Sei sulla pagina 1di 6

Electron tomography reveals details of the internal

microstructure of desalination membranes


Tyler E. Culpa,1, Yue-xiao Shena,1, Michael Geitnera, Mou Paulb, Abhishek Royc, Michael J. Behrd, Steve Rosenberge,
Junsi Guf, Manish Kumara,2, and Enrique D. Gomeza,g,h,2
a
Department of Chemical Engineering, The Pennsylvania State University, University Park, PA 16802; bThe Dow Chemical Company, Lake Jackson, TX 77566;
c
The Dow Chemical Company, Freeport, TX 77541; dAnalytical Science, The Dow Chemical Company, Midland, MI 48667; eWater Solutions, DowDuPont, Edina,
MN 55439; fAnalytical Science, The Dow Chemical Company, Collegeville, PA 19426; gDepartment of Materials Science and Engineering, The Pennsylvania State
University, University Park, PA 16802; and hMaterials Research Institute, The Pennsylvania State University, University Park, PA 16802

Edited by Monica Olvera de la Cruz, Northwestern University, Evanston, IL, and approved July 23, 2018 (received for review March 18, 2018)

As water availability becomes a growing challenge in various regions fully aromatic, highly cross-linked polymer film with a thickness of
throughout the world, desalination and wastewater reclamation 100–200 nm and a root-mean-squared roughness of 50–130 nm
through technologies such as reverse osmosis (RO) are becoming (1, 17, 18).
more important. Nevertheless, many open questions remain re- Because of their critical role in the separation of salt and water,
garding the internal structure of thin-film composite RO membranes. the properties of polyamide films are pivotal to understanding
In this work, fully aromatic polyamide films that serve as the active transport mechanisms and performance in overall modules. The
layer of state-of-the-art water filtration membranes were investi- polyamide active layer has been labeled with a “ridge and valley”
gated using high-angle annular dark-field scanning transmission structure, characterized by techniques such as atomic force mi-
electron microscopy tomography. Reconstructions of the 3D mor- croscopy (AFM), scanning electron microscopy (SEM), and trans-
phology reveal intricate aspects of the complex microstructure not mission electron microscopy (TEM) (15, 18–23). Nevertheless,
visible from 2D projections. We find that internal voids of the active these characterization techniques are largely limited to 2D quali-
layer of compressed commercial membranes account for less than tative analyses mostly at the micrometer scale (23, 24). Polymer
0.2% of the total polymer volume, contrary to previously reported materials with low electron density hamper direct high-resolution
values that are two orders of magnitude higher. Measurements of electron microscope observation, such that measurements using
the local variation in polyamide density from electron tomography quartz crystal microbalance, spectroscopic ellipsometry, electron
reveal that the polymer density is highest at the permeable surface microscopy of thin cross-sections, and other techniques were
for the two membranes tested and establish the significance of employed to characterize polymer voids and thicknesses (25,
surface area on RO membrane transport properties. The same type 26). Another limitation is the focus of most studies on pristine
of analyses could provide explanations for different flux variations membranes (1, 8, 15, 21, 23–25, 27–35) that cannot account for
with surface area for other types of membranes where the density is the influence of membrane compression during operations; such
distributed differently. Thus, 3D reconstructions and quantitative compression could result in swelling, collapsing, and rearrangement
analyses will be crucial to characterize the complex morphology of of the highly heterogeneous polyamide network. Therefore, little
polymeric membranes used in next-generation water-purification quantitative information regarding the internal nanostructure of
membranes. the active layer exists.

|
reverse osmosis polyamide | transmission electron microscopy | Significance
|
tomography voids

The development of membrane materials is limited by our lack


M embrane microstructure plays a key role in determining
transport properties, such as rejection or permeability of a
specific component. As a consequence, characterization of the
of tools to characterize their complex microstructure. We
demonstrate how a combination of careful sample prepara-
tion, electron tomography, and quantitative analysis of 3D
nanoscale morphology has been of intense interest across several models can provide unique insights into the morphology of
areas of membrane research. Most of these studies are focused on polyamide active layers used in reverse osmosis membranes.
measuring the pore size and density, void structure, and spatial Extracting the 3D morphology is required to obtain accurate
locations of component rejection. Nevertheless, resolving the in- estimates of the top surface area and internal void fraction.
terplay between the 3D nanostructure and transport properties Furthermore, mapping the internal heterogeneity shows that
remains challenging, despite notable advances (1). Developing polymer density is highest near the top surface and suggests
approaches to resolve and analyze the 3D structure of membrane new models of how microstructure and membrane perfor-
materials at scales approaching molecular dimensions will likely mance is connected.
be crucial in applications such as ultrafiltration (2), nanofiltration
(3), virus filtration membranes (4), gas separations (5), filtration Author contributions: T.E.C., Y.-x.S., M.P., A.R., M.J.B., S.R., J.G., M.K., and E.D.G. designed
research; T.E.C., Y.-x.S., and M.G. performed research; T.E.C., Y.-x.S., M.G., M.P., A.R.,
for pharmaceuticals (6, 7), and reverse osmosis (RO) (1, 8).
M.J.B., S.R., J.G., M.K., and E.D.G. analyzed data; and T.E.C., Y.-x.S., M.K., and E.D.G. wrote
Of the various available technologies, RO has become the most the paper.
widely used approach for desalination and wastewater reuse due to Conflict of interest statement: M.G., A.R., M.J.B., S.R., and J.G. are all employees of
its economic benefits, high permeability, and selectivity (9–14). RO DowDuPont Inc.
systems rely on a semipermeable membrane to enable water pas- This article is a PNAS Direct Submission.
sage while rejecting ions. Commercial RO membranes are typically Published under the PNAS license.
thin-film composite membranes that incorporate an ultrathin 1
T.E.C. and Y.-x.S. contributed equally to this work.
polyamide top selective layer, a porous 20- to 50-μm polysulfone 2
To whom correspondence may be addressed. Email: manish.kumar@psu.edu or edg12@
support layer, and a nonwoven 50- to 150-μm polyester backing psu.edu.
(10, 15). Polyamide films are synthesized directly on the poly- This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.
sulfone support via an interfacial polymerization reaction between 1073/pnas.1804708115/-/DCSupplemental.
m-phenylenediamine and trimesoyl chloride (16), resulting in a Published online August 13, 2018.

