Enrichment of Metallic Carbon Nanotubes Using a Two-Polymer Extraction Method.

The large-scale enrichment of metallic carbon nanotubes is a challenging goal that has proven elusive. Selective dispersion of carbon nanotubes by specifically designed conjugated polymers is effective for isolating semiconducting species, but a comparable system does not exist for isolating metallic species. Here, we report a two-polymer system where semiconducting species are extracted from the raw HiPCO or plasma-torch nanotube starting material using an electron-rich poly(fluorene-co-carbazole) derivative, followed by isolation of the metallic species remaining in the residue using an electron-poor methylated poly(fluorene-co-pyridine) polymer. Characterization of the electronic nature of extracted samples was carried out via a combination of absorption, Raman, and fluorescence spectroscopy, as well as electrical conductivity measurements. Using this methodology, the metallic species in the sample were enriched 2-fold in comparison to the raw starting material. These results indicate that the use of electron-poor polymers is an effective strategy for the enrichment of metallic species.

Introduction

Due to their excellent physical[1−3] and optoelectronic properties,[4−7] single-walled carbon nanotubes (SWNTs) have received significant attention from the scientific community. Depending on the specific nanotube structure, the SWNTs can be semiconducting (sc-SWNT) or metallic (m-SWNT) in nature.[8] SWNTs have been incorporated into a myriad of applications, including field-effect transistors,[9] sensors,[10−14] high-strength nanocomposites,[15−17] photodetectors,[18] organic photovoltaics,[19,20] touch screens,[21] flexible electronics,[22−24] light-weight conductive wiring,[25] and conductive inks,[26,27] among numerous others.[28,29] Many of these applications require enriched samples of m- or sc-SWNTs. However, the SWNT samples produced using commercially viable methods, which include high-pressure carbon monoxide disproportionation (HiPCO),[30] chemical vapor deposition,[31,32] laser ablation,[33] arc-discharge,[34] and plasma-torch growth,[35] all produce heterogeneous mixtures. Typical as-produced SWNT samples contain approximately one-third m-SWNTs and two-thirds sc-SWNTs.[8] This complicated mixture of SWNT species generally precludes immediate device incorporation; for instance, pure sc-SWNT samples are imperative for field-effect transistors,[9] as m-SWNTs will short-circuit the device.

To address purification issues, several methods have been developed, including density gradient ultracentrifugation,[36,37] agarose gel filtration,[38,39] electrophoresis,[40] two-phase extraction,[41,42] and size-exclusion chromatography.[43,44] Although these methods are capable of separating individual SWNT species, scalability is limited, which is reflected in the still exorbitant costs for enriched m-SWNT samples (>99% purity) from commercial suppliers ($899 USD/mg, Raymor Industries Inc., August 2018). A promising alternative and scalable method is conjugated polymer sorting.[45,46] This method employs the structural tuning of conjugated polymers (side chain, polymer backbone, molecular weight, etc.) to produce polymer–SWNT dispersions enriched with specific SWNT subtypes. The discrimination mechanism, although not yet clear, occurs during the polymer–SWNT sonication step (selective polymer–SWNT interactions) and/or during the centrifugation step (selective SWNT removal). The molecular properties of conjugated polymers can be precisely tuned, and the scaffolds of polyfluorene,[47−51] polycarbazole,[52−54] polythiophene,[55−58] and others[59−61] have been used to produce enriched sc-SWNT samples. Despite this progress, the selective dispersion of m-SWNTs using conjugated polymers remains elusive.

