Nonfluorinated membrane with a decentralized ion-transport network enables efficient and sustainable polysulfide redox flow batteries

Abstract

Polysulfide-based redox flow batteries are promising for long-duration energy storage, owing to ultralow-cost/earth-abundant active materials and full decoupling of power and energy. However, their practical application has been prevented by poor cycle life resulting from polysulfide crossover and a heavy reliance on costly fluorinated membranes (Nafion 117, USD $800 to $3500 per square meter), along with the environmental concerns. Here, we develop a nonfluorinated sulfonated polyethersulfone (SPES)–based membrane with decentralized ion-transport channels, achieving a 20 times higher ionic selectivity at a markedly reduced cost (USD $12 to $66 per square meter) compared to the commercial Nafion membrane. The low-cost SPES-based membrane enabled stable cycling of polysulfide-ferrocyanide redox flow batteries with a high coulombic efficiency (>99.9%) and energy efficiency (average >75%) for 1600 cycles (>6 months). This strategy demonstrated polysulfide-based redox flow batteries with a record longevity using a low-cost and sustainable membrane, paving the way for their practical commercialization.

A cost-effective and nonfluorinated ion-selective membrane for polysulfide flow batteries.

INTRODUCTION

Developing long-duration energy storage systems is imperative to achieving massive deployment of renewable energies (1, 2). Aqueous redox flow batteries (RFBs) are one of the most promising technologies for long-duration storage owing to their intrinsic safety, flexible design, and long lifespan (3, 4). Vanadium RFBs are the most well-developed aqueous RFBs; however, the high cost (USD $89.4 kA hour⁻¹) and scarcity of vanadium limit its widespread deployments (5). Polysulfide-based RFBs (PSFBs) (6, 7) are particularly attractive emerging flow battery systems owing to the ultralow cost (USD $0.15 kA hour⁻¹) (8) and large abundance (5.4 × 10⁷ tons per year) (9, 10) of polysulfide. The major challenge of PSFBs is polysulfide crossover, leading to irreversible capacity loss, and electrode/membrane passivation resulting from polysulfide oxidation to insulating solid sulfur (11). These root causes lead to poor cycle life (<50 cycles), preventing practical application of PSFBs (6, 12). While modification of the commercial perfluorosulfonic acid Nafion membrane effectively mitigated the crossover of polysulfide (13), the high cost of the fluorinated Nafion membrane (USD $800 to $3500 m⁻²) makes it unviable for commercial applications (14, 15). In addition, the by-products of Nafion resin manufacturing, per- and polyfluoroalkyl substances, are referred to as “forever chemicals.” These substances pose increasing concerns because of their human toxicity and ecosystem impacts (16). Low-cost nonfluorinated membranes (e.g., sulfonated polyether ether ketone with a high sulfonation degree) are proposed as the promising candidates to replace the Nafion membrane in PSFB application (17). However, pronounced microphase separation and low dimensional stability could not effectively reduce the risk of polysulfide shuttling, leading to fast capacity decay.

The commonly used Nafion membranes are inadequate to prevent polysulfide crossover because of oversized ion-transport channels resulting from aggregated microphase separation (18, 19). The formation mechanism of ion-transport channels of Nafion is shown in Fig. 1A. Microphase separation is created during the self-assembling process resulting from hydrophobic-hydrophilic phase separation (20, 21). A hydrophobic perfluorinated backbone tends to aggregate to minimize contact with water, forming a hydrophobic fluorocarbon matrix. Meanwhile, the hydrophilic sulfonic acid groups attract water molecules, creating hydrated ionic clusters or hydrophilic domains. Upon hydration, water absorption into these ionic clusters induces membrane swelling, further enhancing the separation between hydrophobic and hydrophilic domains. This swelling process refines the organization of hydrophilic domains into an interconnected cluster network for ion transport across the membrane. However, such an aggregated and centralized cluster network leads to oversized ion-transport channels, leading to high ionic conductivity but poor ionic selectivity with severe polysulfide crossover (22).

Fig. 1. Different phase separation behaviors to form the ion-transport channels.

(A) Aggregated ion-transport channels of the Nafion membrane. (B) Decentralized ion-transport channels adjusted by the comonomer of the sulfone-based hydrocarbon membrane.

Here, we develop nonfluorinated sulfonated polyethersulfone (SPES) with decentralized ion-transport channels to create high-quantity, smaller ion-transport channels to mitigate polysulfide crossover (reduced channel size) while maintaining high ionic conductivity (high-quantity channels). We show that the size and distribution of ion-transport channels can be tuned via microphase separation by controlling the comonomer and degree of sulfonation. Thanks to hydrophilic sulfone groups and the rigid aromatic backbone, the SPES membrane endows high water absorption (high ionic conductivity) with strong structural stability (low swelling ratio). Small-angle x-ray scattering (SAXS) and atomic force microscopy (AFM) were used to characterize the size and distribution of ion-transport channels. ¹H nuclear magnetic resonance (NMR) proton spin-spin relaxation time (T₂) measurement was performed to probe the states and distribution of different water molecules in membranes. The developed low-cost SPES-based membrane (USD $12 to $66 m⁻²) show a much-improved flow battery performance compared to the commercial fluorinated Nafion membrane [Nafion 117 (N117), USD $800 to $3500 m⁻²], enabling stable cycling of a full polysulfide-ferrocyanide (S-Fe) RFB with a high coulombic efficiency (CE; >99.9%) and energy efficiency (EE; average >75%) for 1600 cycles (>6 months).

