Received August 10, 2007

Revised Manuscript Received September 28, 2007

Introduction

Single-walled carbon nanotubes (SWCNT)¹ are of great interest to science and industry because of their electronic and vibronic properties,² which allow the study of one-dimensional physics as well as applications in numerous devices. Among other things, the high aspect ratio and the tuneability of the electronic structure is of advantage. Synthesized SWCNT of high quality, with a narrow diameter distribution and a well-defined mean diameter are desired for several proposed applications as well as fundamental studies. Today, there exist three primary routes for their synthesis. In arc-discharge³, an arc is ignited between two electrodes consisting of carbon mixed with some metal catalyst. Although the yield is relatively good, the as-produced material includes many catalyst particles as well as amorphous carbon. In laser ablation (LA)⁴, a laser beam is used to evaporate a carbon/metal target. LA allows for narrow diameter distributions with appropriate tuning while maintaining a high yield, but the technique is expensive and not suited for mass production. Chemical vapor deposition (CVD)⁵ is the most promising method for a continuous high yield nanotube synthesis. There has been recent progress in CVD synthesis using tailored catalyst particles in order to grow carbon nanotubes in a specific diameter range^{6, 7}. Alexandrescu et al.⁸ did some work toward combining catalytic CVD and lasers in using a floating catalyst method with a continuous wave CO₂ laser beam for heating. Another form of laser-assisted chemical vapor deposition (LACVD) is the pyrolysis of metal−organic compounds⁹. These routes, however, do not use pulsed evaporation. Nishide et al.¹⁰ introduced a pulsed laser evaporation method in which a metal target is evaporated in a hydrocarbon atmosphere, namely, cobalt in ethanol. We further develop this promising approach, which blends the well-known diameter control from pulsed-laser evaporation systems with floating catalyst CVD¹¹. In this study, we present a new approach to achieving enhanced control over the diameter distribution combined with an improved yield in the investigated temperature region. The control over an advanced set of process parameters as well as the combination of two well-established synthesis routes, namely, CVD and LA, allows accurate and stable control of the mean diameter. Furthermore, a study of the mean diameter changes with temperature, pressure, and flow rate provides valuable experimental input for elaborating a consistent microscopic description of the formation mechanism of SWCNT.

Experimental Section

The targets for LA are mounted in the center of a 40 mm wide quartz reactor tube with an optional 16 mm wide constriction around the center. The reactor tube passes through the hot zone of a horizontal oven with possible temperatures up to 1100 °C. A Q-switched high-power Nd:YAG laser (2.5 GW per pulse, pulse width 8 ns)¹² is used at its fundamental wavelength of 1064 nm and focused to a spot of ca. 0.5 cm². The laser beam is in line with the applied gas stream in the reaction zone. Type, pressure, and flow rate of the hydrocarbons (serving as the carbon feedstock), the range of investigated synthesis temperatures, and the different targets are provided in Table 1. The targets were mounted to a molybdenum holder, which in turn is attached to a water-cooled copper coldfinger. The soot is collected on this coldfinger. A Bruker IFS100 Fourier transform−Raman spectrometer (excitation wavelength λ = 1064 nm) and a Renishaw In Via Raman microscope spectrometer (λ = 514 and 785 nm) were used for sample characterization. The samples were prepared by dispersing the products in acetone and then thoroughly sonicating them to ensure homogeneity. Subsequently, the solution was dropped onto heated plates of Al, resulting in a thin film of homogeneous material. For TEM imaging, the solution was dropped on a KBr single crystal and the resulting film was floated off in distilled water and collected on a standard Cu TEM grid. A FEI Tecnai F30 300 kV transmission electron microscope (TEM) was used to assess both the yield and quality of the samples on a local scale. The samples were also characterized by optical absorption spectroscopy (OAS) on a Bruker IFS113V/88 FT-spectrometer.