8694–8699 | PNAS | August 28, 2018 | vol. 115 | no. 35 www.pnas.org/cgi/doi/10.1073/pnas.1804708115


Recent studies determined the void fraction of fully aromatic
polyamide thin films to be 15–32% (specifically 30% and 31%
for the commercial Dow SW30HR and XLE membranes) for a
range of commercially available RO membranes using water
uptake measurements, spectroscopic ellipsometry, and both
plane-view and cross-sectional TEM image analysis (23–25).
While cross-sectional and plane-view TEM micrographs accurately
display the complex morphology of polyamide films, differentiating
between closed voids and permeable area open to the top surface
is difficult due to the finite thickness of any TEM sample. De-
termining the permeable area open to the surface is important
because of its potential influence on membrane transport prop-
erties. Although TEM tomography results suggest the presence
of closed voids within polyamide films, little quantitative infor-
mation is available to enable comparisons across multiple RO
membrane materials (1, 8, 36–38).
Imaging thick polymer films (>100 nm) using TEM tomog-
raphy at high angles may cause variations in defocus and under/
overexposure artifacts across the image, which increase the dif-
ficulties in posttreatment of data (39–41). More importantly,
quantitative analyses regarding the internal structure were not
readily available from these studies due to challenges with the
application of TEM tomography to these low-electron-density
polymeric materials and image analysis. Scanning TEM
(STEM) imaging, which scans a focused probe across a sample,
has the ability to perform a dynamic focus that can account for
variations in defocus at high tilt angles for thick samples (42).
When employed in high-angle annular dark-field (HAADF) Fig. 1. Electron tomography of isolated polyamide films to reveal 3D mi-
mode, HAADF-STEM imaging relies on contrast from differ- crostructure. (A) Schematic of polyamide isolation protocol. (B) AFM micro-
graphs of the polyamide top surface of the pristine and isolated SWHR-C
ences in mass, thickness, or atomic number (Z-contrast) and is
membrane after cross-flow compression showing minimal surface changes
less prone to image artifacts (42, 43). HAADF-STEM is thus during the polyamide isolation process. (C) Schematic of the HAADF-STEM
ideal to resolve vacuum, carbon-based samples, and gold fidu- tomography tilt series acquisition. Two-dimensional projection images are
ciary markers while mitigating under- and overexposure prob- acquired from ±65°. (D) Volume rendering of the SWHR-C membrane.
lems and improving image quality.
In this work, we applied a combination of cross-flow filtration
for sample compression, a versatile polyamide active layer iso- and after polyamide isolation was characterized by AFM. Minimal
lation method, and HAADF-STEM tomography to reconstruct surface changes in terms of root-mean-squared roughness and sur-
3D models of fully aromatic isolated polyamide films from face area are observed. This is contrary to uncompacted membranes,
commercial RO membranes. Surface characterization of pre- and where the roughness and surface area decrease by ∼50% for both
postisolated films suggests minimal perturbation of the polyamide commercial SWHR and BWXLE membranes after isolation (SI
films during sample preparation. Quantitative analyses of 3D recon- Appendix, Figs. S3 and S4 and Tables S2 and S3). We attribute
structions yield structural parameters such as void volume percentage, these changes in roughness and surface area to collapse of the
surface area, and depth profiles of the polymer density in a manner structure in uncompressed membranes due to the isolation
not otherwise possible. Our approach demonstrates the need for procedure. Thus, compressing all membranes before character-
characterizing the 3D morphology to extract crucial parameters ization simulates conditions and the morphology present during
required for developing transport models connecting the micro- module operation.
structure with membrane performance. To demonstrate the variations between dry and hydrated