Previously, we have explored the effect of polymer backbone electron density on SWNT dispersion selectivity.[62−64] We prepared a series of homologous structures and varied only the electron-donating or -withdrawing nature of the pendant substituents. We have observed that under the same dispersion preparation conditions, “electron-rich” conjugated polymers produced enriched sc-SWNT dispersions, whereas relatively electron-poor conjugated polymers produced dispersions with increased amounts of m-SWNTs. In addition to this work, our group has recently shown that the initial ratio of m-/sc-SWNTs can influence the dispersion outcome.[65] Intriguingly, poly(9,9-dioctylfluorene-2,7-diyl), despite its known selectively for sc-SWNTs,[48] was able to disperse m-SWNTs when the m-/sc-SWNT ratio was increased to 3:1. We thus surmised that altering the initial m-/sc-SWNT ratio may enable improved discrimination for m-SWNTs in conjunction with our electron-rich/electron-poor paradigm.

As mentioned above, commercial SWNT samples are relatively homogeneous in their initial proportion of 1:2 m-/sc-SWNTs. Thus, modification of the initial m-/sc-SWNT ratio must be accomplished during the purification process. In a study by Malenfant and co-workers, polyfluorene derivatives were used to discriminate for sc-SWNTs, and then the undispersed SWNT residue was recycled several times to extract as many sc-SWNTs as possible.[66] Conceivably, a conjugated polymer that produces concentrated and enriched sc-SWNT dispersions could be used to bias the initial m-/sc-SWNT ratio toward m-SWNTs, and then a second polymer dispersant can be used to disperse the residual m-SWNTs. In this work, we perform a series of sc-SWNT extractions using an electron-rich conjugated polymer and then disperse the resulting m-SWNT-enriched residue using a relatively electron-poor conjugated polymer. We find that the amount of dispersed m-SWNTs is improved through the combination of initial sc-SWNT removal followed by use of an electron-poor conjugated polymer as the final dispersant.

Results and Discussion

To begin our investigation, we first prepared an electron-rich carbazole-co-fluorene conjugated polymer and an electron-poor methylated fluorene-co-pyridine copolymer. The carbazole monomer was prepared from commercially available 4,4′-dibromobiphenyl via nitration and Cadogan ring closure to form the carbazole nucleus, followed by phase-transfer alkylation to install the branched solubilizing alkyl side chain (see the Supporting Information, Scheme S1). The complementary fluorene monomer was synthesized from commercially available fluorene via bromination with N-bromosuccinimide (NBS), phase-transfer alkylation with 1-bromododecane, and boronate esterification with bis(pinacolato)diboron using Miyaura conditions (see the Supporting Information, Scheme S2).[67] We then polymerized the carbazole and fluorene monomers via Suzuki polycondensation to afford P1 (see the Supporting Information, Scheme S3). Next, polymerization using the fluorene monomer and commercially available 2,5-dibromopyridine generated poly(fluorene-co-pyridine) PF–Py, which could be methylated with iodomethane to produce P2.[64] GPC analysis showed that P1 had a number-average molecular weight (Mn) of 46.3 kDa, corresponding to a degree of polymerization (DP) of ∼46, and a dispersity (Đ) of 1.99, and that P2 had an Mn of 13.5 kDa, corresponding to a DP of ∼23, and a Đ of 1.84. The absorption and emission data for polymers P1 and P2 are provided in the Supporting Information.

With our polymers in hand, supramolecular complexes between P1 and raw HiPCO SWNTs (tube diameter 0.8–1.2 nm)[45] were prepared following literature procedures.[68] During initial studies, we evaluated different dispersion solvents and polymer-to-nanotube mass ratios (Figure S17). It was determined that optimal mass ratios for P1/SWNT and P2/SWNT were 1.5:1 and 1:1, respectively, which provided highly concentrated dispersions in tetrahydrofuran (THF) while consuming minimal amounts of polymer. Briefly, 5 mg of raw SWNTs were added to a solution of 7.5 mg of P1 dissolved in 10 mL of THF. The sample was sonicated for 2 h in an ice-chilled bath sonicator, followed by centrifugation at 8346g for 30 min. The supernatant was carefully removed, filtered through a Teflon membrane with 0.2 μm pore diameter, and thoroughly rinsed with THF until the filtrate did not fluoresce when excited with a hand-held UV lamp at 365 nm. The polymer–SWNT thin film was then redispersed in 10 mL of THF using an ice-chilled bath sonicator for 30 min. Meanwhile, the residue in the centrifugation tube was sonicated in 10 mL of THF for 10 min and then filtered through a 0.2 μm pore diameter Teflon membrane to remove unbound polymer. As shown in Figure , the residue could then be redispersed using either 7.5 mg of P1 (to further extract sc-SWNTs) or 5 mg of P2 (to disperse m-SWNTs). For this study, we use “ExPy” nomenclature, where x indicates the number of extractions (0–4) and y indicates the polymer used to redisperse the residue (1 or 2). Using this process, we produced nine polymer–SWNT dispersions, which were evaluated for sc- and m-SWNT content based on the number of extractions and the final polymer dispersant.

