A systematic approach to mass-production of graphene and other 2D materials is essential for current and future technological applications. By combining a sequential statistical design of experiments with in-situ process monitoring, we demonstrate a method to optimize graphene growth on copper foil in a roll-to-roll rf plasma chemical vapor deposition system. Data-driven predictive models show that gas pressure, nitrogen, oxygen, and plasma power are the main process parameters affecting the quality of graphene. Furthermore, results from in-situ optical emission spectroscopy reveal a positive correlation of CH radical to high quality of graphene, whereas O and H atoms, Ar+ ion, and C2 and CN radicals negatively correlate to quality. This work demonstrates the deposition of graphene on copper foil at 1 m/min, a scale suitable for large-scale production. The techniques described here can be extended to other 2D materials and roll-to-roll manufacturing processes.

Two-dimensional (2D) materials, such as graphene, hexagonal boron nitride (h-BN) and transition metal dichalcogenides (TMDs) have attracted immense attention in the past decade.1,2 2D materials possess a range of favorable properties, from the zero band-gap and high electrical, and thermal conductivity of graphene3,4 to the wide band gap and good dielectric properties of h-BN.5,6 Several studies have considered mass-production of graphene using a variety of techniques.7 For instance, exfoliation of graphite (liquid exfoliation,8 shear mixing9 or chemical reduction of graphite oxide10) is a commonly used method to mass-produce graphene platelets. However, exfoliation produces low yields,9 requires high energy input9 and produces graphene sheets with structural defects. Another method for large-scale production of graphene is chemical vapor deposition (CVD) which has been proven to produce higher quality, large-area, single and few-layer graphene films.11 

Table I summarizes previously reported roll-to-roll CVD processes for large-area graphene film production.12–18 Plasma is more energy-efficient than other energy sources for mass production of graphene because no direct heating of the substrate is required, and hence the supplied power is concentrated in the growth region.19 Also, plasma-based CVD processes are more amenable to scaling due to the higher deposition rate accelerated by active dissociation of the feedgas by the plasma.20 Even though plasma systems produce moderate quality graphene because of ion bombardment, the quality can be increased by applying appropriate optimization methods.

TABLE I.

Summary of available CVD process for large area graphene production.

Heating methodCu foil dimensionsP (mbar)GasSubstrate TemperatureSpeed (mm/min)Residence time (min)ID/IGRef.
Thermal furnace -, 76 cm diagonal 0.6 H2, CH4 1000 oStationary 60 Small 12  
Thermal furnace 25 and 125 μm, 43 X 91 cm2 1013 H2, CH4, Ar 1000 oStationary 300 Small 13  
Thermal furnace 25 μm, 100 cm long 1013 H2, CH4, Ar 1000 o10-400 0.75 – 30 0.2 14  
Thermal furnace 50.8 μm, 0.6 X 60 cm2 H2, C2H4, He 1010 o25-500 0.6 – 12 0.1-1 15  
Thermal furnace 40 μm, 12 X 60 cm2 1013 H2, CH4, N2 1010–1070 o60 Small 16  
Joule heating 36 μm, 23 X 1000 cm2 10 H2, CH4 1000 o100 Small 17  
Remote MW plasma 33 μm, 29 X 48 cm2 0.3 H2, CH4, Ar 400 o300 1.6 0.5 18  
Rf plasma 76 μm, 2.54 cm wide 7-18 H2, CH4, N2, O2, Ar without direct heating 45 - 959 0.2-4.5 0.6-2 This work 
Heating methodCu foil dimensionsP (mbar)GasSubstrate TemperatureSpeed (mm/min)Residence time (min)ID/IGRef.
Thermal furnace -, 76 cm diagonal 0.6 H2, CH4 1000 oStationary 60 Small 12  
Thermal furnace 25 and 125 μm, 43 X 91 cm2 1013 H2, CH4, Ar 1000 oStationary 300 Small 13  
Thermal furnace 25 μm, 100 cm long 1013 H2, CH4, Ar 1000 o10-400 0.75 – 30 0.2 14  
Thermal furnace 50.8 μm, 0.6 X 60 cm2 H2, C2H4, He 1010 o25-500 0.6 – 12 0.1-1 15  
Thermal furnace 40 μm, 12 X 60 cm2 1013 H2, CH4, N2 1010–1070 o60 Small 16  
Joule heating 36 μm, 23 X 1000 cm2 10 H2, CH4 1000 o100 Small 17  
Remote MW plasma 33 μm, 29 X 48 cm2 0.3 H2, CH4, Ar 400 o300 1.6 0.5 18  
Rf plasma 76 μm, 2.54 cm wide 7-18 H2, CH4, N2, O2, Ar without direct heating 45 - 959 0.2-4.5 0.6-2 This work 

In the past, statistical techniques such as Taguchi’s method21,22 and factorial designs23 have been used to find optimum conditions for graphene and carbon nanotube synthesis, respectively. However, basic factorial designs become less effective for large and complex systems. Thus, a holistic framework encompassing design, control and quantification of uncertainty is required. Such an approach based on statistical methods can be further strengthened by data obtained from an in-situ monitoring system of the growth process.

The intensity of the emission lines from optical emission spectroscopy (OES) can be used to understand species concentrations in the plasma qualitatively.24 Previous experimental results indicate that the ID/IG ratio, which is an inverse measure of graphene quality and grain size, improves with increased CH emission intensity from OES.25 In contrast, previous numerical studies have found that the CH species is more favorable than other carbon species for graphene growth on Cu.26,27 Because of the dearth of prior experimental work to determine the roles of different plasma species in graphene growth, this work utilize OES to reveal the effects of different species on graphene quality and to understand and control the deposition at a relatively large scale.

A design of experiments (DOE) technique is implemented in this work to optimize the deposition of graphene on copper foil in a roll-to-roll plasma CVD system. Plasma power, pressure and gas constituents of hydrogen (H2), methane (CH4), nitrogen (N2), oxygen (O2), and argon (Ar) are varied to optimize the growth of graphene. Statistical data-driven models have been built to study the effects of these process parameters on the graphene quality, which is defined by the intensity ratio of the D and G peaks (ID/IG) from Raman spectroscopy. Also, quantitative monitoring of the plasma species during graphene growth is demonstrated using an optical emission spectroscopy (OES) technique. A deposition model of graphene on copper foil in the present system is developed based on the results from OES and Raman correlations. Finally, the effects of the processing speed on graphene deposition are studied and show that plasma CVD is suitable for large-scale production of graphene.

Figure 1 shows the roll-to-roll plasma CVD system in which a sample is placed in the top free-moving winder, passes through the plasma region, and finally collects in the bottom driving winder. The plasma power is supplied from an rf power generator at a frequency of 80 kHz with a maximum power of 5000 W (Diener Electronics Co.). The power is supplied through igniter rods connected to two parallel rectangular electrodes. A capacitively coupled plasma (CCP) is generated between the two electrodes, which are separated by a 4.5 cm gap. The copper sample (76 μm thick and 2.54 cm wide annealed copper 110) is placed 1.5 cm from the right electrode, and the frontal areas of the left and right electrodes are 136.0 and 62.5 cm2, respectively. The asymmetry of the electrodes ensures that the substrate is placed in a higher temperature region. Graphene grows on both sides of the copper foil with different qualities and deposition rates depending on process conditions.

FIG. 1.

(a) Setup of the roll-to-roll plasma CVD system. (b) The plasma region where the copper foil is processed for graphene growth. Graphene is deposited on the right and left sides of the copper foil. (c) Top view of the chamber with the optical emission spectroscopy setup.

FIG. 1.

