Influence of casting substrate on bulk morphology and vanadium ion transport in ionomer nanocomposites

Perfluorosulfonic acid (PFSA) ionomers, particularly Nafion^™,¹ remain the canonical standard for polymer electrolyte membranes (PEMs) utilized in vanadium redox flow batteries (VRFBs).^2,3 When fully hydrated, the hydrophilic sulfonic acid groups phase segregate from the hydrophobic fluorinated backbone, forming a percolated ionic network through which facile transport of protons occurs.^4–6 However, one standing issue with the utilization of PFSA ionomers in VRFBs is the low ion selectivity of these membranes, resulting in deleteriously high vanadium ion crossover, which reduces the lifetime and efficiency of the battery.^7,8 To combat this poor ion selectivity, researchers have borrowed a page from previous work related to the optimization of PFSA membranes for use in direct methanol fuel cells, where composite PFSA materials demonstrated increased selectivity (i.e., reduced methanol crossover).^9–11 Specifically, the incorporation of silica nanoparticles (SiNPs) into PFSA membranes has garnered interest as a promising approach to address this ion crossover issue.^4,12–15

In the literature, pristine Nafion (i.e., no NPs) and Nafion nanocomposite membranes have been fabricated using a wide variety of conditions, where oftentimes, the parameters necessary to repeat the fabrication procedure are ambiguous or completely missing altogether. This is of particular importance for composite membranes utilizing PFSA ionomers as the mechanical and transport properties of these ionomers are known to be highly dependent on the nanostructure of the membrane,² which can be varied through various processing and environmental conditions. Notably, thermal processing of both extruded PFSA ionomers, such as Nafion 117, or solution-cast membranes, also known as “recast” membranes, has been widely investigated.^16–19 Thermal annealing of the PFSA ionomers above their glass transition temperature leads to a higher degree of crystallinity, which can lead to a reordering of the ionic domains during membrane hydration, and long-range ordering of crystalline regions within the polymers.^19,20 This reordering of the ionic domains has been shown to reduce the solubility in polar solvents, while long-range ordering of the crystallites leads to increased durability of the PFSA membranes.^19,21 Thermal annealing of PFSA ionomers has also been shown to affect water uptake, proton conductivity, and vanadium ion permeability, thereby leading to changes in the overall selectivity of the ionomer.^22–24 

Furthermore, the chemical composition of the ionomer membrane surface has been correlated with differences in water uptake between liquid and vapor phases, known as Schroeder's Paradox,²⁶ where increasing the degree of hydrophobicity of the membrane surface leads to reduced water uptake when the membrane is challenged with water vapor vs liquid water.²⁶ Additionally, the specific water activity that the PFSA membrane is challenged with has been shown to augment both the local and bulk ionic conductivities of Nafion 212 membranes.²⁷ This was attributed to the relationship between mass transfer resistance and fraction of conductive surface area, i.e., sulfonic acid sites on the surface of the membrane, where a high concentration of these sites facilitates transport of water from the surface into the bulk membrane, which, in turn, results in accelerated bulk ionic conductivity.²⁸ It is clear from these previous investigations that the chemical composition of the ionomer surface can directly impact transport mechanisms that govern the bulk membrane behavior.² However, to date, these observations have predominantly been limited to extruded Nafion membranes (e.g., Nafion 117, Nafion 212). As the bulk morphology of these pre-processed films is largely fixed due to the extrusion process used to fabricate free-standing membranes, previous literature related to these membranes may not capture the full spectrum of morphological and surface chemistry variations that may be available to solution-cast Nafion membranes.

In the case of solution-cast PFSA membranes, a wide variety of casting conditions (e.g., evaporation rate, temperature) and casting solvents (e.g., ethanol, dimethyl sulfoxide) have been investigated,^{17,18,29–32} where variations in fabrication conditions have been shown to affect the formation of the ionomer nanostructure, resulting in significant changes in the bulk membrane properties.^33–35 For instance, changes in the ionic conductivity of solution-cast PFSA membranes have been previously observed when the casting solvent was varied between 1-butanol and 2-propanol. Researchers observed higher conductivity, as compared to extruded Nafion, in membranes cast with 1-butanol, while membranes cast with 2-propanol demonstrated similar conductivity to their extruded membrane counterparts.³⁶ 

While the impact of fabrication conditions (e.g., thermal annealing, casting solvent) on PFSA ionomer properties has been widely investigated,^30–33 there is little to no information in the literature regarding the influence of casting substrate on the final properties of bulk (>10 μm) PFSA membranes (e.g., morphology, selectivity). In fact, a number of investigations fail to report the casting substrate used in the fabrication of these bulk PFSA films.^3,35,37 In contrast, the impact of various fabrication conditions, such as casting solvent and substrate, rate of evaporation, and thermal annealing, on the resulting properties of PFSA thin films (∼10–100 nm) has been extensively investigated, where the morphology and transport properties have been shown to vary with changes in thin film fabrication conditions.^38–41 Water sorption of Nafion thin films is known to depend on film thickness,^30,31,35 thermal treatment,³⁰ and choice of casting substrate.^28,29 For example, water uptake in PFSA thin films can be substantially altered by the degree of hydrophilicity or hydrophobicity of the casting substrates, where reduced water uptake has been observed in film’s cast on hydrophobic substrates (e.g., gold, platinum).^29,34 The noted differences in water sorption and ionomer swelling have been attributed to variations in the nanostructure of the thin films due to favorable (or unfavorable) interactions with the casting substrate. In the case of thin films cast on hydrophilic substrates (e.g., SiO₂), under hydration, alternating lamellae of water-rich and Nafion-rich layers at the substrate–ionomer interface were observed, though this alternating lamellar structure only propagated ∼10 nm into the total of ∼50 nm of the thin PFSA film.^32,34 The length scales over which the substrate was observed to impact the local nanostructure of the ionomer are comparable to those length scales previously reported for substrate-induced perturbations of the local glass transition temperature (T_g) in thin glassy films (10–100 nm).^42–46 However, recent work from Baglay and Roth^47,48 have challenged our traditional understanding regarding the length scales over, which these interfacial effects can persist. Specifically, suppressed values of the local T_g of the polystyrene were seen to persist up to ∼400 nm in the film when thin polystyrene films were cast onto poly(n-butylmethacrylate) substrates.^47,48 These recent results underscore the fact that our traditional understanding of substrate-induced structuring of polymers, especially nanophase segregated polymers, may not be complete.