RESULTS

Design of decentralized ion-transport channels

To reduce the phase separation and avoid oversized ion-transport channels (Fig. 1A), we aim to create high-quantity, smaller decentralized ion-transport channels by introducing multiple hydrophilic sites into a hydrophobic backbone. On the basis of this principle, we developed SPES, which introduces a hydrophilic sulfone group into the hydrophobic hydrocarbon backbone (Fig. 1B) while carrying hydrophilic diphenyl sulfone as the comonomer (Figs. 1B, 3). The sulfone group, serving as a hydrogen bond acceptor, can form hydrogen bonds with water molecules (23). Embedding the hydrophilic sulfone group in the main chain and comonomer, we aim to create multiple small hydrophilic/hydrophobic building blocks to form high-quantity, smaller ion-transport channels via decentralized phase separation, in contrast to the centralized/aggregated phase separation yielding oversized ion-transport channels in Nafion (Fig. 1A).

To systematically illustrate the critical role of the polymer building block and how it affects the connectivity of hydrophilic regions (24), two more comonomers including hydrophobic bisphenol and hydrophobic hexafluoro bisphenol A are included for comparison. We hypothesize that the sulfonated polyphenylenesulfone (SPPSU) would have the smallest hydrophilic domain among all three polymers owing to the contributions of π-π stacking (25) and charge-transfer complexes (26), leading to a rigid arrangement of molecular chains. In comparison, sulfonated polysulfone (SPSF) and SPES are expected to have a more relaxed structure with an increased free volume owing to the presence of an additional electron-rich comonomer and the dynamic π-flip movements of phenyl rings. Accordingly, the water molecules within these polymers could be more mobile owing to a more loosely organized molecular structure compared to SPPSU (27). Last, the two sulfone groups in an SPES monomer could create more ion-transport channels compared to SPSF.

The degree of sulfonation is another crucial factor controlling the establishment of ion-transport channels (28). At a low degree of sulfonation, the hydrophilic domains in the main chain could become isolated water islands instead of interconnected water channels (29), leading to poor ionic conductivity. On the other hand, at a too high sulfonation degree, large amounts of small ion-transport channels could aggregate to a large ion-transport channels, leading to oversized ion-transport channels with poor ionic selectivity. We compare the pristine membrane properties of SPES with those of SPPSU and SPSF at a fixed sulfonation percentage of 30%, denoted as S30, P30, and F30, respectively, and investigate the influence of degree of sulfonation in SPES at 10 and 50%, denoted as S10 and S50, respectively. Details of membrane fabrication and abbreviations are shown in Materials and Methods and table S1.

Comparison of pristine membrane properties

Water uptake is an essential process for ion-selective membranes to construct interconnected channels for ion transport, which in turn results in varying degrees of membrane swelling depending on the strength of the polymer backbone. We characterized the water/electrolyte uptake (defined as the weight ratio of absorbed water to the dehydrated membrane) (30, 31) and membrane swelling ratio (defined as the volume expansion ratio of absorbed water to the dehydrated membrane) (32, 33) for N117, S30, P30, and F30 (Fig. 2, A and B). We noted that N117 and all the sulfone-based hydrocarbon membranes exhibited reduced electrolyte uptake and swelling ratio in 1 M KOH and 1 M KCl aqueous solutions compared to those in pure water. This behavior could be attributed to the high electrolyte concentration in the external solution, which decreases water activity and lowers the chemical potential of water exterior to the membrane, resulting in a decreased electrolyte uptake and swelling ratio. The results show that S30 exhibits the highest water/electrolyte uptake per volume increased (i.e., “water channel density”), suggesting that S30 could effectively absorb water molecules while maintaining membrane structural integrity (Fig. 2C). This can be attributed to sufficient hydrophilic functional groups (sulfone groups) in the main chain absorbing water molecules while having rigid phenyl groups restricting the expansion of molecular chains, thus increasing the dimensional stability of the membrane (22, 34). Notably, S50 with a high degree of sulfonation (50% sulfonation SPES) has more hydrophilic sulfonic acid groups than S30, leading to a higher water/electrolyte uptake but higher swelling ratio (fig. S1). In contrast, S10 with a low sulfonation degree (10%) shows the least water uptake and swelling owing to the lack of ion-transport channels, which is also supported by the lowest ion exchange capacity (IEC; fig. S2).

Fig. 2. Properties and characterization of the ion exchange membrane.