Results

Optical techniques such as OAS and resonant Raman spectroscopy are popular tools to characterize SWCNT because they allow a broad range of information on yield, diameter, diameter distribution and quality to be obtained. A typical SWCNT Raman spectrum contains the tangential modes viz. G-mode, which is reminiscent of in-plane modes in graphite, the disorder induced D-band, and the radial breathing modes (RBM). The diameters of the SWCNT are related to the RBM modes via ¹⁴, where ν is the Raman shift and d is the diameter of the SWCNT. To determine the yield of the SWCNT, in general, we can use the normalized intensity of the G-mode¹³. Additionally, the ratio of the G-mode intensity to the D-mode intensity (G:D) provides a means to assess the quality of the SWCNT. OAS also provides information on SWCNT yield and on their diameter distribution via analysis of the characteristic absorption peaks^{15, 16}. In these studies we use the first absorption peak, E₁₁^s, (which is corrected for its excitonic nature)¹⁶ to determine the SWCNT yield, the mean diameter and the diameter distribution¹⁶. The yield corresponds to the area of the background-strapped E₁₁^s peak, whereas information on the mean diameter and diameter distribution can be obtained from the fit parameters of a Gaussian distribution¹⁵. The bottom left panel in Figure 1 shows that yield determination from OAS and Raman spectroscopy agree well. Thus the use of the G-mode intensity to determine the yield is particularly useful when the SWCNT yield is insufficient to be accurately determined via OAS. However, because of the resonance enhancement of the G-line, caution should be exercised in using the G-mode on its own for yield determination. However, as we show, in our case, it is useful to show trends. These trends are supported by TEM and OAS data.

Figure 1. Top left: Diameter−temperature dependence obtained from OAS for SWCNT synthesized with NiCoMo (50:40:10 wt %) catalyst in 50 mbar ethanol. Right: Normalized E11s OAS peak of these samples after strapping the background. Inset: Raw spectrum prior to strapping. Bottom left: Yield determined by OAS E11s area (squares) and Raman G-mode intensity (circles). Lines on the left panels are to guide the eye. Click to Enlarge

We investigated a broad variety of different carbon feedstocks; methane, methanol, ethanol, propane, cyclohexane, toluene, and benzylamine with iron as a catalyst¹⁷. Propane, cyclohexane, and ethanol^{18, 19} were also tested with a nickel−cobalt−molybdenum (50:40:10 wt %) composite catalyst. Figure 2 shows the yield as a function of temperature for selected combinations of catalyst and carbon feedstock. For synthesis of these samples, the 16 mm wide reactor tube was used and the flow rate was kept constant at 1 lpm. The partial pressure of methane was set to 50 mbar for all samples shown on the left-hand side (A−C). In A, a composite target consisting of iron and 25 wt % carbon was used instead of pure iron. In C, pure methane was used instead of argon with 5% methane. The optimum temperature is always 900 °C. The extra carbon in A strongly increases the available carbon in the reaction process. The yield is highest in A relative to B and C. The upper two panels on the right-hand column (D, E) compare the yield of samples synthesized using a pure iron target with different carbon feedstocks. The samples in D and E were grown from cyclohexane and propane, respectively. The pressure of cyclohexane was chosen to be 40 mbar and the pressure of propane was set to 80 mbar to provide the same total amount of carbon per unit time. Although the maximum yield is quite comparable in these two cases, the optimum temperature is lower for both cyclohexane and propane as compared to methane. Finally, in part F of Figure 2, we show the yield vs temperature profile of an alloyed metal target (NiCoMo 50:40:10 wt %) under the same conditions as the pure iron target in E. The temperature profile is significantly smoother in the case of the compound target. The TEM images in Figure 2 belong to samples grown at the respective optimized temperatures (marked with an asterisk on panels D−F). The typical overall morphology of all high yield SWCNT material consists of bundles of nanotubes as well as a fair fraction of carbon coated catalyst material. However, for typical low-yield samples like in B and C (not shown), more individual nanotubes are observed than bundles. Figure 3 shows the yield and quality versus temperature profiles for propane and ethanol as carbon sources and NiCoMo (50:40:10 wt %) as the catalyst. Solid symbols are used for the wider 40 mm reaction tube, and open symbols denote the use of the 16 mm wide constriction. The optimum temperature for maximum yield is higher for ethanol than for propane. The wider reaction tube clearly shows improved characteristics regarding the overall yield as well as the quality of the nanotubes. Resonance Raman spectroscopy, as depicted for propane and selected temperatures in Figure 4, reveals a clear increase in the SWCNT mean diameter with increasing temperature (decrease in the RBM frequencies). The two columns on the left-hand side (A, 800 °C; B, 1000 °C) show the RBM at different laser excitation energies. The two columns on the right-hand side show the RBM and high-frequency region (G-line, D-line) for various temperatures at infrared excitation (1.16 eV). The shift in mean diameter can easily be tracked via the relative ratio of the RBM responses at 163 and 265 cm⁻¹ (I₁₆₃:I₂₆₅) in the spectra at 1.16 eV. This specific excitation energy is chosen to compare the SWCNT diameters because its resonances cover the widest spread of diameters. OAS studies as shown in Figure 1 for SWCNT synthesized from ethanol at different temperatures show the same increase in the mean diameter with increasing reaction temperature. In addition, OAS reveals a concomitant increase in the width of the diameter distribution (Figure 1, left side). The well resolved fine structures in the E₁₁^s peak (Figure 1 right side) may be a signature of the predominant diameters¹⁶ or stem from a chirality dependence of the matrix elements. Neither can be excluded, because a specific (n, m) assignment requires a detailed analysis via Raman spectroscopy using multiple-frequency excitation lasers or fluorescence spectroscopy. Figure 6 shows the mean diameter (left panels) and yield (right panels) as a function of either pressure (upper panels) or flow rate (lower panels) for SWCNT synthesized from propane at 800 °C. An increasing pressure decreases the mean diameter and diminishes the yield at the same time. An increasing flow rate increases the mean diameter up to a saturation value, and the yield runs through a clear maximum just before the mean diameter starts to saturate.

