The heterogeneity of as-synthesized single-walled carbon nanotubes (SWNTs) precludes their widespread application in electronics, optics and sensing. We report on the sorting of carbon nanotubes by diameter, bandgap and electronic type using structure-discriminating surfactants to engineer subtle differences in their buoyant densities. Using the scalable technique of density-gradient ultracentrifugation, we have isolated narrow distributions of SWNTs in which >97% are within a 0.02-nm-diameter range. Furthermore, using competing mixtures of surfactants, we have produced bulk quantities of SWNTs of predominantly a single electronic type. These materials were used to fabricate thin-film electrical devices of networked SWNTs characterized by either metallic or semiconducting behaviour.
Carbon nanotubes have recently received extensive attention1 due to their nanoscale dimensions and outstanding materials properties such as ballistic electronic conduction2, immunity from electromigration effects at high current densities1, and transparent conduction3. However, as-synthesized carbon nanotubes vary in their diameter and chiral angle, and these physical variations result in striking changes in their electronic and optical behaviours1,4. For example, about one-third of all possible SWNTs exhibit metallic properties and the remaining two-thirds act as semiconductors. Moreover, the bandgap of semiconducting SWNTs scales inversely with tube diameter. For instance, semiconducting SWNTs produced by the laser-ablation method range from 11 to 16 Å in diameter and have optical bandgaps that vary from 0.65 to 0.95 eV (ref. 5). The currently unavoidable structural heterogeneity of as-synthesized SWNTs prevents their widespread application as high-performance field-effect transistors, optoelectronic near-infrared emitters/detectors, chemical sensors, materials for interconnects in integrated circuits and conductive additives in composites1. Accordingly, their use will be limited until large quantities of these nanomaterials can be produced that are monodisperse in their structure and properties.
To address the problem of heterogeneity and enable future SWNT-based technologies, we have developed a general approach for sorting carbon nanotubes by diameter, bandgap and electronic type (metallic versus semiconducting), using the technique of density-gradient ultracentrifugation. This scalable approach exploits differences in the buoyant densities (mass per volume) among SWNTs of different structures and has been adapted from similar techniques used in biochemistry6,7 for the purification of biological macromolecules, such as nucleic acids and proteins. In this technique, purification is induced by ultracentrifugation in a density gradient. In response to the resulting centripetal force, particles sediment toward their respective buoyant densities and spatially separate in the gradient. This approach differs from previously reported work on the ultracentrifugation of SWNTs9,10, in which density gradients were not used.
Other methods8,10,11,12,13,14,15,16,17,18 for sorting SWNTs have been reported recently; however, none of these techniques has demonstrated the simultaneous sensitivity, scalability and effectiveness that we report here. In addition, we have previously exploited density-gradient ultracentrifugation to enrich DNA-wrapped SWNTs by diameter and bandgap8. However, there are critical drawbacks to using DNA for carbon nanotube functionalization. First, DNA-wrapped SWNTs have limited stability in aqueous density gradients and thus are not amenable to the refinements in enrichment gained from repeated centrifugation. Furthermore, complete removal of the DNA wrapping after enrichment has not been demonstrated, and sensitivity to electronic type has not been observed. Finally, the availability and cost of specific, custom oligomers of single-stranded DNA are prohibitive.
To overcome these obstacles, we have recently explored surfactant encapsulating agents in place of DNA and have discovered that bile salts and their mixtures with other surfactants enable the separation of SWNTs by diameter, bandgap and/or electronic type. Using surfactants, the isolation of specific chiralities of SWNTs can be significantly refined by separating in multiple successive density gradients. Furthermore, the adsorption of surfactants to SWNTs is reversible and compatible with a wide range of tube diameters. For example, we demonstrate here the sorting of SWNTs over the diameter range 7–16 Å. Most importantly, the structure–density relationship for SWNTs can be easily controlled by varying the surfactant itself. For instance, by using mixtures of two surfactants that competitively adsorb to the SWNT surface, we have achieved optimal metal–semiconductor separation.
Comparison of encapsulating agents
The buoyant density of SWNTs in aqueous solution will subtly depend on multiple factors, including the mass and volume of the carbon nanotube itself, its surface functionalization and electrostatically bound hydration layers. To gain insight into the structure–density relationship for SWNTs and the role of encapsulating agents, we first compared two different families of surfactants—anionic-alkyl amphiphiles19 and bile salts20. Specifically, we used two amphiphiles with anionic head groups and flexible alkyl tails: sodium dodecyl sulphate (SDS) and sodium dodecylbenzene sulphonate (SDBS). The three bile salts used were sodium cholate (SC), sodium deoxycholate and sodium taurodeoxycholate. The bile salts are more molecularly rigid and planar amphiphiles with a charged face opposing a hydrophobic one21, which is expected to interact with the SWNT surface (Fig. 1a).