Figure 1

Cartoon representation of the extraction process. The nomenclature used is ExPy, where x is the extraction number (0–4) and y is the polymer used to redisperse the pellet (1 or 2). The chemical structures of electron-rich P1 (blue) and relatively electron-poor P2 (red) are shown.

To characterize the resulting polymer–SWNT complexes, we first performed UV–vis–near-infrared (NIR) absorption spectroscopy. The absorption peaks observed arise from the different SWNT species present within each polymer–SWNT dispersion. For HiPCO SWNTs, there are three different regions: two semiconducting regions, S11 (830–1600 nm) and S22 (600–800 nm), and one metallic region, M11 (440–645 nm).[69] The absorption spectra were normalized to the local minimum at ∼905 nm to highlight the differences in m- and sc-SWNT content for each dispersion. Un-normalized spectra are provided in the Supporting Information (Figure S1). Preliminary studies were performed to ensure that our polymers do not equally disperse sc- and m-SWNTs (details in the Supporting Information). Three dispersions, including a P1–SWNT, P2–SWNT, and a dispersion using SDBS surfactant, were formulated. SDBS is well known to indiscriminately disperse SWNTs, irrespective of their electronic nature. The P1–SWNT and P2–SWNT dispersions possess lesser and greater M11 absorption features, respectively, compared to that of the SDBS–SWNT dispersion (Figure S16), suggesting that P1 shows a degree of selectivity for sc-SWNT dispersions and P2 for m-SWNT dispersions. As shown in Figure A, the intensity of the M11 region increases after each extraction step in the P1–SWNT dispersion series. This suggests that sc-SWNT discrimination is diminished as the m-/sc-SWNT ratio is biased toward m-SWNTs, or as the polymer/SWNT mass ratio increases because the total amount of SWNTs decreases with each extraction. Similarly, the intensity of the M11 region also increases for the P2–SWNT dispersion series as a function of extraction count (Figure B). Intriguingly, the electron-poor P2 disperses more m-SWNTs than the electron-rich P1, as evidenced by the M11 peak intensities when juxtaposing the absorption spectra of polymer–SWNT dispersions at the same extraction number (Figure C,D). This trend was consistent over all extractions (Supporting Information, Figure S2), revealing that the use of an electron-poor conjugated polymer is beneficial in dispersing significant amounts of m-SWNTs, especially when compared to using only P1 as both the sc-SWNT extraction agent and as the final m-SWNT dispersant. The color of the polymer–SWNT dispersions corroborates our analysis. For HiPCO samples, a green color was observed for the P1–SWNT dispersions, indicating sc-SWNT enrichment, whereas a red color was observed for the P2–SWNT dispersions, indicating m-SWNT enrichment (Figure S3).

Figure 2

UV–vis–NIR absorption spectra for the polymer–SWNT extraction study using HiPCO SWNTs. Polymer–SWNT dispersions after 0–4 extractions for (A) P1 and (B) P2. A direct comparison of the absorption spectra for polymer–SWNT dispersions after (C) 0 extractions and (D) 3 extractions are shown. The spectra are normalized to the local minimum at ∼905 nm to show a relative m-SWNT and sc-SWNT content.

