Report

Sub–10 nm polyamide nanofilms with ultrafast solvent transport for molecular separation

See allHide authors and affiliations

Science  19 Jun 2015:
Vol. 348, Issue 6241, pp. 1347-1351
DOI: 10.1126/science.aaa5058

Composite membranes for filtering solvents

Much research has focused on finding membranes that can purify water or extract waste carbon dioxide. However, there is also a need for the removal of small molecules from organic liquids. Many existing processes are energy-intensive and can require large quantities of solvents. Karan et al. grew confined polymer layers on a patterned sacrificial support to give rippled thin films that were then placed on ceramic membranes (see the Perspective by Freger). The composite membrane showed high flux for organic solvents and good stability and was able to separate out small molecules with high efficiency.

Science, this issue p. 1347; see also p. 1317

Abstract

Membranes with unprecedented solvent permeance and high retention of dissolved solutes are needed to reduce the energy consumed by separations in organic liquids. We used controlled interfacial polymerization to form free-standing polyamide nanofilms less than 10 nanometers in thickness, and incorporated them as separating layers in composite membranes. Manipulation of nanofilm morphology by control of interfacial reaction conditions enabled the creation of smooth or crumpled textures; the nanofilms were sufficiently rigid that the crumpled textures could withstand pressurized filtration, resulting in increased permeable area. Composite membranes comprising crumpled nanofilms on alumina supports provided high retention of solutes, with acetonitrile permeances up to 112 liters per square meter per hour per bar. This is more than two orders of magnitude higher than permeances of commercially available membranes with equivalent solute retention.

Many separation processes used by industry require evaporation and distillation, which have high energy consumption due to the latent heat of vaporization. Membrane technology would require only one-tenth as much energy to process an equivalent amount of liquid (1), but for industrial processes involving large quantities of organic liquids, such membranes should be stable in organic solvents and have high permeance to enable processing within a reasonable time. Thin-film composite (TFC) membranes for water desalination (2, 3), which comprise a polyamide separating layer formed by interfacial polymerization on top of a porous ultrafiltration support membrane, have been adapted for organic solvent nanofiltration (OSN). However, the permeance of TFC-OSN membranes is still relatively low (~2.5 liters m−2 hour−1 bar−1 for membranes that reject >90% of solutes with molecular weight <300 g mol−1) (4), demanding large membrane areas for industrial applications. Meanwhile, for OSN, Karan et al. (5) used chemical vapor deposition to prepare diamond-like carbon (DLC) nanosheet membranes that showed ultrafast permeance when the thickness of the DLC separation layer was decreased to 30 nm; the DLC layer remained mechanically robust. In contrast, when the same thickness-reduction approach was followed for solution-cast films of a linear polymer (PIM-1), a maximum heptane permeance of ~18 liters m−2 hour−1 bar−1 was measured for a film 140 nm thick, after which permeance decreased with decreasing thickness. This was attributed to structural relaxation of the films for thicknesses less than 100 nm (6).

We postulated that thin films comprising network polymers will achieve ultrafast solvent permeance when their thickness is reduced below 100 nm, in contrast to films comprising linear polymers. Hence, we decided to explore the permeation of solvents through highly cross-linked sub–10 nm polyamide nanofilms fabricated via interfacial polymerization of diamine and acid chloride. By controlling the rate of interfacial reaction, m-phenylenediamine (MPD)–trimesoyl chloride (TMC) nanofilms were prepared on a sacrificial layer of cadmium hydroxide nanostrands. The fabrication process is shown in Fig. 1, A and B. The nanostrand layer was formed on an ultrafiltration support membrane [either cross-linked polyimide XP84 (figs. S1 and S2) or porous alumina] via vacuum filtration of nanostrand solution (5, 7, 8) (Fig. 1, C and D, and fig. S3). Polyamide nanofilms with controlled morphology were formed on the nanostrand layer through controlled release of diamine at the water-hexane interface (fig. S4). The nanostrand layer was then removed by acid dissolution, using a dilute aqueous solution of hydrochloric acid, or, for prolonged interfacial polymerization reactions, by virtue of the acid formed in the reaction (Fig. 1B). This resulted in the formation of ultrathin nanofilms comprising a highly cross-linked polyamide network with terminal carboxylic acid groups on the surface facing the hexane (figs. S5 to S7).