(A) Comparison of water/electrolyte uptake. (B) Swelling ratio of pristine membranes in different aqueous solutions. (C) Water channel density of pristine membranes. (D) SAXS profiles of membranes. d represents the Bragg spacing. a.u., arbitrary units. (E) AFM tapping mode phase images of N117 and S30. The dark regions present hydrophilic domains of the polymer matrix. (F) ASR of pristine membrane measured in different electrolyte solutions. (G) Permeability of ions across the membrane. The diffusion rate was tested at room temperature. h, hours. (H) Mechanical properties of pristine membrane. The error bars in the figure represent the standard deviation derived from at least three independent measurements for each membrane and condition.

We examine the size and distribution of ion-transport channels via SAXS and tapping mode AFM. The aggregation of sulfonic acid groups (─SO₃H) in Nafion (N117) promotes aggregated microphase separation, forming well-defined nanoscale ionic clusters and water channel networks in the membrane, showing a sharp SAXS peak, as observed in Fig. 2D. In contrast, sulfonic acid groups in sulfone-based membranes are random and dispersive. In addition, sulfone-based membranes have the rigid backbone and decentralized microphase separation, which leads to a weak and broad SAXS peak compared to that of Nafion (16). According to the Bragg equation (35), the sizes of hydrophilic clusters for N117, S30, F30, and P30 were 3.87, 1.42, 1.38, and 1.40 nm, respectively (Fig. 2D and fig. S3). All the sulfone-based membranes exhibit a much smaller hydrophilic cluster size than N117, which directly supports our design concept (Fig. 1). The hydrophilic cluster size of S30 is slightly larger than those of F30 and P30, which is consistent with the hypothesis that S30 has more hydrophilic domains in the polymer matrixes than F30 and P30. Tapping mode AFM phase images (Fig. 2E) show that N117 has large and centralized/aggregated hydrophilic domains, while the S30 membrane has small and decentralized hydrophilic domains, which is in line with our design concept (Fig. 1). We further conducted positron annihilation lifetime spectroscopy (PALS) measurement to elucidate the microphase separation characteristics of N117 and S30 (fig. S4). The PALS spectrum of N117 reveals two distinct ortho-positronium (o-Ps) lifetimes, including a shorter lifetime (τ₃, fixed at 1.8 ns) and a longer lifetime (τ₄) (36). The fixed τ₃ of 1.8 ns, typical of o-Ps annihilation in water, corresponds to annihilation within the ionic-water clusters, formed by sulfonic acid groups and absorbed water. The longer τ₄ is attributed to o-Ps annihilation in larger free volume cavities within the hydrophobic perfluorinated backbone and the amorphous interfacial regions between the backbone and ionic-water clusters. These lifetimes reflect a well-defined and aggregated microphase-separated morphology. In contrast, SPES exhibits a single o-Ps lifetime (τ₃) in its PALS spectrum, suggesting that the o-Ps annihilates in both the polymer backbone and hydrophilic domains because of its decentralized microphase separation.

The ionic conductivity and selectivity are directly affected by the size, distribution, and structure of the ion-transport channels (37). While the ionic conductivities of the sulfone-based membranes are slightly lower than that of N117 (figs. S5 and S6), the area-specific resistance (ASR) of S30 (2.39 ohm·cm² in 1 M KOH) is comparable with that of N117 (2.33 ohm·cm² in 1 M KOH) (Fig. 2F) thanks to the lower membrane thickness of the sulfone-based membranes (table S1). We note that the S30 membrane shows a lower ASR than F30 and P30 (2.99 and 3.02 ohm·cm² in 1 M KOH, respectively). This is attributed to the additional sulfone groups and higher IEC of SPES, which result in increased water uptake and water channel density, facilitating enhanced ion transport across the membrane. For ionic selectivity, benefiting from the decentralized ion-transport channels, all sulfone-based membranes endow much lower permeability than N117 (Fig. 2G and table S2). We investigated the effect of sulfonation degree on the ionic conductivity and selectivity of the membrane. Compared to S30, S10 shows extremely low ionic conductivity because of the lack of interconnected water channels (fig. S6E). In contrast, S50 shows severe ion permeation (figs. S7 to S9), which could be attributed to the presence of too many hydrophilic sulfonic acid groups, leading to aggregated water clusters. Together, S30 exhibits a comparable ASR and much lower polysulfide permeability compared with N117. Consequently, the ionic selectivity (defined as the ratio of ionic conductivity to ionic permeability) of S30 is more than 20 times higher than that of N117 (fig. S10), making it a promising nonfluorinated ion-selective membrane to replace fluorinated N117. Furthermore, all sulfone-based membranes exhibit a much higher tensile strength compared to N117, while S30 is the most ductile among sulfone-based membranes (Fig. 2H and fig. S11). Thermogravimetric analysis (TGA) and Fourier transform infrared spectroscopy were performed to confirm the thermal (fig. S12) and chemical stability (figs. S13 and S14) of sulfone-based hydrocarbon polymers. The high chemical stability of SPES in S-Fe RFB electrolytes is in strong contrast to the poor stability of SPES in vanadium RFB resulting from the high oxidation potential of pentavalent vanadium ions (38, 39). Considering the performance indicators for ion-selective membranes for RFB applications (40), the S30 membrane shows a low ASR, high ionic selectivity, high thermal/chemical stability, and a balanced mechanical strength, making it a promising candidate as a low-cost membrane for polysulfide-based RFBs.