ENVIRONMENTAL
polyamide films, we use in situ ellipsometry to determine changes

SCIENCES
Results in thickness of isolated polyamide films when the relative humidity
Polyamide Isolation Process. Fig. 1A shows a schematic of the poly- (RH) is changed from 0 RH to 88% RH. We find that the swollen,
amide isolation process; details can be found in SI Appendix. Briefly, hydrated thickness of the SWHR-C and BWXLE-C membranes
we spin-coat poly(3,4-ethylenedioxythiophene)-poly(styrenesulfonate) increases by 23% and 10% from their dry-state morphologies,
as a sacrificial layer onto a silicon substrate (44). The RO membrane respectively (SI Appendix, Fig. S5). Thus, we focus on character-
top surface is applied to the substrate and organic solvents dissolve ization of dry membranes, as our results suggest significant mor-
the support layers. After drying in a vacuum oven (SI Appendix, Fig. phological changes for hydrated membranes are not expected.
S1), the sacrificial layer is dissolved in water and the polyamide film
is removed. Thus, the use of a sacrificial layer enables character- HAADF-STEM Tomography and 3D Model Quantification. We obtained
ization of the top surface of isolated polyamide films. a tilt series by the acquisition of HAADF-STEM images at var-
We compress membranes in a cross-flow system at 150 psi for ious tilt angles (Fig. 1C) up to ±65° (SI Appendix, Figs. S6 and
12 h to simulate operating conditions in RO modules. SI Ap- S7). Tilt series of the polyamide film were aligned and the 3D
pendix, Fig. S2 and Table S1 show that membrane thicknesses volume was reconstructed using FEI Inspect3D software. The
decrease by 25% and 8% due to compression for brackish water reconstructed grayscale 3D model was segmented using Avizo 3D
(BWXLE) and seawater high-rejection (SWHR) membranes, software to obtain a binary volumetric model and visualize the
respectively. We denote membranes that are compressed before polyamide film (Fig. 1D). Video reconstructions as a function of
analysis with a “-C” after the membrane abbreviation. We propose z-height for each tomogram are shown in Movies S1 and S2. The
that examining the effects of the isolation procedure is important reconstruction is segmented through thresholding individual XY
to ensure minimal introduction of artifacts. The polyamide top planes and is therefore dependent on the grayscale intensity of
surface of a compressed membrane (SWHR-C, Fig. 1B) before each voxel (Fig. 2A). Gold nanoparticles, applied to the surface of

Culp et al. PNAS | August 28, 2018 | vol. 115 | no. 35 | 8695
the membrane before tomography, serve as fiduciary markers and
permit high-quality volume reconstructions.
The segmented binary volumetric model separates the selected
tomogram space into three parts: polymer film, closed voids within
the film, and vacuum (in the microscope). Fig. 2A shows an ex-
ample of the resulting 3D reconstruction and corresponding
thresholding for SWHR-C. This segmentation enables analyses of
the 3D volume in terms of multiple microstructural parameters; in
this work, we focus on the closed void fraction (voids within the
membranes), overall surface area, and void fraction along the
depth of the film.
Previous work based on cross-sectional TEM images of 60- to
100-nm-thick ultramicrotomed sections suggests the presence of Fig. 3. Quantitative analyses of 3D reconstructions are insensitive to
closed voids throughout polyamide active layers (25, 45). These thresholding procedure. (A) Single-slice cross-sections of the SWHR-C to-
features are subject to the limitations imposed by the projection mogram for five intensity values used for thresholding (1–5 from top to
bottom; 5 is most aggressive while 1 is least aggressive). (B) Porosity as a
of the structure onto 2D images and overlapping features. From
function of film thickness (Z-height) and thresholding intensity value de-
our 3D reconstructions, we can select regions where a close pore termined from polymer areal fraction at each individual XY plane of the
is apparent in the cross-sectional view, such as in Fig. 2B. Fig. 2C SWHRC-tomogram. The polyamide thickness is ∼170 nm. (C) Cross-section
shows a series of images along the y axis (in XZ plane) of three TEM micrograph of a ∼70-nm-thick SWHR-C membrane section obtained
1-nm-thick cross-sectional slices with a spacing distance of 4 nm from ultramicrotomy. The rough, top surface of the polyamide active layer is
based on the region identified in Fig. 2B. The selected region apparent and the underlying layer is the polysulfone support layer.
(left slice) indicates a closed void. The visualization of the same
region a 4-nm step further along the y axis (middle slice) reveals
this enclosed “pore” becomes larger, and another 4-nm extension here and thus seems to have little significance on membrane
(right slice) shows that the pore opens to the top surface, indicating transport properties.
it contributes to the surface area of the polyamide film. Further- We calculate the top surface area of each segmented model
more, the polymer volume completely separates vacuum on either and normalize it by the projected area. The normalized surface
side, indicating that there are no observable pores at the current areas for the SWHR-C and BWXLE-C tomograms are 3.48 ± 0.15
resolution (1–5 nm depending on local variations in polyamide and 3.07 ± 0.04, respectively, indicating that the unevenness of the
thickness) that percolate through the thickness of the polyamide surface contributes about three times more surface area
membrane (46). per unit projected area. In contrast, values measured from AFM
If we define the void fraction as the ratio of the volume of the are about a factor of two lower, only 1.63 ± 0.08 for SWHR-C
closed voids to the total polymer volume, we can compute this and 1.27 ± 0.05 for BWXLE-C membranes (SI Appendix, Tables
measure of the closed porosity for our entire 3D reconstructions. S2 and S3); AFM values agree with previously reported values (18,
We find the closed void fraction is 0.12% and 0.04% for SWHR- 21, 49). A possible explanation for the reduced surface area by
C and BWXLE-C membranes, respectively. These results are in AFM measurements is that an AFM tip cannot reach all of the
contrast to, and more than two orders of magnitude smaller than, accessible surface area due to the complex, or “leaf-like,” folding
recent reported values (25). We speculate that overestimation of polyamide morphology formed during interfacial polymerization,
void fractions in the literature are from artifacts due to the as highlighted in SI Appendix, Fig. S8 (50, 51). The larger surface
thickness of the 2D projection cross-sectional images or from the area from 3D reconstructions compared with AFM images is
use of pristine instead of compressed membranes. A near-zero consistent with previous work where nanoparticle filtration and
3D SEM data suggest that top surface nodules contribute to the
void fraction of the SWHR-C and BWXLE-C membranes sug-
permeable area (8).
gests many features that were interpreted as closed voids in
Analyses along the z axis yields polymer areal fraction as a
previous studies (1, 23, 25, 45, 47, 48) are actually accessible to
function of depth and thus membrane thickness can be defined by
the top surface and thus contribute to overall filtration surface
this profile. First, we define a “low-porosity” polyamide region as
area. Further, the small amount of internal voids appears to have
when the porosity, denoted as 1-polymer fraction (by volume), is
little variance between commercial RO membranes examined
less than 5% for a given z value. Thus, the first z-slice on the
bottom of the tomogram is defined as z = 0 nm, and we evaluate
the porosity as a function of height within the polymer film (Fig.
3C and SI Appendix, Fig. S9), where a porosity of 1 indicates all
vacuum and 0 indicates all polymer. An initial decrease from 1 to
0 represents the roughness on the polyamide bottom surface (near
the polysulfone interface), and the increase back to 1 is represen-
tative of the polyamide roughness on the top surface. The porosity
profiles indicate a portion of the polyamide film that is completely
free of voids (within the tomographic resolution of about 5 nm)
near the bottom polyamide surface, confirming the results from
other studies (23, 36, 52, 53). By our definition of a low-porosity
region, we can quantify the low-porosity polymer layer thickness for
each polyamide film (SI Appendix, Tables S4 and S5). The low-
Fig. 2. Analyses of 3D reconstructions reveal that most pores in cross- porosity layer thicknesses are 21 nm and 62 nm for the SWHR-C
sectional images are connected (open) to the top surface of the mem-
and BWXLE-C membranes, respectively, and are consistent with
brane. (A, Left) Single slices of the XY (Top), XZ (Middle), and YZ (Bottom)
planes. (A, Right) Corresponding thresholding needed for segmentation
apparent dense layers observed in previous electron micrographs
procedure. (B) Single-slice cross-section of SWHR-C tomogram near the tilt of cross-sections (23, 24). As discussed below, this definition of
axis (XZ). (C) Zoomed-in region of the boxed area in B with subsequent slices a low-porosity layer does not address the possibility of varying
4 nm in depth along the y axis showing that the pore visible in B is open to density within polymer regions. The 1-polymer fraction profiles
the top surface 8 nm further into the film. can also be used to estimate the maximum thickness of the