Fig. 1 Description of the controlled interfacial polymerization process and the resulting sub–10 nm nanofilms.

(A) A sacrificial Cd(OH)2 nanostrand layer was prepared on top of an ultrafiltration support membrane [porous alumina or cross-linked polyimide (XP84)]. The nanostrand layer was saturated with an aqueous solution of diamine and contacted with a hexane layer containing TMC, enabling the synthesis of polyamide nanofilms via interfacial polymerization. (B) The nanostrand layer was removed by acid dissolution, resulting in a free-standing nanofilm that was then attached to a support membrane. (C and D) Cross-sectional and surface SEM images of a nanostrand layer, 120 nm thick, formed on an alumina support. The nanofilms with controlled surface morphology were fabricated from MPD-TMC on XP84 and then transferred onto alumina. (E and F) Cross-sectional SEM images of smooth nanofilm (MPD-0.1%-10min) (E) and crumpled nanofilm (MPD-3%-1min) (F). Insets in (E) and (F) are high-magnification images. (G and H) AFM height image and corresponding height profile of a section of a smooth nanofilm (MPD-0.1%-10min) on top of a silicon wafer. A scratch was made to expose the wafer surface and allow measurement of the height from the silicon wafer surface to the upper nanofilm surface. (I) AFM images of the wrinkles formed when the MPD nanofilms are transferred onto an elastomer substrate and subjected to an applied compressive stress. Top, smooth nanofilm; bottom, crumpled nanofilm. (J) Photograph of a smooth nanofilm ~8 nm thick (MPD-0.1%-10min) transferred to a wire lasso. (K) Aspiration of a crumpled nanofilm through a pipette tip 320 μm in diameter.

The surface morphology of the nanofilms varies with MPD concentration (fig. S8); with increasing MPD concentration, the nanofilm appears crumpled, with feature sizes of 100 to 500 nm (figs. S9 to S13 and table S1). We believe that the crumpling phenomenon occurs when heat generation resulting from high-rate interfacial reactions leads to local temperature rises. These then create interfacial instabilities in the hexane layer through buoyancy driven by Rayleigh-Bénard convection (9). This causes the nanofilm to bend and crumple up, and generates additional interfacial area over the same time scale as the interfacial reaction itself. The resulting nanofilm is thus a “mold” of the interface (10).

Scanning electron microscopy (SEM) images of smooth and crumpled nanofilms (MPD-0.1%-10min and MPD-3%-1 min in Table 1, respectively) are shown in Fig. 1, E and F. Atomic force microscopy (AFM) images of a smooth free-standing nanofilm (MPD-0.1%-10min; fig. S14) transferred onto a silicon wafer and scratched to reveal the wafer surface give a thickness of ~8.4 nm (Fig. 1, G and H). Layer-by-layer growth of an MPD-TMC film on a silicon wafer (11) has suggested that the thickness of each cycle is 0.9 nm, and if similar dimensions apply to the MPD-TMC nanofilms formed by interfacial polymerization, these would be 8 to 10 polyamide units thick.

Table 1 Composition and surface properties of free-standing nanofilms fabricated by interfacial polymerization (IP).

Root mean square (RMS) roughness and thickness of free-standing nanofilms were measured on silicon wafers. The atomic composition was assessed by XPS measurements from free-standing nanofilms transferred onto gold-coated silicon wafers. The density of carboxylic acid groups (–COOH) was calculated from the core-level C1s XPS spectra (table S6). The degree of network cross-linking was calculated from the ratio of network to linear cross-linked portion of the polymer (table S7). The contact angle of the nanofilms was measured on XP84. Nanofilms with the suffix ACT were activated with DMF by immersion for 4 hours followed by washing in methanol (immersion for 15 min). Nanofilm with the suffix Flip refers to the reverse side. ND, not determined.