(a) Setup of the roll-to-roll plasma CVD system. (b) The plasma region where the copper foil is processed for graphene growth. Graphene is deposited on the right and left sides of the copper foil. (c) Top view of the chamber with the optical emission spectroscopy setup.

Close modal

Gas mixtures of hydrogen (H2), methane (CH4), nitrogen (N2), oxygen (O2) and argon (Ar) flow in the chamber with maximum flow rates of 1000, 1000, 100, 100 and 5000 standard cubic centimeters per minute (sccm), respectively. The pressure in the chamber is monitored by 10 and 100 mbar pressure gauges. The process parameters (i.e., plasma power, pressure, and mole fractions of H2, CH4, N2, O2, and Ar) vary between lower and upper limits to explore the parametric space during the optimization process, as listed in Table II. Limits and the plasma/power constraint are selected to sustain a stable plasma during the growth process (see Figure S1 in the supplementary material).

TABLE II.

Plasma operating conditions and constraints implemented in this work.

ParameterLower LimitUpper LimitConstraint
Hydrogen (H220% 60%  
Methane (CH410 % 30 % H2%+CH4%+N2% 
Nitrogen (N20 % 10 % +O2(%)+Ar(%)=100% 
Oxygen (O20 % 10 %  
Argon (Ar) 20 % 50 %  
Pressure 7 mbar 18 mbar 0<Pressurembar+PowerWatt50<10 
Power 500 Watt 1400 Watt  
ParameterLower LimitUpper LimitConstraint
Hydrogen (H220% 60%  
Methane (CH410 % 30 % H2%+CH4%+N2% 
Nitrogen (N20 % 10 % +O2(%)+Ar(%)=100% 
Oxygen (O20 % 10 %  
Argon (Ar) 20 % 50 %  
Pressure 7 mbar 18 mbar 0<Pressurembar+PowerWatt50<10 
Power 500 Watt 1400 Watt  

In-situ emission of the plasma in the range of 300-933 nm was detected by a spectrometer during the optimization process experiments (Fig. 1 (c)). The optical emission spectroscopy consists of lenses to focus the plasma emission to a fiber optic which is connected to a spectrometer (Princeton Instruments, Acton SP-2756). The spectrometer has a focal length of 750 mm and a grating of 500 nm Blaze. The diffracted light is detected by a CCD camera (Princeton Instruments, PIXIS 256E). The wavelength of the spectrometer was calibrated using a Mercury light source. Emission intensity was also calibrated using an intensity calibration light source to account for grating and CCD camera efficiencies.

Graphene films were analyzed using a scanning electron microscopy (SEM) (Hitachi Corp., s-4800) by measuring secondary electrons (SE) and backscattering electrons (BSE).28 Optical transmission measurements of graphene transferred to a glass slide were performed in the visible wavelength range (PerkinElmer, Lambda 950) to estimate graphene thickness. Also, Raman spectroscopy (Horiba Ltd.) was used to study the quality of graphene on both sides of the copper foil. The wavelength of the laser excitation is 532 nm with a magnification of 100x. The Raman spectrum is obtained directly on the Cu substrate to eliminate effects induced by the transfer of graphene from the copper substrate to background-free substrates (i.e., SiO2/Si). Several graphene samples were transferred to SiO2/Si substrates and show similar results to the direct measurements from the Cu substrate after background subtraction (see Figures S2 and S3 in the supplementary material).

In the optimization process, the speed of the moving copper foil in the plasma (web speed) is set to 45 (±3) mm/min to minimize the effects of ion bombardment. Figure 2 presents Raman spectra of graphene showing D, G, D’ and 2D peaks at 1350, 1580, 1620 and 2700 cm-1, respectively.29 The G peak represents the sp2 carbon structure due to C-C in-plane vibrations, whereas the D and D’ peaks appear because of defects in the graphene lattice. The 2D (or G’) peak is an overtone of the D peak and occurs because of a double-resonance process. The peak intensity ratio of the 2D and G peaks (I2D/IG) and the lineshape of the 2D peak can be related to the number of layers of graphene.30 Nevertheless, graphene samples produced from CVD processes have been reported to have weak coupling between layers resulting in disordered layer stacking.29 Therefore, interpretation of (I2D/IG) and the lineshape of the 2D peak from Raman results of these samples should be made with care. Alternatively, the D and G peak ratio (ID/IG) has stronger correlation to the defect density level and the crystalline size (La).31–33 For instance, the ID/IG ratio decreased from 2.6 to 1.6 upon lowering the residence time from 22.0 to 4.5 minutes (Fig. 2). The defect density increases with residence time due to increased ion bombardment that initiate defects in the lattice. Therefore, ID/IG is preferred to I2D/IG for representing the quality of graphene, and the ID/IG ratio is used in this work as an objective function in our statistical optimization process as explained next.

FIG. 2.

Raman spectra of graphene that show more defects with increased plasma residence time. Other conditions are kept constant at 60% H2, 25% CH4, and 15% Ar at 15 mbar.

FIG. 2.

Raman spectra of graphene that show more defects with increased plasma residence time. Other conditions are kept constant at 60% H2, 25% CH4, and 15% Ar at 15 mbar.

Close modal

A statistical design methodology34 is applied to minimize the ID/IG ratio from Raman spectroscopy for the right and left sides of the foil. We minimize both objectives by casting the problem as a two-objective stochastic optimization problem:

x^=arg maxE[Oi(x)],   i=1,2
(1)

where x is the process control parameter (plasma power, gas pressure and gas constituents), Oi(x), is the experimentally measured ID/IG ratio of the graphene deposited on the left side (i = 1) and on the right side (i = 2) of the Cu foil, and E denotes the expectation over the measurement noise. Figure 3 summarizes the optimization methodology that initially starts by exploring the effects of the process parameters (x) on the quality of both sides. Due to the asymmetric physical conditions on the left side and right side of the Cu foil, the two objectives may be competing, i.e., choosing a process input that increases the quality of graphene deposited on the left side of the Cu foil may result in decreased quality of graphene deposited on the right side. We say that a set of process inputs x1 dominates a different set of process inputs x2 if E[Oix1]E[Oix2]fori=1,2. In words, x1 dominates x2 if x1 results in higher graphene quality on both sides of the Cu foil. The set of non-dominated process inputs is called the Pareto efficient frontier (PEF) of the two-objective stochastic optimization problem, as shown in Fig. 3. For any dominated operating point, there is always an operating point on the Pareto efficient frontier that results in higher quality of graphene on both sides of the Cu foil. In this sense, the PEF defines the optimal operating points of the plasma CVD system for high-quality graphene deposition. The details of the statistical methods are provided in Sec. III of the supplementary material.

FIG. 3.

The optimization process methodology starts with performing experiments at different design variables (process parameters). The results are presented in a Pareto efficient frontier with epistemic uncertainty. If this uncertainty is not acceptable, a new set of experiment is selected to increase the information and to decrease the ID/IG ratio. Data-driven models are developed from these experimental results. The process iterates until convergence initiated by small epistemic uncertainty. As a result, the final Pareto efficiency frontier of the objective (ID/IG) is obtained.

FIG. 3.

The optimization process methodology starts with performing experiments at different design variables (process parameters). The results are presented in a Pareto efficient frontier with epistemic uncertainty. If this uncertainty is not acceptable, a new set of experiment is selected to increase the information and to decrease the ID/IG ratio. Data-driven models are developed from these experimental results. The process iterates until convergence initiated by small epistemic uncertainty. As a result, the final Pareto efficiency frontier of the objective (ID/IG) is obtained.