8696 | www.pnas.org/cgi/doi/10.1073/pnas.1804708115 Culp et al.


Evaluating Tomogram Reproducibility. The choice of thresholding
values is critical and can have large impact on the segmented
binary models and the final obtained quantitative information
from reconstructed 3D tomograms. Previous work has explored
the thresholding dependence on structural parameters of 3D
reconstructions and in some cases even used an algorithm to
determine the appropriate thresholding (55–57). Nevertheless,
ambiguity could exist in accurately identifying the boundary be-
tween polyamide and vacuum or voids given the complex struc-
ture and modest contrast. As such, we propose to systematically
vary the intensity where thresholding takes place. The different
thresholding procedures create a set of segmented models for a
given tomogram. We controlled the thresholding starting from a
conservative value that retains more of the polymer fraction
(model 1) to a limit where the polymer layer was thin and broken
through (model 5); details of the thresholding procedure can be
found in Materials and Methods. From segmented models 1
through 5, we observe a decreasing polyamide film thickness in
1-nm-thick XZ planes of the segmented models of the SWHR-C
tomogram (Fig. 3A). To verify our segmented models accurately
represent the morphology of polyamide thin films, we compare
with a cross-sectional TEM image of the identical membrane
Fig. 4. (A) XZ planes of the thresholded volume of SWHR-C and BWXLE-C (Fig. 3B). SI Appendix, Fig. S10 displays the results of a similar
tomograms with intensity displayed as a heat map (red corresponds to higher
analysis performed for the BWXLE-C tomogram.
intensity). (B) Average voxel intensity from tomograms as a function of z-
height for the SWHR-C and BWXLE-C membranes. Each data point repre-
We further investigate the dependence of void fraction, surface
sents the average intensity of each voxel in a given XY plane only where area, low-porosity layer thickness, and porosity profile along the
polyamide is located (i.e., vacuum contributions are neglected). Contributions z direction on the subsequent segmented binary models using dif-
from gold fiduciary markers have been removed. The bottom of the tomo- ferent thresholding values. The systematic thresholding variations
gram (below the polyamide bottom surface) is defined as z = 0. The poly- show only slight variations in low-porosity layer thickness and porosity
amide top surface is at approximately z = 280 nm for both membranes. profiles that is due to the influence of thresholding on the overall
polyamide film thickness (Fig. 3C and SI Appendix, Fig. S9 and
Tables S4 and S5). Other parameters such as the void fraction and
polyamide film, or when the polymer fraction goes to zero. The
surface area show negligible dependence on tomogram thresholding,
thicknesses of the SWHR-C and BWXLE-C films are about 170 nm
indicating that our thresholding procedure accurately represents the
and about 190 nm, consistent with previous reports (26). Moreover, structure of the polyamide film (SI Appendix, Tables S4 and S5).
these results present a different way of envisioning RO membranes; Tomogram reproducibility was analyzed by examining the var-
instead of a void-filled, 200-nm-thick active layer, the polyamide thin iability of the polyamide surface area of the SWHR-C tomogram
film may be a void-absent film of 60 nm or less with a surface area (SI Appendix, Table S6). We progressively separate each tomo-
more than double what was previously measured (18, 21). gram into smaller tomograms with equivalent projection areas and
HAADF-STEM image intensities are directly proportional to examine the resulting variability in surface area. As expected, the
the density of the sample at a given position, enabling a local standard deviation of the surface area increases as the subdivision
analysis of polymer density within membranes. Thus, if we ex- area is reduced due to the higher variability in the structure.
amine a cross-section of the tomogram reconstruction such that Nevertheless, when the subdivision area is 3.4 μm2 (four subdivi-
only the polymer regions are visible (obtained via thresholding), sions) or 1.5 μm2 (nine subdivisions), the standard deviations are
we can examine local variations in density within polyamide thin nearly equivalent, suggesting a 1.5-μm2 projected area is sufficient
films. As shown in Fig. 4A and Movies S3 and S4, significant to statistically capture the heterogeneity in the microstructure; our

