Separation of Metallic from Semiconducting Single-Walled Carbon Nanotubes

See allHide authors and affiliations

Science  18 Jul 2003:
Vol. 301, Issue 5631, pp. 344-347
DOI: 10.1126/science.1086534


We have developed a method to separate metallic from semiconducting single-walled carbon nanotubes from suspension using alternating current dielectrophoresis. Our method takes advantage of the difference of the relative dielectric constants of the two species with respect to the solvent, resulting in an opposite movement of metallic and semiconducting tubes along the electric field gradient. Metallic tubes are attracted toward a microelectrode array, leaving semiconducting tubes in the solvent. Proof of the effectiveness of separation is given by a comparative Raman spectroscopy study on the dielectrophoretically deposited tubes and on a reference sample.

After more than 10 years of extensive research on carbon nanotubes, it is widely recognized that single-walled carbon nanotubes (SWNTs) have considerable potential as building blocks in future nanoscale electronics (1). Metallic SWNTs could function as leads in a nanoscale circuit (2), whereas semiconducting SWNTs could perform as nanoscale Schottky-type field-effect transistors (35). For the realization of nanotube-based electronics, it is necessary to manipulate metallic and semiconducting SWNTs separately. Unfortunately, they are typically grown together as bundles. We have developed a method to extract the metallic tubes based on alternating current (ac) dielectrophoresis. SWNTs develop an induced dipole moment when subjected to an electric field (6). The induced dipole moment allows one to align and to move suspended tubes in a specific manner by designing an appropriate inhomogeneity of the electric field. Site-selective deposition of aligned carbon nanotubes, and in particular of bundles of SWNTs, has been demonstrated using applied electric fields (711). In these studies, bundles of SWNTs were either already present or formed in suspension because of van der Waals interactions. Recently, a method to prepare stable suspensions with high yields of individual SWNTs has become available (12). By using a similar suspension, we report here the separation of metallic from semiconducting SWNTs via ac dielectrophoresis.

For the preparation of an individual suspension of SWNTs, 50 mg of raw SWNT soot, prepared by high-pressure decomposition of CO (HiPco) (13), was suspended in 100 ml of D2O containing 1 weight % of the surfactant sodium dodecyl sulfate (SDS) under sonication, a procedure similar to that described in (12). After sonication, the suspension was centrifuged at 170,000g for 4 hours, and the upper 90% of the supernatant was than carefully decanted. The resulting nanotube suspension has a typical mass concentration of roughly 10 mg/liter. The suspension is characterized by sharp features in its electronic absorption spectrum associated with pairwise transitions between van Hove singularities (vHs) in the electronic density of states. The suspension also shows strong fluorescence. Both observations indicate that the suspension contains mainly micelle-coated individual nanotubes rather than bundles (12).

Microelectrodes were prepared with standard electron-beam lithography and wired to a function generator as schematically depicted in Fig. 1. For the ac dielectrophoresis, the generator was operated at a frequency f of 10 MHz and a peak-to-peak voltage (Vp-p) of 10 V. After the generator was switched on, a drop of suspension of individual SWNTs (≈10 μl) was applied to the chip. After a delay of 10 min, the drop was gently blown off the surface by a stream of nitrogen gas, and the generator was switched off. The resulting sample was characterized with a confocal Raman microscope (model CRM-200; Witec) excited with an Ar+ ion laser (at 514.5 nm excitation; Spectra Physics). The laser spot had a diameter of ∼1.25 μm at the sample, with a power density of ∼105 W/cm2. When Raman spectra of the dielectrophoretically deposited aligned SWNTs were taken, the polarization of incident light was generally chosen to be in parallel to the aligned tubes. All Raman spectra were recorded using a grating of 1800 grooves/mm and a corresponding spectral resolution of 1 cm1, except for the polarization dependence, measured with a grating of 600 grooves/mm and 3.75 cm1 resolution. We also studied the sample with an incident-light dark-field microscope (Leica) using an unpolarized halogen lamp. A reference sample was prepared by depositing SWNTs onto a Si substrate by simply applying a drop of suspension of individual SWNTs (∼10 μl) and letting it dry.

Fig. 1.

Illustration of the experimental setup, showing microelectrodes wired to a chip carrier. The metallic nanotubes (black) are deposited from a drop of nanotube suspension onto the electrodes by ac dielectrophoresis, leaving the semiconducting tubes (white) in suspension. Gold electrodes are 30 nm thick and 50 μm wide, with a 50-μm pitch on a p-type silicon substrate with 600 nm of thermally oxidized SiO2. A thin Ti adhesion layer was used.

Figure 2 shows a dark-field micrograph of the sample after deposition of SWNTs via ac dielectrophoresis. The accumulated SWNTs appear green to the eye because of strong Rayleigh scattering in the green wavelength range. This is directly related to the optically allowed excitations, which are explained later on. The high degree of alignment of the tubes along the electric field lines, present during deposition, is clearly visible. Rayleigh scattering is strongly suppressed for light polarized perpendicular to the SWNT axis. Hence, by inserting an analyzer (polarization filter), we can visualize the homogeneity of the nanotube alignment. The scattered light is not visible over large areas if the analyzer is perpendicular to the deposited SWNTs.