Close modal

Utilizing the advantages of the roll-to-roll process, multiple experiments were conducted in one set (batch) with negligible influence of experimental sequence on graphene deposition (see Figure S4 in the supplementary material). The total number of experiments in the optimization process is 101, conducted sequentially in 6 sets. After starting with initial sets of random measurements, called Sets 1-3 in Fig. 4, the methodology selects the next conditions of the seven process parameters to maximize the quality and information in the PEF. Three more experimental sets were conducted to improve the graphene quality, labeled Sets 4, 5, and 6 in Figure 4. The PEF for Set 6 is included in Fig. 3, in which the locations of the optimized conditions are marked with green symbols. The ID/IG ratio decreases for Sets 4-6 compared to the initial sets (Sets 1-3). Despite the presence of high-energy ions in the low-frequency rf CCP, the experimental design improves the ID/IG ratio from 1.4 to 0.7, as shown in Fig. 4.

FIG. 4.

Sequential optimization of graphene on both sides of Cu foil. SEM images are shown for representative low and high qualities of graphene.

FIG. 4.

Sequential optimization of graphene on both sides of Cu foil. SEM images are shown for representative low and high qualities of graphene.

Close modal

The optimized process parameters from the above sequential sets are 31% H2, 25% CH4, 4% N2, 1% O2, and 39% Ar at 9.2 mbar and 750 W. This condition produces the minimum ID/IG ratio on both sides as suggested by our model in three different sets of the sequential optimization process. The Raman spectra of graphene on both sides at this condition are shown in Fig. 5 (a). The ID/IG ratio for both sides is 0.7 which remains uniform across the width and length of the substrate as presented in Fig. 5 (b). The optimized condition produces uniform growth coverage as evident from secondary electron (SE) and backscattering electrons (BSE) SEM images from the same region, as illustrated in Fig. 5 (c) and (d). In addition, oxidation tests of the optimized condition and another sample obtained using high plasma power were conducted (both cases were used to study the web speed effects on graphene growth as described in Section D). The results from the oxidation tests confirm the uniformity of graphene growth and quality, as it acts as a corrosion barrier on copper foil due to the gas impermeability and thermal stability of graphene (Fig. S5 in the supplementary material).35 

FIG. 5.

(a) Raman spectrum of the optimized condition at 31% H2, 25% CH4, 4% N2, 1% O2, 39% at 9.2 mbar and 750 W. (b) Spatial dependence of ID/IG along the width of the 1-inch Cu foil representing uniform growth of graphene for the optimized condition, whereas the quality is not uniform for graphene growth at 50% H2, 21% CH4, 0% N2, 0% O2, 29% Ar at 18.8 mbar and 1250 W. SEM images of graphene on copper obtained from (c) secondary electrons (SE) and (d) backscattering electrons (BSE) of a uniform deposition whereas (e) shows a non-uniform deposition from backscattering electrons (BSE) taken from a non-optimized condition.

FIG. 5.

(a) Raman spectrum of the optimized condition at 31% H2, 25% CH4, 4% N2, 1% O2, 39% at 9.2 mbar and 750 W. (b) Spatial dependence of ID/IG along the width of the 1-inch Cu foil representing uniform growth of graphene for the optimized condition, whereas the quality is not uniform for graphene growth at 50% H2, 21% CH4, 0% N2, 0% O2, 29% Ar at 18.8 mbar and 1250 W. SEM images of graphene on copper obtained from (c) secondary electrons (SE) and (d) backscattering electrons (BSE) of a uniform deposition whereas (e) shows a non-uniform deposition from backscattering electrons (BSE) taken from a non-optimized condition.

Close modal

On the other hand, partial coverage of graphene was detected for other samples grown at non-optimal conditions. For instance, a non-uniform ID/IG ratio across the width of the substrate is shown in Fig. 5 (b) with higher values near the edges due to the non-uniform temperature and/or ion flux across the substrate that result in higher ID/IG ratio for this sample. Also, since backscattering electrons originate from elastic scattering processes that increase with larger atomic number (Z), the difference between copper with Z=29 and carbon with Z=6 can be detected depending on the contrast of BSE SEM images.28 Indeed, partial coverage of graphene on copper for a non-optimized condition is observed from the BSE SEM image as evident in Fig. 5(e). The image indicates partial coverage of graphene with darker regions compared to brighter regions of copper due to its higher Z.

The high defect density is a consequence of limitations imposed by the low-frequency CCP discharge. For instance, graphene quality from several samples, including the above optimized condition, remains similar when decreasing the length of the plasma from 20 to 5 cm using two symmetric small electrodes (Fig. S6). Even though the plasma size decreases by 75%, the quality of graphene does not improve due to the severe effects of ion bombardment. These results suggest that high-energy ions, which exist at this low frequency, have large effects on graphene quality regardless of the plasma size. Keeping the substrate for longer residence time in the plasma was found to enhance the graphitization processes and thus lead to higher I2D/IG as evident in Fig. 2. Similar graphene quality was observed by Terasawa and Saiki25 with an ID/IG area ratio of approximately 0.7 using a higher frequency of 13.56 MHz rf CCP CVD system at 300 W, 1 Pa, 80% H2 and 20% CH4. Even though CCP discharges have higher ion energy and larger sheath thickness as plasma frequency decreases,36,37 the optimization process improved the quality of graphene in our low-frequency plasmas by exploring the process parameter space more efficiently as discussed below.

The plasma parameters and their interactions play important roles in optimizing the quality of graphene. For example, Mehedi et al.21 found that inclusion of interactions among process parameters is important to reach optimum conditions for high-quality graphene. Thus, using statistical design of experiments, the influence of each of the plasma parameter on the graphene quality can be included in the surrogate models. These data-driven models for the two objectives (ID/IG ratio of graphene on both sides of Cu foil) are developed and validated with the experimental results. Figure 6 shows a comparison between the model and experimental results from Raman spectroscopy at three different positions. Initially, the model in Fig. 6 (a) deviates from the experiments because of the relatively small number of experiments that intend to cover the seven dimensions (i.e., the process parameters). The data-driven models are improved greatly with increased experiments (Fig. 6 (b)), and are then used to derive the importance of each process parameter on graphene quality as discussed below.

FIG. 6.

Sequential batch runs during the exploration and optimization process. The model results in (b) Validation 6 agree better with the experimental results than the results in (a) Validation 4 due to the increased number of experiments.

FIG. 6.

Sequential batch runs during the exploration and optimization process. The model results in (b) Validation 6 agree better with the experimental results than the results in (a) Validation 4 due to the increased number of experiments.

Close modal

From the models, a sensitivity analysis is performed to determine the relative influence of the plasma parameters on the two objectives. The sensitivity factor is derived from length scales that represent the variation of the measured values from the surrogate model values (refer to the supplementary material for further details on length scales). The larger the length scale of a process parameter, the less variation of the objective caused by this parameter. Hence, the inverse of the length scales is used to represent the relative sensitivity of the objective to each process condition. Figure 7 identifies the relative importance of the process parameters to the quality of graphene on both sides of the Cu foil. The most important input conditions affecting graphene quality on the right side (in decreasing order of their effect) are pressure, nitrogen mole fraction and oxygen mole fraction. On the other hand, the variables influencing the quality on the left side are oxygen mole fraction, plasma power, and finally pressure and methane mole fraction. These important input parameters affect the plasma properties and consequently the quality of graphene as discussed in the following paragraphs.

FIG. 7.

The sensitivity analysis of the process inputs that affect the ID/IG extracted from the statistical surrogate models. The right side is more sensitive to pressure, then nitrogen mole fraction (N2) and thirdly to oxygen mole fraction (O2). On the other hand, the left side is more sensitive to oxygen mole fraction (O2) followed by power and finally methane mole fraction (CH4) and pressure.

FIG. 7.

