Layer exchange synthesis of multilayer graphene

Since the discovery of LE in the Si–Al system more than 20 years ago, the growth mechanism has been well studied both experimentally and theoretically [26–33]. In addition to the Si–Al system, many LE combinations have been reported for semiconductors, such as Si–Ag [34, 35], Si–Au [36, 37], Si–Zn [38], Ge–Al [39, 40], Ge–Ag [41], Ge–Au [42, 43], Ge–Zn [44], and Ge–Sb [45]. These metals are also useful for the LE of amorphous SiGe alloys [46–51]. The LE mechanism is considered to be common for all materials, and carbon is no exception. The LE process is illustrated in figure 1. First, a metal layer and an amorphous carbon (a-C) layer are sequentially prepared on an arbitrary substrate (figure 1(a)). During annealing, carbon atoms diffuse from the amorphous layer into the metal layer, mainly through the metal grain boundaries (figure 1(a)). When the carbon concentration in the metal is supersaturated, MLG nucleates in the metal (figure 1(b)). Later, the carbon atoms dissolved in the metal contact the nuclei, which induces lateral growth (figure 1(c)). The lateral growth stresses the metal and pushes it to the upper layer (figure 1(c)) in a process called the push-up phenomenon. In some studies of the LE in the Si–Al system, in situ microscopic cross-sections and planar images showing push-up phenomena during the LE process were directly observed [29, 52, 53]. Eventually, MLG forms the bottom layer, while the metal forms the upper layer (figure 1(d)). Therefore, MLG is formed during annealing in the LE method, while the starting structure is similar to that in the conventional precipitation during cooling. Thermodynamically, the driving force of the LE process is the difference in the Gibbs free energy between the amorphous and crystalline phases [28, 31]. The metastability and high free energy of the amorphous layer cause supersaturation, which leads to nucleation in the metal layer [29, 32]. Based on the mechanism, there are three conditions for inducing LE during annealing rather than precipitation during cooling. (i) The initial carbon layer should be amorphous or at least poorly crystalline [25, 54], while there are no restrictions on the deposition methods of the carbon layer. (ii) The a-C layer should be thick enough to cause supersaturation of carbon atoms in the metal layer. This means that the film thickness ratio of a-C and metal is important [25]. (iii) The annealing time should be sufficient to induce MLG nucleation in the metal layer, which depends on the annealing temperature. Once the LE is completed, the process does not recur because the MLG formed is completely crystalline. LE occurs even if the order of the layers is inverted [55–57], which enables self-organized formation of the lower electrode [58, 59]. For semiconductors, the inverted-LE has been studied for application in devices with vertical structures such as solar cells [60, 61].

Zoom In Zoom Out Reset image size

After LE, the thickness of the MLG formed is almost determined by the initial thickness of the metal layer. Basically, the initial bottom and top layers serve as 'molds' for the eventual top and bottom layers, respectively [62, 63]. We define this phenomenon as LE, which differs from conventional precipitation methods. For the LE of semiconductors, when the amorphous layer is thicker than the metal layer, semiconductor 'islands' form on the bottom-crystalline semiconductor layer, and when the amorphous layer is thinner than the metal layer, the bottom-crystalline semiconductor layer has 'holes', which reduces the coverage of the semiconductor layer on the substrates [64, 65]. Therefore, to obtain uniform MLG, the thicknesses of the a-C and the metal layers should almost be the same. The best way to obtain a high-quality MLG layer with high surface coverage will be to remove the islands in the island-rich samples, which has proven in the LE of semiconductors [66, 67].