Herein, we investigate the influence of the casting substrate on the nanostructure and vanadium ion transport (specifically, the vanadyl ion, VO²⁺) in PFSA and PFSA nanocomposite membranes. Specifically, dense Nafion and Nafion nanocomposite membranes were formed by solution casting dispersions on top of both polished quartz and polytetrafluoroethylene (PTFE) substrates. In the case of Nafion nanocomposites, membranes containing unfunctionalized (bare silica) and surface-functionalized (sulfonic acid- and amine-functionalized) nanoparticles were fabricated. To gain insight into the impact of the casting substrate on the nanostructure of the ionomer, the bulk morphology of hydrated (in H₂O) PFSA and PFSA nanocomposite membranes was characterized using small-angle neutron scattering (SANS). Furthermore, the chemical composition of the substrate-adjacent membrane surface was determined by scanning electron microscopy (SEM) with energy dispersive x-ray spectroscopy (SEM-EDS). Additionally, the vanadium ion permeability of the ionomer films was measured via ultraviolet–visible spectroscopy. Note, during the VRFB operation, transport of vanadium ions across the PFSA ionomer is caused by both the concentration gradient and the applied electrical field. Several articles investigate the effect of an electrical field, as well as charging and discharging on vanadium ion crossover;^49–52 however, this work only focuses on concentration gradient-driven vanadyl ions (VO²⁺) transport.

Ethanol (pure, 200 proof, anhydrous), sulfuric acid (95% to 98%, ACS Reagent), fuming sulfuric acid (reagent grade), (3-mercaptopropyl)trimethoxysilane (95%), vanadium(IV) oxide sulfate hydrate (97%), magnesium sulfate (anhydrous), Nafion stock dispersion (Nafion DE 2021, w_PFSA = 20% in a mixture of lower aliphatic alcohols and water), and tin (II) chloride (reagent grade, 98%) were purchased from Sigma Aldrich. Methyl isobutyl ketone (certified ACS solvent) was purchased from Fisher Scientific. Glycidyl phenyl ether (99%) was purchased from Acros Organics. Hydrogen peroxide (30%) was purchased from VWR Analytical. Unfunctionalized SiNPs (colloidal silica in methanol; MTST grade; D_P,avg = 11 nm) were obtained from Nissan Nanomaterials. Reverse osmosis (RO) water (resistivity ≈ 18 MΩ cm) was used for all experiments.

The SiNPs were functionalized as previously described.¹⁵ In general, the appropriate reagents were added to a suspension of unfunctionalized SiNPs, allowed to react, then washed, and separated by centrifugation. All membranes were cast from the as-received Nafion dispersion. To incorporate the SiNPs, 5 wt. % of either the functionalized or unfunctionalized SiNPs (mass of SiNPs/total mass of solids × 100%) were suspended in the Nafion dispersion by sonication for at least 30 min prior to casting. Pristine Nafion suspension (i.e., no SiNPs) was also sonicated for at least 30 min prior to casting. The Nafion suspensions were then cast onto either a polished quartz or a polytetrafluoroethylene (PTFE) substrate, covered by a funnel with Kim-wipe flue, and allowed to evaporate overnight on the benchtop. The dried hybrid films were then annealed at 140 °C for 2 h under dynamic vacuum, after which the oven was shut off and dynamic vacuum was pulled for an additional 30 min. After 30 min, the valve to the vacuum pump was closed and the films were left to cool down to room temperature under static vacuum. Prior to beginning measurements, the films were hydrated in RO water overnight before further use. Hydrated film thicknesses were on the order of ∼100 μm.

Vanadium ion crossover was measured as previously described.¹⁵ Briefly, a tailor-made diffusion cell (Permegear Franz cell; Bethlehem, PA; shown as Fig. S1 in the supplementary material) was used. The receiving cell (volume 15 ml) was filled with 1.5 mol l⁻¹ MgSO₄ in 3 mol l⁻¹ H₂SO₄, and the donating cell (volume 1 ml) was filled with 1.5 mol l⁻¹ VOSO₄ in 3 mol l⁻¹ H₂SO₄, and the membrane was sandwiched between the two cells. Unless otherwise noted, the membrane was placed such that the substrate-adjacent surface was facing the vanadium-rich cell (i.e., the donating cell). Aliquots were taken via the sidearm of the receiving cell at regular time intervals, and the concentration of vanadium (IV) ions (VO²⁺) was measured using a ultraviolet–visible (UV–vis) spectrometer (VWR UV-3100PC), which scanned from wavelengths of 1100 nm to 400 nm. The peak associated with the presence of VO²⁺ can clearly be observed around a wavenumber of 760 nm. Immediately following UV–vis characterization, the aliquots were placed back into the receiving cell. From these data, the permeability of vanadium ions can be calculated from the following equation:

where $CD$ and $CR(t)$ are the vanadium ion concentration in the donating and receiving cells (mol l⁻¹), respectively, A and L are the area (cm²) and thickness of the membrane (cm), respectively, P is the permeability of VO²⁺ ions (cm² s⁻¹), and $VR$ is the volume of the receiving cell (l). This expression assumes the following: (1) the permeation in the membrane has reached pseudo-steady state, (2) vanadium ion permeability is independent of ion concentration, (3) $CD≫CR(t)$⁠, and (4) the reduction in $CD$ over the length of the experiment is negligible.⁵³ 

IEC experiments were performed according to the literature.⁵⁴ Briefly, the membrane was dried under vacuum at 80 °C for 24 h, massed, and immersed in 1 mol l⁻¹ NaCl for 24 h. Next, the membrane was removed from the NaCl solution, and the remaining solution was titrated with 0.01 mol l⁻¹ NaOH with phenolphthalein [1% (w/v) in a mixture of 1:1 water:ethanol]. The IEC for each membrane (mmol g⁻¹) was calculated as follows:

where $VNaOH$ is the volume of the titrated NaOH solution (l), $CNaOH$ is the concentration of the NaOH solution (mmol l⁻¹), and $mdry$ is the dry mass of the membrane (g).