Fig. 2.

Rayleigh scattered light from the dielectrophoretically deposited SWNTs and the electrodes, recorded with an incident-light dark-field microscope. The scattered light from the aligned SWNTs appears green to the eye (A) and is polarized perpendicular to the electrodes (B).

Figure 3A shows the radial breathing mode (RBM) regions obtained by averaging 10 single Raman spectra measured on different spots of the aligned SWNTs and on the reference sample. The variations between the individual spectra are within a few percent owing to the large number of tubes measured within the laser spot. The reference spectrum shows two bands, with one band (red) dominated by a feature at 187 cm1 and the other band (blue) consisting of three peaks at 247, 263, and 271 cm1. The frequency of the RBM ωRBM is proportional to the inverse diameter: ωRBM = c1/d + c2, where c1 = 223.5 nm cm1 and c2 = 12.5 cm1 are empirically derived parameters (14, 15), with d = a(n2 + m2 + nm)1/2/π and a = 0.249 nm. The integers (n,m) define the structure of a SWNT in terms of its diameter d and chiral angle (16). The lower panel of Fig. 4 shows the RBM frequencies given by the above formula for the experimentally relevant diameter range for HiPco tubes (14).

Fig. 3.

Raman spectra of SWNTs deposited via ac dielectrophoresis compared to a reference sample deposited on Si without the application of an electric field. (A) RBMs associated with metallic (blue) and semiconducting (red) SWNTs. (B) G-mode region.

Fig. 4.

Correlation between energy E0 (M1, S3), RBM frequency ωRBM, and SWNTdiameter. Metallic SWNTs [open circles, (nm) mod 3 = 0; solid circle, n = m] and semiconducting SWNTs (x's) are easily distinguishable by their RBMs at the excitation wavelength λ = 514.5 nm. Horizontal solid line splits the figure into top and bottom panels. Squares indicate calculated RBM frequencies from (14). Blue and red lines indicate observed RBMs of metallic and semiconducting SWNTs, respectively.

Note that M1 is split from quasi-metallic SWNTs.

For a given excitation wavelength, only those SWNTs contribute to a Raman signal that can be resonantly excited between optically allowed vHs in the electronic density of states (16), which are separated by the energy E. For a certain (n,m) nanotube, E can be determined rather reliably from a tight-binding calculation using curvature-modified hopping parameters (17). Furthermore, the uncertainty in the tight-binding parameter γ0 ranging between 2.5 and 3 eV can be taken into account by considering E0. The upper panel of Fig. 4 shows a region of E0 versus diameter d of those SWNTs that can possibly be excited with the laser energy Eexc = 2.41 eV used in our experiment. Owing to the above selection rules, three types of SWNTs can possibly be resonantly excited under the experimental conditions: (i) semiconducting SWNTs, excited over the third pairwise vHs (S3); (ii) quasi-metallic SWNTs with nm divided by 3, leaving a remainder (mod) of = 0; and (iii) metallic SWNTs with n = m, both excited over the first pairwise vHs (M1). Within the experimentally accessible region of excitation, all metallic SWNTs have diameters d < 1.1 nm and all semiconducting SWNTs have diameters d > 1.1 nm. Hence we conclude that all metallic SWNTs have RBM frequencies in the range between 218 and 280 cm1, well separated from the RBM frequencies of the semiconducting SWNTs, ranging between 175 and 213 cm1.

It is thus straightforward to conclude from our measurements on the reference sample that the lower RBM band originates from at least one type of semiconducting SWNT and that the higher RBM band originates from at least three types of metallic SWNTs. A possible assignment of the observed RBM frequencies to specific SWNTs is given in Table 1. However, the above general distinction between metallic and semiconducting SWNTs is independent of this assignment.

Table 1.

Possible assignment to observed RBM frequencies. Error in RBMs is ±1 cm-1.

Observed RBM (cm-1) Predicted RBM (cm-1) Assignment (n,m)
272View inline 273.2 9,3
262View inline 260.8 8,5
247View inline 247.5 12,0
245.1 7,7
245.1 11,2
187View inline 188.7 16,0View inline
187.7 15,2View inline
187.7 13,5
184.8 14,4View inline
  • View inline* Full width at half maximum (FWHM) ≈ 10 cm-1.

  • View inline FWHM ≈ 13 cm-1. Predicted RBMs were calculated from (14).

  • View inline Not observed in fluorescence measurements (14).