ENVIRONMENTAL
heterogeneity in polymer density is present throughout both

SCIENCES
SWHR-C and BWXLE-C active layers. As shown in Fig. 4B, in-
tegrating the tomogram intensities for only the polymer regions
leads to a polymer density profile as a function of film thickness.
For both the SWHR-C and BWXLE-C membranes, polymer
density is highest near the top surface, and in particular within
protruding features that contribute to the surface roughness, or
“polyps.” The profiles shown in Fig. 4B also demonstrate that the
bottom surface is on average of higher density compared with the
middle of the film. As a consequence, we speculate that limiting
barrier layers of two high-performance, commercial RO mem-
branes are near the polyamide top surface, although bottom layers
may contribute as well. Indeed, the larger increase in density
near the top surface for SWHR-C membranes compared with
BWXLE-C membranes suggests a larger barrier layer, which is Fig. 5. Schematics of models for location of layers with high polymer
consistent with the lower permeability of SWHR membranes (30, density in polyamide thin films. (A) If a high-density layer is at the poly-
sulfone interface, then the projected area Am is crucial to establish transport
54). The location of regions of high polymer density may differ for
properties. (B) Layer with high polymer density in the middle of the mem-
other commercial and experimental membranes, and we propose brane. Again the projected area is the most accurate representation of
that the density analysis developed here is needed to pinpoint membrane area. (C) A layer of high polymer density located at the water
crucial transport elements and barriers. interface, where now the surface area is needed to obtain flow properties.

