Catalyst-Free Growth of Networked Nanographite on Si and SiO 2 Substrates by Photoemission-Assisted Plasma-Enhanced Chemical Vapor Deposition

We have developed a photoemission-assisted plasma-enhanced chemical vapor deposition (CVD) process, where DC discharge plasma is assisted by photoelectrons emitted from the substrate under ultraviolet (UV) light irradiation. Under Ar gas atmosphere and in vacuum, plasma current was measured as a function of sample bias voltage to clarify the mechanism, by which photoemission-assisted plasma is generated. Owing to the advantages of the photoemission-assisted plasma-enhanced CVD, where the plasma is generated close to the substrate and in a controllable volume, we have succeeded in growing shiny black ﬁlms of networked nanographite, without any catalyst, on Si(001) and SiO 2 (350 nm)/Si(001) substrates at a deposition rate of ∼ 2 µ m/min despite of low electric power consumption of plasma, ∼ 4 W. Cross-sectional transmission electron microscopy (TEM) and diﬀraction (TED) observations revealed that samples grown at ∼ 700 ◦ C with Ar-diluted CH 4 were composed of multilayer graphene particles (diameter of ∼ 10 nm) that were closely connected to each other and shared some graphene sheets between them, although their crystallographic conﬁgurations were randomly oriented. In bulk-sensitive C 1 s photoelectron spectra using synchrotron radiation at 7933 eV, a chemically-shifted component of sp 2 -bonded carbon atom was dominant and the π – π ∗ transition loss peak was clearly observed for the samples grown on both substrates, indicating that the multilayer graphene particles substantially contain high-quality graphene sheets in agreement with evaluations by microscopic Raman spectroscopy. We named the layered carbon structure “carbon mille-feuille.”


I. INTRODUCTION
Graphene, a single layer of graphite, has recently attracted attention because of its interesting properties originating from the band structure with massless Dirac fermions [1,2]. For example, electron and holes can be induced in concentrations up to ∼10 13 cm −2 by applying gate voltage [3], and carrier mobilities nology nodes of 22 nm.
For practical application of graphene to the devices and interconnections, a starting point is to develop a process of synthesizing single-layer and multilayer graphene. A candidate process involves peeling highly oriented pyrolytic graphite (HOPG) onto SiO 2 surfaces, which is convenient for obtaining high-quality single-layer graphene with large domain sizes up to a few micrometers. Samples so produced are electrically isolated, making it possible to measure the electric performances as demonstrated by Novoselov et al. [3]. However, they are limited in size to a few tens of micrometers and therefore are difficult to position on a SiO 2 surface. Another candidate process, recently of interest, involves surface-functionalized highly solution-processable graphene nanosheets [12,13], produced by oxidation of graphite with mixed acid solutions of H 2 SO 4 /HNO 3 and KClO 3 followed by exfoliation and surface-functionalization. However, this process fails to completely remove groups such as O, OH, and NH from the graphene nanosheets, and thus the electronic properties of the product are not those inherent to just graphene. In yet another candidate process, inch-scale patterned growth of graphene films has been achieved by chemical vapor deposition (CVD) using catalysts such as platinum [14], titan carbide [15], and nickel substrates [16,17]. Problems arise with these metallic catalysts, however, because, for electronic and optical applications, it is necessary to completely remove the metallic catalyst layer, make the graphene layers self-standing, and then attach the layers to an insulating substrate [16]. Heating SiC substrates at high temperature also produces large-scale epitaxially grown graphene layers [18][19][20], whereas annealing at temperatures above ∼1400 • C, close to or above the melting point of Si, ( 1414 • C), is necessary to grow high-quality graphene layers with large domain sizes.
Alternative CVD processes, however, show some promise. Plasma-enhanced CVD is of practical for growing graphene layers at relatively low temperatures without any catalyst over a large-scale substrate [21,22]. In fact, free-standing subnanometer graphite sheets have been synthesized on various substrates such as Si, W, Mo, SiO 2 , and Al 2 O 3 without any special treatment by means of plasma-enhanced CVD inductively coupled with radiofrequency energy (13.56 MHz) [21].