View this table:

These sub–10 nm nanofilms are robust, flexible, and defect-free over a few square centimeters in area (fig. S15). Wrinkling-based measurements (12, 13) under compressive stress induced by an elastomer substrate (Fig. 1I and fig. S16) confirmed that nanofilms have a Young’s modulus in the range 0.11 to 2.71 GPa, depending on the interfacial polymerization conditions (table S2). Figure 1J shows a smooth nanofilm (MPD-0.1%-10min) transferred to a wire lasso; although the film is only ~8 nm thick, it forms an integral surface across the whole 1.5-cm diameter of the lasso. As demonstrated in Fig. 1K, a nanofilm can be aspirated through a narrow pipette tip without visibly tearing or fragmenting.

The strength and flexibility of the nanofilm make it feasible to produce composite membranes in a two-stage process in which the separating nanofilm layer is formed by interfacial polymerization as a free-standing entity and is then placed on top of a support to form a composite membrane. Further, if the crumpled texture is robust under pressurized filtration, then (relative to a smooth film) it will provide a higher permeable nanofilm area per unit area of composite membrane support. This will enhance composite membrane permeance, which is calculated according to flow normal to the area of the support membrane.

Properties of nanofilms fabricated under different interfacial reaction conditions and with both aromatic and semi-aromatic diamines are listed in Table 1. The thickness of the smooth MPD nanofilms was approximately constant after 1 min, whereas their mass measured with a quartz crystal microbalance increased by a factor of 3 with prolonged reaction time (1 to 10 min) (table S3). The increasing density explains the increase in elastic modulus of the nanofilm (table S2). The top (hexane-facing) surfaces of the nanofilms were characterized by AFM (figs. S17 to S22 and table S4). For crumpled MPD nanofilms, the reported thickness reflects the apparent thickness of the crumpled layer (up to 94 nm, MPD-3%-1 min in Table 1) rather than the actual nanofilm thickness. In fig. S23, we show that the conditions that result in a crumpled nanofilm formed on the nanostrand layer give a smooth nanofilm when applied at an aqueous-organic interface between free liquids, which suggests that the crumpling occurs when the film formation is constrained by the nanostrand layer. Under AFM, the reverse (aqueous-facing) surfaces of the nanofilms revealed imprints of the nanostrand layer as well as the porous backside of the crumpled sheet (fig. S24 and table S5).

We estimated the carbon, oxygen, and nitrogen content in the nanofilms by x-ray photoelectron spectroscopy (XPS) (Table 1 and fig. S25). Deconvolution of the C1s spectrum revealed the signature of amide and carboxyl groups; the percentages of carboxylic acid groups (resulting from the hydrolysis of unreacted acyl chloride groups) and unreacted amine groups were estimated from the N1s and O1s spectra (figs. S26 and S27 and table S6). The extent of chemical cross-linking was calculated from the relative values of N and O measured from XPS (table S7). The thicknesses for smooth MPD nanofilms were also estimated by Ar sputtering in XPS and spectroscopic ellipsometry, confirming a sub–10 nm thickness (fig. S28). Estimated values of C, N, and O from energy-dispersive x-ray (EDX) measurements under transmission electron microscopy (TEM) (table S8) concur with XPS measurements. The contact angles for all MPD nanofilms were in the range 50° to 60°, which suggests similar polarity (surface energy) of the surfaces of all the nanofilms formed. Furthermore, TEM observations confirmed that the nanofilms were amorphous polymers (figs. S29 to S35). Piperazine (PIP) nanofilms required longer reaction times (10 min) to become defect-free (figs. S36 to S38).