Mechanism of ion transport and selection in the SPES membrane

We further investigated the differences in ion-transport behaviors among various ions to demonstrate the high ion selectivity of the SPES membrane. Ion transport through ion-transport channels might occur via three mechanisms, including vehicle mechanism, Grotthuss mechanism, and ion exchange mechanism (41). The vehicle mechanism involves the diffusion of hydrated ions through hydrophilic nanochannels, intimately related to the mobility and spatial distribution of water molecules within ion-transport channels. The vehicle mechanism dominates for most ions, except protons (H⁺) and hydroxyl ions (OH⁻), where water molecules serve as carriers, facilitating ion movement via solvation shells (42). The factors, including the hydration level of the membrane, microphase separation, and channel morphology, determine the formation and connectivity of ion-transport channels, affecting the ion diffusion rate via the vehicle mechanism. The Grotthuss mechanism enables rapid H⁺ or OH⁻ transport via hydrogen-bonded water clusters, leveraging proton hopping and water reorientation for increased conductivity (43). In addition, the ion exchange mechanism is counterion hopping between the adjacent ion exchange functional groups, requiring high energy to overcome coordination barriers (44). The presence of ion exchange groups creates an electrostatic potential barrier that restricts the transport of co-ions (e.g., anions in cation exchange membranes), resulting in Donnan exclusion. A smaller channel size and higher surface charge density could amplify co-ion repulsion, further strengthening Donnan exclusion and improving ion selectivity. For K⁺ ions (the major charge carriers in PSFBs) and polysulfide ions (the redox-active species in PSFBs), the vehicle and ion exchange mechanisms are the two major ion-transport pathways in membranes (Fig. 3A).

Fig. 3. Ion conduction mechanism in the SPES membrane.

(A) Schematic illustration of the ion conduction mechanism in Nafion and SPES membranes. (B and C) ¹H NMR T₂ measurements of hydrated (B) N117 and (C) S30 membranes. (D) ¹H NMR cryoporometery measurements of N117 and S30 membranes. (E to G) Calculation of relative energy for the K⁺ ion transport through the (E) SPES, (F) SPSF, and (G) SPPSU polymer matrixes.

States and distribution of water molecules

We first conducted ¹H NMR T₂ measurements to reveal the states and distribution of different water molecules in S30 and N117 (Fig. 3, B and C). The shorter spin-spin relaxation time (T₂) indicates slower water movement and, thus, stronger interaction with the polymer chains. There are three types of water in ion-transport channels, including strongly bound water (time domain of 0 to 10 ms), loosely bound water (time domain of 10 to 200 ms), and free water (time domain of 200 to 1000 ms) (45, 46). Bound water refers to water molecules that interact with sulfonic acid groups or sulfone groups, while free water refers to the nonbound bulk water in the polymer matrix (47, 48). The ¹H NMR T₂ measurement results show that S30 has very limited free water and much more loosely and strongly bound water compared to N117, which could be attributed to strong interactions between hydrophilic sulfone groups with water molecules in smaller channel size resulting from decentralized phase separation in the polymer matrix. Consequently, the reduced mobility of water molecules might decrease the ion diffusion rate (including polysulfide ions) via the vehicle mechanism, consistent with the fact that S30 exhibits lower ionic permeability than N117 (fig. S9).

We further use differential scanning calorimetry (DSC) to investigate the amount of loosely bound water (melting point between −30° and 0°C) and free water (melting point at 0°C) in the polymer matrix (fig. S15). N117 shows a sharp peak at 0°C, while S30 shows a peak at around −20°C, suggesting that S30 has a large amount of loosely bound water but a low fraction of free water, which is consistent with the ¹H NMR T₂ measurements. Compared to S30, the S10 membrane shows no detectable peak in the DSC melting curve, likely due to the predominance of water molecules existing in a strongly bound, nonfreezing state. In contrast, S50 shows a distinct peak at around −3°C, attributed to its high sulfonation degree, which promotes the formation of aggregated ion-transport channels, weakening the interaction between water molecules and sulfonic acid groups. Although the vehicle mechanism primarily relies on the diffusion of ions in a hydrated form within the ion-transport channels, with interactions with water molecules being predominant, bound water might alter the electric field environment within the channels, thereby decreasing the diffusion rate of polysulfide ions via the vehicle mechanism. These observations suggest that ion transport through loosely and strongly bound water plays a crucial role in ion conduction in SPES, which is different from the Nafion membrane where free water dominates (49).