To further probe the electronic nature of the nine polymer–SWNT dispersions produced, resonance Raman spectroscopy was performed. This technique allows for the examination of both m- and sc-SWNT species within a given sample[70] and utilizes laser excitation wavelengths that overlap with the van Hove singularities present in the one-dimensional density of states for a particular SWNT type.[71] As the electronic transitions depend on nanotube species and diameter, only a subset of the total nanotube population will be observed for each individual excitation wavelength, necessitating the use of multiple excitation wavelengths for complete characterization.[72] Thin films of our polymer–SWNT dispersions were prepared by drop-casting the dispersion onto silicon wafers, followed by evaporation at RT. A reference SWNT sample was prepared by briefly sonicating the raw starting material in chloroform and using this suspension to prepare a sample using the same drop-casting method. For HiPCO SWNTs, three laser excitation wavelengths were used: 514, 633, and 785 nm. These wavelengths have previously been shown to adequately characterize the HiPCO electronic character, as both sc- and m-SWNTs can be separately probed.[73]Figure shows the radial breathing mode (RBM) for P1–SWNT and P2–SWNT dispersions at 633 nm. All the spectra were normalized to the G-band (at approximately 1590 cm–1) for comparative analysis. Using a 633 nm excitation wavelength, both m- (175–230 cm–1) and sc-SWNTs (240–300 cm–1) are in resonance with HiPCO samples.[74] As shown in Figure A, the P1–SWNT samples possess sharp peaks in the sc-SWNT region. As the extraction number increases, the intensity of the m-SWNT region also increases, albeit minimally. This suggests that after extracting a critical amount of sc-SWNTs, P1 begins to disperse small amounts of m-SWNTs. Likewise, the P2–SWNT samples exhibit increasing propensity for dispersion of m-SWNT as the number of extractions increase. As seen in Figure B, P2 disperses substantially more m-SWNTs than P1 at the same extraction number, as evidenced by the increased peak intensity in the m-SWNT region. This result corroborates the conclusions of the absorption spectra (vide supra).

Figure 3

Raman spectra of the RBM region (λex = 633 nm) for polymer–SWNT samples after sc-SWNT extractions (0–4 times) with HiPCO SWNTs prepared using (A) P1 and (B) P2. Pink boxes represent signals arising from m-SWNTs, whereas gray boxes represent signals arising from sc-SWNTs.

To further characterize the samples, we examined them using an excitation wavelength of 785 nm, which is primarily in resonance with sc-SWNTs (175–280 cm–1). As shown in Figure S4A,B, all polymer–SWNT samples possessed sc-SWNTs. When HiPCO SWNTs are excited at this wavelength, an intense peak (∼265 cm–1) originating from bundled (10, 2) SWNTs is observed.[75] This “bundling peak” allows for the evaluation of the bundling degree in a sample. Figure S4A,B shows that all the polymer–SWNT samples have significant bundling peak suppression, in contrast to the raw nanotube control sample. This suggests that both P1 and P2 efficiently exfoliate SWNTs. Lastly, the polymer–SWNT samples were probed at 514 nm. For sc-SWNTs, a broad peak is centered at 180 cm–1, whereas for m-SWNTs, the peaks arise between 225 and 290 cm–1.[76] As shown in Figure S4C,D, m-SWNTs are present in all the samples. However, based on the G-band analysis, it appears that the m-SWNT signals are more intense in the P2–SWNT samples (for full Raman spectra refer to the Supporting Information, Figure S5). The G-band consists of two peaks: a lower-frequency G– and a higher-frequency G+. For sc-SWNTs, the G– and G+ both have Lorentzian line shapes, but for m-SWNTs, the G– exhibits a broader Breit–Wigner–Fano (BWF) line shape.[77] The BWF line shape is more prominent in the P2–SWNT samples, suggesting that they possess a higher m-SWNT content. Based on the aforementioned analysis, the P2–SWNT samples possess more m-SWNTs than P1–SWNT samples for all extractions, although P2 cannot completely remove all sc-SWNTs.