Our aim is to develop a process for synthesizing multilayer graphene that is applicable to interconnections in advanced LSI devices. Four criteria are required for such a process: (i) large deposition area, with potential extension to the 12-in wafers for practical usage; (ii) low (< ∼400 • C) deposition temperature, to avoid thermal damage to interlayer insulators made of low-k materials; (iii) catalyst free, to reduce whole interconnection electric resistivities; and (iv) capability for impurity doping, to increase the carrier concentration. According to these criteria, plasma-enhanced CVD, as mentioned above, is a promising candidate process. For synthesis of carbon materials such as diamond, graphene, fullerene (C 60 ), carbon nanotubes (CNTs), and diamond-like carbon (DLC), however, the conventional plasma-enhanced CVD processes have several problems, including slow deposition rate (∼200 nm/h for carbon nanowall growth [22]), unintended carbon deposition on the chamber wall and sample holder that requires frequent maintenance to wipe out it, and a small deposition area in comparison to the large electric power consumption of plasma. All of these problems originate from the fact that plasma is generated far from the substrate. Therefore, it is thought to be a key concept to generate plasma as close as possible to the substrate.
Accordingly, we have proposed [24] and developed [25] a photoemission-assisted plasma-enhanced CVD process. In our CVD process, DC discharge plasma is assisted by photoelectrons emitted from the substrate under irradiation with ultraviolet (UV) light, making it possible to generate a high volume of photoemission-assisted plasma close to the substrate and thus to maximize growth rate, minimize power consumption, and suppress unintended carbon deposition. In this study, following a detailed description of the experimental apparatus, we investigate the mechanism of plasma generation by examining the relationship between plasma current I P and substrate bias voltage V B for photoemission-assisted plasma; we report the synthesis of networked nanographite on Si and SiO 2 (350 nm)/Si substrates by photoemissionassisted plasma-enhanced CVD; and we observe the deposited layers by cross-sectional transmission electron microscopy (TEM) and diffraction (TED), bulk-sensitive Xray photoelectron spectroscopy (XPS) using synchrotron radiation, and microscopic Raman spectroscopy.
II. EXPERIMENTAL Figure 1 shows a schematic diagram of the experimental apparatus constructed for the photoemission-assisted plasma-enhanced CVD. The apparatus includes a reaction chamber, UV light source, sample holder, plasma control and monitoring circuits, gas supply system, and gas exhaust system. The base pressure of the chamber before the CVD growth was ∼10 −4 Pa without baking and ∼10 −6 Pa after the long-term baking. During the CVD growth, the chamber was evacuated by rotary pump and isolated from turbo molecular pump to maintain the gas pressure at a designated value under the constant gas flow rates regulated by two mass flow meters of argon (Ar) and methane (CH 4 ). The gas pressure was measured with a capacitance manometer (Baratron Type 626, MKS) during the CVD growth, while a penning gauge was utilized during evacuating.
The UV light source was a Xe excimer lamp (UER20H-172A, USHIO), which provides semi-monochromatic UV lights (peak wavelength 172 nm, power density ∼50 mW/cm 2 ) through a fused quartz window. To inhibit absorption of UV lights by O 2 in the air as well as to cool the lamp, the lamp was continuously purged with N 2 gas of 99.999% purity. It is noteworthy that both Ar and CH 4 used for the CVD growth show no photoabsorption at hν = 172 nm, so 172-nm UV light can reach the substrate surface without loss of intensity independent of the gas pressure in the reaction chamber. Thus photoelectron emission from the substrate is highly efficient, even though the substrate is far from the lamp, which resides outside the chamber.
The sample holder consists of a Si heater, sample, and facing electrode. The distance between sample and facing electrode is 5.6 mm. The sample is mounted on the Si heater with a Mo foil clamp for thermal and electrical connections. The electrode, also made of Mo, has an opening (∼5×8 mm 2 ) slightly larger than the size of the sample. A few Mo wires are attached over the opening to compensate for disturbances in the electric field while maintaining the electrode's UV light transmittance above 95%. For generating photoemission-assisted plasma, the sample was subjected to a negative electric potential relative to the electrode. Resistor R (5 kΩ) was connected in series to limit the plasma current, because gas resistance would otherwise be considerably reduced upon the start of glow discharge. Plasma current I P and sample bias voltage V B were monitored in real time by two digital multi-meters (7352A, ADCMT).