Membrane surface topography was obtained using a Hitachi SU-6600 Variable Pressure Scanning Electron Microscope. Dried membrane samples were cut to size, mounted via a double-sided carbon tape to aluminum slabs, and left uncoated for both SEM and SEM-EDS imaging. Images were collected under a chamber pressure of 15 Pa to minimize sample charging. Backscattered electron (BSE) imaging was then performed at 20 kV. SEM-EDS data were obtained under these same parameters, where maps were analyzed with Aztec EDS microanalysis software (Oxford Instruments).

Small-angle neutron scattering (SANS) experiments were performed on the NG-B 10 m SANS instrument (all samples cast on the polished quartz) and the NG-7 30 m SANS instrument (all samples cast on the PTFE substrate) at the National Institute of Standards and Technology Center for Neutron Research (NCNR). Nanocomposite films were cast and annealed as described in Sec. II B. At least 24 h prior to SANS experiments, the membranes were hydrated in liquid H₂O. The hydrated films were then placed in a demountable cell, where the distance between quartz windows was 1 mm. A circular aperture with a diameter of 0.5 cm was utilized for all samples. The incoming neutron wavelength and the sample-to-detector distance were varied to collect a range of Q values $(Q=4πsinθ/λ)$⁠, where $θ$ and $λ$ are the scattering angle and wavelength of the neutrons, respectively. In this study, SANS data were collected over Q values ranging from 0.0035 Å⁻¹ to 0.5 Å⁻¹ for the 10 m SANS instrument and 0.0009 Å⁻¹ to 0.5 Å⁻¹ for the 30 m SANS instrument. The total collection time for each sample was approximately 3 h. The SANS data were reduced using the software package developed at the NIST Center for Neutron Research,⁵⁵ where the thickness of the cell was used for all reduction calculations.

Captive bubble contact angle measurements were performed on hydrated membranes that were equilibrated and immersed in RO water. The contact angle measurements were performed using a Krüss DSA 10 Mk2 goniometer (Krüss GmbH, Hamburg, Germany) with Drop Shape Analysis software (ver. 1.80.0.2, Krüss GmbH, Hamburg, Germany). The images were analyzed using the Captive Bubble plugin for ImageJ (NIH), where the contact angles were measured with the circle best-fit.⁵⁶ 

Table I lists the nomenclature for each of the Nafion-SiNP nanocomposite films, including the casting substrate. For instance, the nomenclature “Naf-UF-T” indicates a Nafion membrane containing 5 wt. % UF-SiNP that was solution cast onto a PTFE substrate. Note that this nomenclature will be utilized for the remainder of the paper. The TA-SiNP and PS-SiNP functionality have been characterized previously.¹⁵ 

Vanadium ion permeation and ion exchange capacity (IEC) measurements were carried out on pristine Nafion (i.e., membranes containing no SiNPs) and Nafion nanocomposite membranes containing 5 wt. % SiNPs, cast on both polished quartz and PTFE substrates. The results of these experiments, for each SiNP surface functionalization, are summarized in Fig. 1. As noted, the vanadium ion permeability and IEC of Naf-Q, Naf-UF-Q, Naf-TA-Q, and Naf-PS-Q have been previously published.^14,15

As seen from Fig. 1, the vanadium ion permeabilities (blue and red bars, left y axis) and IEC values (connected blue and red data points, right y axis) of pristine PFSA and PFSA ionomer nanocomposites vary based on the casting substrate, which to the best of our knowledge, has never been previously demonstrated. It is important to note that substrate effects have been previously reported for PFSA thin films (<100 nm), where substrate effects were observed to propagate 5 to 10 nm into the thin film.^{26,27,38–41,57–59} However, the results reported in this paper are for ionomer membranes with thicknesses ranging from 50 to 70 μm, which is several orders of magnitude larger than the substrate effects previously reported. There are three possible explanations for this substrate-dependent VO²⁺ permeability: (1) the casting substrate has modified the nanostructure of the bulk membrane, thereby altering the diffusivity of vanadyl ions; (2) the casting substrate has caused a chemical reorganization of the membrane surface directly adjacent to the substrate, altering how ions partition into the ionomer; or (3) a combination of (1) and (2). Focusing on the vanadyl ion permeabilities and IECs for PFSA and PFSA ionomer nanocomposites cast on quartz (blue bars and blue data points in Fig. 1, respectively), we observe reduced VO²⁺ permeabilities and IECs for Naf-UF-Q and Naf-TA-Q, when compared to Naf-Q. However, this trend of suppressed permeability with lower IEC does not hold, as the highest VO²⁺ permeability was observed for Naf-PS-Q, which had a lower IEC than Naf-Q. Additionally, all membranes and nanocomposite membranes cast on PTFE substrates showed higher vanadium ion permeabilities but lower IECs compared to Naf-Q. Compared to their counterparts cast on quartz substrates, Naf-UF-T and Naf-TA-T demonstrated vanadium ion permeabilities approximately twice that of Naf-UF-Q and Naf-TA-Q, respectively, though lower or similar IECs were observed for these nanocomposite membranes. As IEC is a measure of the available sulfonic acid groups (i.e., charged groups) in the ionomer, changes to this value allude to the possibility that the casting substrate is perturbing the bulk morphology within the PFSA membrane.

In principle, changes to the IEC of an ionomer can significantly alter vanadyl ion permeability, as it is well known that the presence and concentration of charged groups strongly influence sorption and transport in charged polymers.^60–62 Without considering morphological changes, equilibrium sorption (or portioning) of charged species (i.e., ions) in charged membranes (i.e., ionomers) has traditionally been viewed through the lens of the Donnan theory (or Donnan potential). According to the Donnan theory,^63–66 an increase in the amount of fixed negative charges in the membrane should inhibit the sorption of negatively charged coions (in this case, SO₄²⁻ ions) into the membrane, while simultaneously inhibiting the desorption of positively charged counterions (in this case, VO²⁺ and H⁺ ions, if we only consider the vanadyl-rich solution) from the membrane. However, an increase in the amount of fixed negative charges (i.e., higher IEC) should also lead to an increase in the sorption (or partitioning) of counterions into the membranes, where at low external ion concentrations, the concentration of counterions may be two times of that in the adjacent electrolyte solution.⁶⁷ However, correlating changes in IEC to changes in ion permeability through Donnan theory arguments, may not be valid for these systems, given the high concentration of coions and counterions in the ionomer-adjacent, vanadyl ion solution.