The sensitivity analysis of the process inputs that affect the ID/IG extracted from the statistical surrogate models. The right side is more sensitive to pressure, then nitrogen mole fraction (N2) and thirdly to oxygen mole fraction (O2). On the other hand, the left side is more sensitive to oxygen mole fraction (O2) followed by power and finally methane mole fraction (CH4) and pressure.

Close modal

The pressure in the chamber has the largest influence on the quality of graphene on the right side because the ion bombardment on the substrate is more affected by pressure in low frequency rf plasmas.38 This is also attributed to the large influence of the pressure on the right electrode sheath that is adjacent to the substrate’s right side. With a rise in pressure, the sheath thickness decreases and the ID/IG ratio declines due to the absence of direct contact between the substrate and the sheath, which contains high-energy ions that are accelerated toward the substrate. However, as the pressure further increases, the plasma changes from alpha to gamma mode39 which is more energetic and thus leads to higher ID/IG ratios (Fig. S7 in the supplementary material). Therefore, an optimum pressure exists at which the sheath thickness is small in the alpha mode without any apparent transition to the gamma mode. This optimum pressure value is 9.2 mbar for the condition of 31% H2, 25% CH4, 4% N2, 1% O2, and 39% Ar at 750 W, as indicated by the response surface in Fig. 8 (a). The low quality for pressures below 9.2 mbar is due mainly to the large sheath thickness resulting from a lower collision rate and higher mean free path that promote ion bombardment.40 However, the quality decreases when the pressure increases beyond 9.2 mbar due to the transition to gamma mode, which has higher current density and electron number density39 that enhance ion flux to the substrate and thus degrade quality.

FIG. 8.

ID/IG response surfaces as a function of (a) pressure, (b) nitrogen, (c) oxygen and (d) plasma power. The line shows the mean values of the response surface for the predictive model. The shaded uncertainties represent the lack of knowledge (blue) and noise from the experiment (pink). The response surfaces are obtained for fixed conditions of the optimized condition 31% H2, 25% CH4, 4% N2, 1% O2, and 39% Ar at 9.2 mbar and 750 W, as indicated by the red and yellow arrows. Experimental results closer to this condition have larger symbols size.

FIG. 8.

ID/IG response surfaces as a function of (a) pressure, (b) nitrogen, (c) oxygen and (d) plasma power. The line shows the mean values of the response surface for the predictive model. The shaded uncertainties represent the lack of knowledge (blue) and noise from the experiment (pink). The response surfaces are obtained for fixed conditions of the optimized condition 31% H2, 25% CH4, 4% N2, 1% O2, and 39% Ar at 9.2 mbar and 750 W, as indicated by the red and yellow arrows. Experimental results closer to this condition have larger symbols size.

Close modal

The second important factor for the right side ID/IG ratio is nitrogen mole fraction, which affects the quality as shown in the response surface (Fig. 8 (b)). The ID/IG ratio decreases with increased nitrogen mole fraction until about 4% when keeping other conditions constant at the optimized condition of 31% H2, 25% CH4, 1% O2, and 39% Ar at 9.2 mbar and 750 W. Adding nitrogen in Ar and H2 plasmas, which are used as bath gases, results in higher dissipated power in the plasma because of the active vibrational modes of nitrogen gas. Thus, the excitation processes and transfer of energy between electrons and the neutral gas are enhanced with increased N2 mole fraction. For instance, the OES results suggest higher emission intensities with increased N2 mole fraction and hence more reactive species present in the plasma to enhance the deposition process. Indeed, the dissociation of methane increases with the addition N2 to produce HCN species from vibrational excitation reactions as reported for a 13.56 MHz rf plasma.41 Similarly, the addition of N2 to a H2/CH4 microwave plasma for diamond growth leads to higher production of CH species, indicating enhanced methane decomposition in N2-containing plasmas.42 

The addition of oxygen is the most important factor for graphene quality on the left side. Figure 8 (c) shows the response surface of the ID/IG ratio as a function of the oxygen concentration at the optimized condition of 31% H2, 25% CH4, 4% N2, and 39% Ar at 9.2 mbar and 750 W. Graphene quality increases when the concentration of oxygen increases till about 4% after which the quality decreases with further increases in oxygen mole fraction.43 The initial rise in quality (decreased ID/IG ratio) could be attributed to the consumption of hydrogen atoms by oxygen, and, thus, a corresponding decrease in graphene etching.44 Furthermore, oxygen reacts with amorphous carbon deposited on the Cu surface and results in better graphitization of the deposited carbon film. Previous reports indicate that the presence of oxygen atoms on copper enhances the growth of large graphene domain due to decreased nucleation density and result in higher growth rate.45 Similar results have been obtained when introducing oxygen in the gas mixture during graphene growth, and show that higher growth rate occurs at an optimum value of oxygen mole fraction.46 However, further increase in oxygen concentration will degrade graphene quality because of etching as a result of carbon oxidation.47 In addition, oxygen can bond with the graphene lattice to form oxygen-doped graphene. Also, energetic oxygen atoms can break C-C bonds of the deposited graphene and result in a disordered structure.48 Therefore, the optimized condition of graphene includes 1% oxygen in the gas mixture to increase graphene domain size and growth rate, without etching or damaging graphene films.

Figure 8 (d) presents the response surface of the ID/IG ratio as a function of the plasma power at the optimized condition. The quality decreases as power increases because of increased energetic ion concentrations and species fluxes to the substrate that induce defects in the deposited films.49–51 The rise in density of excited species with increased power is corroborated by measurements from OES as explained in the next section. On the other hand, the quality decreases at lower plasma power because of the low production of active species and decreased heating of the substrate. Therefore, an average power value at 750 W results in higher quality. The effects of plasma power on graphene growth at higher web speeds are discussed in Section D for two cases: this optimized condition at 750 W, and a higher deposition rate case at 1250 W.

Figure 9 shows an emission spectrum from a probe volume near the substrate material where the most prominent species are present (see Table SI). The species emission intensities are correlated to the quality of graphene using a statistical mapping analysis as described in the supplementary material (Fig. S8). The emission intensities of the species CN, N2+, CH, Ar+, C2, H2, Hα, Ar, O, C, and N measured using OES are mapped into a 1D space in order to extract important correlations to graphene quality. Figure 10 (a) presents the results of the analysis in which the negative of the importance factor “wi” indicates a decrease of ID/IG ratio, whereas a positive wi factor indicates an increase in ID/IG ratio. The emission of CH is correlated with an increase in graphene quality in agreement with previous theoretical results26,27 because the chemical chain -CH is more energetically favorable on Cu(111) than other carbon clusters.52 Furthermore, other species that have minor contributions to the increased quality are N, N2+, and H2, as more than 75% of the ranges of their importance factors indicate higher quality (i.e., negative wi).

FIG. 9.

A plasma emission spectrum sample measured during the growth process. Important species in the plasma are indicated in the spectrum.

FIG. 9.

A plasma emission spectrum sample measured during the growth process. Important species in the plasma are indicated in the spectrum.

Close modal
FIG. 10.

(a) The importance of species emission from OES affecting ID/IG peak ratio. The box plots show the median in red between the first and third quartiles. The species CH indicates higher quality of graphene, whereas CN, Ar+, C2 and O degrade the quality. The emission intensities variations of (b) Ar+ and (c) C2 with plasma power. (d) The increase of Ar+ emission intensity with increased gas pressure. Higher plasma power and/or gas pressure enhances the energy input in the plasma that leads to higher ion bombardment (Ar+) and thus more sputtering of carbon as C2. The optimum condition has minimum emissions intensities of Ar+ and C2 at the indicated plasma power and gas pressure of 750 W and 9.2 mbar, respectively. The scattering in (b)-(d) is due to the variations of other plasma conditions. The emission intensity of Ar+ in (b) and (d) are normalized by the mole fraction of Ar. The high emission intensities of C2, as indicated by the circle, are from samples with thick graphene deposition.