Figure 2. Yield−temperature dependence for different synthesis conditions: (A) 75% Fe, 25% C target in a 1 bar Ar:5% CH4 atmosphere, (B) 100% Fe in a 1 bar Ar:5% CH4 atmosphere, (C) 100% Fe in a 50 mbar CH4 atmosphere, (D) 100% Fe in 40 mbar cyclohexane atmosphere, (E) 100% Fe in an 80 mbar propane atmosphere, (F) NiCoMo (50:40:10 wt %) in an 80 mbar propane atmosphere. Stars denote samples of which TEM images were taken. Lines are to guide the eye. Click to Enlarge

Figure 3. Yield−temperature dependence for the NiCoMo target on carbon feedstock and reactor tube diameter. (A) In 40 mbar, 1 lpm propane in the 40 mm diameter tube; dashed curve and hollow symbols show comparison with the results from the 16 mm diameter tube (100 mbar/0.5 lpm), (B) In 50 mbar, 0.5 lpm ethanol in the 40 mm diameter tube. The shift in the optimum temperature is clearly visible. Lines are to guide the eye. Click to Enlarge

Figure 4. (A, B) Multifrequency Raman spectra for samples synthesized with a NiCoMo catalyst in 100 mbar/0.5 lpm propane at (a) 800, (b) 1000 °C. The shift to larger mean diameters can be tracked via the relative increase (or decrease) in the abundance of large (small) diameter SWCNT (arrows). (C) Raman spectra of the RBM and high-frequency region for different synthesis temperatures normalized to the G-mode. Click to Enlarge

Figure 5. Proposed nucleation mechanism after ref 21. Right: Overlap of the nucleation window with the precipitating cluster distribution determines the SWCNT diameter distribution. The change in the upper and lower limits arises from changes in the carbon availability and therefore different V:A constraints. Left: With a shift in the nucleation window with increasing carbon availability, the size of the nucleating catalyst clusters is reduced. Clusters with larger size (V:A ratio) are encapsulated by carbon, whereas small clusters cannot precipitate enough carbon to form a nucleating cap.21 Click to Enlarge

Figure 6. Diameter (A ,C) and yield (B, D) dependence on pressure and flow rate for SWCNT synthesized with NiCoMo (50:40:10 wt %) catalyst in propane at 800 °C. The flow rate and pressure were fixed to 1 lpm and 40 mbar, respectively; the reactor tube diameter was 40 mm. Lines are to guide the eye. Click to Enlarge

The various experimental observations on the yield and diameter distribution reported above are brought together in a detailed microscopic model as discussed in the following section.