Initially, we explored the sorting of SWNTs in the 7–11 Å diameter range synthesized by the CoMoCAT method, using SC and SDBS encapsulations, as depicted in Fig. 1. For the case of SC encapsulation, multiple regions of separated SWNTs are visible throughout the density gradient (Fig. 1b). The most buoyant region is characterized by SWNTs that have been sorted into bands of various colours, corresponding to the different bandgaps of the semiconducting tubes. In contrast, for the case of SDBS-encapsulated SWNTs, all of the SWNTs are compressed into a narrow black band (Fig. 1d).
After centrifugation, the separated SWNTs can be removed from the centrifuge tubes, layer by layer, using established techniques for fractionation, and each layer can be optically characterized to determine quantitatively the mode and quality of separation (see Supplementary Information, Figs S1 and S2, and Methods). For the case of SC-encapsulated SWNTs, the amplitudes of optical absorbance for different transitions in the 900–1,340 nm range (first-order semiconducting transitions) indicate separation by diameter and bandgap. More specifically, the spectra illustrate that SWNTs of increasingly larger diameters are enhanced at increasingly larger densities. A similar correlation between diameter and density was also observed for the cases of sodium deoxycholate and sodium taurodeoxycholate. However, for the case of SDBS (Fig. 1e) and SDS (see Supplementary Information, Fig. S3) encapsulations, separation as a function of diameter was absent.
This trend of increasing density with increasing diameter also extends to SC-encapsulated SWNTs in the 11–16 Å diameter range that were synthesized by laser ablation (Fig. 1f). In the optical spectra, separation by diameter is observed as a red shift in the second-order optical absorbance transitions for semiconducting SWNTs, 800–1,075 nm (ref. 5), with increasing density (Fig. 1g). Moreover, an enrichment of these SWNTs by electronic type is also detected. In the most buoyant fractions, we observe an enhancement in concentration of semiconducting SWNTs with respect to metallic SWNTs, which have first-order optical transitions ranging from 525 to 750 nm (ref. 22).
Repeated, refined sorting
Using density-gradient centrifugation with bile salts such as SC, it is clear that we can enrich SWNTs by both their structure and electronic properties. However, the degree of isolation achieved after a single step of the technique is limited by the diffusion of SWNTs during ultracentrifugation, mixing during fractionation, and statistical fluctuations in surfactant encapsulation. To overcome these limitations and improve the sorting process, the centrifugation process can be repeated for multiple cycles. For example, an enriched fraction of SWNTs sorted in a density gradient can be further enriched in a second density gradient. This enables the optimal isolation of a targeted electronic type or a specific chirality of SWNT. To demonstrate the approach, we targeted the enrichment of the (6,5) and (7,5) chiralities of semiconducting SWNTs (7.6 and 8.3 Å in diameter, respectively). In Fig. 2, photoluminescence emission–excitation matrices depict the photoluminescence intensity of semiconducting SWNTs as a function of excitation and emission wavelengths for SWNTs before and after each of three iterations of density-gradient centrifugation (see Supplementary Information, Fig. S4, for plots of the corresponding absorbance spectra). After each iteration, the relative concentrations of the (6,5) and (7,5) chiralities of semiconducting SWNTs increase (Fig. 2). After enriching the (6,5) chirality (7.6 Å) three times, we achieved bulk solutions of the SWNTs in which >97% are within 0.2 Å of the mean diameter (see Supplementary Information, Fig. S5, and Methods). Further improvements in the isolation of individual chiralities of SWNTs may be possible with additional cycles.
Tuning of the structure–density relationship
Although the separation of SWNTs can be significantly enhanced via multiple cycles of ultracentrifugation, further improvements can be realized by optimizing the effectiveness of a single cycle through tuning of the structure–density relationship for SWNTs. For example, by adjusting pH or by adding competing co-surfactants to a gradient, the isolation of a specific diameter range or electronic type can be targeted.