Size exclusion and ion exchange

We conducted ¹H NMR cryoporometery measurements to examine the size of fast ion diffusion regions in S30 and N117. Figure 3D shows that the fast ion diffusion region of S30 (6.9 Å) is smaller than that of N117 (10.4 Å). The sizes of bare polysulfide (S₄²⁻) and potassium ion are ⁓9.0 and ⁓1.4 Å, respectively (fig. S16). On the basis of these findings, we believe that the crossover of hydrated polysulfide ions (diameter of hydrated polysulfide ion: ~10.8 Å) through the ion-transport channels of the SPES membrane is much more restrained via size exclusion compared with N117. We observed that the small channel size in the SPES membrane also influences the transport of K⁺ ions. However, the abundance of ion-transport channels (Fig. 2C) contributes to the maintenance of the rapid transport of K⁺ ions, resulting in a comparable ASR and much higher K⁺/S_x²⁻ ion selectivity than N117. We confirmed that the “decentralized” ion-transport network in the S30 membrane is characterized by a distributed array of high-quantity, smaller channels, as opposed to an aggregated and centralized cluster network in the N117 membrane (Figs. 1 and Figs. 3A).

Not only the channel size but also surface charge and interaction between ions and channel walls could influence the ion selectivity of the membrane. We further conducted zeta potential measurements to characterize the surface charge of N117 and S30 (fig. S17). The S30 membrane shows a higher negative surface charge (−22.3 mV) at pH 10 in a 1 mM KCl solution compared to N117 (−15.1 mV). This stronger negative surface charge in S30, attributed to its higher density of sulfonic acid groups, enhances electrostatic interactions with cations and strengthens the Donnan exclusion of polysulfide ions (13). In addition, SAXS (Fig. 2D) and ¹H NMR cryoporometery (Fig. 3D) analyses confirmed that S30 has a smaller channel size compared to N117. The combination of smaller channel sizes and higher surface densities in S30 could present a stronger electrostatic repulsion strength against polysulfide ions compared to N117, which notably improves ion selectivity.

To understand the influence of the monomer chemical structure on ion transport, we conducted density functional theory (DFT) calculations to study the interactions between water molecules, K⁺ ions, and polymer monomers (fig. S18). SPES shows the strongest interaction with water molecules (lowest binding energy of −29.18 kcal mol⁻¹), which is consistent with the experimental results that SPES has the highest water content in the conduction channel (Fig. 2, A and C) to assist ion transport. We calculate the binding energy of K⁺ ions with different sulfone-based polymers (fig. S18). SPES shows the lowest binding energy (weakest binding) as the presence of negatively charged diphenyl sulfone weakens the negative electrostatic potential of the polymer chain (fig. S19) (50), leading to weakened electrostatic attraction with K⁺ on SPES and a lower energy barrier for K⁺ hopping (51, 52), which is consistent with Fig. 3 (G to F) showing that SPES has the lowest energy barrier for K⁺ ion exchange. Furthermore, in addition to the sulfonic acid groups, SPES having the comonomer of diphenyl sulfone offers additional ion hopping sites for K⁺ ion conduction on the polymer matrix surface (fig. S20).

Comparisons in S-Fe RFBs

We study all membranes in S-Fe RFBs using polysulfide as the negolyte and ferrocyanide as the posolyte (Fig. 4A) and investigate the differences between N117 and S30 for long-term flow battery stability. We first examine their self-discharge behaviors via monitoring the open-circuit voltage (OCV) of the charged S-Fe RFBs [50% state of charge (SOC)]. The self-discharge evaluation was conducted at 50% SOC as the evaluation condition, selected as it falls within the medium SOC range of 40 to 80%, which is representative of typical operating conditions for RFBs in grid energy storage applications (53). Maintaining RFBs within this SOC range optimizes readiness for charge/discharge cycles while minimizing performance degradation, consistent with standard grid operation protocols. As can be seen in Fig. 4B, the OCV of the S-Fe RFB using N117 decayed to nearly 0 V after 150 hours. In contrast, the S30 membrane shows a much lower self-discharge rate (>700 hours), which can be attributed to reduced polysulfide crossover owing to decentralized ion-transport channels. Elevated SOC levels may influence the self-discharge rate of RFBs, particularly in applications requiring prolonged storage at high charge states. We further investigated the self-discharge behaviors of the S-Fe RFB using N117 and S30 membranes at higher SOC levels (e.g., 75 and 100%). N117 shows a notable accelerated self-discharge rate (<120 hours) at elevated SOC compared to 50% SOC. Meanwhile, the S30 membrane maintains a high OCV for more than 500 hours, indicating superior polysulfide selectivity (fig. S21). Rechargeable LIB technologies face an inherent challenge of self-discharge, especially at high operating voltages, resulting in energy losses over time for long-duration energy storage applications (54). Aqueous RFBs leverage decoupled energy and power scaling to provide superior resistance to self-discharge (55). By using highly stable redox-active species, combined with strategies to increase the size of active materials and improve membrane selectivity to minimize crossover, RFBs could achieve robust energy retention and low self-discharge rates (56, 57). These advancements position RFB technologies as a promising alternative to conventional lithium-ion batteries for long-duration, grid-scale energy storage applications.

Fig. 4. Electrochemical measurement and cell performance of pristine membranes.