As a substrate for CVD growth, Si(001) wafers with native oxides (∼1 nm) and SiO 2 (350 nm)/Si(001) wafers were used after cutting into suitable size (3×10 mm 2 and 6×11 mm 2 , respectively). Samples were cleaned in situ in exposure to Ar plasma for 5 min before the CVD growth. The heater was operated by Joule heating using a DC power supply that is electrically isolated from the ground. A typical sample heating voltage V s was ∼10 V at 700 • C. Sample temperature was measured by infrared pyrometer calibrated in advance with a chromel-alumel thermocouple. Ar (99.9999% purity) and CH 4 (99.999% purity) were used as carrier and source gases, respectively. Typical flow rates of Ar and CH 4 were 1.7 and 0.5 sccm, respectively. Gas pressure was 2000-4000 Pa and growth temperature was ∼700 • C. The reference sample was highly oriented pyrolytic graphite (HOPG) (ZYA, NT-MDT).
Crystallographic information on crystal structure, lattice constant, and grain size were obtained by crosssectional TEM and TED (H-9000UHR, Hitachi) at an acceleration voltage of 300 keV. C-C bonding configuration was observed by a microscopic Raman spectrometer (SPEX 500M, HORIBA JOBIN YVON) combined with an Ar + ion laser with conditions as follows: laser light wavelength = 488 nm; laser power = 0.3 W; probing area diameter = ∼2 µm; objective lens magnification = ×50. Chemical analysis of the sample was performed by bulk-sensitive XPS using synchrotron radiation at BL47XU, SPring-8, Sayo, Japan; details of the experimental setup are described elsewhere [26]. Analysis conditions were as follows: photon energy = 7933 eV; take-off angle = 89 • ; estimated escape depth of a C 1s photoelectron = ∼10 nm; total energy resolution, including the contributions of the excitation light and an electron energy analyzer = 0.25 eV from the spectral feature of the Fermi level of Au.

III. RESULTS AND DISCUSSION
A. IP-VB characteristics of photoemission-assisted plasma Figure 2(a) shows plots of current I vs. bias voltage V B for the cleaned Si(001) substrate. For photoelectron current I PE vs. V B under vacuum at room temperature ( Fig. 2(a), black open squares), I PE rapidly increases with increasing V B up to ∼10 V and then remains almost at ∼0.2 µA independent of V B . The work function of Si is ∼5 eV [27], so excited photoelectrons can escape from the surface for 172-nm (7.2 eV) UV light irradiation, but their kinetic energies are < ∼2 eV. Such small kinetic energies of emitted photoelectrons account for the rapid increase followed by a plateau observed in the figure. Thus we can set ∼0.5 µA/cm 2 as the value that triggers generation of photoemission-assisted plasma for the conventional Xe excimer lamp combined with a fused quartz window. I PE is convenient for monitoring the lifetime of the lamp, the degree of contaminant deposition on the fused quartz window, and the status of the N 2 purge.
For plasma current I P vs. V B under Ar gas atmosphere at 400 Pa ( Fig. 2(a), red open circles), I P is much larger than I PE in vacuum and increases exponentially with increasing V B . This indicates that collision-induced ionization of Ar takes place under this discharge condition, leading to steady generation of plasma. The plasma completely disappears when the lamp is turned off. This confirms that plasma generation is assisted by photoelectrons. At V B = ∼190 V, I P increases steeply from ∼10 −5 A to ∼10 −4 A, then remains in the region 0.1-1 mA, which is still not governed by the series resistor R. At V B > ∼220 V, I P is limited by R. The steep increase in I p implies the appearance of so-called glow-discharge plasma. In the absence of 172-nm UV light irradiation, however, glow discharge appears not at V B = ∼190 V, but only at V B > ∼250 V. As previously reported [28], I P increases very gradually in the Townsend plasma until, with glow discharge at V B = ∼250 V, it jumps by about seven orders of magnitude. The observed difference in V B for the start of glow discharge indicates that glow discharge in the V B region of 190-250 V is also assisted by photoelectrons. Figure 2b shows a typical feature size of glow discharge at V B = 292 V in the presence of UV light irradiation. Glow discharge is restricted to only above the substrate. The I P -V B characteristic in the presence of UV light irradiation can therefore be divided into two regions: region A (V B < ∼190 V), in which photoemission-assisted Townsend discharge occurs, and region B (V B > ∼190 V), in which photoemission-assisted glow discharge occurs.
