Ultra‐Permeable Single‐Walled Carbon Nanotube Membranes with Exceptional Performance at Scale

Abstract Enhanced fluid transport in single‐walled carbon nanotubes (SWCNTs) promises to enable major advancements in many membrane applications, from efficient water purification to next‐generation protective garments. Practical realization of these advancements is hampered by the challenges of fabricating large‐area, defect‐free membranes containing a high density of open, small diameter SWCNT pores. Here, large‐scale (≈60 cm2) nanocomposite membranes comprising of an ultrahigh density (1.89 × 1012 tubes cm−2) of 1.7 nm SWCNTs as sole transport pathways are demonstrated. Complete opening of all conducting nanotubes in the composite enables unprecedented accuracy in quantifying the enhancement of pressure‐driven transport for both gases (>290× Knudsen prediction) and liquids (6100× no‐slip Hagen–Poiseuille prediction). Achieved water permeances (>200 L m−2 h−1 bar−1) greatly exceed those of state‐of‐the‐art commercial nano‐ and ultrafiltration membranes of similar pore size. Fabricated membranes reject nanometer‐sized molecules, permit fractionation of dyes from concentrated salt solutions, and exhibit excellent chemical resistance. Altogether, these SWCNT membranes offer new opportunities for energy‐efficient nano‐ and ultrafiltration processes in chemically demanding environments.


Carbon Nanotube Synthesis
Vertically aligned carbon nanotubes were grown on 100 mm Si (100) wafers that were coated with a Fe/Mo/Al2O3 (5.5/0.5/300 Å) catalyst stack by electron-beam evaporation without breaking vacuum between layers (base pressure ≤ 1.6 x 10 -6 mbar). The thicknesses of the catalyst layers were recorded in situ by a quartz crystal balance. CNT synthesis was performed by low pressure chemical vapor deposition in a cold-wall furnace (AIXTRON ® Black Magic Pro 6 in.) with acetylene as the hydrocarbon feedstock, as described in detail elsewhere. [1] The average CNT number density N (cm -2 ) of produced forests was calculated using the weight gain method with Equation (S1) where ρ is the volumetric mass density, SSAg is the specific surface area of graphene (1315 m 2 g -1 ), and d is the mean SWCNT diameter from TEM measurements. [2] This method has been shown to agree well with other techniques such as X-ray attenuation. [3] The volumetric mass density was obtained from the mass gain of the catalyst-coated silicon wafer after CNT growth and calculated according to Equation (S2) where mCNT is the mass of the CNT forest, hCNT is the height of the CNT forest, and ACNT is the CNT growth area. Approximately 30 mg of VACNTs were grown on each wafer and measured with 0.1 mg precision. Possible error on mCNT quantification due to amorphous carbon content amounts to < 5%, as revealed by previous TGA measurements. [1] The CNT forest height (~40 µm) was measured by optical microscopy as the difference between the z-location of the water surface (without CNTs) and the top of the CNT forest when each plane was in focus (0.1 µm resolution). A weighted average of 5 height measurements at different locations along the wafer meridian was used for calculations.

Carbon Nanotube and Composite Membrane Imaging
The SWCNT diameter distribution was determined using high-resolution transmission electron microscopy (HRTEM, 2100-F field-emission analytical TEM, JEOL) from a sample of more than 100 CNTs (0.01 nm resolution). CNT samples were harvested from pristine forests prior to membrane fabrication, sonicated in ethanol, and drop cast onto Formvar coated Cu TEM grids.
To extract the CNT diameter, defined here as the distance between the CNT wall centers, collected TEM images were analyzed using a custom MATLAB script. [1] Note that the inner pore diameter is actually 0.34 nm smaller (~1.4 ± 0.7 nm) than the reported average CNT diameter defined above (1.7 nm ± 0.7).
Cross-sectional scanning electron microscopy images of the composite membranes were taken by an Apreo (Thermo Scientific) SEM at 3 kV accelerating voltage and spot size 6. ImageJ software was used to measure the membrane thickness. CNT composites employed in this work were 36 ± 4 µm thick.

Carbon Nanotube Membrane Fabrication
The composite membranes were created by vapor deposition of poly-para-xylylene (parylene N) within vertically aligned CNT forests as described in detail elsewhere [3][4][5] . The polymer coating fills the interstitial gaps between the tubes and leaves a small excess layer on the top of the membrane surface. Excess parylene N and the CNT caps were removed using an inductive super magnetron (ISM) generated oxygen plasma (NE-550EXa, ULVAC

Control Membranes with Blocked CNTs
Control membranes with internally clogged CNTs were fabricated following the same protocol described above, with the only difference being that the CNT tips of the forests were removed via a short air plasma etch after growth (5 min, 30 W RF, Harrick PDC-001). Thus, during the following parylene N deposition step, both the inter-tube spaces and the inner volume of the uncapped CNT tubes were filled by polymer. As detailed in our separate work, [4] these control membranes with intentionally clogged CNT channels did not transport fluids even after extensive etching beyond the limit required to open membranes in this study. Together with the detailed transport and rejection tests, data collected from these control membranes strongly suggests that all recorded flow in our standard CNT membranes is through the CNTs only.