LE can be identified by examining the cross-sectional structure using focused-ion-beam microscopy, scanning electron microscopy (SEM), and transmission electron microscopy (TEM) with energy dispersive x-ray (EDX) analysis. Raman spectroscopy is also a useful tool for easily identifying LE, especially when the substrate is transparent [22]. Before LE, a-C exists on the front surface and metal on the back surface (figure 2(a)). Therefore, a broad band corresponding to a-C is observed in the Raman spectrum of the front surface, while no peaks are observed in the spectrum of the back surface because of the reflection of the Raman laser by the metal (figure 2(b)). After the LE process, D, G, and G' peaks corresponding to MLG [68] are observed in the Raman spectrum of the back surface, while no peaks are observed in the spectrum of the front surface owing to the metal reflection (figure 2(b)). When the initial a-C layer is thicker than the metal layer, the Raman spectra of the front side of the sample after LE exhibits relatively weak peaks owing to the formation of MLG islands in the upper metal layer [69]. Similar to the LE of semiconductors [67, 70], island MLG has much poorer crystallinity than the bottom MLG layer, which is reflected in the intensity ratio of G to D peaks in the Raman spectra [69]. Depending on the type of metal, LE can be identified by naked eye because the color of the sample surface (and the back surface when the substrate is transparent) changes before and after LE (figure 2(c)) [71, 72]. Metal removal is also useful for identifying LE. When LE is successful, the MLG layer remains on the substrate after the removal of metal by wet etching, which is also visible to the naked eye. More reliably, the remaining layer on the substrate should be evaluated using EDX. The residual metal concentration in the resulting MLG layer is limited by the solid solubility limit of the metal in MLG and is therefore, usually below the detection limit of EDX (∼1%) [22, 71].

Zoom In Zoom Out Reset image size

Most metal-catalyzed graphene syntheses have used transition metals because they are highly reactive owing to their open d-orbitals [73–76]. To date, 15 transition metals (Ni, Co, Fe, Cr, Mn, Ru, Ir, Pt, Ti, Mo, W, Pd, Cu, Ag, and Au) have been examined for LE with carbon [71]. When the stacked structure of a-C/metal/SiO₂ substrate, in which the thicknesses of a-C and metal layers are almost the same, is annealed at temperatures up to 1000 °C, the solid-phase reaction that takes place between carbon and metal can be classified into four groups depending on the metal species: (1) LE, (2) carbonization, (3) local MLG formation, and (4) no graphitization (figure 3(a)). The difference in the solid-phase reaction causes large differences in the Raman spectra; a sharp G peak corresponding to highly crystalline MLG appears only in the case of LE (figure 3(b)). Currently, LE for carbon has been confirmed in eight transition metals (Co, Ni, Cr, Mn, Fe, Ru, Ir, and Pt), as summarized in the periodic table in figure 3(c). Generally, early transition metals form more stable carbides [73, 74]; therefore, Ti, Mo, and W form carbides without LE. In contrast, late transition metals tend to form solid carbon solutions. Groups 8−10 have higher carbon solubility than group 11, reflecting the number of electrons in the d orbital [75, 76]. Most of the metals in groups 8−10 induce LE because these metals can dissolve sufficient carbon to cause MLG nucleation in the metal layer. Pd is an exception and cannot induce LE owing to the agglomeration of the Pd layer during annealing. The ease of agglomeration is generally related to the wettability between the metal layer and the substrate. Because Co also easily aggregates, it is difficult to obtain a uniform MLG layer. However, by depositing the Co layer while heating the sample (200 °C) to increase its thermal stability, the LE process is completed before Co is dotted [77]. Changing the substrate material and inserting the underlayer will also be effective in suppressing the aggregation of the metal layer. Therefore, to induce LE of carbon, the metal should (i) have a high solid solubility of carbon, (ii) be stable during annealing, and (iii) not form a carbide. Some metals (Cr, Mn, and Fe) that can form carbides are preceded by LE before carbonization [78]. Considering that only Al, Au, Ag, Zn, and Sb are known to induce LE for Si and Ge, many metals can be used for LE of carbon, despite its short history. This is because carbon is less likely to form compounds than the highly reactive semiconductors.