In fact, recent work from Kamcev and co-workers^65–67 suggests that the parameter $CAm,w$⁠, expressed as moles of fixed charge groups per liter of sorbed water (here $A=SO3−$⁠) may be a more accurate representation of the electrostatic environment the diffusing ions experience inside the ionomer. Furthermore, they found that, at high ion concentrations in the external solution, the environment inside the membrane and in the contiguous solution become thermodynamically similar, that is, at high total ion concentrations in the external solution, $CTs$⁠, the fixed charges in the membrane can be sufficiently screened by sorbed counterions. Here, $CTs$ is the total ion concentration and is the sum of the concentration of coions, $CSO42−s$⁠, and counterions, $CVO2+,H+s$⁠. Doing a back-of-the-envelope calculation (assuming 20 wt. % liquid water uptake for all membranes),⁶⁸ values of $CSO3−m,w$ for membranes cast on thquartz range from ∼4–5 moles of sulfonic acid groups per liter of sorbed water. For all membranes cast on PTFE, $CSO3−m,w≈4mol$ $SO3−$ per liter of sorbed water. With respect to the 1.5 M VOSO₄ in 3.0 M H₂SO₄ solution, we calculate a value of $CTs>10mol$ ions/l water, which is over twice that of the $CSO3−m,w$ values calculated for both PTFE- and quartz-cast PFSA and PFSA ionomer nanocomposites. Given this, changes to the partitioning of the vanadyl ions into the PFSA ionomers, and their effect on ion permeability, cannot be accurately described by the traditional Donnan theory. Furthermore, previous investigations have reported increased vanadium ion permeabilities with increased ionomer IECs,^60–62 which is contrary to the trends observed in this work, where higher vanadyl ion permeabilities were observed for PTFE-cast membranes that had lower or similar IECs to their quartz-cast counterparts. These results further underscore that considering variations in IEC alone cannot adequately explain the observed differences in vanadium ion permeabilities among various PFSA nanocomposites.

To help elucidate the impact of the membrane surface chemistry (i.e., surface hydrophilicty) on the vanadium ion permeability of the ionomers, contact angle measurements on all PFSA and PFSA ionomer nanocomposites were performed. Specifically, captive bubble contact angle experiments were performed. This particular experimental setup was selected as it allowed for contact angle measurements to be performed on hydrated membranes, which is consistent with the state the membrane would be in during the VRFB operation. However, it is important to note that the experimentally measured angles were converted into “traditional” contact angle measurements using ImageJ. That is, contact angles below 90° indicate a hydrophilic surface.

As seen from Table II, the contact angles for all membranes indicated that both the air- and substrate-adjacent surfaces were hydrophilic, though some were more hydrophilic than others. Both the air- and substrate-interfaces of Naf-Q and Naf-T have contact angles that are greater than their nanocomposite counterparts, indicating that the Naf-Q and Naf-T surfaces are more hydrophobic than the nanocomposite membranes. Additionally, the substrate-adjacent surfaces for all membranes were more hydrophilic than the air-adjacent surfaces. Furthermore, the surface-adjacent surface for all of the nanocomposite membranes was lower than both the Naf-Q and Naf-T. This result suggests that some portion of the SiNP assemble (i.e., settle) at the ionomer–substrate interface, irrespective of the chemistry of the interface, increase the hydrophilicity of the nanocomposite substrate-adjacent surface.

All vanadium ion permeability measurements shown in Fig. 1 were performed such that the substrate-interface was facing the vanadium-rich donating cell. Thus, the contact angle measurements of the substrate-interface may provide some insight into the kinetics of the vanadyl ion partitioning into and out of the membrane. For the most part, when the contact angle data are viewed in conjunction with the vanadium ion permeability data, we observe that vanadium ion permeability is higher in membranes with more hydrophilic substrate-adjacent surfaces. Furthermore, when the hydrophilicity of the air- and substrate-adjacent surfaces are similar, as is the case for Naf-PS-Q and Naf-PS-T, we observe a similar vanadium ion permeability for these membranes. As seen from Table II, the only series of membranes that does not follow this trend are those containing UF-SiNPs, where membranes cast on quartz showed a lower vanadyl ion permeability but had a more hydrophilic substrate-adjacent surface than their PTFE-cast counterparts.

It is curious to note that the VO²⁺ permeabilities (red bars in Fig. 1) and the IECs (solid red circles in Fig. 1) of membranes cast on PTFE were similar, regardless of the presence and specific surface chemistry of the SiNPs in the PFSA membrane. When comparing VO²⁺ permeabilities of PTFE-cast membranes to their polished quartz-cast counterparts, we observe an ∼100% and ∼70% increase for Naf-UF-T and Naf-TA-T membranes, respectively, while an ∼25% increase was observed for Naf-T membranes. However, as seen in Fig. 1, the VO²⁺ permeability of Naf-PS-T membranes slightly decreased (<10%) when compared to Naf-PS-Q membranes. This underscores the need to characterize potential substrate-dependent changes to the structure of the PFSA membranes.

As mentioned in Sec. III A, given the high total ion concentration in the adjacent vanadyl solution, the Donnan theory may not be useful in adequately explaining the observed differences in vanadium ion permeabilities between PTFE- and polished quartz-cast membranes. This result alludes to the potential of substrate-induced changes to the ionomer morphology and/or substrate-induced changes to the chemical composition of the membrane surface directly adjacent to the casting substrate, both of which can alter the permeation of vanadyl ions into and across the ionomer membrane. To investigate potential changes in the bulk morphology of the PFSA nanocomposites, SANS experiments were carried out on hydrated PFSA ionomer and PFSA ionomer nanocomposites (membranes were hydrated in H₂O at least 24 h prior to collecting scattering data). Furthermore, a combination of AFM and SEM-EDS was employed to interrogate the roughness and chemical composition of the membrane surface adjacent to the casting substrate, respectively. Note that SANS data for membranes cast on the polished quartz have been previously published.¹⁴ 