Photoluminescence (PL) maps of the polymer–SWNT dispersions were recorded and juxtaposed with the locations of sc-SWNT fluorescence maxima, assigned using literature data.[6] The dispersions for E0P1, E0P2, and E3P2 were concentration-matched (absorption of ∼0.27 at 1135 nm) using UV–vis–NIR spectroscopy (Supporting Information, Figure S6). The PL maps of all the extractions for P1 (Figure S7) and P2 (Figure S8) are provided in the Supporting Information. As shown in Figure , it is apparent that when intensity scales are identical, the E0P1 dispersion has the most intense fluorescence in comparison to the other polymer–SWNT samples. When comparing E0P2 to E3P2, the former possesses a slightly higher fluorescence intensity than the latter. These differences in fluorescence intensity are likely due the presence of m-SWNTs, which are known fluorescence quenchers.[5] Although the Raman analysis indicates that bundles are not dispersed to any appreciable amount in these systems, their presence cannot be entirely ruled out. Thus, the likely source of the observed fluorescence quenching is the presence of m-SWNTs that are either individually wrapped with polymer or present in small bundles. This suggests that E3P2 possesses more m-SWNTs than E0P2, and that both E0P2 and E3P2 possess more m-SWNTs than E0P1. These data are consistent with the information provided by the UV–vis–NIR and Raman spectra.

Figure 4

PL maps for concentration-matched samples of (A) E0P1, (B) E0P2, and (C) E3P2. Maps are set to identical intensity scales. Samples were concentration-matched using UV–vis–NIR spectroscopy (absorbance of ∼0.27 at 1135 nm).

Lastly, we characterized our SWNT samples using electrical conductivity measurements via the van der Pauw method.[78] We first prepared square-shaped thin films with dimensions of 0.5 × 0.5 cm for our polymer–SWNT samples by filtering 300 μL of polymer–SWNT dispersion through a Teflon filtration membrane having a 0.2 μm pore diameter, clamped between two solvent-resistant aluminum masks under vacuum. The samples were concentration-matched using UV–vis–NIR spectroscopy (Figure S9). The thin films were dried in an oven at 60 °C for 1 h and then resistance was measured by direct contact with four platinum probes placed in the corners of the square-shaped thin film. Voltages from 0 to 250 mV were applied to one pair of contiguous electrodes and the current was measured on the opposite pair of electrodes (e.g., if the thin film corners were labeled from 1 to 4 clockwise, the first measurement would apply V12 and measure I43). The thin film resistance (R) was calculated from the slope of the resulting I–V curve. The measurements were repeated for all four electrode combinations (i.e., V12, V23, V34, and V41) and the total resistance of the thin film (RT) was calculated as (measured in triplicate). The sheet resistance (Rs) was then calculated as . It was found that the Rs of the E0P2 thin film was (3.6 ± 1.7) × 107 Ω/sq, and that the Rs of the E4P2 thin film was (2.1 ± 0.7) × 106 Ω/sq. The Rs difference of an order of magnitude between the E0P2 and E4P2 samples is consistent with E4P2 possessing a higher concentration of m-SWNTs than E0P2, which is corroborated by the above analyses.