In general, DC glow discharge can be interpreted in terms of a Townsend's theory [28] of α and γ regimes. In the α regime, ionization of an Ar atom is caused by inelastic collision with an electron, and the resultant electron produced, together with the Ar + ion, are also responsible for further ionization of Ar atoms after their sufficient acceleration en route from cathode to anode electrode. If the α regime were to occur repeatedly, electron and Ar + ion densities would increase in a cascade manner. It is important to consider what causes the first α regime event. Under no UV light irradiation, an electron resulting from a cosmic-ray-induced ionization or a collision-induced ionization between atoms and/or molecules could cause the accidental first event. In addition, Ar + ions produced in the α regime can impinge on the cathode electrode, leading to emission of electrons (γ regime). The α regime is known to depend strongly on the electric field, and α and γ regimes occur in a feedback loop between them, so glow discharge appears suddenly at critical conditions. The critical voltage V C applied between electrodes can be represented as a function of the product pd with universal curves given by Paschen's law [28], where p and d are the gas pressure and gap distance, respectively, between cathode and anode electrodes.
Under 172-nm UV light irradiation, I PE of ∼0.5 µA/cm 2 (∼3×10 12 electrons/cm 2 s) is available to cause the first γ regime event even in the absence contribution from the γ regime. In region A, I PE seems to be much larger than the number of electrons caused by the γ regime, because I P falls to the background level when the lamp is turned off. In other words, I p is governed by two factors: I PE and the α regime. Under constant UV light irradiation, I PE remains almost unchanged over a wide range of V B (Fig. 2(a)). In addition, the first Townsend coefficient of the α regime increases rapid with increasing electric field [28]. Thus, it is likely that the dependence of I P on V B in region A is predominantly determined by the α regime. Therefore we name this interesting phenomenon in region A "photoemission-assisted Townsend discharge." With increasing V B , the number of Ar + ions impinging on the cathode electrode increases exponentially, making the effect of the γ regime significant. Eventually glow discharge can result as observed at V B = ∼190 V (Fig. 2(a)). It is noteworthy that glow discharge does not appear at all at ∼190 V in the absence of UV light irradiation, indicating that the jump in I P with glow discharge at ∼190 V occurs with some assistance from I PE . In fact, glow discharge disappears when the lamp is turned off near the critical voltage V B = V C = ∼190 V. Therefore, glow discharge at relatively low V B can be categorized as "photoemission-assisted glow discharge". DC discharge plasma in regions A and B is hereafter collectively referred to as "photoemission-assisted plasma." At V B sufficiently > ∼250 V, the electron and Ar + ion densities of glow discharge increase, so UV light of 8-11 eV generated from the recombination between them becomes intense, eventually overcoming the electron density I PE from the lamp. At this condition, glow discharge can be persistently sustained when the lamp is turned off. Glow discharge in the absence of UV light irradiation is known to exhibit a hysteresis loop in the associated I P -V B curves. Namely when V B rises to above V C and then falls to below V C , glow discharge is still present due to the UV light irradiation from the plasma itself. In photoemission-assisted glow discharge, a similar effect is observed with 172-nm UV light irradiation.
From the plasma generation mechanism in regions A and B described above, we conclude that photoemissionassisted plasma offers two advantages. First, we can precisely set I P to designated values up to that of glow discharge, and thus optimize reaction efficiency by controlling plasma power. Second, we can apply photoemissionassisted plasma-enhanced CVD of carbon materials by creating photoemission-assisted plasma close to the substrate with intentionally limited volume between sub- strate and facing electrode, as described in section B below. This second advantage is explained as follows. From Fig. 2(a), we know that, at V B > ∼250 V, I P in the absence of UV light irradiation is the same as in the presence of UV light irradiation. However, plasma feature sizes differ under those two conditions. Photoemission-assisted glow discharge is restricted to only above the substrate, whereas common glow discharge in the absence of UV light irradiation tends to appear locally at the sharp corner of the electrode due to the enhancement effect of the electric field and to be widely spread over the rear side of the substrate. Therefore irradiating the substrate with UV light has the practical effect of restricting plasma to within the required area as well as making it uniform over that area.