Discussion

Going back to Figure 2, we want to explain the differences in yield using the various target-carbon feedstock combinations. These experiments were conducted with the 16 mm diameter tube. It is evident that Ar:5% CH₄ does not provide enough carbon to allow high-yield carbon nanotube nucleation and growth and that this can be compensated for by adding carbon to the target. The effect of using pure methane at 50 mbar versus diluted methane (partial pressure 50 mbar, total pressure 1 bar) is not so easy to understand, as the use of 50 mbar pure methane does not change the carbon to metal ratio in the reaction zone at first sight. However, it does change the dynamics of the evaporation plume because a much lower background gas pressure leads to a larger expansion of the plume²⁰, slower cooling and better intermixing with the hydrocarbon gas. This changes both the nucleation point in time and the respective numbers and sizes of catalyst particles, which nucleate SWCNT. It was recently suggested that not all catalyst particles can nucleate SWCNT²¹. In brief, particles that are too small cannot take up enough carbon to form a stable SWCNT cap because of the low volume to surface area ratio (V:A ratio). Particles that are too big take up too much carbon and become encapsulated by carbon before they can nucleate SWCNT. This imposes a V:A ratio constraint on the catalyst particles available for nucleation, thus defining a nucleation window. SWCNT nucleation can occur only when the nucleation window and the precipitating catalyst particle distribution overlap²¹. The compounds cyclohexane and propane have relatively higher molar carbon content. In addition, they also have C−C bonds, which are energetically easier to break as compared to the C−H bonds in methane. A radical induced decomposition is also likely, as the carbon to hydrogen ratio is much higher in these compounds compared to methane²². The lower optimum temperatures for these compounds can be explained as a reduction of available carbon at high temperatures because of self-pyrolysis and self-polymerization and condensation to light and heavy oils^{23, 24}. Propane proved to be a little bit more stable than cyclohexane. The increased yield−temperature curve width when using a NiCoMo catalyst (Figure 2F) can be explained by a wider spread of particle compositions²¹ and thus slightly different eutectic temperatures. For each of these temperatures, a part of the precipitating catalyst distribution lies in the nucleation window, and these catalyst particles are able to nucleate SWCNT²¹. Figure 3 shows the yield and quality results for two different carbon feedstocks, propane and ethanol. It is easy to see that the yield is much higher when using the 40 mm diameter reactor tube, in contrast to what one might expect²⁵. This can be explained by the larger volume available for plume expansion and, more importantly, by the slower gas velocity in the reaction zone. The carbon required for nanotube growth and nucleation is not directly available, as found in LA, but has to be obtained from the decomposition of hydrocarbons, which requires some time. The optimum temperature does not change with the width of the reactor, pointing to a strict dependence on thermal aspects of catalyst particle formation²¹. The optimum temperature using ethanol (with NiCoMo) as a carbon feedstock is restored to 900 °C, comparing well with the work of Nishide et al.¹⁰ for a pure Co target. This is because self-polymerization and self-pyrolysis do not play a significant role for ethanol below 1000 °C, an explanation being OH radical formation²⁶. OH radicals are also known to etch amorphous carbon species²⁷, resulting in a better G:D ratio for the ethanol-synthesized SWCNT (Figure 3). OAS and Raman studies show that the width of the diameter distribution and the mean diameter increase with temperature (Figure 1). Their evolution with temperature corresponds well with similar LA studies at these temperatures;^{12, 21, 28} however, the width of the diameter distribution is expected to saturate at roughly twice the width of a comparative LA sample. The shift in mean diameter can be explained by a shift in the nucleation window against the precipitation window and therefore a different overlap²¹. At higher temperatures, the cooling is slower because of a smaller temperature gradient and the resulting SWCNT diameters will increase because the catalysts particles have more time to coalesce/ripen. If the overlap is large, the width of the distribution is large²¹. This model is also applicable to the changes in the mean diameter upon a change of pressure and flow rate. These changes are not as pronounced as compared to the changes with temperature. Figure 6 shows these results, where a decrease of the mean diameter (small I₁₆₃:I₂₆₅) with increasing pressure and decreasing flow is evident. On the right side of Figure 6, the yield of the respective samples is shown. Increased pressure means a higher carbon:metal ratio because of a higher thermal conductivity and higher precursor density at the point of nucleation. The higher carbon availability shifts the nucleation window to smaller catalyst particles, as bigger ones are encapsulated because of an excess of carbon (Figure 5, right panel). The yield is found to decrease with increasing carbon content, which is contrary to what one might expect. This is because the overlap of the nucleation window and precipitation window is shifted toward the tail of the precipitating catalyst cluster distribution (Figure 5, left panel). Concomitant with these results, a decrease in SWCNT diameters with increasing pressure has also been reported for ferrocene-based floating catalyst CVD²⁹. The flow-rate dependence of diameter and yield follows a similar argumentation. Low gas velocities mean that the residence time of the catalyst particles in the active zone is prolonged, leading to a higher carbon:metal ratio because there is more time available for precursor decomposition, resulting in CNT with smaller diameters. In turn, the mean diameter increases with increasing gas flow. This mechanism is known from LA²¹. However, at very high gas velocities, it is expected that the gas is not fully heated (lower gas temperature). This effect counteracts the increase in diameter and leads to a stabilization of diameters in the high-flow region investigated. The decrease in yield at too high flow rates can again be explained by the fact that the overlap of the two respective windows diminishes.