FIG. 10.

(a) The importance of species emission from OES affecting ID/IG peak ratio. The box plots show the median in red between the first and third quartiles. The species CH indicates higher quality of graphene, whereas CN, Ar+, C2 and O degrade the quality. The emission intensities variations of (b) Ar+ and (c) C2 with plasma power. (d) The increase of Ar+ emission intensity with increased gas pressure. Higher plasma power and/or gas pressure enhances the energy input in the plasma that leads to higher ion bombardment (Ar+) and thus more sputtering of carbon as C2. The optimum condition has minimum emissions intensities of Ar+ and C2 at the indicated plasma power and gas pressure of 750 W and 9.2 mbar, respectively. The scattering in (b)-(d) is due to the variations of other plasma conditions. The emission intensity of Ar+ in (b) and (d) are normalized by the mole fraction of Ar. The high emission intensities of C2, as indicated by the circle, are from samples with thick graphene deposition.

Close modal

On the other hand, increased C2 and O emission indicates lower graphene quality. Similarly, the importance factors of CN, Hα, C and Ar+ emission lines lie within the low-quality region in Fig. 10 (a). The oxygen atoms affect the quality due to their high energy in the plasma, which can break carbon bonds of the graphene deposited on the substrate.53 Also, ion bombardment by Ar+ and the etching effects of Hα result in disorder in the carbon films deposited on the Cu substrate, likely resulting in sputtering of C2 and C species from the surface. Emission from C2 and C species are higher for samples with thicker carbon films, as evidenced by the corresponding increase in their emission intensities as discussed next. Another possible negative effect of C2 in graphene quality is the creation of vacancies due to C2 saturation on the copper substrate.52 

The process inputs for the optimized condition (31% H2, 25% CH4, 4% N2, 1% O2, and 39% Ar at 9.2 mbar and 750 W) provide a plasma with lower concentrations of C2, CN, O, Ar+ and H and thus result in deposition of high-quality graphene. For instance, the emission intensity of Ar+ increases with power due to higher energy input to the plasma (Fig. 10 (b)). The increase of Ar+ concentration can lead to higher defects in the lattice54 which could result in sputtering of C2 as suggested by the increase of C2 emission intensity in Fig. 10(c). Therefore, the optimum power of 750 W is high enough to heat the substrate and ionize the gas mixture, but at the same time has lower concentrations of Ar+ and C2 (and O and H, which are not shown here). Similar results are observed with the optimum pressure value of 9.2 mbar which, in addition to its important role in determining sheath thickness and plasma type, produces lower Ar+ emission as shown in Fig. 10(d).

The O atom emission intensity is relatively low at the optimized condition of 1% O2 and increases linearly with increased oxygen mole fraction as shown in Fig. 11(a). Higher concentration of oxygen affects the quality of graphene negatively due to etching as discussed above. Furthermore, oxygen consumes carbon species in the plasma which in turn decreases the deposition rate.46 For example, the emission of C2 increases with oxygen mole fraction initially, reaches a maximum near 5% O2 and then decreases with higher oxygen mole fraction (Fig. 11(b)), likely due to the consumption of carbon species in the plasma by oxygen to form CO2 or CO.

FIG. 11.

The effects of oxygen mole fraction on the emission intensities of (a) O and (b) C2. As the emission intensity of O increases with O2 mole fraction, more C2 is emitted initially as a result of carbon atoms breakage. The emission of C2 starts to decrease beyond 5% O2 due depletion of carbon in the plasma by reactions with oxygen. The optimum condition has 1% O2 to increase the deposition rate and decrease the nucleation density. (c) The variation of CN emission intensity with N2 mole fraction. After 4%, which is the optimum value, the emission intensity of CN rises rapidly to cause defects in graphene. The scattering in (a)-(c) is due to the variations of other plasma conditions.

FIG. 11.

The effects of oxygen mole fraction on the emission intensities of (a) O and (b) C2. As the emission intensity of O increases with O2 mole fraction, more C2 is emitted initially as a result of carbon atoms breakage. The emission of C2 starts to decrease beyond 5% O2 due depletion of carbon in the plasma by reactions with oxygen. The optimum condition has 1% O2 to increase the deposition rate and decrease the nucleation density. (c) The variation of CN emission intensity with N2 mole fraction. After 4%, which is the optimum value, the emission intensity of CN rises rapidly to cause defects in graphene. The scattering in (a)-(c) is due to the variations of other plasma conditions.

Close modal

Similarly, the optimized condition includes 4% N2 that produces a low concentration of CN as shown in Fig. 11(c). The CN emission intensity rises rapidly for N2 mole fraction above 4%. From a mapping analysis, CN is considered to contribute negatively to graphene growth due to the consumption of CH by N2. For instance, the concentration of CN increases more rapidly than CH with increasing pressure and power due to the reaction: CH+ N2 HCN + N.42 Hence, an optimum amount of 4% N2 is needed to increase the dissociation of methane through vibrational excitations reactions, without consuming important carbon radicals and ions that contribute to high-quality graphene. The emission intensity of CN has lower values at 9% and 10% N2 due to variations of other process inputs. These lower values of CN could explain the other local minimum in ID/IG ratio at 9% and 10% N2 in the response surface of Fig. 8(b).

The variations in the ID/IG ratio with CH/O, CH/C2 and H2/H emission intensity ratios are presented in Fig. 12. These ratios represent the combined effects of species that indicate higher quality (i.e., CH or H2) to the species that degrade quality (i.e., O, C2 or H). The quality decreases almost linearly with increasing CH/O (Fig. 12 (a)). The wide range of ID/IG (i.e., 0.75-1.20) occurs when CH/O is 1.0 or less. However, when the CH/O ratio increases from 1.0 to 4.0, the ID/IG ratio decreases linearly from 0.85 to 0.75. Furthermore, Figure 12 (b) shows a decreased ID/IG ratio with increased CH/C2, indicating higher quality with higher CH and lower C2. Finally, the effects of the H2/H emission ratio on the quality of graphene is presented in Fig. 12 (c). The quality of graphene improves with increased H2/H. The ID/IG ratio varies between 0.52 and 1.40 at lower ratios of H2/H whereas ID/IG remains constant at 0.8 as H2/H increases from 0.025 to 0.06. These emission intensity ratios exhibit higher values at the optimized condition (see Fig S9). Thus, these results suggest that higher CH/O, CH/C2 and H2/H emission intensity ratios are signatures for high quality and thus can be used for online monitoring for large production of graphene.

FIG. 12.

(a) The effects of the emission intensity ratio of CH/O to the ID/IG ratio which decreases linearly with increased CH/O. (b) The ID/IG ratio decreases exponentially with increased CH/C2 emission intensity ratio. (c) The ID/IG ratio decays exponentially as the intensity ratio H2/H increases.

FIG. 12.

(a) The effects of the emission intensity ratio of CH/O to the ID/IG ratio which decreases linearly with increased CH/O. (b) The ID/IG ratio decreases exponentially with increased CH/C2 emission intensity ratio. (c) The ID/IG ratio decays exponentially as the intensity ratio H2/H increases.