Figure 2 shows the SANS data of fully hydrated, in H₂O, PFSA membranes [Fig. 2(a)], as well as electron microscopy images [Figs. 2(b) and 2(c)], and SEM-EDS maps [Figs. 2(d) and 2(e)] of Nafion cast on both polished quartz and PTFE substrates. There are two correlation peaks in the SANS curves for Nafion shown in Fig. 2(a). The peak at lower Q can be assigned to the spacing between crystalline domains in the hydrophobic PTFE region (henceforth referred to as the “hydrophobic peak”), while the peak at high Q can be assigned to the period spacing of the hydrophilic ionic domains/clusters within the ionomer (henceforth referred to as the “ionic peak”).⁴ As seen in Fig. 2(a), there are a number of important differences between the SANS curves for Naf-Q and Naf-T. First, there is a shift of the hydrophobic peak to a lower Q when the casting substrate is changed from the polished quartz to PTFE. This shift to a lower Q indicates that the spacing between the crystalline domains has increased, that is, the amorphous PTFE regions between the crystallites have expanded, indicating less efficient packing within this hydrophobic region of the nanophase segregated PFSA ionomer. Second, there is a low-$Q$ upturn in the SANS curve for Naf-T that is not observed in the SANS data for Naf-Q. This low-$Q$ upturn seen in the SANS data for Naf-T indicates the presence of a larger structural feature, likely some types of longer-range crystallization in the hydrophobic region.^14,69–71 Finally, there is a little change in the spacing of the ionic groups, as the location of the ionic peak did not change appreciably between the two casting substrates. This observation indicates that, for the most part, the difference in casting substrates does not significantly alter the evolution of the ionic domains during membrane formation (i.e., during solvent evaporation, forming a dense membrane). However, as mentioned, differences in the shape of the scattering curve at lower Q highlight that the choice of casting substrate does alter the formation of the hydrophobic PTFE region of the ionomer, which may play some role in the observed increase (∼25%) in VO²⁺ permeability for membranes cast on PTFE substrates.

While the vast majority of the literature suggests that vanadium ion transport is entirely linked to morphology and interconnectivity of the ionic domains within PFSA ionomers, a work by Di Noto and co-workers^72,73 indicates that changes to the PTFE matrix of the ionomer may have a greater effect on transport that occurs within the ionic domains than was originally believed. In their work, they described a peristaltic-like motion of ion transport within PFSA ionomers, which can be altered due to perturbations within the hydrophobic domain of the ionomer. If we consider that vanadium ion transport may also require such peristaltic motions, then the changes to the morphology of the PTFE domain, observed in the SANS curves, are likely to alter these peristaltic motions within membranes cast on PTFE, which could explain the observed increase in permeability.

The SEM images of the substrate-adjacent sides of Naf-Q and Naf-T, shown in Figs. 2(b) and 2(c), respectively, indicate that the roughness of membrane surface varies with the casting substrate. The root mean square (RMS) roughness values (as determined by atomic force microscopy; see the supplementary material and Table S1) indicate that the substrate-adjacent surface of Naf-T is an order of magnitude higher rougher than that of Naf-Q, which is corroborated by the images shown in Fig. 2. This result was not unexpected given the differences in roughness between the polished quartz and PTFE substrates. However, the corresponding SEM-EDS maps of these SEM images, shown in Figs. 2(d) and 2(e), respectively, indicate that the chemical composition of the membrane surface remains similar regardless of the substrate on which the membranes were cast. In Figs. 2(d) and 2(e), the colors orange, pink, neon blue, red, blue, and neon green correspond to the elements calcium (Ca), potassium (K), oxygen (O), carbon (C), sulfur (S), and fluorine (F), respectively. The EDS maps of the substrate-adjacent surfaces of the films provide useful information regarding the chemical composition, where the distribution and amount of S and F present on this surface are noteworthy, with signatures from S and F being indicative of the hydrophobic and hydrophilic domains, respectively, of the PFSA membrane. The presence and distribution of Ca, K, O, and C are not relevant to the current discussion and can be attributed to leftover catalyst,⁷⁴ dust,⁷⁵ and the partial vacuum atmosphere. Remember that the vanadium ion permeabilities shown in Fig. 1 are calculated from experiments, where the substrate-adjacent side of the PFSA membrane is facing the concentrated vanadium solution (i.e., the vanadium solution in the donating cell). Given that the chemical composition of this surface is similar between Naf-Q and Naf-T, the observed change in vanadium ion permeability is likely related to changes in the nanostructure of the PTFE matrix of the PFSA membrane, which are alluded to by the differences in the SANS curves between the two membranes.

To elucidate potential changes in the structure and surface composition of PFSA membranes containing unfunctionalized and functionalized SiNPs, SANS and SEM experiments were performed on the three nanocomposite membranes listed in Table I. Figure 3 shows the SANS data of fully hydrated, in H₂O, PFSA nanocomposites [Fig. 3(a)], as well as electron microscopy images [Figs. 3(b) and 3(c)], and SEM-EDS maps [Figs. 3(d) and 3(e)] of Naf-UF cast on both polished quartz and PTFE substrates. As shown in Fig. 3(a), there is not a measurable change in the location of the hydrophobic and ionic peaks, when the casting substrate is changed from the polished quartz to PTFE, indicating that the periodic spacing of these structural features has not changed appreciably for the Naf-UF membranes. Note that, as previously observed,¹⁴ the introduction of UF-SiNP into the PFSA membrane, when cast on the polished quartz, results in a shift of the hydrophobic peak to lower Q, as compared to pristine PFSA ionomers. In this case, the introduction of UF-SiNPs leads to an increase in the periodic spacing of the PTFE crystallites in the hydrophobic domain of the ionomer. From Fig. 3(a), we observe that this shift to lower Q is independent of the casting substrate. Interestingly, the increased spacing of the PTFE crystallites within the hydrophobic domains for both Naf-UF-Q and Naf-UF-T affect the vanadium ion permeability oppositely, where an increase in vanadium ion permeability was observed for Naf-UF-T and a decrease in vanadium ion permeability was observed for Naf-UF-Q (see Fig. 1).