Measurement of the MPD nanofilm thickness has been a challenge in understanding the polyamide separating layer in TFC desalination membranes; current estimates are in the range of several hundred nanometers (3, 14, 15). Our nanofilms can be made free-standing without deforming their morphology, thereby enabling close examination. We assert that the rough morphology [length scales ~100 to 500 nm with hollow features (fig. S33)], which results from moderate MPD-TMC concentrations, comprises crumpled nanofilms ~8 nm thick (fig. S39) and that it is the thickness of the whole crumpled zone that has been reported previously as the film thickness of TFC membranes.

Figure 2A shows the rejection performance of a typical nanofilm composite membrane. Organic solvent permeance of polyamide TFC membranes is known to be enhanced by dimethylformamide (DMF) activation (4), and the same trend is observed for our nanofilm composite membranes (Table 2, Fig. 2B, and figs. S40 and S41). Seeking to explain this phenomenon, we examined surface morphology, chemical structure, film thickness, and mass per unit area before and after activation with DMF. None revealed any significant differences, and so we attribute the permeance increase to molecular-level modification of the membrane material (see supplementary materials and table S9). Before and after activation, a similar rejection was observed for charged dyes and polystyrene oligomers, whereas neutral dyes exhibited slightly lower rejection (Table 2 and figs. S42 to S45). MPD nanofilm membranes were tight, with molecular weight cutoff (MWCO) below 246 g mol−1 based on the rejection of 6-hydroxy-2-naphthalenesulfonic acid sodium salt (HNSA; molecular weight 246.2 g mol−1, size 0.59 nm3; table S10), whereas 4-(aminomethyl)piperidine (AMP) and PIP nanofilms were progressively looser with higher MWCO. These membranes are at least equivalent in selectivity to TFC-OSN membranes reported to date (4). Acetonitrile, with viscosity of 0.32 × 10−3 Pa·s and solubility parameter due to dipole force (δp) of 18.0 MPa1/2 (table S9), gave the highest permeance of 112 liters m−2 hour−1 bar−1. Methanol (0.49 × 10−3 Pa·s, 12.3 MPa1/2) gave the second highest permeance of 52.2 liters m−2 hour−1 bar−1. Methyl ethyl ketone (MEK; 0.38 × 10−3 Pa·s, 9.0 MPa1/2), with a molar diameter 20% greater than that of acetonitrile, gave a permeance one-third that of acetonitrile; the relatively nonpolar heptane and toluene gave the lowest permeances. Methanol flux was linear with transmembrane pressure (fig. S15).

Fig. 2 Nanofiltration performance of smooth and crumpled nanofilm composite membranes.

(A) Ultraviolet-visible absorption spectra of acid fuchsin (ACF) dye in methanol to estimate the separation performance of the MPD nanofilm composite membrane. Inset shows the ACF molecular structure. (B) Plot of solvent permeances against the combined solvent property (viscosity, molar diameter, and solubility parameter) for crumpled nanofilm (MPD-3%-1min) on alumina. (C) Variation of methanol permeance through crumpled nanofilm (MPD-4%-1min) composite membranes with time. Membranes were activated via DMF filtration followed by washing in methanol. (D) Plot of methanol permeance with time for the MPD nanofilm (after DMF activation) on alumina (smooth, MPD-0.1%-10min; crumpled, MPD-3%-1min). Nanofiltration was conducted in a dead-end stirred cell (500 rpm) at 30°C under 10 bar. (E and F) AFM height image and schematic representation of the estimated length from AFM, showing actual and superficial length of smooth nanofilm (E) and crumpled nanofilm (F).

Table 2 Organic solvent nanofiltration performance of nanofilm composite membranes.