B. Synthesis by photoemission-assisted plasma-enhanced CVD
Both the surface cleaning and CVD growth process were performed in region B (Fig. 2(a)). The plasma color changed from bluish-purple for pure Ar during surface cleaning to bluish-white after mixing Ar with CH 4 for CVD growth. The sample grown on Si substrate is shiny black (Fig. 3(a)), indicating the presence of graphite. The measured thickness of the deposited layer was ∼40 µm for a growth time of 20 min, for a growth rate of ∼2 µm/min. Figure 2(b) shows that the plasma is as thin as ∼2 mm in the vertical direction and that electric power consumption of the plasma is as small as ∼4.3 W, where I P = 30 mA and plasma voltage is equal to V B − I p · R = 142 V, although electric power of 37 W (100 V, 0.37 A) and 30 W (∼10 V, ∼3 A) are additionally consumed for lamp operation and sample heating at ∼700 • C, respectively. Because of the plasma's small volume and small electric power consumption, it seems impossible to perform carbon deposition with practical growth rates. To the contrary, we achieved growth rates as high as ∼2 µm/min, because the photoemission-assisted plasma is very close to the substrate surface. At gas pressures of ∼100 Pa, the mean free path of radicals such as CH 3 contributing to carbon deposition is roughly ∼0.1 mm. For conventional plasma-enhanced CVD processes coupled with radio frequency or microwave power, the plasma is of a few to a few tens of centimeters in size and far from the substrate compared with the mean free path of radicals. As a result, only a portion of the radicals arrive at the substrate without loss of chemical reactivity. In contrast to the situation for conventional plasma-enhanced CVD processes, most of the radicals can contribute to carbon deposition because the photoemission-assisted plasma is very thin (∼1 mm) and very close to the substrate (within ∼1 mm), as seen in Fig. 2(b). These parameters are comparable to the mean free path of radicals in plasma and therefore the present value of ∼2 µm/min seems to be reasonable.
Thus we have accomplished high growth rate with low electric power consumption of plasma by means of the photoemission-assisted plasma-enhanced CVD as expected. Furthermore we have succeeded in achieving photoemission-assisted plasma-enhanced CVD on SiO 2 (350 nm)/Si(001) substrate as well as on Si(001) substrate. Figure 3(b) shows that carbon deposition was uniformly achieved at ∼700 • C over the SiO 2 (350 nm)/Si(001) substrate except for both longitudinal sides, where the substrate surface was covered with a Mo foil clamp as mentioned above. The growth rate is estimated to be ∼112 nm/min. The deposited sample also appears shiny black, suggesting the presence of graphite. Successful deposition on SiO 2 (350 nm)/Si substrate without any catalyst enables measurement of the electric performances because of the electric isolation of the deposited layer.
C. Observations by cross-sectional TEM and TED, bulk-sensitive XPS, and microscopic Raman spectroscopy Figure 4 shows a cross-sectional TEM image and TED patterns of a sample grown on Si(001) substrate with layer thickness of ∼40 µm and, for comparison, an image of peeled HOPG on SiO 2 substrate. A regularly arranged layer structure is observed for HOPG ( Fig. 4(b)) over a wide area and only 0 0 2n diffraction spots of graphite appear in the TED pattern of the inset. Portions of the sample at ∼2 µm below the surface (Fig. 4(a)) and ∼0.3 µm from the interface between deposited layer and Si(001) substrate (Fig. 4(c)) were observed to explore the crystallographic nature of the depth profile. For both portions, TED patterns obtained in selected-area-diffraction (SAD) mode (electron beam size = 150 nm) show rings with almost uniform radial intensity distributions, ascribed to 002, 100, and 004 diffractions of graphite ( Fig. 4(a) insert) [29,30]. At other depths, TED patterns are similar to those in Fig. 4(a), indicating that the deposited layer is composed of polycrystalline graphite with no specific configurations regarding both the a and c axes independent of depth over ∼40 µm.