(A) Schematic illustration and optical photograph of an aqueous S-Fe RFB assembled with an ion-selective membrane. The negolyte is polysulfide, and the posolyte is ferrocyanide. K⁺ ions serve as the primary charge carriers. (B) OCV curves of S30 and N117 membranes. (C and D) Voltage profiles of (C) S30 and (D) N117 membranes during charge-discharge cycling test at a current density of 20 mA cm⁻². mAh, mA·hour. (E) Cell performance of S30 (red) and N117 (blue) membranes for long-term cycling measurement. h, hours.

We investigated all membranes in S-Fe RFBs between 10 and 50 mA cm⁻². All sulfone-based membranes exhibited a much higher CE (>99.5%) than the N117 membrane (CE: 97.1 to 99.2%). In addition, the EE of the S-Fe RFB using the S30 membrane is higher than that using F30 and P30 and comparable to that using N117 (fig. S22). Furthermore, S50 shows higher EE but faster capacity fade than S30 and N117 at high current densities, which can be attributed to aggregated water clusters (figs. S23 to S25).

We compared the voltage profiles of the S-Fe RFBs using S30 and N117 at 20 mA cm⁻² in Fig. 4 (C and D). The S-Fe RFBs using the N117 membrane showed increasing overpotential and fast capacity fading within less than 100 cycles (Fig. 4D). During cycling, soluble polysulfide ion crossover through the membrane to the posolyte, where they are oxidized to form insulating solid sulfur species, which deposit on both the positive electrode and membrane surface. The deposition of solid sulfur increases the charge transfer resistance and obstructs ion-transport channels, contributing to a rise in overpotential. The increased overpotential and electrode/membrane passivation result in the rapid capacity fade. We note that aggregated ion-transport channels of the Nafion membrane lead to severe polysulfide crossover, resulting in the deposition of insulating solid sulfur on the membrane and positive electrode (fig. S26). In contrast, the S-Fe RFBs using the S30 membrane showed high cycling stability with negligible changes on the voltage profile (Fig. 4C). The S-Fe RFBs using S30 can be stably cycling for more than 360 cycles (1100 hours) with a CE > 99.6% and low capacity decay rates of 0.004% per cycle and 0.034% per day (Fig. 4E). Membrane thickness is a critical factor influencing performance indicators such as ionic conductivity, polysulfide crossover, and overall cell efficiency. We also evaluated the performance of S30 membranes (~150 μm thick) in comparison to that of Nafion N115 (~140 μm thick). The S-Fe RFBs using the N115 membrane presented a low CE (~98.4%) and fast capacity decay within 60 cycles compared to those using S30 (fig. S27), suggesting that SPES membranes with decentralized ion-transport channels have improved polysulfide selectivity than commercial Nafion membranes. To examine the potential of the SPES membrane for practice applications, we conducted a S-Fe RFB stack (100 cm²) with two cells in series (fig. S28). The cell stack presented a CE as high as 99.6% and cycled stably for more than 350 cycles at 20 mA cm⁻².

Surface modification of SPES for long-life S-Fe RFBs

Long-term cycling stability is one of the most important indicators of practical applications. While the S-Fe RFBs using the S30 membrane show much improved cycling stability compared to commercial Nafion N117, its CE is still lower than 99.9%, which is required for practical applications. Our previous report showed that the crossover rate of Nafion can be markedly reduced with carbon coating modification (13). We therefore performed carbon coating modification on the S30 membrane (denoted as S30–C) and compared it with the reported carbon-coated N117 (denoted as N117–C) (fig. S29) (13). As shown in Fig. 5 (A to C) and fig. S30, the S-Fe RFBs using the S30–C membrane showed a much lower overpotential than those using the reported N117–C membrane at all current densities from 10 to 50 mA cm⁻². This is attributed to a much lower ASR and higher ionic conductivity of S30–C compared to those of N117–C (fig. S6, C and F). The inverse trend of ionic conductivity between S30–C and N117–C results from a 50% reduction in ionic conductivity of N117 after carbon coating (8.32 mS cm⁻¹ versus 4.16 mS cm⁻¹ in a 1 M KOH aqueous solution). In contrast, the changes in ionic conductivity before and after carbon coating for S30–C are minimum (6.25 mS cm⁻¹ versus 5.79 mS cm⁻¹ in a 1 M KOH aqueous solution). These could be attributed to the much stronger membrane strength of S30 to withstand solvent/slurry penetration during carbon coating (figs. S31 and S32). In the S-Fe RFBs, the choice of an alkaline negolyte and neutral posolyte is driven by the need to optimize the high stability of the redox-active species. Considering the different pH environments of S-Fe RFBs, we investigated the migration of hydroxide ions (OH⁻) through the membrane (in an H-cell) and the variations of pH value during the battery operation (figs. S33 and S34). The carbon-coated SPES membrane (S30–C) shows a mitigating effect on OH⁻ migration across the membrane compared to the pristine SPES membrane (S30), contributing to the high cycling stability of S-Fe RFBs. We acknowledge that elevated temperatures can increase the diffusion rate of polysulfide ions and potentially affect the swelling behaviors and selectivity of membranes. The S-Fe RFBs using S30–C show a high CE > 99.7% for more than 350 cycles (>1000 hours) even at 50°C (fig. S35), which demonstrates that the S30–C membrane effectively mitigates polysulfide crossover, enabling stable battery operation.