Gas Permeation
The gas transport rates through the composite membranes were measured in custom-built deadend permeation cells at ambient conditions (21 °C). The upstream of the membrane was flushed several times and the system was allowed to come to equilibrium before measurement. The feed was pressurized, and the downstream flow rate was measured using mass flow meters.
The obtained permeation data for each gas was used to calculate the enhancement factor, defined as the relative ratio of the measured CNT membrane permeance over the Knudsen diffusion prediction. The Knudsen permeance Pk is given by Equation (S3) where ε is the porosity, d is the CNT diameter, τ is the tortuosity, R is the universal gas constant, T is temperature, L is the membrane thickness, and M is the molar mass of the gas. In these calculations, we assumed a tortuosity of 1.25 based on our previous results. [3] For estimating gas and liquid flow enhancements, we assumed that all CNTs in the membranes are open to fluid flow at the gas permeance plateau. While we cannot experimentally validate that 100% of the CNTs are conducting fluids via other methods, we believe that our assumption is reasonable. Indeed, since the CNTs are capped during polymer infiltration, significant pore clogging by the vapor-deposited parylene N is unlikely. SWCNTs cannot form bamboo structures and do not display the tendency to be internally blocked by catalyst particles.

Liquid Permeation
The liquid permeation experiments were conducted in the same custom-built dead-end permeation cells. The membranes were first soaked in isopropyl alcohol for 3 minutes before being rinsed thoroughly with deionized water to ensure wetting of all the hydrophobic CNT pores. The feed was stirred using a magnetic stirrer suspended above the membrane and pressurized using nitrogen gas. The permeate weight was collected in a glass vial and automatically recorded on a digital balance (Adventurer AX, Ohaus).
The pure water flux was measured after the gas permeation experiments, but before any dye or salt experiments to avoid potential pore blocking or fouling. The enhancement factor for liquid permeation was defined as the relative ratio of the measured per-CNT flow rate

Rejection Measurements
To exclude the presence of membrane defects in both small and large area samples, we For NaCl and Na2SO4 rejection quantification, the absorbances in Equation (4) were replaced by the permeate and feed solution conductivities measured at room temperature (21 °C) with a 712 Conductometer (Metrohm AG).

Sorption Measurements
Sorption experiments were performed using Rose Bengal (1 x 10 -5 M) and 5 nm PEG-coated gold nanoparticle (0.05 mg mL -1 ) solutions. 1 cm 2 CNT composite membrane samples were soaked in 20 mL of each feed for 72 hours. This contact time is nearly two orders of magnitude longer than a typical rejection test. The UV-vis spectra along with images of the solution vials before and after membrane soaking are shown in Figure S1. The solution concentrations remained unchanged or increased only slightly after exposure to the membrane, suggesting that minor water loss due to evaporation around the seal of the vial is more significant than any dye adsorption onto the CNT membrane itself.
Furthermore, the membrane rejections before and after sorption tests were measured.
Membranes showed > 99% rejection for both the Rose Bengal and 5 nm PEG-coated gold nanoparticle solutions even after soaking in solution for 72 hours.
These results demonstrate that dye/particle adsorption on the membrane is negligible and cannot account for the complete dye/particle rejection observed during filtration experiments ( Figure 2f).

Donnan Exclusion Modeling
The Donnan model was used to predict the rejection behavior of ions through a charged CNT membrane according to Equation (S6) where ci m is the anion concentration in the membrane, ci is the anion concentration in solution, cx m is the membrane charge concentration, and zi and zj are the anion and cation charge, respectively. [7] The membrane charge concentration was fit to the experimental salt rejection data at the lowest salt concentration (1 mM NaCl and 0.33 mM Na2SO4) and then kept constant to estimate the rejection at higher concentrations.

Raman Spectroscopy
Micro-Raman spectroscopy (inVia™ Qontor ® confocal Raman microscope, Renishaw) with an excitation wavelength of λ = 633 nm and grating of 1200 lines mm -1 was used to quantify the quality of the CNTs by determining the G-band (~1590 cm -1 ) to D-band (~1310 cm -1 ) ratio. The radial breathing mode peak intensities were normalized to the D-band peak area. Raman spectra were measured at the center and edge of the CNT wafer before coating with parylene N. High G/D band values (6 and 6.8 at the center and edge, respectively) indicate good structural quality and spatial uniformity of the CNTs across large area.  Table S1. Direct Blue 71 and 5 nm PEG-coated gold nanoparticle solution rejection data at 0.28 bar applied pressure for the 1 cm 2 membrane in Figure 1f.

Tabulated Data
Feed Solution Rejection (%) Direct Blue 71 99.4 ± 0.3 PEG-coated Gold Nanoparticles 99.8 Table S2. Experimentally measured average enhancement factors for gas flow through CNT membranes shown in Figure 2b.