  • We now compare the RBM regions of the SWNTs deposited via ac dielectrophoresis with the reference spectra. Again we observe the two bands associated with metallic and semiconducting SWNTs. However, the intensity ratio is very different, which we quantify by the ratio η of the integrated RBM bands of metallic and semiconducting SWNTs. We obtain on the reference sample ηref ≈ 5, compared to ηsample ≈ 40 on the aligned SWNTs. The ratio ηsample /ηref ≈ 8 reflects the relative abundance of metallic SWNTs. With the general assumption that ≈33% of the suspended SWNTs are metallic, we estimate that 80 ± 5% of the dielectrophoretically deposited SWNTs are metallic (18). The 5% error is derived from the uncertainty in the subtraction of the baseline. One could argue that ηsample is large because semiconducting SWNTs may be present in our sample but are not aligned. This possibility is dismissed by the polarization dependence of the sample measurement, which shows a complete suppression of both RBM bands if excited with light polarized perpendicular to the alignment (Fig. 5). A second reference sample exposed to a metallic electrode structure without an applied electric field showed similar (to within 10%) Raman spectra compared to the electrode-free reference. We can therefore exclude an influence of the metal electrodes on the spectra. Furthermore, we do not see a dependence of η on the position of the laser spot relative to the electrodes.

    Fig. 5.

    Raman spectra of SWNTs deposited via ac dielectrophoresis with the incident light polarization parallel to (0°) and perpendicular to (90°) the deposited SWNTs. (Left) RBM region; (right) G-mode region.

    Comparison of the G band regions derived from graphite-like in-plane mode of both spectra in Fig. 3B provides further support for the enrichment of metallic SWNTs in the dielectrophoretically deposited sample. It has been shown that there are characteristic differences between the G bands for metallic and semiconducting nanotubes (16). Both types of tubes show two dominant features between 1500 and 1600 cm1: a low-frequency component ωG–, which is associated with vibrations along the circumferential direction, and a high-frequency component ωG+, which is attributed to vibrations along the direction of the nanotube axis. In semiconducting nanotubes, both ωG– and ωG+ show a Lorenzian lineshape, with ωG+ being stronger in intensity than ωG–. In metallic nanotubes, both components are of equal intensity and ωG– is much broader, exhibiting an asymmetric Breit-Wigner-Fano lineshape (16). In this context, the data in Fig. 3B are in agreement with the enrichment of metallic tubes in our sample prepared by ac dielectrophoresis.

    We now discuss how ac dielectrophoresis is operative in separating metallic from semiconducting SWNTs in aqueous suspension with the help of a simple electromechanical model. In our experiment, using the frequency f = 10 MHz, electrophoretic forces [due to charging (19)] vanish because of time averaging, and only the induced dipole moment interacting with the inhomogeneous electric field gives rise to a translational motion along the electric field gradient. This is, in fact, ac dielectrophoresis (20, 21). To consider only the simplest case, the time-averaged dielectrophoretic force for a dielectric sphere is expressed as Embedded Image(1) where ϵp and ϵm are the dielectric constants of the particle and the solvent medium, respectively, and Erms is the average field strength (21). Equation 1 is a valid approximation for our experiment, because metallic and semiconducting SWNTs are ballistic conductors and insulators, respectively (22). Positive dielectrophoresis, the attraction toward the higher field strength, occurs for ϵp > ϵm. The static dielectric constant for semiconducting SWNTs, ϵs, has been calculated in (6) and was found to be inversely proportional to the square of the band gap EG, the energetic distance between the first pairwise vHs (S1) Embedded Image(2) where ℏωp ≈ 5 eV is the energy of the plasma oscillation along the nanotube axis (23). From the tight-binding calculation mentioned above (17), we find for our tubes with d = 0.8 to 1.4 nm the smallest possible band gap values EG between 0.19 γ0 and 0.32 γ0. With γ0 = 2.5 to 3.0 eV, we obtain a minimum band gap EG ≈ 0.5 eV for the semiconducting tubes in our samples. Translating band gaps into dielectric constants according to Eq. 2, we obtain finite dielectric constants for semiconducting SWNTs with ϵs < 5. For metallic SWNTs, owing to the mobile carriers, we expect a very large absolute value of the dielectric constant. In fact, it has been suggested that the polarizability of metallic SWNTs is effectively infinite (6). This statement is also true for quasi-metallic SWNTs [(nm) mod 3 = 0] with energy gaps smaller than the thermal energy. Returning to Eq. 1, we calculate the sign of the ac dielectrophoretic force for individual SWNTs in D2O with ϵD2O = 80 and assuming that ϵ (10 MHz) ≈ ϵ (0). For semiconducting SWNTs, we derive a negative dielectrophoresis, whereas for the metallic and quasi-metallic SWNTs, the dielectrophoresis is positive, explaining why metallic SWNTs are deposited on our electrodes whereas the semiconducting SWNTs remain in suspension.

    In principle, a complete separation between metallic and semiconducting SWNTs should be possible if all tubes in the suspension are present as individual tubes (no bundles). Otherwise, bundles, which contain semiconducting tubes and at least one metallic tube dominating the dielectric constant of the bundle, will be attracted to the electrodes as well. In our experiment, we achieved an enrichment of metallic tubes up to 80%, indicating that only a small fraction of tubes is present as bundles.

    We estimated that about 100 pg of metallic tubes were deposited, out of roughly 100 ng of tubes present in the drop. Certainly this is a small quantity (too small to see a corresponding effect in the leftover solution). However, we are confident that the separation method can be substantially scaled up by the use of microfluidic dielectrophoretic separation cells, as commonly used in biology (24).

    References and Notes

    View Abstract

    Navigate This Article