Fig. 5. Surface modification to enhance cycling stability.

(A to C) Voltage profiles of modified SPES and Nafion membranes with various current densities from 10 to 50 mA cm⁻². (A) Galvanostatic voltage profiles of the modified Nafion membrane (N117–C). (B) Galvanostatic voltage profiles of the modified SPES membrane (S30–C). (C) Comparison of voltage profiles of N117–C (blue) and S30–C membranes (red) at current densities of 20 mA cm⁻² (solid line) and 50 mA cm⁻² (dotted line). (D) Cell performance of N117–C and S30–C membranes with long-cycle testing. (E) Cell performance of the modified SPES membrane with the “EE recovery” process for 6 months at a current density of 20 mA cm⁻². h, hours.

We examined the long-term cycling stability of the S-Fe RFBs using the S30–C membrane at 20 mA cm⁻². As shown in Fig. 5D, the S-Fe RFBs using the S30–C membrane can stably operate with a high CE > 99.9% and high EE > 77.8% for more than 500 cycles (>1700 hours) with low capacity decay rates of 0.0004% per cycle and 0.0032% per day, which is comparable to those of N117–C. We then explored the use of the “EE recovery” process using an electrolyte to remove the sulfur deposition (fig. S36) and continued cycling the RFBs with the original electrolyte (5, 58, 59). During the first 90 days, we observed that the EE of a cell decreased from 80.2 to 72.9% (capacity decay from 134.4 to 127.1 mA·hour). We believe that this can be attributed to polysulfide oxidation on the positive electrode/membrane, forming elemental/insulating sulfur. Subsequently, the EE of the S-Fe RFBs increased to 78.2% and they continued to cycle for another 90 days (and counting) with an average EE of more than 75% (Fig. 5E). This strategy demonstrated PSFBs with record longevity with high EE at a markedly reduced cost (tables S3 and S4). We further explored the applicability of the SPES-based membrane to other polysulfide-based RFB systems, such as polysulfide-iodide RFBs (fig. S37). Thanks to high polysulfide selectivity, the carbon-coated S30–C membrane shows stable cycling performance (more than 100 cycles) with a high CE (>99.7%) at 20 mA cm⁻², suggesting that the SPES-based ion-selective membrane becomes a promising candidate as a low-cost membrane for polysulfide-based RFBs.

DISCUSSION

We developed low-cost SPES with decentralized ion-transport channels to create high-quantity, smaller ion-transport channels to mitigate polysulfide crossover while maintaining high ionic conductivity. SAXS reveals that all sulfone-based membranes exhibit a much smaller water cluster size than Nafion, confirming our design strategy. ¹H NMR T₂ measurement results showed that SPES has much more loosely bound water (contributing to fast ion transport) and very limited free water (suppressed polysulfide crossover) compared to N117. ¹H NMR cryoporometery results suggest that the crossover of hydrated polysulfide ions (larger than 9.0 Å) through the ion-transport channels of the SPES membrane (6.9 Å) is much more restrained via size exclusion compared with N117 (10.4 Å). Owing to the balanced design of hydrophilic sulfone groups and the rigid aromatic backbone, the SPES membrane demonstrated 20 times higher ionic selectivity (2.06 × 10⁶ S·min cm⁻³) at a markedly reduced cost (USD $12 to $66 m⁻²) compared to the commercial fluorinated N117. The low-cost SPES-based membrane enabled stable cycling of full S-Fe RFBs with an industrially relevant high CE (>99.9%) and EE (average >75%) for 1600 cycles (>6 months) at 20 mA cm⁻². This study demonstrated a rational approach to effectively tackle polysulfide crossover, removing the necessity for costly fluorinated membranes, paving the path for the practical commercialization of sustainable polysulfide RFBs. This strategy could be broadly applicable to other nonfluorinated ion-selective membranes, including ketone-based, triphenylphosphine oxide–based, or imidazole-based systems. By adjusting the type and density of comonomers and functional groups, microphase separation and channel morphology could be optimized, enhancing selectivity for specific ions in separation and purification applications.

MATERIALS AND METHODS

Materials

All the chemicals were used as received. Potassium (poly)sulfide (K₂S_x, ≥42% K₂S basis, where x is determined to be ~2.0) (60), sulfur powder (purified by sublimation, ~100 mesh particle size), N,N-dimethylacetamide (≥99%), 2-propanol (IPA; anhydrous, ≥99.5%), 1-methyl-2-pyrrolidinone (anhydrous, 99.5%), sulfuric acid (H₂SO₄; 95 to 98%), and hydrogen peroxide (H₂O₂; 30 wt % in H₂O) were received from Sigma-Aldrich. Potassium ferrocyanide (K₄[Fe(CN)₆]·3H₂O; ≥99.5%), potassium ferricyanide (K₃[Fe(CN)₆]; ≥99.5%), sodium hydroxide (NaOH; 99.7%), potassium chloride (KCl; 99.8%), potassium iodide (KI; ≥99%), sodium chloride (NaCl; 99.9%), and riboflavin sodium phosphate (FMN-Na; 93%) were received from Aladdin. Polytetrafluoroethylene preparation (PTFE; 60 wt % dispersion in H₂O), potassium hydroxide (KOH; 95%), and phenolphthalein (pH indicator) were received from Macklin. Sulfonated sulfone–based polymers with different degrees of sulfonation, including SPES, SPSF, and SPPSU, were received from Dechi Technology. Polyvinylidene fluoride (HSV900) was received from Arkema. The Nafion membrane (N117 and N115, Dupont) was received from H2FLOW (Shanghai) Advanced Materials International Trade Co., Ltd. Carbon felts (GFD 4.6EA) were received from SGL Carbon SE. Ketjen black (KB; ECP-600JD) was received from Lion Corporation.