The TEM image (Fig. 4(a)) also shows a laminar struc- ture as well as domains with random orientations. These domains are not isolated but rather are complicatedly connected to one another via some graphene sheets. The averaged layer spacing is ∼0.38-0.40 nm, slightly larger than that of HOPG (0.34 nm; Fig. 4(b)) and close to that of the onion-like carbon structure (∼0.4 nm) [31]. However, each domain is not the onion-like carbon structure itself, although graphene sheets are considerably curved, suggesting large strain. The TEM image suggests that the domain size is ∼10 nm. To confirm this estimation, we obtained TED patterns in nanobeam diffraction (NBD) mode (Fig. 4(c)). When the electron beam size is 30 nm, the pattern still shows rings, on which several spots are superposed, suggesting that part of the observed area contains wellordered domains of graphene sheets. When the beam size is decreased to 5 nm, the pattern changes remarkably into much narrow spot patterns, although the 002 and 100 diffraction spots are slightly elongated along the radial direction and large compared to those of HOPG ( Fig. 4(b)), which implies that the domain size is at least larger than 5 nm, in accordance with TEM observation. Thus the deposited layer is comprised of microscopically well-layered graphene sheets. Due to the structural features observed in Fig. 4(a), we have named this structure "carbon mille-feuille. " We consider now the presence of impurities in the deposited layer. The amount of oxygen was estimated by O 1s photoelectron intensity to be ∼5%, and is perhaps due to residual H 2 O and O 2 because of no baking before CVD deposition. Other impurities such as metals are below the detection limits of XPS. Figure 5 compares the C 1s photoelectron spectrum of the sample grown on Si(001) with that of HOPG. An asymmetric C 1s spectral feature of HOPG originates from excitation of electron-hole pairs in the final state peculiar to the metallic material [33]. A broad peak at binding energy ∼291.5 eV is ascribed to the π-π * transition, giving rise to an energy loss of ∼6.5 eV relative to the main C 1s peak [34]. Full width at half maximum (FWHM) of C 1s for the deposited layer is narrow and almost equal to that for HOPG. In addition, a well-separated π-π * transition loss peak is observed for both the deposited layer and HOPG. These features clearly indicate that the main part of the sample grown on Si(001) substrate is comprised of highquality graphene sheets, in accordance with TEM observation ( Fig. 4(a)). For the sample grown on SiO 2 (350 nm)/Si(001) substrate, FWHM of C 1s is almost the same as for the sample grown on Si(001) substrate, although not shown here.
Next we consider the large tail disagreements at the high and low binding energy (E B ) sides of the C 1s spectra between the deposited layer and HOPG. The chemically shifted component present at the high E B side is thought to arise from graphene-sheet defects, as reported for HOPG [35,36]. On the other hand, a dominant component of the high-E B disagreement is associated with carbon atoms having an sp 3 configuration, for which the chemical shift is ∼0.5 eV [37,38]. In the present case, the sp 3 configuration is due not to synthesis of diamond [38], but rather to concomitant growth of amorphous carbon [37] because no TED patterns for diamond are observed in Fig. 4. Such amorphous carbon is likely distributed at domain boundaries ( Fig. 4(a)). The increasing amount of amorphous carbon results in a significant broadening of C 1s compared to FWHM of HOPG as previously reported [39]. In the high-E B region of 2-5 eV relative to the main peak, there are contributions due to oxides such as C-O, C=O and C-O-O [38,40], while the amount of oxygen is very small (∼5%), as evaluated by careful analysis of C 1s.
Finally we consider C-C bond configurations of the deposited layer. In the Raman spectra of HOPG (Fig. 6), only a sharp peak appears at ∼1566 cm −1 [32]. This is the so-called G band associated with graphene sheet. In contrast, a doublet peak appears at ∼1340 cm −1 and ∼1580 cm −1 for the deposited layers on Si and SiO 2 substrates. The G band has a shoulder at the high-wavenumber side, indicating the existence of an additional component. This shoulder, which we call the D' band, is likely due to graphene-sheet defects [41,42]. It hardly appears at all for carbon black [43] and more clearly for relatively oriented domains of graphite [22,42]. Therefore we can use the D' band as a measure of crystallinity for graphitic samples. An intense peak at ∼1340 cm −1 , which we call the D band, is presumably due to amorphous carbon, because it grows when HOPG is exposed to Xe + ions [44].