Zoom In Zoom Out Reset image size

In the LE of carbon, the metal species also affect the crystallinity of the resulting MLG and the growth temperature, similar to the LE of semiconductors [21, 44]. Annealing at 1000 °C completes the LE of carbon for the above-mentioned eight metals, while the intensity ratio of the G to D peaks in the Raman spectra, which indicates the crystal quality of MLG [79], varies with the metals (figure 4(a)). In addition, the temperature range for LE also depends on the metal species (figure 4(b)). Ni is one of the best metals for low-temperature LE synthesis of MLG. This tendency can be summarized as follows: metals with high carbon solid solubility produce MLG with relatively low crystallinity at low temperatures, whereas metals with low carbon solid solubility produce high-quality MLG at high growth temperatures.

Zoom In Zoom Out Reset image size

As an example, there is a difference in the cross-sectional TEM images of the MLG layers formed by LE with Ni (figures 5(a)–(d)) and Pt (figures 5(e)–(h)). In both cases, MLG is formed on the substrate as a result of LE; however, the shape of the MLG layers is different in the Ni and Pt samples. In the Ni sample, the MLG layer is rippled and contains randomly oriented nanocrystalline carbon in the spaces between the rippled MLG and the Ni layer or substrate (figures 5(a) and (b)) [22]. In contrast, the MLG layer between the Pt layer and the substrate is uniform (figures 5(e) and (f)) [71]. These results account for the difference in crystallinity derived from the Raman spectra (figure 4). The MLG layers are {002} oriented along the film surface for both the Ni and Pt samples (figures 5(c) and (g)). This behavior is consistent with the conventional precipitation methods of MLG: {002} oriented graphene precipitates from metals owing to the minimization of surface or interface free energy [80, 81]. While LE is well known to control the crystal orientation of semiconductor layers even on amorphous substrates [82, 83], the effect is particularly remarkable for MLG, probably because of its large surface energy anisotropy. The selected-area electron diffraction pattern of the Ni sample is arced (figure 5(d)) while that of the Pt sample is spotty (figure 5(h)), reflecting the shape of the MLG layer in each sample. Thus, the type of metal affects the quality of MLG, but the relatively poor crystallinity obtained by Ni-induced LE can be significantly improved by controlling other factors, as described below.

Zoom In Zoom Out Reset image size

According to the G/D intensity ratio of the Raman spectra (figure 4(b)), the annealing temperature affects the crystallinity of MLG formed by LE. This behavior is well known for any MLG precipitation method [8–10, 13, 19, 25, 81]. For most metals, a higher annealing temperature provides a higher G/D intensity ratio, i.e. higher crystallinity of MLG. This corresponds to the general principle of crystallization: defects in grains decrease as the growth temperature increases, which makes the atoms easier to move. However, this behavior does not apply to some metals, including Ni (figure 4(b)). This is likely because the G/D intensity ratio of the Raman spectra is determined by the balance of intragrain defects and grain boundary defects; lowering the growth temperature increases both the grain size (reducing grain boundaries) of MLG and the intragrain defects. This behavior is well known in solid-phase crystallization, including LE, of semiconductors [21, 84, 85]. In addition, high growth temperatures agglomerate the unstable metal layer, resulting in a rough surface (figure 6) because the metal layer acts as a mold for the MLG (figure 1). Therefore, for some metals, lowering the LE temperature is important for improving the quality of MLG and for a wide range of applications.

Zoom In Zoom Out Reset image size

The rate of temperature decrease and the types of inert gases forming the annealing atmosphere are considered as important factors in conventional MLG precipitation methods using solid-phase reaction between carbon and metals [86–88]. The rate of temperature decrease does not affect LE, because the synthesis of MLG is completed during annealing. In addition, the type of inert gas has not been observed to affect the LE synthesis of MLG, similar to the LE of semiconductors. However, atmospheric exposure/non-exposure of the a-C layer has a great influence on the subsequent growth characteristics because the amorphous layer easily absorbs oxygen from the atmosphere. For the Ni-induced LE of carbon, the MLG synthesis temperature is lowered to 350 °C by avoiding air exposure and annealing in a vacuum deposition chamber [89]. The MLG layer formed at 350 °C exhibits higher crystallinity than the MLG layer grown at high temperatures (≥ 600 °C) after air exposure. This growth temperature is lower than the heat resistant temperature of a flexible polyimide substrate, which opens up the possibility of flexible device applications. In fact, MLG can be synthesized on a flexible polyimide substrate by coating a SiO₂ layer to prevent the contamination of the a-C layer from the plastic, as shown in figures 7(a) and (b) [90]. In this example, LE with an inverted structure was employed to self-organize the electrodes under MLG (figures 7(c) and (d)) for rechargeable battery applications, as described later.