Close modal

Correlations between the D-peak position, D-peak full width at half maximum (FWHM), and G-peak FWHM with the OES data were made using the mapping analysis as shown in Fig. 13 (a), (b) and (c), respectively. The results from these correlations indicate that emission from O and C2 (among others) is associated with increasing the D-peak position, D-peak FWHM, and G-peak FWHM values. The dependence of D-peak position and the G- and D-peak FWHMs on emission ratios of the important species are presented in Fig. 14. The D-peak position decreases exponentially with increased CH/O (Fig. 14 (a)). At lower CH/O ratio, the position of the D peak ranges from 1360-1372 cm-1, but then decreases to an asymptotic value around 1345 cm-1 when the CH/O ratio is higher than 2.0. Such a result agrees with the increase of graphene quality with increased CH/O ratio (Fig. 12 (a)). Similarly, the effects of the Ar+/O ratio on the FWHM of the D peak are shown in Fig. 14 (b). The FWHM of the D peak are 160-200 cm-1 when the values of the Ar+/O ratio are less than or equal to 5. However, the D-peak FWHM decreases to a value around 140 cm-1 when the Ar+/O ratio is higher than 10. Lastly, the FWHM of the G peak decays exponentially with increased CH/O ratio and reaches an asymptotic value of 60 cm-1 (Fig. 14 (c)).

FIG. 13.

The importance of species emission from OES affecting: (a) D-peak position, (b) D-peak FWHM and (c) G-peak FWHM. The box plots show the median in red between the first and third quartiles. The species CH indicates higher quality of graphene, whereas C2 and O degrade the quality.

FIG. 13.

The importance of species emission from OES affecting: (a) D-peak position, (b) D-peak FWHM and (c) G-peak FWHM. The box plots show the median in red between the first and third quartiles. The species CH indicates higher quality of graphene, whereas C2 and O degrade the quality.

Close modal
FIG. 14.

(a) The D-peak position decreases when the intensity ratio of CH/O increases, indicating higher quality of graphene. (b) The effects of the Ar+/O ratio on the FWHM of the D-peak which show lower values of the FWHM with increased Ar+/O ratio. (c) The exponential decay of the G-peak FWHM with increased CH/O ratio presenting higher quality graphene at narrower G-peak.

FIG. 14.

(a) The D-peak position decreases when the intensity ratio of CH/O increases, indicating higher quality of graphene. (b) The effects of the Ar+/O ratio on the FWHM of the D-peak which show lower values of the FWHM with increased Ar+/O ratio. (c) The exponential decay of the G-peak FWHM with increased CH/O ratio presenting higher quality graphene at narrower G-peak.

Close modal

Based on the forgoing results, we propose a conceptual model of graphene deposition on copper foil in our system (Fig. 15). Considering the left side of the copper foil, the left electrode is powered during a half-cycle from the rf generator. The supplied electric field energizes the electrons in the plasma. Electrons are then accelerated toward the right electrode away from the cathode sheath. Gas molecules react with electrons to produce active radicals, ions and atoms. Intermediate species then reach the copper substrate with CH that favors growth of high-quality graphene. However, energetic species, such as O, Ar+, or H can break, sputter, or etch the carbon deposited on the copper, respectively. Consequently, C2 and C species are sputtered from the graphene deposited on the copper substrate. These processes occur during the rf cycle and result in graphene deposition with different thicknesses depending on the web speed as explored in the following section.

FIG. 15.

Proposed deposition mechanism of graphene on copper substrate during the roll-to-roll plasma CVD process. Intermediate species are produced in the plasma in which high-quality graphene is deposited on the Cu foil from CH, whereas energetic atoms and ions (i.e., O, H and Ar+) react with the deposited graphene, causing in carbon atoms breakage.

FIG. 15.

Proposed deposition mechanism of graphene on copper substrate during the roll-to-roll plasma CVD process. Intermediate species are produced in the plasma in which high-quality graphene is deposited on the Cu foil from CH, whereas energetic atoms and ions (i.e., O, H and Ar+) react with the deposited graphene, causing in carbon atoms breakage.

Close modal

After optimization of graphene at a lower web speed (45 mm/min) in the previous sections, the effect of web speed on graphene deposition was studied for two cases for speeds up to 959 mm/min (Fig. 16). The condition for Case 1 is similar to the optimized condition (31% H2, 25% CH4, 4% N2, 1% O2, 39% Ar, 9.2 mbar, and 750 W), whereas the condition for Case 2 is 33% H2, 30% CH4, 10% N2, 8% O2, 19% Ar, 15 mbar, and 1250 W. The quality in the optimized condition (Case 1) decreases with increased web speed (Fig. 16(a)). At lower plasma power (750 W), longer residence time is needed to increase the substrate temperature, and supply sufficient carbon for deposition. The decrease of graphene quality with web speed for Case 1 is consistent with previous reported results, but for a thermal CVD system.15 However, the ID/IG ratios of both sides in Case 2 are independent of web speed because of the high-power plasma (Fig. 16 (a)). The high plasma power of Case 2 is sufficient to heat the substrate and produce active carbon species for graphene deposition, even at higher web speed.

FIG. 16.

The effects of speed variation on graphene quality: (a) The variation of ID/IG ratio with web speed. (b) Raman spectra of Case 2 samples at different speeds. Lower intensities of Raman spectra, measured with a similar setup, with increased speed. (c) Optical transmittance of graphene of Case 2 as a function of web speed. The number of deposited graphene layers decreases almost linearly with web speed in the studied time range with deposition rate of 7 (±1) layers/min as plotted in the inset.

FIG. 16.

The effects of speed variation on graphene quality: (a) The variation of ID/IG ratio with web speed. (b) Raman spectra of Case 2 samples at different speeds. Lower intensities of Raman spectra, measured with a similar setup, with increased speed. (c) Optical transmittance of graphene of Case 2 as a function of web speed. The number of deposited graphene layers decreases almost linearly with web speed in the studied time range with deposition rate of 7 (±1) layers/min as plotted in the inset.

Close modal

Figure 17 presents SEM images for Case 2 at different speeds compared to a static experiment conducted at a longer residence time of 30 min. The SEM image of the longest residence time shows an area where the carbon film has peeled off, indicating the deposition of thicker carbon layers. We also noticed the appearance of microcracks in the graphene film for Case 2 at speed of 45 mm/min. These negative features could be attributed to the tension during the roll-to-roll process, since the cracks are perpendicular to the direction of rolling.55 A higher resolution SEM image of the crack region is shown in Fig. 17 in which wrinkles and microcracks exist. The tension during the roll-to-roll processing accompanied with heating of the Cu foil could lead to microcracks due to the different thermal expansion coefficients between the copper substrate and the graphene.17 

FIG. 17.

SEM images of graphene on copper foil substrate as a function of web speed (or residence time which is equal to the length of the electrode divided by the web speed). Thicker graphene was deposited when the substrate was stationary in the plasma for 30 min due to the high deposition rate in plasma CVD. Also, copper grains exist at higher speeds, indicating that the plasma provides sufficient power to heat the substrate even at a low residence time of 0.2 min.

FIG. 17.

SEM images of graphene on copper foil substrate as a function of web speed (or residence time which is equal to the length of the electrode divided by the web speed). Thicker graphene was deposited when the substrate was stationary in the plasma for 30 min due to the high deposition rate in plasma CVD. Also, copper grains exist at higher speeds, indicating that the plasma provides sufficient power to heat the substrate even at a low residence time of 0.2 min.

Close modal

On the other hand, copper grains are apparent at high speeds, indicating a lower thickness of the deposited films. Moreover, the presence of the copper grain boundaries for the higher speed sample suggest that the substrate was heated to a temperature sufficient to recrystallize the Cu grains. Furthermore, Raman spectra show a decrease of the D- and G-peak intensities with increased web speed for samples from Case 2 as shown in Fig. 16 (b). These results suggest that graphene can be grown at about 1 m/min using plasma as a heat source.