Encouraged by these results, we next investigated our methodology using plasma-torch SWNTs, which have a higher average diameter (tube diameter of 1.1–1.5 nm)[45] than HiPCO SWNTs. Supramolecular complexes were prepared with plasma-torch grown SWNTs according to the literature procedures (see the Supporting Information for details).[68] Photographs of these dispersions can be found in the Supporting Information (Figure S10). For plasma-torch SWNTs, there are four regions of interest in the absorption spectrum: three semiconducting regions, including S11 (1400–1900 nm), S22 (750–1150 nm), and S33 (420–580 nm), and one metallic region, M11 (600–750 nm).[66] The spectra, depicted in Figure , were normalized to the local maximum at ∼936 nm to highlight the differences in m-/sc-SWNT content. Un-normalized spectra are provided in the Supporting Information (Figure S11). Because the M11 and S22 regions do not overlap with each other or the conjugated polymer absorption, it is possible to estimate the m-/sc-SWNT ratio for the plasma-torch SWNT samples using UV–vis–NIR spectroscopy.[79] This analysis entails measuring the ratio of the integrations for the M11 and S22 regions (as depicted in Figure S14, Supporting Information). In the P1–SWNT dispersions, >90% of the sample was composed of sc-SWNTs. Meanwhile, for the P2–SWNT dispersions, >65% m-SWNT content was present after the extraction process. This indicates a 2-fold increase in the m-SWNT content compared to the initial starting material (Supporting Information, Table S1).

Figure 5

UV–vis–NIR absorption spectra for the polymer–SWNT extraction study using plasma-torch SWNTs. Polymer–SWNT dispersions after 0–4 extractions for (A) P1 and (B) P2. Spectra are normalized to the local maximum at ∼936 nm to show relative m-SWNT and sc-SWNT content.

For plasma-torch SWNTs, only the 633 and 785 nm excitation wavelengths are in resonance for Raman analysis.[7] As before, all the spectra were normalized to the G-band (at approximately 1590 cm–1) for comparative analysis. At 633 nm (Figure S12A,B), a BWF line shape arises from m-SWNTs.[77] Relative to the raw SWNTs, the suppressed BWF line shape for the P1–SWNT dispersions suggests enrichment of sc-SWNTs. As the number of extractions increase, the P1–SWNT dispersions begin to disperse m-SWNTs, as evidenced by the increase in the G–-band intensity. For the P2–SWNT dispersions, each sample exhibits an intense G–-band, indicating the presence of m-SWNTs. The 785 nm excitation wavelength (Figure S12C,D) was used to probe the presence of both m-SWNTs (135–175 cm–1) and sc-SWNTs (190–240 cm–1).[71] As before, the P1–SWNT dispersions exhibit m-SWNT signals only after multiple sc-SWNT removal steps. The P2–SWNT dispersions, meanwhile, exhibit intense m-SWNT signals after each extraction step. The UV–vis–NIR and Raman data overall suggest that, similar to HiPCO SWNTs, m-SWNT enrichment from plasma-torch SWNT samples is the highest when both the initial m-/sc-SWNT ratio is high and an electron-poor conjugated polymer is used as the final dispersant. This two-polymer system demonstrates a proof of concept for producing enriched m-SWNT dispersions using conjugated polymer sorting.

Conclusions

The design of an enrichment methodology for m-SWNT species with the potential for scalability has been explored. Using an electron-rich conjugated polymer that produces enriched and concentrated sc-SWNT dispersions, the initial m-/sc-SWNT ratio can be biased toward m-SWNTs. Subsequent dispersion of the residue using an electron-poor conjugated polymer produces dispersions with more m-SWNTs than possible using only the electron-rich polymer. Characterization of HiPCO polymer–SWNT samples using UV–vis–NIR, Raman, and fluorescence spectroscopy indicate that the P2–SWNT samples possessed more m-SWNTs than the P1–SWNT samples. The results were corroborated with electrical conductivity measurements that showed that the E4P2 sample had an order of magnitude lower sheet resistance than the E0P2 sample. Characterization of the plasma-torch polymer–SWNT samples via UV–vis–NIR and Raman spectroscopy indicated that the same result could be obtained, irrespective of the SWNT diameter. These results demonstrate a proof of concept for m-SWNT enrichment using a two-polymer, electron-rich/electron-poor polymer system.