In our electric arc-discharge apparatus5, the arc is generated between two electrodes in a reactor under a helium atmosphere (660 mbar). The cathode was a graphite rod (16 mm diameter, 40 mm long) and the anode was also a graphite rod (6 mm diameter, 100 mm long) in which a hole (3.5 mm diameter, 40 mm deep) had been drilled and filled with a mixture of a metallic catalyst and graphite powder. The arc discharge was created by a current of 100 A; a voltage drop of 30 V between the electrodes was maintained by continuously translating the anode to keep a constant distance (3 mm) between it and the cathode. Typical synthesis times were 2 min. As the catalyst we used mixtures such as Ni–Co, Co–Y or Ni–Y in various atomic percentages; these are known to yield a series of interesting carbon nanostructures6. The mixture used by Guo et al.7 during their laser ablation process (Co and Ni, both at 0.6 at.%) did not produce a good yield of nanotubes in our case. However, we found that a mixture of 1 at.% Y and 4.2 at.% Ni gave the best results. In this case we observed (in a total carbon mass of 2 g): (1) large quantities of rubbery soot condensed on the chamber walls; (2) web-like structures between the cathode and the reactor walls (no webs when either Y or Ni were absent); (3) a cylindrical deposit at the cathode's end; and (4) a small ‘collar’ (20% of the total mass) around the cathode deposit, as a black, very light and porous but free-standing material.

Within all these products it was possible to observe by scanning electron microscopy (SEM; using a JEOL JSM 6300F instrument) filament-like structures that are more or less dense, depending on where in the reactor they were deposited. The ‘collar’ deposit was densest; the soot was the least dense. A characteristic SEM image of the collar deposit (Fig. 1) shows large amounts of entangled carbon filaments, homogeneously distributed over large areas (here at least a few square millimetres) and with diameters ranging from 10 to 20 nm. The average length between two entanglement points is several micrometres; we could not identify any filament ends. From several SEM images we estimate the yield of these filaments (with respect to the total volume of the solid material) to be of the order of 80%. Higher magnification confirms the existence of entangled filaments although it is not possible at this stage to identify the nature of the crosslinks.

Figure 1: Scanning electron microscopy image of the light and porous material that formed as a collar around the cathode deposit in our apparatus, s.
figure 1

howing a high density of entangled carbon filaments. Scale bar, 1 µm.

High-resolution transmission electron microscopy (HRTEM; using a JEOL 4000FX instrument) images of the collar deposit (Fig. 2) show that each filament consists of smaller aligned SWNTs self-organized into bundle-like crystallites with diameters ranging from 5 to 20 nm. As these bundles are often bent, some portions are orientated parallel to the electron beam and are therefore seen edge-on (Fig. 2). The tube diameters are around 1.4 nm, and the average distance between them is 1.7 nm. Evidence for periodic packing is also given by electron diffraction patterns obtained from an assembly of bundles (not shown).

Figure 2: High resolution transmission electron microscopy view of one bundle from the collarette bent in such a way that it is seen edge-on.
figure 2

The bundle has a uniform width and consists of 20 SWNTs of almost uniform diameter, packed according to a triangular lattice. The focusing conditions are such that the lattice is imaged as a triangular arrangement of strong white dots (which are located at the centres of the SWNTs) on a dark background. Scale bar, 4 nm.

The number of tubes within a bundle is typically of the order of 20, and their section can be either roughly circular (Fig. 2) or elongated along a dense direction of the lattice. Larger-diameter bundles consist mostly of an assembly of smaller bundles separated by twin-like boundaries. The intra-bundle organization is also evidenced by the existence of diffraction fringes whose spacing is directly related to the lattice parameter and depends on the orientation of the crystalline bundle with respect to the electron beam.

Besides the bundles, we see randomly distributed spherical nanoparticles ranging from 3 to 20 nm in diameter, usually embedded in amorphous carbon. These nanoparticles probably consist of the remaining metallic catalyst. The SWNTs pass through them with no obvious evidence of a direct link. We have seen no other metallic or carbonaceous structures in the deposits.