Negatively charged dyes with varying molecular weight dissolved in methanol were used to study solute retention. Nanofilm membranes were prepared on the prefabricated nanostrand layer deposited on cross-linked polyimide (XP84) or alumina. Composite membranes with the suffix ACT were activated with DMF filtration for 15 to 30 min followed by filtration with methanol under 10 bar in a dead-end cell. Feed solution concentration was 20 mg liter−1 in methanol. Feed volume was 50 ml or 100 ml, half of which was collected as permeate to determine the rejection value. Nanofiltration experiments were conducted in a dead-end stirred cell (500 rpm) at 30°C under 10 bar. Permeance values are expressed as liters m−2 hour−1 bar−1.

View this table:

From these observations, we propose a phenomenological transport model (16) describing the permeance (Ps,i) of solvent s through each nanofilm i:Embedded Image (1)where Ki is a proportionality constant for nanofilm i (m3 Pa−0.5), δp,s is the solubility parameter (Pa0.5), ηs is the solvent viscosity (Pa·s), and dm,s is the molar diameter of the solvent s (m). Equation 1 describes solvent permeance well for both activated and nonactivated nanofilms (Fig. 2B and fig. S41).

Methanol permeance for various nanofilm membranes is plotted as a function of time in Fig. 2C. The activated polyamide nanofilm maintained a constant permeance when supported on the aging-resistant alumina. With XP84, permeance decreased by 60% over the first 4 hours; both composite membranes exhibited similar rejection (Table 2). This permeance decline is presumably a result of physical aging and compaction of the XP84 support membrane with DMF filtration. If so, then these highly cross-linked nanofilms must be adhered to an aging-resistant support material to enable sustained performance. This result is especially interesting because these observations could only have been made through the formation of free-standing nanofilms that can be attached to either support.

Figure 2D shows that an alumina supported crumpled MPD nanofilm provided higher permeance than an alumina-supported smooth MPD nanofilm by a factor of >4 (Table 2). This is consistent with the assertion that the crumpled nanofilm provides an effective permeable area greater than the area of support it covers. Crucially, it appears that this crumpled texture does not collapse and fail under pressure. Linear flux over the pressure range 0.3 to 50 bar (fig. S15D) suggests that the permeable area of the crumpled nanofilm is constant, with no folding or flattening to form overlapping layers as pressure is increased. For a smooth film, the distance traveled by an AFM tip is the same as the distance across the scanned area, whereas the distance traveled by the tip as it moves up and down the ridges of the crumpled film is greater by a factor of 1.8 (Fig. 2, E and F). The permeance data suggest that the permeable area of a crumpled film is at least 4 times that of the smooth film; the AFM data support this, and the schematic in Fig. 2F illustrates why the AFM value of 1.8 is an underestimate.

The methanol flux through the crumpled MPD nanofilm on a porous alumina support (52.2 liters m−2 hour−1 bar−1) is more than 20 times the value reported for TFC membranes (4) and is two orders of magnitude higher than commercially available OSN membranes (fig. S46). This ultrafast permeation through a polymeric thin film, without compromising selectivity, improves the potential for energy efficiency in OSN applications (2, 17). A range of chemical processes have a strong demand for solvent-resistant nanofiltration membranes with superior permeance. We provide here a potential candidate in the form of sub–10 nm crumpled nanofilm composite membranes with ultrafast solvent permeance, excellent selectivity, and sufficient mechanical strength for nanofiltration applications.

Supplementary Materials

www.sciencemag.org/content/348/6241/1347/suppl/DC1

Materials and Methods

Supplementary Text

Figs. S1 to S46

Tables S1 to S10

References (1859)

References and Notes

  1. A thin nanostrand containing water is not a good heat transfer medium, and so heat generated at the interface will be dissipated by heating of the hexane, which has low heat capacity (see supplementary materials and fig. S39).
  2. Acknowledgments: Supported by UK Engineering and Physical Sciences Research Council platform grant EP/J014974/1 and the BP International Centre for Advanced Materials. We thank J. T. Cabral for his suggestions on determining the modulus values of the nanofilms.
View Abstract

Navigate This Article