These results reveal a tradeoff between quality and throughput of graphene production for samples prepared at lower plasma power (e.g., Case 1 in this work) or by thermal CVD. For example, the ID/IG ratio increases from 0.1 at 25 mm/min to 1.0 at 500 mm/min in thermal CVD as reported by Polsen et al15 The decrease of the quality with increased speed is due to limitations in growth kinetics and graphene nucleation in thermal CVD systems or lower plasma power (and different plasma conditions) as shown for Case 1. However, due to active species in the plasma that accelerate the graphene deposition rate, the plasma CVD produce graphene with similar quality as a function of speed until about 1 m/min for Case 2 (Fig. 16).

Graphene deposition in our roll-to-roll plasma CVD system has higher throughput due to a high growth rate of graphene on copper foil. In our system, the estimated growth rate of graphene is 7 (±1) layers/min for Case 2. This rate is estimated from optical transmittance measurements as evident in Fig. 16(c) by considering a 2.3% decrease of light transmittance for each layer of graphene.56 This high growth rate is due to the abundantly active radical species in the plasma. Similarly, Kato et al.20 showed that growth rate in plasma CVD is two to three orders of magnitudes higher than thermal CVD. Thus, plasma resources have the advantages of higher throughput for graphene deposition with lower input power compared to thermal CVD, as summarized in Table I. For instance, graphene can be deposited at a rate of 5-500 mm/min for thermal CVD systems whereas this rate can be raised to about 1000 mm/min using plasma CVD as reported here. However, graphene quality from plasma CVD is lower than for thermal CVD systems with negligible values of ID/IG compared to ID/IG values of 0.5 and 0.7 in plasma CVD systems as reported by Yamada et al.18 and the results of this work, respectively. Such moderate film quality limits the benefits of graphene production from plasma CVD to applications with less stringent demand of quality, such as oxidation barriers for metals, but not for high quality applications in electronics and photonics.57 

This work takes a statistical approach to scale up graphene production in a custom-built roll-to-roll rf plasma CVD system. Even though the quality of graphene is limited by the plasma in this process, a wealth of information obtained demonstrates the tremendous benefits of combining statistical analysis and in-situ process monitoring. Based on the developed surrogate models, the quality of graphene is largely influenced by gas pressure, nitrogen, oxygen, and plasma power. Moreover, the results from OES suggest correlations of CH species to low defects, while the presence of Ar+, C2, Hα, CN, and O correlate to high defects. This work provides fundamental physical insights for the design and characterization of high-throughput plasma CVD systems for graphene and other 2D materials manufacturing. Further development is needed to obtain higher quality of graphene in plasma systems by minimizing and controlling ion and molecular fluxes to the substrate.

See supplementary material for the images of plasma stability conditions as well as plasma types, the Raman spectroscopy analysis of graphene, the details of the statistical methods, the uniformity of graphene growth, the effects of plasma size on graphene quality, the details of the optical emission spectroscopy and the correlations between OES species emission ratios with process parameters are included in the supplementary material.

This work was supported by the US National Science Foundation through its Scalable Nanomanufacturing Program (Grant: 1344654). We thank Guy Telesnicki and the staff of Birck Nanotechnology Center at Purdue University for support during the design, start-up and operation of the roll-to-roll plasma CVD system. We also thank Prof. Xianfan Xu for the use of Raman spectroscopy equipment.