In Fig. 3a–c we demonstrate diameter tunability. The relative concentration of several different diameters (7.6, 8.3 and 9.8/10.3 Å) of SWNTs is plotted against density for the cases of SC-encapsulated SWNTs at pH 7.4, SC-encapsulated SWNTs at pH 8.5, and for a mixture of 1:4 SDS/SC (by weight) at pH 7.4. By tuning the structure–density relationship, the differences in density among SWNTs of these diameters can be modified. For example, by increasing the pH to 8.5, the SWNTs near 8.3 Å in diameter shift to more buoyant densities, enabling optimal separation of SWNTs in the 9.8/10.3 Å range (Fig. 3b). Alternatively, by adding SDS to compete with the SC for non-covalent binding to the nanotube surface, the SWNTs in the 8.3 and 9.8/10.3 Å diameter regime shift to significantly larger buoyant densities, enabling optimal separation of SWNTs near 7.6 Å in diameter (Fig. 3c). At the highest densities, the relative concentration of SWNTs can appear anomalously high in some cases (for example, the blue curve of Fig. 3a) due to contributions to the absorbance spectrum from high-density SWNT aggregates.
Co-surfactant populations have an even greater effect on the optimization of metal–semiconductor separation for SWNTs in the 11–16 Å diameter regime. By adding one part SDS for every four parts SC (by weight, 2% by weight overall) to a gradient, much more distinct metal–semiconductor separation is made evident. For SWNTs separated in 1:4 SDS/SC mixtures, only three bands are observed (Fig. 3d). We can deduce from measured optical absorbance spectra that the top band (orange hue) consists of predominantly semiconducting SWNTs (Fig. 3d) and that the band just below the top band (green hue) is highly enriched in metallic SWNTs (see Supplementary Information, Fig. S6, for a plot of the absorbance spectrum). Further tuning of the co-surfactant mixture to a 3:2 SDS/SC ratio permits significantly improved isolation of metallic SWNTs. In Fig. 3e, spectra corresponding to primarily metallic (3:2 SDS/SC) and primarily semiconducting (1:4 SDS/SC) SWNTs are shown.
To demonstrate the applicability of SWNTs sorted in density gradients and to confirm their separation by electronic type, field-effect transistors were fabricated (see Supplementary Information, Methods) consisting of percolating networks of thousands of metallic or semiconducting SWNTs (Fig. 4). At negative gate biases, both networks exhibited similar sheet resistances of about 500 kΩ per square. However, by varying the voltage applied across the gate dielectric capacitor (100 nm SiO2), the resistivity of the semiconducting network was increased by over four orders of magnitude (on/off ratio >20,000). In contrast, the metallic networks were significantly less sensitive to the applied gate bias characterized by on/off ratios of less than two. The two distinct behaviours of the semiconducting and metallic films independently confirm the separation by electronic type initially observed by optical absorption spectroscopy of the sorted materials presented in Fig. 3e. Additionally, the two films establish the applicability of this method in producing usable quantities of sorted, functional material. For example, a single fraction of semiconducting SWNTs (150 µl) contains enough SWNTs for 20 cm2 of a thin-film network similar to that demonstrated in Fig. 4, corresponding to >1011 SWNTs (see Supplementary Information, Fig. S7). Such thin-film networks have possible applications as flexible and transparent semiconductors and conductors.
We believe that surfactant-based separation using density-gradient ultracentrifugation is largely driven by how surfactants organize around SWNTs of different structures and electronic types. The energetic balance among nanotube–surfactant, water–surfactant and surfactant–surfactant interactions as well as the packing density, orientation, ionization and the resulting hydration of the surfactants should all be critical parameters affecting buoyant density and the quality of sorting. Additionally, the capacity of a surfactant to disperse SWNTs in aqueous solution should also play a role in determining the degree of separation. However, this characteristic alone is not a good predictor. For example, sorting of SDBS-encapsulated SWNTs was not observed here, despite the fact that Wenseleers and co-workers20 have demonstrated that both SDBS and SC disperse SWNTs equally well in aqueous solution.
Differences in the organization of anionic-alkyl surfactants and bile salts around carbon nanotubes are expected based on previous studies of these amphiphiles in other systems. On graphene, which is the closest analogue to an SWNT, anionic-alkyl surfactants organize into hemicylindrical micelles with liquid-like hydrophobic cores19,23, whereas bile salts form well-structured monolayers with their less-polar sides facing the hydrophobic surface24. Bile salts also order to form well-defined host–guest structures around small hydrophobic molecules21,25. Accordingly, the rigidity and planarity of bile salts, in contrast with anionic-alkyl surfactants, are expected to result in encapsulation layers that are sensitive to subtle changes in the underlying SWNT. Furthermore, the observed metal–semiconductor selectivity indicates a coupling of the surfactant and/or its hydration with the electronic nature of the underlying SWNT. Lu and co-workers have suggested that metallic SWNTs interact more strongly with adsorbates via π interactions than semiconducting SWNTs, due to their larger electronic polarizability26. Additionally, the packing density of the surfactants and their hydration are likely to be sensitive to electrostatic screening by the underlying SWNT. Other effects, such as partial charge transfer between metallic SWNTs and surfactants induced by CH–π interactions could also be important.