We performed a least-squares curve-fitting analysis of the Raman spectra using the three components D, G, and D' after subtracting a linear background. Analysis gives the following D, G, and D' peak positions: 1339, 1574, and 1605 cm −1 for the sample on Si(001) and 1341, 1577, and 1605 cm −1 for the sample on SiO 2 /Si(001). The D and D' bands vary significantly in peak position according to the photon energy of the excitation laser [41]. This behavior is interpreted by the double resonance Raman mechanism in graphite [20,41]. The D band is more intense than G band for both samples. This does not indicate that the amount of amorphous carbon exceeds that of graphene sheet. Rather, it is due to the difference in sensitivity to the excitation energy of Raman spectroscopy between the two bands. In apparent contradiction to the D/G intensity ratio measured by Raman spectroscopy, graphene sheet is actually the predominant component, as reported for carbon nanowalls [22,43]. This trend is also supported by C 1s XPS observation (Fig. 5). The D' band is clearly observed for the samples on Si(001) and SiO 2 /Si(001) substrates, while the G band is relatively weak. The peak positions of the G band, 1574 and 1577 cm −1 for the samples on Si(001) and SiO 2 /Si(001) substrates, respectively, are higher than that for HOPG, 1566 cm −1 . The shifts reach ∼8 cm −1 toward the high-wave-number side. It is well known that the peak position of the G band changes toward the high-wave-number (blue shift) or low-wavenumber (red shift) side depending on domain size [45], strain due to lattice mismatch [20,45,46], and graphenesheet curvature [47,48]. In particular, the blue shifts have been reported for carbon nanotubes [47,48] and relatively small domain sizes [45]. Therefore the blue shift observed in Fig. 6 is thought to be associated not only with nanosize domains, but also with graphene-sheets curvature. From the D/G intensity ratio [41,43], we evaluated the domain sizes to be ∼9.3 and ∼9.9 nm for the samples on Si and SiO 2 substrates, respectively. These values are rather close to the values estimated by TEM estimation (∼10 nm).
As described above, all observations by cross-sectional TEM and TED, bulk-sensitive XPS, and microscopic Raman spectroscopy are consistent with and complementary to one another. Consequently, we conclude that networked nanographite with a nanosize domain structure was synthesized on Si(001) and SiO 2 (350 nm)/Si(001) substrates without any catalyst by photoemission-assisted plasma-enhanced CVD using a Xe excimer lamp as a UV source.

IV. SUMMARY
We have developed a photoemission-assisted plasmaenhance CVD process to synthesize multilayer graphene applicable for LSI interconnections. By examining the relationship between plasma current I P and substrate bias voltage V B for photoemission-assisted plasma, we clarified that photoelectrons emitted from the substrate in the presence of UV irradiation play an important role in increasing plasma current in the Townsend discharge region and in decreasing the critical substrate bias voltage V C for inducing glow discharge. Therefore the term "photoemission-assisted plasma" includes two kinds of phenomena: photoemission-assisted Townsend discharge and photoemission-assisted glow discharge. In both discharge modes, plasma is observed to be close to the substrate surface in the presence of UV irradiation within the spacing between substrate and facing electrode used for accelerating electrons.
For the ∼2×10 mm 2 deposition area on the Si substrate, a deposition rate of ∼2 µm/min was achieved with the following conditions: plasma power consumption = 4 W; electric power to heat the sample at ∼700 • C = ∼30 W; electric power to operate the Xe excimer lamp = ∼37 W. This attempt succeeded because photoemission-assisted plasma is generated very close to the substrate, so most radicals can reach the substrate to contribute to deposition and unintended carbon deposition on electrodes and chamber walls is suppressed. Cross-sectional TEM and TED, bulk-sensitive XPS, and microscopic Raman spectroscopy observations reveal that the deposited layer is comprised of nanoscale domains of graphite complicatedly connected with one another via some graphene sheets, which prompted us to name this structure "carbon millefeuille."