Zoom In Zoom Out Reset image size

The addition of an interlayer between the metal layer and the amorphous layer has been actively studied in the LE of semiconductors [91–94]. This interlayer controls the mutual diffusion of the semiconductor and the metal atoms during annealing. LE does not occur when the metal atoms diffuse into the semiconductor layer before the semiconductor atoms diffuse into the metal layer. Employing an appropriate material as an interlayer can solve this problem and induce LE [95]. Even in a system where LE occurs without an interlayer, diffusion control using an interlayer increases the grain size [96, 97] or aligns the crystal orientations [98, 99]. The use of an interlayer has also been reported to increase the grain size of MLG in conventional precipitation methods [100].

In Ni-induced LE of carbon, SiO₂ and AlO_x were investigated as an interlayer [69]. In particular, an AlO_x interlayer was found to significantly improve the grain size of the resulting MLG layer (figures 8(a) and (b)). The G/D intensity ratio in the Raman spectra, corresponding to the crystallinity of MLG, has a peak depending on the interlayer thickness for both SiO₂ and AlO_x interlayers (figures 8(c) and (d)). According to TEM analyses, a 1 nm thick AlO_x interlayer increases the grain size of MLG to a few micrometers, which is one order of magnitude larger than that of the sample without an interlayer. This is because the interlayer delays the diffusion of carbon atoms into the metal and suppresses MLG nucleation, which gives time for lateral growth of MLG and enlarges the grain size of MLG. Therefore, the improvement in the G/D intensity ratio is considered to basically reflect the increase in the grain size of MLG. When the Al₂O₃ interlayer is prepared by sputtering an Al₂O₃ target instead of air exposure of an Al membrane, the G/D intensity ratio increases further (figure 9(a)) [101]. Therefore, the material, thickness, and fabrication process of the interlayer affect the LE properties because they change the degree of interdiffusion between carbon and metal. The crystallinity of MLG can be further improved by controlling these parameters.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

The film thickness of MLG after LE corresponds to the initial film thickness of the metal layer which works as a mold [69]. Therefore, the MLG thickness can be controlled over a wide range by modulating the thickness of the metal layer; this is an advantage over conventional precipitation methods. In fact, Ni-induced LE can be used to synthesize MLG with film thicknesses ranging from 5 to 200 nm [101], which are not the limits of thin and thick layers, respectively.

According to the G/D intensity ratio in the Raman spectra, the crystal quality of MLG is relatively low for MLG thicknesses ≤20 nm and considerably improves at thicknesses ≥50 nm, especially when a diffusion-controlling Al₂O₃ interlayer is introduced between the a-C and Ni layers (figure 9(a)). It is important to note that the Al₂O₃ interlayer here is directly deposited using an Al₂O₃ sputtering target, which provides a higher G/D intensity ratio in the Raman spectra compared to the AlO_x interlayer formed by atmospheric exposure (figure 8). The maximum G/D intensity ratio reaches 20, which is higher than that of most MLG layers synthesized on insulators using metal-induced solid-phase crystallization techniques [102–104]. Although the crystal quality of the thin MLG layer is low, it can be formed uniformly on the substrate (figures 9(b) and (c)). The Ni particles on the MLG (figure 9(c)) are attributed to the agglomeration of a thin Ni layer during annealing, which accumulated on the MLG layer after LE. After removing the Ni layer, the thin MLG layer exhibits a high light transmittance (figure 9(d)), which is important for application as transparent conductive films. According to our experiments, it is possible to synthesize MLG with film thickness >500 nm using Ni-induced LE; however, in principle, there is an upper limit to the film thickness that can be obtained by LE. For example, when the Al thickness exceeds 1500 nm in the Si–Al system, the upper Si layer is crystallized before the Si atoms reach the substrate through the Al layer, which inhibits the complete LE [105]. The parameters for realizing thick films using LE are not yet known even in the case of semiconductors; however, the key is to accelerate the diffusion of carbon into the metal. The need for thick MLG layers in various devices will instigate further research in this area.