It is difficult to determine the SWNT diameters accurately from electron microscopy, but the tube configuration can be studied in detail using Raman spectroscopy (Fig. 3). The spectrum exhibits three main zones at low (140–200 cm−1), intermediate (300–1,300 cm−1) and high (1,500–1,600 cm−1) frequencies, and is characteristic of SWNTs (other carbon materials produce other types of spectra8,9). The high-frequency bands can be decomposed into two main peaks around 1,589 and 1,562 cm−1 with shoulders at 1,550 and 1,530 cm−1. These features have previously been assigned to a splitting of the E2g mode of graphite10. The very-high-resolution spectrum obtained in the low-frequency domain (Fig. 3 inset) shows at least five components at 149, 164, 172, 178 and 183 cm−1. These two spectra were recorded for exactly the same experimental conditions and sample, except that the laser beam explored two different areas. Thus the spectrum in this frequency domain is very sensitive to the sample area investigated, in contrast with that in the range 1,500–1,650 cm−1. In addition, different excitation wavelengths produce different relative intensities of these bands, indicating strong resonance behaviour. According to earlier calculations11, these modes are expected to be of Ag symmetry, and their frequency increases with decreasing tube diameter. Therefore our results clearly reflect a distribution in diameters (or of helical pitch for a given diameter). It can be shown theoretically10 that only ‘armchair’ tubes (with diameter and pitch parameters n = m ranging from 6 to 12) generate Raman modes between 750 and 800 cm−1. In Fig. 3 the peaks in the frequency range 300–1,200 cm−1 can be mostly identified as overtones and combinations of lower-frequency modes; but the modes of 750 and 782 cm−1 are characteristic of such armchair nanotubes.

Figure 3: Raman spectrum of SWNTs recorded at room temperature, using an excitation wavelength of 514.
figure 3

5 nm. The intensity in the frequency range between 250 and 1,200 cm−1 has been magnified 70 times. Inset, the low-frequency range (120–200 cm−1) recorded with high resolution at two different spots on the sample.

X-ray diffraction was performed using an INEL CPS 120 detector in Debye–Scherrer geometry. A small flake from ground collar material was attached to an etched Si(111) wafer moistened with ethyl alcohol. After background subtraction, the data show an intense peak near diffusion vector Q = 0.43 å−1 and four weaker peaks up to 1.8 å−1 (Fig. 4 trace a). These low-Q scattering peaks indicate the existence of a two-dimensional lattice of SWNTs organized in bundles2. The width of the peak at Q = 0.43 å−1 yields a coherence length of the order of 100 å, in very good agreement with the average transverse size of the bundles. The region above Q = 3 å−1 (Fig. 4 inset) is dominated by two peaks at 3.1 and 3.6 å−1 which correspond to the (111) and (200) reflections of pure face-centred cubic nickel. The analysis of their shapes (using the Warren–Averbach12 method) results in a particle size distribution ranging from 2 to 21 nm, with an average size of 11.2 nm. Our X-ray diffraction data are entirely consistent with the observations made by HRTEM.

Figure 4: X-ray diffraction patterns at low angle of our collar sample (trace a) and of the sample obtained by the laser ablation technique b.
figure 4

y Thess et al.2 (trace b). The graphite peak, due to remaining graphitic particles, has been removed for clarity; its position is shown by an asterisk. Inset, diffraction pattern associated with the Ni nanoparticles.

A direct comparison of the X-ray diffraction data obtained by Thess et al.2 (Fig. 4 trace b) and by us reinforces the great similarity between both sample characteristics: yields in the range 70–90%, diameters around 1.4 nm, crystalline bundles of a few tens of tubes and only a few isolated SWNTs. Our results therefore imply that there is an unique growth mechanism for these nanotubes, which does not strongly depend on the details of the experimental conditions, but which depends much more on the kinetics of carbon condensation in a non-equilibrium situation. Temperature, and temperature gradients in space and time, must play an important role, as can be seen by the fact that most of the SWNTs were found in a very specific zone of the reactor (a few centimetres around the cathode). Additionally, the use of a second element beside the catalyst Ni or Co during the evaporation process is also critical. In our experiments it is yttrium that strongly favours the growth of SWNTs.

Our electric-arc technique is able to produce gram quantities of well-defined single-walled nanotubes in the form of highly crystalline bundles. It is a cheap method which could be scaled up easily, and so seems to be a very promising alternative to the double-laser-ablation technique2.