Culp et al. PNAS | August 28, 2018 | vol. 115 | no. 35 | 8697
projected area of our 3D reconstruction is nearly 10 times larger properties should be analyzed and predicted based on the actual
(13.4 μm2). surface area as determined from techniques such as tomography.
Given that this area is much larger than the projected area, reported
Discussion flux values are currently significantly overestimated. Additionally,
The production rate of water from an RO membrane (Qw) is a the length scale for resistance in transport models such as that
function of both the area of the membrane, Am, as well as the represented in Eqs. 1 and 2 will likely correlate with the average
membrane thickness, L, as described by widely used solution thickness of the high-density layer adjacent to the top surface of the
diffusion transport models that simplify to Eqs. 1 and 2 for most membrane and could be determined using techniques described in
RO applications [see, e.g., Geise et al. (58)]: this study. Qualitatively, we show that the higher increase in polymer
density for SWHR-C membranes near the top surface is consistent
Pw M w with lower permeability compared with BWXLE-C membranes.
Qw = Jw Am = Am ðΔP − ΔπÞ [1]
L RT In summary, we demonstrate that compression of RO mem-
branes via cross-flow filtration before characterization is both more
Pw Mw representative of the membrane structure under operational con-
permeance = A = , [2]
L RT ditions and minimizes structural rearrangement during solvent-
based active layer isolation procedures. We present an active layer
where Jw is the water production rate per unit area, Pw is the pure isolation procedure that enables the confirmation of minimal
water permeability of the membrane, Mw is the molecular weight polyamide structural rearrangements of compressed membranes
of water, R is the gas constant, T is temperature in degrees using AFM. We establish that HAADF-STEM tomography is an
kelvin, ΔP is the pressure drop across the membrane, and the excellent tool to both visualize and quantify the internal nano-
Δπ is the osmotic difference across the membrane. structure of polyamide films. We find that closed void fractions are
Despite the apparent simplicity of Eqs. 1 and 2, the challenge
0.12% and 0.04% for the SWHR-C and BWXLE-C membranes,
with applying these equations is obtaining values for Am and L
respectively, and are more than two orders of magnitude lower
for commercial RO membranes. Thus, the permeance A, com-
than previously reported values. We also find that the surface area
monly referred to as the “A value,” is currently used to characterize
of the polyamide top surface is double what has been conven-
membrane performance; but, this lumps multiple parameters to-
gether and prohibits the development of structure–function rela- tionally measured with AFM, due to the increase in accessible area
tionships. In this work, we demonstrate that the surface area of from presence of voids connected to the top surface. We extract a
membranes is significantly higher than the projected area, and for polymer region on the polyamide bottom surface with a porosity of
any given point on the membrane the effective thickness varies less than 5%, which can be calculated by the porosity curves from
significantly at the nanoscale. Our work presents a method to the tomograms. The low-porosity layer thicknesses for the SWHR-C
identify and describe the relevant dense-layer thickness of RO and BWXLE-C membranes were 21 nm and 62 nm, respectively,
membranes using quantitative analysis of RO tomography data. and agree with previous studies; nevertheless, we propose that
We propose that either the projected area or the actual surface these values are not the key step in connecting structure to
area (that we determine using tomography) should be used to transport. Instead, we propose that polymer density and its dis-
model flux depending on the distribution and location of the actual tribution dictate flow, wherein membranes that show highest
dense layer (that provides the highest resistance to flow) of the polyamide density near the top surface as shown here will require
membrane (Fig. 5). If a dense layer is identified close to the in- measurements of the actual surface area to determine relevant
terface between the polyamide active layer and polysulfone support flux values. Altogether, using RO membranes as an example, our
layer, as evidenced by high polymer density that is separated from results demonstrate the importance of 3D characterization at the
the top rough surface (as depicted in Fig. 5 A and B), we hypothesize nanoscale to elucidate fundamental underpinnings of membrane
that the use of projected area to model transport is appropriate properties. We propose that local variations in porosity, density,
because the top layers would then provide little resistance to and surface area will lead to heterogeneity in flux within membranes,
flow compared with the layer of high density. If, however, there such that connecting chemistry, microstructure, and performance of
is evidence that a barrier layer could be adjacent to the top membranes for RO, ultrafiltration, virus and protein filtration, and
surface area (Fig. 5C), as based on recent nanoparticle filtration gas separations will require 3D reconstructions from techniques such
experiments (32), then the actual surface area would be more as electron tomography.
relevant to determine transport correlations. If the density of the
membrane is homogeneous, then membrane properties must be Materials and Methods
defined locally (at the nanoscale) to account for different Materials and methods for membrane compression in a cross-flow device,
thicknesses. We propose that detailed tomographic studies on a polyamide film isolation, RH ellipsometry data acquisition, AFM data acqui-
wide range of membranes, possibly with higher resolution than sition, cross-section image acquisition, and electron tomography acquisition
those achieved in this work, will be crucial to unequivocally and analysis can be found in SI Appendix.
establish the location of the dense layer and enable modeling of
transport properties as well as development of robust structure– ACKNOWLEDGMENTS. The authors thank Prof. Andrew Zydney and Robert
function relationships. Cieslinski for educational discussions and Jennifer Grey for assistance with
tomography data acquisition and software support. This work was supported
In the case of commercial RO membranes examined here, we by Dow Chemical Company Agreement 225559AK/177526 and NSF Grant
hypothesize that the higher polymer density near the top surface DMR-1609417 (to T.E.C. and E.D.G.). Flat-sheet membranes were provided
(Fig. 4B) establishes a dense layer and suggest that membrane courtesy of Dow Water & Process Solutions.

1. Pacheco F, Sougrat R, Reinhard M, Leckie JO, Pinnau I (2016) 3D visualization of the 4. Nazem-Bokaee H, et al. (2018) Probing pore structure of virus filters using scanning
internal nanostructure of polyamide thin films in RO membranes. J Membr Sci 501: electron microscopy with gold nanoparticles. J Membr Sci 552:144–152.
33–44. 5. Wang X, et al. (2017) Reversed thermo-switchable molecular sieving membranes composed
2. Zhou C, et al. (2017) Fabrication of nanoporous alumina ultrafiltration membrane of two-dimensional metal-organic nanosheets for gas separation. Nat Commun 8:14460.
with tunable pore size using block copolymer templates. Adv Funct Mater 27: 6. Bellona C, Drewes JE, Xu P, Amy G (2004) Factors affecting the rejection of organic
1701756. solutes during NF/RO treatment–A literature review. Water Res 38:2795–2809.
3. Choi H, Sofranko AC, Dionysiou DD (2006) Nanocrystalline TiO2 photocatalytic 7. Ganiyu SO, van Hullebusch ED, Cretin M, Esposito G, Oturan MA (2015) Coupling of
membranes with a hierarchical mesoporous multilayer structure: Synthesis, charac- membrane filtration and advanced oxidation processes for removal of pharmaceuti-
terization, and multifunction. Adv Funct Mater 16:1067–1074. cal residues: A critical review. Sep Purif Techol 156:891–914.