While there are no significant differences in the location of the hydrophobic and ionic peaks between the two membranes shown in Fig. 3(a), we do observe the presence of an appreciable low-$Q$ upturn in the SANS data for both Naf-UF-Q and Naf-UF-T, that was not present in the SANS curves of the pristine PFSA membranes [see Fig. 2(a)]. With respect to nanocomposites, this low-$Q$ upturn is traditionally attributed to nanoparticle aggregation/agglomeration within the polymer.⁷⁶ The low-$Q$ upturn observed for Naf-UF-Q is steeper and begins at a higher Q as compared to its PTFE-cast counterpart. This suggests that the aggregation of UF-SiNPs in Naf-UF-T results in, on average, slightly larger nanoparticle aggregates than when these membranes are cast on the polished quartz. As previously shown, the low-$Q$ upturn seen in the SANS data of both Naf-UF-Q and Naf-UF-T is indicative of fractal scattering, where the steeper slope observed for Naf-UF-Q suggests Porod-like scattering between the smoother interfaces of the SiNP aggregates and Nafion.^14,77

In contrast, the lower slope of the low-$Q$ upturn in the SANS data for Naf-UF-T is likely indicative of rougher interfaces between the SiNP aggregates and Nafion. This difference in the low-$Q$ upturn between Naf-UF-Q and Naf-UF-T may explain the ∼100% increase in vanadium ion permeability observed, when the casting substrate is changed to PTFE. Additionally, when comparing the SANS data of Naf-T and Naf-UF-T, we observe a similar upturn in the low-$Q$ data for Naf-T [albeit less prominent than what is observed in Fig. 3(a)]. This is interesting to note since the vanadium ion permeabilities for Naf-T and Naf-UF-T are not significantly different from one another (∼20% difference). This observation hints at the possibility that the PTFE substrate is governing the assembly of these larger (>100 nm) structural features as the solvent evaporates to form the dense PFSA membranes, regardless of the presence of UF-SiNPs.

In addition to changes in the aggregation of the UF-SiNPs within the PFSA membrane, the SEM images and SEM-EDS maps of the substrate-adjacent surface of Naf-UF-Q and Naf-UF-T shown in Figs. 3(b)–3(e) help elucidate how the casting substrate impacts the preferential segregation of the UF-SiNP within the membrane. First, from Figs. 3(b) and 3(c), we see that the substrate-adjacent surface of Naf-UF-T is significantly rougher than that of Naf-UF-Q, where this roughness cannot solely be explained by the difference in roughness of the two casting surfaces [as observed in Figs. 2(b) and 2(c)]. The roughness we observe in Fig. 3(c) appears to be caused by the preferential segregation of the UF-SiNP to this interface. We can more confidently conclude that this is the case when we look at the SEM-EDS maps shown in Figs. 3(d) and 3(e). Note, here the elemental signal colors utilized are as follows: the colors orange, yellow, blue, fuchsia, neon green, neon blue, and red correspond to the elements copper (Cu), K, S, silicon (Si), F, O, and C, respectively. In addition to the elemental signals from S and F, the elemental signal for Si is now relevant for this discussion as it provides information regarding the preferential segregation (or lack thereof) of the unfunctionalized and functionalized SiNPs to this interface.

Seen in Fig. 3(d), the S, F, and Si signals are relatively well-dispersed across the substrate-adjacent surface of Naf-UF-Q, that is, there is no measurable preferential segregation of either hydrophobic or hydrophilic groups, nor the UF-SiNPs, to the polished quartz surface. This can also be seen in Fig. S3, where the EDS maps for each individual element are shown. In contrast, the SEM-EDS map in Fig. 3(e), along with the corresponding individual elemental maps shown in Fig. S3, indicates slightly more segregation between the individual elements, where that the concentrations of S [compare Figs. S3(c) and S3(g)] and Si [compare Figs. S3(d) and S3(h)] at this surface have decreased and increased, respectively. Furthermore, there are one or two areas of high Si concentration that span 10 s of micrometers along the surface of Naf-UF-T. This confirms that the roughness of the substrate-adjacent surface of Naf-UF-T is a result of the UF-SiNP preferentially segregating (and aggregating) to the PTFE–Nafion interface. This result also correlates well with the increased hydrophilicity of the substrate-adjacent surface, as observed by the contact angles shown in Table II. The roughness of this interface was further confirmed by the RMS roughness of the membrane surface (shown in Table S1 in the supplementary material), where the RMS roughness value for Naf-UF-T is over an order of magnitude higher than that for Naf-UF-Q. As this is the membrane surface that is exposed to the concentrated vanadium ion solution of the donating cell, the higher concentration of hydrophilic groups (i.e., presence of either S or Si) at this surface may explain the observed increase in vanadium ion permeability between Naf-UF-Q and Naf-UF-T, as the partitioning of vanadyl ions into the membrane is accelerated. This result alludes to the fact that the chemical composition of the membrane surface that is directly in contact with the donation cell may play a larger role in governing the vanadium ion transport through the membrane, further supported by the comparison of contact angle and vanadium ion permeability. We will come back to this notion later on in this section.

Figure 4 shows the SANS data of fully hydrated, in H₂O, PFSA nanocomposites [Fig. 4(a)], as well as electron microscopy images [Figs. 4(b) and 4(c)], and SEM-EDS maps [Figs. 4(d) and 4(e)] of Naf-TA cast on both polished quartz and PTFE substrates. As seen in the SANS data in Fig. 4(a), the casting substrate does not significantly affect the formation of the ionic domains as the location of the ionic peak does not measurably change between Naf-TA-Q and Naf-TA-T. However, there are a couple of notable differences between the SANS curves for Naf-TA-Q and Naf-TA-T in the mid-$Q$ region. First, the hydrophobic peak for Naf-TA-T is less discernable than that of the hydrophobic peak of Naf-TA-Q, though neither of these peaks is as prominent as those observed in Figs. 2(a) and 3(a). Second, there appears to be the emergence of an entirely new correlation peak between the hydrophobic peak and the low-$Q$ upturn for Naf-TA-T. The presence of this new scattering peak could be indicative of an increased dispersity in the periodic spacing of the crystallites or could indicate a more periodic spacing of the TA-SiNP aggregates within the PFSA membrane.