Hence the pulsed-laser-assisted CVD system inherits the dynamics of catalyst particle formation of LA and extends them by an independently tunable carbon feedstock. This hybrid synthesis approach leads to SWCNT yields smaller than state-of-the-art LA but allows synthesis at lower temperatures. Further, for our system, no growth promoters like sulfur³¹ or water vapor³² were used. The relative purity of samples synthesized with our system as determined with the method developed by Itkis et al.³⁰ (NiCoMo 50:40:10 wt % in 50 mbar ethanol) is about 12%, which is approximately 25% of those obtained from LA with a total of 10 wt % catalyst. Further, this hybrid route allows lower synthesis temperatures than in LA and affords controlled mean diameters and narrow diameter distributions, which other CVD routes do not offer. A study of the effects of the laser power and power density as well as the use of growth promoters known from CVD like sulfur³¹ or water³² would provide further insight into the nucleation and growth processes and clarify the role of these remaining optimization parameters. If we assume this to be known, a scale-up is in principle possible in the same manner as LA may be scaled up. This route has a further advantage, in that the use of pure metal targets have a much longer lifetime as compared to mixed carbon:metal targets as used in conventional LA.

Conclusion

We have demonstrated that it is possible to synthesize SWCNT by laser evaporation in a hydrocarbon gas atmosphere. The hybrid route inherits the positive aspects of LA and CVD systems. SWCNT with narrow diameter distributions, as found in LA, are obtained. The use of a hydrocarbon feedstock allows in principle for continuous synthesis through a constant supply of fresh catalyst particles. The obtained data show a correlation between the diameter of the nanotubes and the carbon availability, enabling the mean diameter of the SWCNT to be fine-tuned via the background pressure and flow rate of the precursor. The results are fully explained by our previously suggested volume to surface area constraint growth model²¹, which we now extend to incorporate the effects of carbon availability on the SWCNT diameter. Our findings provide new insight into understanding the nucleation processes governing the synthesis of carbon nanotubes. Furthermore, the use of this hybrid synthesis route provides a means to bridge the gap between CVD and laser ablation, enabling improved yield and diameter control as compared to conventional supported catalyst CVD in this temperature region.

* Corresponding author. E-mail: markus.loeffler@ifw-dresden.de.