8698 | www.pnas.org/cgi/doi/10.1073/pnas.1804708115 Culp et al.


8. Kłosowski MM, et al. (2016) Micro-to nano-scale characterisation of polyamide 33. Melián-Martel N, Sadhwani J, Malamis S, Ochsenkühn-Petropoulou M (2012) Struc-
structures of the SW30HR RO membrane using advanced electron microscopy and tural and chemical characterization of long-term reverse osmosis membrane fouling
stain tracers. J Membr Sci 520:465–476. in a full scale desalination plant. Desalination 305:44–53.
9. Greenlee LF, Lawler DF, Freeman BD, Marrot B, Moulin P (2009) Reverse osmosis 34. Freger V (2003) Nanoscale heterogeneity of polyamide membranes formed by in-
desalination: Water sources, technology, and today’s challenges. Water Res 43: terfacial polymerization. Langmuir 19:4791–4797.
2317–2348. 35. Tang CY, Kwon Y-N, Leckie JO (2007) Characterization of humic acid fouled reverse
10. Jacangelo JG, Rhodes Trussell R, Watson M (1997) Role of membrane technology in osmosis and nanofiltration membranes by transmission electron microscopy and
drinking water treatment in the United States. Desalination 113:119–127. streaming potential measurements. Environ Sci Technol 41:942–949.
11. Petersen RJ (1993) Composite reverse osmosis and nanofiltration membranes. 36. Kurihara M, Hanakawa M (2013) Mega-ton water system: Japanese national research
J Membr Sci 83:81–150. and development project on seawater desalination and wastewater reclamation.
12. Wang J, et al. (2014) A critical review of transport through osmotic membranes. Desalination 308:131–137.
J Membr Sci 454:516–537. 37. She F, et al. (2010) 3-Dimensional characterization of membrane with nanoporous
13. Elimelech M, Phillip WA (2011) The future of seawater desalination: Energy, tech- structure using TEM tomography and image analysis. Desalination 250:757–761.
nology, and the environment. Science 333:712–717. 38. Nunes SP, et al. (2011) Switchable pH-responsive polymeric membranes prepared via
14. Lee KP, Arnot TC, Mattia D (2011) A review of reverse osmosis membrane materials block copolymer micelle assembly. ACS Nano 5:3516–3522.
for desalination—Development to date and future potential. J Membr Sci 370:1–22. 39. Hart RG (1968) Electron microscopy of unstained biological material: The polytropic
15. Cahill DG, Freger V, Kwak S-Y (2008) Microscopy and microanalysis of reverse-osmosis montage. Science 159:1464–1467.
and nanofiltration membranes. MRS Bull 33:27–32. 40. De Rosier DJ, Klug A (1968) Reconstruction of three dimensional structures from
16. Morgan PW, Kwolek SL (1959) Interfacial polycondensation. II. Fundamentals of electron micrographs. Nature 217:130–134.
polymer formation at liquid interfaces. J Polym Sci 40:299–327. 41. Hoppe W, Gassmann J, Hunsmann N, Schramm HJ, Sturm M (1974) Three-dimensional
17. Hu Y, et al. (2016) Enhancing the performance of aromatic polyamide reverse osmosis reconstruction of individual negatively stained yeast fatty-acid synthetase molecules
membrane by surface modification via covalent attachment of polyvinyl alcohol from tilt series in the electron microscope. Hoppe Seylers Z Physiol Chem 355:
(PVA). J Membr Sci 501:209–219. 1483–1487.
18. Kwak S-Y, Ihm DW (1999) Use of atomic force microscopy and solid-state NMR 42. Kübel C, et al. (2005) Recent advances in electron tomography: TEM and HAADF-
spectroscopy to characterize structure-property-performance correlation in high-flux STEM tomography for materials science and semiconductor applications. Microsc
reverse osmosis (RO) membranes. J Membr Sci 158:143–153. Microanal 11:378–400.
19. Petersen RJ, Cadotte JE (1995) Thin film composite reverse osmosis membranes. 43. Ercius P, Alaidi O, Rames MJ, Ren G (2015) Electron tomography: A three-dimensional
analytic tool for hard and soft materials research. Adv Mater 27:5638–5663.
Handbook of Industrial Membrane Technology, ed Porter MC (Noyes Publications,
44. Kozub DR, et al. (2011) Polymer crystallization of partially miscible polythiophene/
Park Ridge, NJ), pp 307–348.
fullerene mixtures controls morphology. Macromolecules 44:5722–5726.
20. Cadotte JE (1985) Evolution of composite reverse osmosis membranes. Materials
45. Ma X-H, et al. (2018) Nanofoaming of polyamide desalination membranes to tune
Science of Synthetic Membranes, ACS Symposium Series (Am Chemical Soc, Wash-
permeability and selectivity. Environ Sci Technol Lett 5:123–130.
ington, DC), Vol 269, pp 273–294.
46. Heidari Mezerji H, Van den Broek W, Bals S (2011) A practical method to determine the
21. Kwak SY, Jung SG, Yoon YS, Ihm DW (1999) Details of surface features in aromatic
effective resolution in incoherent experimental electron tomography. Ultramicroscopy
polyamide reverse osmosis membranes characterized by scanning electron and
111:330–336.
atomic force microscopy. J Polym Sci B Polym Phys 37:1429–1440.
47. Kolev V, Freger V (2014) Hydration, porosity and water dynamics in the polyamide
22. Karan S, Jiang Z, Livingston AG (2015) MEMBRANE FILTRATION. Sub-10 nm polyamide
layer of reverse osmosis membranes: A molecular dynamics study. Polymer (Guildf) 55:
nanofilms with ultrafast solvent transport for molecular separation. Science 348:
1420–1426.
1347–1351.
48. Xu J, Yan H, Zhang Y, Pan G, Liu Y (2017) The morphology of fully-aromatic poly-
23. Yan H, et al. (2015) The porous structure of the fully-aromatic polyamide film in re-
amide separation layer and its relationship with separation performance of TFC
verse osmosis membranes. J Membr Sci 475:504–510.
membranes. J Membr Sci 541:174–188.
24. Pacheco FA, Pinnau I, Reinhard M, Leckie JO (2010) Characterization of isolated
49. Kwak S-Y, Jung SG, Kim SH (2001) Structure-motion-performance relationship of flux-
polyamide thin films of RO and NF membranes using novel TEM techniques. J Membr
enhanced reverse osmosis (RO) membranes composed of aromatic polyamide thin
Sci 358:51–59.
films. Environ Sci Technol 35:4334–4340.
25. Lin L, Lopez R, Ramon GZ, Coronell O (2016) Investigating the void structure of the
50. Ghosh AK, Jeong B-H, Huang X, Hoek EM (2008) Impacts of reaction and curing
polyamide active layers of thin-film composite membranes. J Membr Sci 497:365–376. conditions on polyamide composite reverse osmosis membrane properties. J Membr
26. Lin L, Feng C, Lopez R, Coronell O (2016) Identifying facile and accurate methods to Sci 311:34–45.
measure the thickness of the active layers of thin-film composite membranes–A 51. Zou H, et al. (2010) Synthesis and characterization of thin film composite reverse
comparison of seven characterization techniques. J Membr Sci 498:167–179. osmosis membranes via novel interfacial polymerization approach. Sep Purif Technol
27. Tang CY, Kwon Y-N, Leckie JO (2007) Probing the nano-and micro-scales of reverse 72:256–262.
osmosis membranes—A comprehensive characterization of physiochemical properties 52. Tsuru T, et al. (2013) Multilayered polyamide membranes by spray-assisted 2-step
of uncoated and coated membranes by XPS, TEM, ATR-FTIR, and streaming potential interfacial polymerization for increased performance of trimesoyl chloride (TMC)/m-
measurements. J Membr Sci 287:146–156. phenylenediamine (MPD)-derived polyamide membranes. J Membr Sci 446:504–512.
28. Hirose M, Ito H, Kamiyama Y (1996) Effect of skin layer surface structures on the flux 53. Kamada T, Ohara T, Shintani T, Tsuru T (2014) Optimizing the preparation of multi-
behaviour of RO membranes. J Membr Sci 121:209–215. layered polyamide membrane via the addition of a co-solvent. J Membr Sci 453:
29. Tang CY, Kwon Y-N, Leckie JO (2009) Effect of membrane chemistry and coating layer 489–497.
on physiochemical properties of thin film composite polyamide RO and NF mem- 54. Combernoux N, et al. (2017) Treatment of radioactive liquid effluents by reverse
branes: I. FTIR and XPS characterization of polyamide and coating layer chemistry. osmosis membranes: From lab-scale to pilot-scale. Water Res 123:311–320.