Previous studies have shown that the TA-SiNP are known to aggregate within the PFSA matrix, forming aggregates of sizes ranging from ∼10 nm (few particle clusters) to several 100 s nm (tightly bound aggregates).¹⁵ As SiNP aggregates are posited to reside at the interface of the ionic and hydrophobic domains,^78,79 the higher concentration of TA-SiNP aggregates may distort the spacing of the PTFE crystallites within the hydrophobic domains more significantly than the other SiNP functionalities. Alternatively, higher concentration of TA-SiNP aggregates may also contribute to the mid-$Q$ scattering, resulting in the presence of an additional scattering peak, which convolutes the “traditional” hydrophobic peak of the PFSA ionomer. While we cannot completely resolve the origin of this additional mid-$Q$ peak seen in the SANS data for Naf-TA-T, it is clear that the effect of the casting substrate on the membrane nanostructure has propagated through the entire thickness of the film, resulting in ∼70% increase in the vanadium ion permeability as compared to Naf-TA-Q. The presence of the additional peak in the Naf-TA-T scattering curve is likely related to changes in the ordering/morphology of the PTFE domains. As mentioned earlier, such alterations could lead to changes in the peristaltic motions within the ionomer, ultimately leading to changes in the transport of vanadium ions through the ionomer. Note that the slopes (and location) of the low-$Q$ upturn appear similar for Naf-TA-Q and Naf-TA-T, suggesting that the larger-scale aggregation and interfaces between the TA-SiNP aggregates and PFSA matrix are similar between the two films.

In addition to the bulk structural changes captured by the SANS data in Fig. 4(a), electron microscopy was employed to capture differences in the structure and chemical composition of the substrate-adjacent surface of both Naf-TA-Q and Naf-TA-T. As seen in Figs. 4(b) and 4(c), the structure of this surface is quite different between the two casting substrates. Most notably, the substrate-adjacent surface for Naf-TA-Q appears to have numerous “pits” that are not present in Naf-TA-T. The individual pits range in size from ∼10 μm to ∼50 μm, though in some areas of the membrane surface, these pits have loosely aggregated in structural features on the order of 100 μm. Furthermore, in contrast to the previous two membranes reported, the RMS roughness values of the substrate-adjacent surface for both casting surfaces are on the same order of magnitude (see Table S1 in the supplementary material). We can gain insight into the chemical composition of these two surfaces through the SEM-EDS maps shown in Figs. 4(d) and 4(e). Note that the elemental signal colors utilized in Figs. 4(d) and 4(e) are as follows: the colors blue, fuchsia, neon green, neon blue, and red correspond to the elements S, Si, F, O, and C, respectively. As previously mentioned, only the signals from S, F, and Si will be considered for this discussion.

The SEM-EDS maps of Naf-TA-Q and Naf-TA-T, shown in Figs. 4(d) and 4(e), respectively, are quite different than those previously shown for pristine Nafion or Nafion containing 5 wt. % UF-SiNP. Specifically, the SEM-EDS maps show a distinct separation between the elemental signal for Si and those for S and F. This separation is more easily seen in Fig. S4, where the SEM-EDS maps for each of the individual elements are shown. Notably, there is a high concentration of Si and almost no S or F in the pits of the substrate-adjacent surface of Naf-TA-Q. While there are no pits observed in the substrate-adjacent surface of Naf-TA-T, a similar distinct separation between the elemental signal for Si and those for S and F can be seen in Fig. 4(e). However, unlike Naf-TA-Q, the concentration of Si at the substrate-adjacent surface of Naf-TA-T is significantly lower [compare Figs. S4(d) and S4(h)]. Additionally, the size of the high Si concentration regions on the surface of Naf-TA-T is smaller (∼5 to 10 μm in size) than those seen on the surface of Naf-TA-Q (∼10 to 50 μm in size). This result indicates that while the TA-SiNPs preferentially aggregate at both the polished quartz and PTFE substrates, the protonated amine surface of TA-SiNP leads to more favorable energetic interaction between NP and the polished quartz surface. The high concentration of TA-SiNP, and consequent higher concentration of hydrophilic groups (described by the contact angles, see Table II), at this surface may help explain the significantly lower vanadium ion permeability observed in Naf-TA-Q when compared to Naf-TA-T.

Finally, Fig. 5 shows the SANS data of fully hydrated, in H₂O, Naf-PS-Q, and Naf-PS-T [Fig. 5(a)], as well as electron microscopy images [Figs. 5(b) and 5(c)] and SEM-EDS maps [Figs. 5(d) and 5(e)] of Naf-PS-Q and Naf-PS-T, respectively. This set of membranes is of particular interest as unlike the pristine Nafion and the other Nafion nanocomposites cast on both polished quartz and PTFE substrates, Naf-PS-Q and Naf-PS-T demonstrate similar vanadium ion permeabilities and IEC values (see Fig. 1). As seen in Fig. 5(a), the SANS curves of the hydrated membranes are almost identical to each other. The locations of the hydrophobic and ionic peak are analogous between the two PFSA nanocomposites. There is a slight difference in the low-$Q$ region of the SANS curves, where the low-$Q$ upturn for Naf-PS-T starts at slightly larger Q, though this difference is quite small. Interestingly, although the bulk properties and morphologies are similar for Naf-PS-Q and Naf-PS-T, the substrate-adjacent surface, as analyzed by SEM-EDS [Figs. 5(d) and 5(e)], there is a significant phase separation and preferential segregation of the functionalized nanoparticles to the ionomer–substrate interface. The elemental signal colors utilized in Figs. 5(d) and 5(e) are as follows: the colors blue, fuchsia, neon green, neon blue, and red correspond to the elements S, Si, F, O, and C, respectively.

As seen in Figs. 5(b) and 5(d), the substrate-adjacent surface for Naf-PS-Q shows the presence of similar pits (both in size and in number) as seen in Naf-TA-Q. However, in contrast to Naf-TA-Q, the RMS roughness value for the substrate-adjacent surface of Naf-PS-Q is ∼75% lower than that of Naf-TA-Q (see Table S1 in the supplementary material). Similar to Naf-TA-Q, there is a high concentration of Si present in and around these pits on the membrane surface, indicating that the PS-SiNPs have a slight preference to segregate to the polished quartz–PFSA nanocomposite interface. However, this preferential segregation of the functionalized nanoparticles is most noticeable for Naf-PS-T, where the SEM image shown in Fig. 5(c) shows the largest amount of preferential nanoparticle segregation (and aggregation) at this interface, when compared to any of the other membranes investigated. As seen in SEM-EDS map in Fig. 5(e), the substrate-adjacent surface of Naf-PS-T shows the highest Si concentration of any of the membranes investigated (also see Fig. S5).

