO)4 is gaseous under the reaction conditions. Problems with wet chemical etching methods were avoided, namely that residues of the etching agents remain on the graphene layers and that water becomes incorporated during transfer. Furthermore, mechanical transfer steps were avoided since the graphene resided on the insulating YSZ buffer layer after removal of the Ni film. Mechanical transfer of the graphene layer is generally problematic because of the mechanical stress caused in the layer, and it usually leaves residues of the applied polymer supports on the graphene67-69 and is not easily implemented into a technical processes. In contrast, no transfer is required here because, after the CO etching, the Si support could, in principle, directly be used as back gate in a device. The experiments were performed in a glass tube filled with ∼1 bar of CO and traces of sulfide components such as Na2S x n H2O or H2S.16
To monitor the removal of nickel, XP survey spectra were taken from the graphene-covered Ni(111) samples before the reaction and after several hours in the glass tube reactor (Figure 9a). The spectrum recorded before shows the XP and Auger signals of Ni and the C 1s peak of graphene (black spectrum in Figure 9a). After reaction (about 90 h) the C 1s peak was unchanged, whereas the Ni signals had almost completely vanished (blue spectrum in Figure 9a). Instead, strong signals of oxygen and zirconium, the main components of the oxide buffer layer, had appeared. The reaction had thus successfully removed most of the nickel from underneath the graphene layer. However, in high-resolution spectra small Ni signals could be detected and a small S 2p peak had appeared (detection limit of the employed XP spectrometer some % of a monolayer, depending on the element). These peaks were explained by small, three-dimensional nickel sulfide particles formed by reaction with the added sulfide compound, whereas metallic Ni was absent (see Ref. [16] for details). Wet chemical etching methods for metal-grown graphene layers also leave metal residues, even though in smaller amounts, as has been reported recently.70 In the present case, the explanation is that, while sulfur initially acted as catalyst, small particles of almost stoichiometric NiS built up as the nickel content decreased that were no longer reactive, a problem that may be solved in the future by lowering the sulfur content at the end of the process.
The samples were furthermore investigated by Raman spectroscopy (Figure 9b). Before the reaction, no graphene Raman signals were detected, although the Ni(111) film was definitely covered by a graphene layer (black spectrum in Figure 9b). However, the strong interaction of the graphene layer with the Ni substrate is known to shift the Dirac cone to approximately 2.8 eV below EF,71 so that the resonant Raman transitions in the graphene layer are forbidden by Pauli blocking.72 After the reaction, the characteristic strong signals of graphene appeared (Figure 9b). They occurred at 1328 cm−1 (D band), 1588 cm−1 (G band), 1623 cm−1 (D' band), and 2651 cm−1 (2D band). The 2D band could be fitted with a single Lorentzian (FWHM = 41 cm−1), and the G-to-2D area ratio was 0.29, typical values for a graphene monolayer. Rotated bilayer or multilayer graphene that would give very similar Raman spectra 73-75 could be excluded because under optimized conditions exclusively lattice-aligned graphene was obtained. The small feature at 1440 cm−1 could be attributed to the YSZ/Si substrate. The D band indicates a considerable density of defects. A detailed analysis showed that these were caused by point defects or domain boundaries rather than by sp3-hybridized defects, and that they were not formed during the removal of the Ni film by the etching reaction with CO, but must have been present before the reaction.16 The Raman data were thus consistent with the UHV analysis and the growth conditions and confirmed that the monolayer graphene was successfully transferred to the insulating buffer layer. In the literature, in which exclusively a defined monolayer was transferred from nickel, the D band in the Raman spectra varied from not detectable 58, 76-78 to a strong signal.21, 23, 79-81 Certainly further efforts are needed to improve the method, in particular to remove the NiS particles, but it was shown that the method is able to remove the Ni film from underneath the graphene sheet in a pure gas phase reaction.
4 Summary
Thin, single-crystalline metal films of Ru(0001), Ir(111), and Ni(111) on 4 inch Si(111) wafers with YSZ buffer layers were investigated as possible substrates for the epitaxial growth of graphene. After deposition, the metal films were still relatively rough and displayed holes. However, by an annealing step sufficiently flat surfaces could be obtained. The morphologies were then fully comparable to well-prepared surfaces of bulk crystals of the same metals. The only qualitatively different feature, observed for the Ni(111) films, and to a lower degree for the Ir(111) layers, was a high density of monoatomic straight steps. These are slip lines formed by thermal stress-induced plastic deformations (gliding along 
planes) due to the different thermal expansion coefficients of the metals and the Si substrate. However, they did not affect the quality of the graphene layers. They even prevented the formation of wrinkles.
Graphene could be successfully grown by CVD of ethylene on all three metal films. On the Ru(0001) film the same well-ordered (23 × 23) moiré structure as on bulk Ru was observed. For iridium, the preferred high temperatures that have formerly been revealed for graphene growth on bulk Ir(111) crystals were above the stability range of the Ir/YSZ/Si multilayer structure. However, by lowering the ethylene pressure well-ordered overlayers of the known incommensurate graphene structure could be obtained. On the Ni(111) films ethylene CVD led to domain mixtures of the lattice-aligned (1 × 1) graphene, rotated phases, and small bilayer domains. By application of a suitable annealing sequence, this domain structure could be healed and a complete transformation into the lattice-aligned graphene could be achieved. It was shown that the healing process is based on the equilibrium between graphene on the surface and carbon dissolved in the bulk of the Ni film. Controlling this equilibrium was possible for the thin films because of the defined amount of dissolved carbon, something that would not have been achieved with bulk Ni samples.
For the Ni(111) films a new method of detaching the graphene layer from the metal was analyzed, using the reaction of Ni with CO to give gaseous Ni(CO)4. Under quite mild conditions, ∼1 bar of CO at 75°C, and using catalytic amounts of sulfide compounds, the Ni films could be etched from underneath the graphene layer. The method avoids the difficulties connected with the usually applied chemical etching methods, and it avoids a mechanical transfer step to an insulating support as the graphene sheet, after etching, already resides on the insulating YSZ layer.
Acknowledgments
We thank the German research foundation DFG for supporting this project within the priority program 1459 “Graphene” (WI 1003/7-2, SCHR 479/3-1, and SE 1087/10-1/2).
Conflict of Interest
The authors have declared no conflict of interest.