LE does not occur when the film thickness of the metal or a-C is smaller than a few nanometers. According to the study on LE of semiconductors [106], this is attributed to the difficulty in nucleation of MLG in the metal layer. Therefore, multilayer structures containing thin a-C or metals can induce LE, resulting in the formation of a uniform MLG layer molded on the initial thick Ni layer (figures 10(a) and (b)). This phenomenon was originally discovered in the LE of semiconductors [107–109] and has also been demonstrated for Ni-induced LE of carbon [110, 111].

Zoom In Zoom Out Reset image size

Inserting a 1 nm thick a-C underlayer between Ni and the substrate promotes the supersaturation of carbon atoms in Ni and then accelerates the nucleation (figure 10(a)) [110]. This not only lowers the LE temperature, but also enhances the crystal quality of MLG (figure 10(c)). The thickness of the a-C underlayer is important: LE will occur at both the top and bottom sides if the a-C layer is too thick (>5 nm). The a-C/Ni laminated structure also results in uniform MLG as long as the laminated Ni layer is sufficiently thin (figure 10(b)) [111]. The crystal quality of MLG improves dramatically as the number of a-C/Ni multilayers increases: 15 laminated layers of 4 nm thick a-C and 0.5 nm thick Ni result in high G/D intensity ratio in the Raman spectra (figure 10(c)). Thus, the multilayer structure improves the crystal quality of MLG and thereby, its electrical conductivity. This phenomenon is explained by the fact that the interface is the diffusion site of carbon atoms; it promotes the lateral diffusion of carbon atoms and increases the grain size of the MLG [109].

The factors introduced above have already been demonstrated to be effective in the LE of carbon. According to studies on the LE of semiconductors, there are many other parameters that affect the LE of carbon. For example, the underlayer material generally affects LE [112–115]. The qualities of the initial carbon and metal layers are also important factors. Because the carbon atoms diffuse mainly through metal grain boundaries, a metal layer with smaller grains will induce a faster rate (lower growth temperature) of LE [116–118]. Introducing defects [119] or impurities [120, 121] into the a-C layer will also make the carbon bonds in a-C easier to break and promote diffusion of carbon atoms into metals.

MLG is a promising material for various applications, including electrodes and heat spreaders, owing to its high electrical and thermal conductivities [122, 123]. Transparent electrodes are the most actively studied application for thin MLG [124–126], while thick MLG is suitable for wiring in integrated circuits because its current-carrying capacity exceeds that of Cu [127–129]. Applications to flexible devices are particularly attractive because MLG is soft and bendable [130, 131]. The ability of LE to synthesize uniform MLG on arbitrary substrates at low temperatures is an excellent feature that helps to incorporate MLG on plastics and in existing devices without damaging them.

Because LE can form MLG uniformly on an insulator, its electrical characteristics can be evaluated easily using Hall effect measurements with the Van der Pauw method [101]. The lower growth temperature tends to provide a higher electrical conductivity (figure 11(a)), which is in contrast to that of most MLG formed by cooling precipitation methods [132–134]. There are two reasons for this: in solid-phase crystallization of amorphous layers, including LE [21, 85], a lower growth temperature provides (i) more uniform crystalline layers (figure 6) and (ii) larger grain size owing to the lower nucleation frequency. This results in lower grain boundary scattering of carriers. It is important to note that, LE is not completed at low temperatures (≤400 °C) unless vacuum annealing avoids air exposure, as mentioned before (figure 7). Both avoidance of air exposure and formation of a-C underlayer appear to be effective in improving the electrical conductivity of MLG (figure 11(a)) as well as lowering the growth temperature [89].