1.
A. K.
Geim
and
I. V.
Grigorieva
, “
Van der Waals heterostructures
,”
Nature
499
,
419
425
(
2013
).
2.
S. Z.
Butler
 et al, “
Progress, challenges, and opportunities in two-dimensional materials beyond graphene
,”
ACS Nano
7
,
2898
2926
(
2013
).
3.
A. K.
Geim
and
K. S.
Novoselov
, “
The rise of graphene
,”
Nat. Mater.
6
,
183
191
(
2007
).
4.
A. K.
Geim
, “
Graphene: Status and prospects
,”
Science
324
,
1530
1534
(
2009
).
5.
K. K.
Kim
 et al, “
Synthesis and characterization of hexagonal boron nitride film as a dielectric layer for graphene devices
,”
ACS Nano
6
,
8583
8590
(
2012
).
6.
N. R.
Glavin
 et al, “
Amorphous boron nitride: A universal, ultrathin dielectric for 2D nanoelectronics
,”
Advanced Functional Materials
26
,
2640
2647
(
2016
).
7.
F.
Bonaccorso
 et al, “
Production and processing of graphene and 2d crystals
,”
Mater. Today
15
,
564
589
(
2012
).
8.
Y.
Hernandez
 et al, “
High yield production of graphene by liquid phase exfoliation of graphite
,”
Nat. Nanotechnol.
3
,
563
568
(
2008
).
9.
K. R.
Paton
 et al, “
Scalable production of large quantities of defect-free few-layer graphene by shear exfoliation in liquids
,”
Nat. Mater.
13
,
624
630
(
2014
).
10.
S.
Stankovich
 et al, “
Synthesis of graphene-based nanosheets via chemical reduction of exfoliated graphite oxide
,”
Carbon N. Y.
45
,
1558
1565
(
2007
).
11.
X.
Li
 et al, “
Large-area synthesis of high-quality and uniform graphene films on copper foils
,”
Science
324
,
1312
1314
(
2009
).
12.
S.
Bae
 et al, “
Roll-to-roll production of 30-inch graphene films for transparent electrodes
,”
Nat. Nanotechnol.
5
,
574
578
(
2010
).
13.
I.
Vlassiouk
 et al, “
Large scale atmospheric pressure chemical vapor deposition of graphene
,”
Carbon N. Y.
54
,
58
67
(
2013
).
14.
T.
Hesjedal
, “
Continuous roll-to-roll growth of graphene films by chemical vapor deposition
,”
Appl. Phys. Lett.
98
(
2011
).
15.
E. S.
Polsen
,
D. Q.
McNerny
,
B.
Viswanath
,
S. W.
Pattinson
, and
A.
John Hart
, “
High-speed roll-to-roll manufacturing of graphene using a concentric tube CVD reactor
,”
Sci. Rep.
5
,
10257
(
2015
).
16.
G.
Zhong
 et al, “
Growth of continuous graphene by open roll-to-roll chemical vapor deposition
,”
Appl. Phys. Lett.
109
,
193103
(
2016
).
17.
T.
Kobayashi
 et al, “
Production of a 100-m-long high-quality graphene transparent conductive film by roll-to-roll chemical vapor deposition and transfer process
,”
Appl. Phys. Lett.
102
,
23112
(
2013
).
18.
T.
Yamada
,
M.
Ishihara
,
J.
Kim
,
M.
Hasegawa
, and
S.
Iijima
, “
A roll-to-roll microwave plasma chemical vapor deposition process for the production of 294mm width graphene films at low temperature
,”
Carbon N. Y.
50
,
2615
2619
(
2012
).
19.
A.
Kumar
,
A. A.
Voevodin
,
D.
Zemlyanov
,
D. N.
Zakharov
, and
T. S.
Fisher
, “
Rapid synthesis of few-layer graphene over Cu foil
,”
Carbon N. Y.
50
,
1546
1553
(
2012
).
20.
R.
Kato
,
S.
Minami
,
Y.
Koga
, and
M.
Hasegawa
, “
High growth rate chemical vapor deposition of graphene under low pressure by RF plasma assistance
,”
Carbon N. Y.
96
,
1008
1013
(
2016
).
21.
H.-A.
Mehedi
 et al, “
Synthesis of graphene by cobalt-catalyzed decomposition of methane in plasma-enhanced CVD: Optimization of experimental parameters with Taguchi method
,”
J. Appl. Phys.
120
(
2016
).
22.
S.
Santangelo
 et al, “
Taguchi optimized synthesis of graphene films by copper catalyzed ethanol decomposition
,”
Diam. Relat. Mater.
41
,
73
78
(
2014
).
23.
Á.
Kukovecz
,
D.
Méhn
,
E.
Nemes-Nagy
,
R.
Szabó
, and
I.
Kiricsi
, “
Optimization of CCVD synthesis conditions for single-wall carbon nanotubes by statistical design of experiments (DoE)
,”
Carbon N. Y.
43
,
2842
2849
(
2005
).
24.
J.
Ma
,
M. N. R.
Ashfold
, and
Y. A.
Mankelevich
, “
Validating optical emission spectroscopy as a diagnostic of microwave activated CH4/Ar/H2 plasmas used for diamond chemical vapor deposition
,”
J. Appl. Phys.
105
,
43302
(
2009
).
25.
T.
Terasawa
and
K.
Saiki
, “
Growth of graphene on Cu by plasma enhanced chemical vapor deposition
,”
Carbon N. Y.
50
,
869
874
(
2012
).
26.
H.
Shu
 et al, “
What are the active carbon species during graphene chemical vapor deposition growth?
,”
Nanoscale
7
,
1627
1634
(
2015
).
27.
W.
Zhang
,
P.
Wu
,
Z.
Li
, and
J.
Yang
, “
First-principles thermodynamics of graphene growth on Cu surfaces
,”
J. Phys. Chem. C
115
,
17782
17787
(
2011
).
28.
J.
Kim
,
F.
Kim
, and
J.
Huang
, “
Seeing graphene-based sheets
,”
Mater. Today
13
,
28
38
(
2010
).
29.
L. M.
Malard
,
M. A.
Pimenta
,
G.
Dresselhaus
, and
M. S.
Dresselhaus
, “
Raman spectroscopy in graphene
,”
Phys. Rep.
473
,
51
87
(
2009
).
30.
A. C.
Ferrari
 et al 
Raman spectrum of graphene and graphene layers
,”
Phys. Rev. Lett.
97
,
187401
(
2006
).
31.
F.
Tuinstra
and
J. L.
Koenig
, “
Raman spectrum of graphite
,”
J. Chem. Phys.
53
,
1126
1130
(
1970
).
32.
K.
Sato
, “
D-band Raman intensity of graphitic materials as a function of laser energy and crystallite size
(
2006
).
33.
L. G.
Caņado
 et al, “
General equation for the determination of the crystallite size la of nanographite by Raman spectroscopy
,”
Appl. Phys. Lett.
88
,
163106
(
2006
).
34.
P.
Pandita
, “
Stochastic multi-objective optimization on a budget: Application to multi-pass wire drawing with quantified uncertainties
,” arXiv:1706.01665 (
2017
).
35.
C.
Jia
,
J.
Jiang
,
L.
Gan
, and
X.
Guo
, “
Direct optical characterization of graphene growth and domains on growth substrates
,”
Sci. Rep.
2
,
707
(
2012
).
36.
A.
Perret
,
P.
Chabert
,
J.
Jolly
, and
J.-P.
Booth
, “
Ion energy uniformity in high-frequency capacitive discharges
,”
Appl. Phys. Lett.
86
,
21501
(
2005
).
37.
A.
Salabas
,
L.
Marques
,
J.
Jolly
,
G.
Gousset
, and
L. L.
Alves
, “
Systematic characterization of low-pressure capacitively coupled hydrogen discharges
,”
J. Appl. Phys.
95
,
4605
4620
(
2004
).
38.
Y.
Catherine
and
P.
Couderc
, “
Electrical characteristics and growth kinetics in discharges used for plasma deposition of amorphous carbon
,”
Thin Solid Films
144
,
265
280
(
1986
).
39.
Y. P.
Raizer
,
M. N.
Shneider
, and
N. A.
Yatsenko
,
Radio-frequency capacitive discharges
(
CRC Press
,
1995
).
40.
N.
Mutsukura
,
K.
Kobayashi
, and
Y.
MacHi
, “
Plasma sheath thickness in radio-frequency discharges
,”
J. Appl. Phys.
68
,
2657
2660
(
1990
).
41.
S. Y.
Savinov
,
H.
Lee
,
H. K.
Song
, and
B. K.
Na
, “
The effect of vibrational excitation of molecules on plasmachemical reactions involving methane and nitrogen
,”
Plasma Chem. Plasma Process.
23
,
159
173
(
2003
).
42.
B. S.
Truscott
,
M. W.
Kelly
,
K. J.
Potter
,
M. N. R.
Ashfold
, and
Y. A.
Mankelevich
, “
Microwave plasma-activated chemical vapor deposition of nitrogen-doped diamond. II: CH4/N2/H2 plasmas
,”
J. Phys. Chem. A
120
,
8537
8549
(
2016
).
43.
A. J.
Strudwick
 et al, “
Chemical vapor deposition of high quality graphene films from carbon dioxide atmospheres
,”
ACS Nano
9
,
31
42
(
2015
).
44.
L.
Gan
and
Z.
Luo
, “
Turning off hydrogen to realize seeded growth of subcentimeter single-crystal graphene grains on copper
,”
ACS Nano
7
,
9480
9488
(
2013
).
45.
Y.
Hao
 et al, “
The role of surface oxygen in the growth of large single-crystal graphene on copper
,”
Science
342
,
720
723
(
2013
).
46.
T.
Terasawa
and
K.
Saiki
, “
Effect of vapor-phase oxygen on chemical vapor deposition growth of graphene
,”
Appl. Phys. Express
8
,
35101
(
2015
).
47.
E.
Starodub
,
N. C.
Bartelt
, and
K. F.
McCarty
, “
Oxidation of graphene on metals
,”
J. Phys. Chem. C
114
,
5134
5140
(
2010
).
48.
A.
Nourbakhsh
,
M.
Cantoro
,
T.
Vosch
,
G.
Pourtois
,
F.
Clemente
,
M. H.
Van Der Veen
,
J.
Hofkens
,
M. M.
Heyns
,
S.
De Gendt
, and
B. F.
Sels
, “
Bandgap opening in oxygen plasma-treated graphene
,”
Nanotechnology
21
,
435203
(
2010
).
49.
Y. S.
Kim
 et al, “
Methane as an effective hydrogen source for single-layer graphene synthesis on Cu foil by plasma enhanced chemical vapor deposition
,”
Nanoscale
5
,
1221
1226
(
2013
).
50.
L.
Van Nang
and
E.-T.
Kim
, “
Controllable synthesis of high-quality graphene using inductively-coupled plasma chemical vapor deposition
,”
J. Electrochem. Soc.
159
,
K93
(
2012
).
51.
L.
Cheng
 et al, “
Low temperature synthesis of graphite on Ni films using inductively coupled plasma enhanced CVD
,”
J. Mater. Chem. C
3
,
5192
5198
(
2015
).
52.
T.
Niu
,
M.
Zhou
,
J.
Zhang
,
Y.
Feng
, and
W.
Chen
, “
Growth intermediates for CVD graphene on Cu(111): Carbon clusters and defective graphene
,”
J. Am. Chem. Soc.
135
,
8409
8414
(
2013
).
53.
J. M.
Grace
and
L. J.
Gerenser
, “
Plasma treatment of polymers
,”
Journal of Dispersion Science and Technology
24
(
2003
).
54.
M. M.
Lucchese
 et al, “
Quantifying ion-induced defects and Raman relaxation length in graphene
,”
Carbon N. Y.
48
,
1592
1597
(
2010
).
55.
S. R.
Na
 et al, “
Cracking of polycrystalline graphene on copper under tension
,”
ACS Nano
10
,
9616
9625
(
2016
).
56.
R. R.
Nair
 et al, “
Fine structure constant defines visual transparency of graphene
,”
Science
320
,
1308
(
2008
).
57.
K. S.
Novoselov
 et al, “
A roadmap for graphene
,”
Nature
490
,
192
200
(
2012
).

Supplementary Material