ENVIRONMENTAL
Desalination 242:149–167. 55. Ostadi H, Jiang K, Prewett P (2008) Micro/nano X-ray tomography reconstruction fine-
30. Tang CY, Kwon Y-N, Leckie JO (2009) Effect of membrane chemistry and coating layer tuning using scanning electron microscope images. Micro Nano Lett 3:106–109.

SCIENCES
on physiochemical properties of thin film composite polyamide RO and NF mem- 56. Goris B, Roelandts T, Batenburg KJ, Heidari Mezerji H, Bals S (2013) Advanced re-
branes: II. Membrane physiochemical properties and their dependence on polyamide construction algorithms for electron tomography: From comparison to combination.
and coating layers. Desalination 242:168–182. Ultramicroscopy 127:40–47.
31. Mitchell GE, Mickols B, Hernandez-Cruz D, Hitchcock A (2011) Unexpected new phase 57. Ostadi H, et al. (2010) Influence of threshold variation on determining the properties
detected in FT30 type reverse osmosis membranes using scanning transmission X-ray of a polymer electrolyte fuel cell gas diffusion layer in X-ray nano-tomography. Chem
microscopy. Polymer (Guildf) 52:3956–3962. Eng Sci 65:2213–2217.
32. Li Y, et al. (2017) Probing flow activity in polyamide layer of reverse osmosis mem- 58. Geise GM, Park HB, Sagle AC, Freeman BD, McGrath JE (2011) Water permeability and
brane with nanoparticle tracers. J Membr Sci 534:9–17. water/salt selectivity tradeoff in polymers for desalination. J Membr Sci 369:130–138.

Culp et al. PNAS | August 28, 2018 | vol. 115 | no. 35 | 8699

Potrebbero piacerti anche