Density-gradient ultracentrifugation provides a scalable approach for sorting carbon nanotubes by diameter, bandgap and electronic type. This strategy has been demonstrated for SWNTs encapsulated by bile salts and mixtures of bile salts with anionic-alkyl surfactants for SWNTs between 7 and 16 Å in diameter. By successive iterations of ultracentrifugation, sharp diameter distributions have been achieved in which more than 97% of semiconducting SWNTs are within 0.2 Å of the mean diameter. Furthermore, the structure–density relationship for SWNTs has been engineered to achieve exceptional metal–semiconductor separation by using mixtures of competing co-surfactants, thus enabling the isolation of bulk quantities of SWNTs that are predominantly a single electronic type (typical yields are shown in the Supplementary Information, Fig. S8, and Methods). Because SWNTs sorted by this method are highly compatible with subsequent processing techniques and can be integrated into devices, it is expected that density-gradient ultracentrifugation will directly impact the large number of technological applications that require SWNTs with monodisperse structure and properties.
To disperse SWNTs in solutions of bile salts or other surfactants, 1 mg ml−1 SWNTs were dispersed in solutions of 2% weight per volume (w/v) surfactant via ultrasonication (see Supplementary Information, Methods, for more details). In co-surfactant experiments, SWNTs were initially dispersed in a 2% w/v SC surfactant solution and then diluted into 2% w/v co-surfactant solution. For example, in a 1:4 SDS/SC co-surfactant experiment, one part 2% w/v co-surfactant was added to four parts of a 2% w/v SC solution containing dispersed SWNTs. Furthermore, density gradients were loaded homogeneously from top to bottom with these surfactants at the same ratios. Thus, no gradients in surfactant concentrations were present.
Density gradients were formed from aqueous solutions of a non-ionic density gradient medium, iodixanol6,8,27,28, purchased as OptiPrep 60% w/v iodixanol, 1.32 g cm−3 (Sigma-Aldrich). Gradients were created directly in centrifuge tubes by one of two methods: by layering of discrete steps and subsequent diffusion into linear gradients or by using a linear gradient maker (see Supplementary Information, Figs. S9–S11, and Methods).
Centrifugation and Fractionation
Centrifugation was carried out in two different rotors, a fixed angle TLA100.3 rotor and a swing bucket SW41 rotor (Beckman-Coulter), at 22 °C and at 64,000 r.p.m. and 41,000 r.p.m., respectively, for 9–24 h, depending on the spatial extent and initial slope of the gradient. At the average radii (37.9 mm and 110 mm, respectively), these rotational velocities result in centripetal accelerations of 174,000 g and 207,000 g, respectively. To fractionate TLA100.3 tubes, a modified Beckman Fractionation System (Beckman-Coulter) was utilized in an upward displacement mode using Fluorinert FC-40 (Sigma-Aldrich) as a dense chase media. 25 µl fractions were collected. To fractionate SW41 centrifuge tubes, a Piston Gradient Fractionator system was utilized (Biocomp Instruments, Canada). Fractions of 0.5–3.0 mm were collected (70–420 µl in volume). In both cases, fractions were diluted to 1 ml in 2% w/v surfactant solution for optical characterization.
This work was supported by the US Army Telemedicine and Advanced Technology Research Center, the National Science Foundation and the Department of Energy. A National Science Foundation Graduate Student Fellowship (M.S.A.), a Natural Sciences and Engineering Research Council of Canada Postgraduate Scholarship (A.A.G.), and an Alfred P. Sloan Research Fellowship (M.C.H.) are also acknowledged. Furthermore, J. Suntivich, X. Du and M. Disabb are gratefully recognized for measurement of optical absorbance spectra (J.S., X.D.) and evaporation of Au electrodes (M.D.). We thank J. Widom and the Keck Biophysics Facility for use of their ultracentrifuges, J. Chen for providing laser-ablation-grown SWNTs, and L. Palmer and Ph. Avouris for useful discussions.
Supplementary methods and figures S1-S11
About this article
Nature Nanotechnology (2018)