Although the SEM-EDS maps shown in Figs. 5(d) and 5(e) seem to indicate a higher concentration of Si at the substrate-adjacent surface of Naf-PS-T, the contact angle measurements for Naf-PS-Q and Naf-PS-T are identical (see Table II). Given the similar SANS curves and contact angles, it follows that the vanadium ion permeabilities for both Naf-PS-Q and Naf-PS-T are equivalent (within less than 10% of each other). Unlike the Naf-TA system, where significant changes to the SANS curves are observed, the choice of casting substrate, either polished quartz or PTFE, does not measurably impact the nanostructure of the Naf-PS membranes. Taking all these into account, it appears that the differences in vanadyl ion permeability are primarily due to substrate-induced changes to the bulk morphology, which could ultimately impact the polymer dynamics, i.e., peristaltic motions that govern vanadium ion transport in the PFSA ionomer nanocomposites. However, changes to the hydrophilicity of the substrate-adjacent surface may play some role in the observed differences, though its effect on vanadium ion transport appears to be much less significant than changes to the overall nanostructure of the ionomer. Additionally, it is important to highlight that the choice of the casting substrate had a significant impact on the assembly of nanoparticles in the PFSA membranes, as well as a significant impact on the intricate nanophase segregation of the bulk ionomer. To date, the propagation of substrate effects across the entire thickness of the membrane is something that has only been demonstrated for extremely thin PFSA films (<20 nm).^26,27,57 However, as previously mentioned, recent results from Baglay and Roth^47,48 alluded to the possibility of substrate effects propagating much further away from the substrate than had previously been thought.

Note, up to this point, all of the vanadium ion permeabilities were calculated from crossover experiments that were performed with the substrate-adjacent surface facing the vanadium-rich solution (i.e., facing the donating cell). As seen with the PFSA nanocomposites, the chemical composition of the substrate-adjacent surface had a lesser effect on the crossover of vanadium ions [compare Naf-TA-Q, Naf-PS-Q, and Naf-PS-T; see Figs. 1, 4(d), 5(d), and 5(e)], and it was, in fact, the bulk nanostructure of the membrane that played the largest role in governing vanadium ion transport across the membranes. This was also the case with Naf-Q and Naf-T, as the hydrophilicity of their substrate-adjacent surfaces was similar (see Table II) but exhibited different bulk nanostructures [see Fig. 2(a)], ultimately resulting in ionomers with different vanadium ion permeabilities (∼25% higher for Naf-T vs Naf-Q; see Fig. 1). To interrogate the impact of the PFSA surface chemical composition on vanadium ion partitioning and crossover, permeability experiments were performed on Naf-Q and Naf-T with the air-adjacent surface facing the vanadium-rich solution (i.e., the substrate-adjacent surface facing away from the donation cell).

Figure 6 shows the vanadium ion permeabilities calculated through Naf-Q and Naf-T with both the substrate-adjacent surface and the air-adjacent surface facing the vanadium-rich solution in the permeability cell. Note that the permeabilities shown for the substrate-adjacent surfaces are the same as those shown in Fig. 1. Surprisingly, the permeability of vanadyl ions through Naf-Q and Naf-T with the air-adjacent surfaces facing the vanadium-rich solution are identical and are lower than those reported earlier (∼15% lower compared to Naf-Q and ∼30% lower compared to Naf-T). The lower vanadyl ion permeability may be, in part, attributed to the increased hydrophobicity of these surfaces vs the substrate-adjacent surfaces, as shown by the contact angles in Table II.

Notably, the vanadium ion permeabilities reported in Fig. 6 show that the transport of vanadium ions can be altered just by simply changing the orientation of the PFSA films. This result, in part, suggests that for pristine PFSA membranes (i.e., ionomers containing no SiNPs), changes to the surface chemistry of the membrane are just as important in governing vanadium ion crossover as the nanostructural changes induced by the casting substrate. This is an important observation as it has direct implications for operating flow batteries, where membrane orientation could unknowingly impact battery performance.

In conclusion, it was demonstrated, for the first time, that the choice of the casting substrate (e.g., hydrophobic vs hydrophilic) can significantly impact the morphology, as well as both the substrate-adjacent and air-adjacent surface chemistries, of bulk (>50 μm) PFSA ionomer and PFSA nanocomposite membranes. Furthermore, the permeability of vanadium ions across the membranes was shown to be a function of the casting substrate. Except for Naf-PS, membranes cast on the polished quartz demonstrated lower ion permeabilities when compared to their PTFE-cast counterparts. In the case of Naf-PS, the casting substrate had little impact on the vanadium ion permeability, where the vanadium ion permeabilities were the same. This was attributed to the almost identical nanostructures and surface chemistries of the substrate-adjacent surfaces of Naf-PS-Q and Naf-PS-T, as shown by hydrated SANS and captive bubble contact angle measurements, respectively. For PFSA nanocomposites, it was also shown that both the substrate-adjacent surface chemistry and the bulk morphology of the membranes governed transport of vanadium ions, though changes to the bulk morphology appeared to have a more significant impact on vanadyl ion transport.

Interestingly, it was shown that the surface chemistry of the pristine PFSA membranes has a measurable impact on the vanadium ion crossover performance of the ionomer, where the vanadium ion permeability was identical for pristine PFSA membranes cast on either the polished quartz or PTFE when the air-adjacent membrane surface was challenged by the high concentration of vanadium ions of the donating cell. This result highlights that, unlike the PFSA nanocomposite films, the surface chemistries of the different interfaces of pristine PFSA membranes play a significant role in governing ion transport, in addition to the nanostructural changes induced by the choice of the casting substrate. The results from this study underscore the importance of providing detailed synthesis information when fabricating and testing bulk, solution-cast PFSA, and PFSA nanocomposite membranes.

See the supplementary material for a schematic of the permeation cell, the RMS roughness values of the substrate-adjacent and air-adjacent surfaces, and the individual elemental SEM-EDS maps of the substrate-adjacent surface of various membranes.

The authors gratefully acknowledge support from the National Science Foundation (NSF) CAREER Program (No. DMR-1848347) and the Clemson University Advanced Materials Research Lab for using their electron microscopy facilities. Allison Domhoff would like to acknowledge support of the National Science Foundation Graduate Research Fellowship Program (Award No. 1246875) and the Hitachi High Technologies Electron Microscopy Fellowship. The authors would also like to acknowledge the National Institute of Standards and Technology, U.S. Department of Commerce, for providing the neutron research facilities used in this work. The authors would like to specifically thank Tyler Martin for his assistance with neutron experiments and numerous fruitful discussions.