Zoom In Zoom Out Reset image size

The electrical conductivity of MLG formed by LE also depends on the thickness of MLG (figure 11(b)), which may reflect the balance between contamination from the substrate in the case of thin-film and grain size reduction in the case of thick film [105, 106]. The introduction of the Al₂O₃ interlayer strongly improves the electrical conductivity owing to grain size enlargement [69, 101]. These behaviors are consistent with the crystal quality of MLG estimated from the G/D intensity ratio in the Raman spectra. The t = 50 nm sample with the interlayer exhibits the highest electrical conductivity of 2700 S cm⁻¹ and showed the highest quality among the LE samples according to the Raman study (figure 4(b)). The electrical conductivity exceeds that of highly oriented pyrolytic graphite (HOPG) with a low mosaic degree of 0.4°, synthesized at 3000 °C or higher. The carrier mobility of MLG reached 550 cm² V⁻¹ s⁻¹, which is higher than that of most MLG layers formed on insulators, but lower than that of HOPG (960 cm² V⁻¹ s⁻¹) [101]. The high electrical conductivity of MLG reflects its high carrier density (10¹⁹–10²⁰ cm⁻³); however, it does not imply that the MLG has higher crystallinity than HOPG. Although the origin of the high carrier density in MLG is still unclear, defects and residual Ni in MLG or surface doping by HNO₃ solution during Ni removal are possible reasons [135, 136]. The electrical conductivity can be further improved by combining the interlayer and multilayer structures, which will pave the way for application of MLG in electrodes.

Graphite has been used as an anode in conventional Li-ion batteries (LIBs) with liquid electrolytes and also shows good anode performance for solid electrolytes [137]. Therefore, an all-solid-state battery using a graphite thin-film, which is a laminated structure of graphene, as the anode increases the feasibility of flexible rechargeable batteries. Although graphene has a high specific surface area and a corresponding high specific capacity [138, 139], MLG with sufficient thickness is desirable for actual anode applications, from the perspective of capacity per area. Because LE enables the synthesis of MLG films much thicker than those obtained from conventional precipitation methods, applications to thin-film rechargeable battery anodes are highly expected [78, 90].

To focus on the evaluation of LIB anode characteristics, MLG needs to be synthesized on a metal substrate. Ni-induced LE at 350 °C with an inverted structure allows formation of MLG on a Mo foil without alloying (figure 12(a)). This structure, with Ni as the lower electrode, is also useful when constructing a thin-film rechargeable battery on an insulating substrate such as plastic. When the sample is packaged in a coin-type cell with a Li foil, an electrolyte, and a separator, the MLG exhibits charge/discharge operation (figure 12(b)). After 50 cycles, the discharge capacity and the Coulombic efficiency reach 6.2 μAh cm⁻² and 99%, respectively, and remain stable. Such stable operation is an excellent feature of carbon-based anodes, while there are various candidates for anode materials in all-solid-state batteries. The rate performance shows that the capacity is over 400 mAh g⁻¹ at a current rate of 0.1–2 C (figure 12(c)), which is higher than the theoretical capacity of graphite (372 mAh g⁻¹). This behavior is attributed to the capacitive contribution and reduced diffusion distance of Li ions in local short-range ordered structures [140–142]. Moreover, the capacity retention of the sample is excellent; the capacity assessed during the second 0.1 C measurement almost recovered to that assessed during the first. Although the areal capacity is still small owing to the thin (100 nm) layer of MLG, it can be improved by increasing the MLG thickness. The synthesis temperature (350 °C) for MLG is low enough for the use of plastic substrates, enabling the development of flexible rechargeable batteries.

Zoom In Zoom Out Reset image size