Membrane preparation

All the polymer powders were stored in a dry box for desiccation. The membrane was manufactured by the solution-casting method. First, the polymer was dissolved in N,N-dimethylacetamide to form the homogeneous solution and then defoamed overnight. The concentration of casting solution was 30 wt %. Then, the polymer solution was poured onto a clean glass plate and casted by a doctor blade at a speed of 0.12 m min⁻¹ followed by drying at 80°C in a casting coater for 8 hours. Afterward, the membrane was peered off and ion exchanged to K-type by immersing in 5 wt % H₂SO₄ for 1 hour and a 1 M KOH aqueous solution for 2 hours at 80°C sequentially. Last, the membrane was rinsed and stored in deionized (DI) water. The thickness of the wet membrane was ~150 μm. The Nafion membrane was treated with 5 wt % H₂O₂, 5 wt % H₂SO₄, and 1 M KOH sequentially (13).

The modified membrane was prepared by drop coating. For the sulfonated hydrocarbon membrane, IPA, DI water (the volume ratio of IPA:water was 7:3), KB, and diluted PTFE dispersion (5 wt %) (the weight ratio of KB:PTFE was 8:1) were mixed and sonicated in ice water for 20 min to form the uniform slurry. The mass loading of KB was about 0.23 mg cm⁻² on both sides of the membrane. The membrane was heated at 80°C in an oven for 12 hours and soaked in DI water for usage. The Nafion-based modified membrane was prepared using a previously reported method (13).

Characterization

Ion exchange capacity

The IEC was determined by using the titration method (61). First, the H-type dry membrane was suspended in a saturated NaCl aqueous solution for 24 hours to liberate the H⁺ ions. The solution was titrated with a 0.1 M NaOH solution by using phenolphthalein as the pH indicator. The IEC of the polymer was calculated by the following equation

$$\text{IEC}\left(\text{mmol}{\mathrm{g}}^{–1}\right)=\frac{V_{\text{NaOH}}\times C_{\text{NaOH}}}{W_{\text{dry}}}$$ (1)

where V_NaOH is the volume (μl) of NaOH solution when phenolphthalein turns red, and V_dry is the weight (mg) of the dry membrane.

Water/electrolyte uptake and swelling ratio

The K-type sample membrane was fully dried in a vacuum oven at 80°C for 24 hours, named the dehydrated membrane. The dehydrated membrane was immersed in DI water or salt aqueous solution under room temperature for 24 hours, named the hydrated membrane. The water/electrolyte uptake (ω) was determined as the weight ratio of absorbed water to the dehydrated membrane

$$\mathrm{\omega }=\frac{m_{\mathrm{h}}-m_{\mathrm{d}}}{m_{\mathrm{d}}}\times 100$$ (2)

where m_h and m_d are the weights of hydrated and dehydrated membranes, respectively.

The swelling ratio (φ) was calculated as the volume expansion ratio of absorbed water to the dehydrated membrane

$$\mathrm{\phi }=\frac{V_{\mathrm{h}}-V_{\mathrm{d}}}{V_{\mathrm{d}}}\times 100\%$$ (3)

where V_h and V_d are the dimensional sizes of hydrated and dehydrated membranes, respectively. The length, width, and thickness of the membrane were measured using a micrometer screw.

The water channel density (ρ) was defined as the weight change to volume change of the membrane in pure water

$$\mathrm{\rho }=\frac{m_{\mathrm{h}}-m_{\mathrm{d}}}{V_{\mathrm{h}}-V_{\mathrm{d}}}\times 100$$ (4)

Mechanical properties

The mechanical properties of the K-type wet membrane were tested on an electromechanical universal testing machine (INSTRON 3367). The size of the sample membrane was 1 cm by 3 cm. The tensile strength and elongation of the membrane were obtained by the stress-strain curve.

Morphology

The membrane morphology was characterized using a SEM (scanning electron microscopy) Quanta 400F. The membranes were frozen and broken in liquid nitrogen for cross-sectional observation. The samples were spray coated with platinum powder for 60 s in advance. Tapping mode AFM images were obtained by a Bruker Dimension Icon. The scan range was 400 by 400 nm.

SAXS

SAXS was tested on Xenocs Xeuss 2.0 with the detector (Pilatus 3R 300K) and x-ray wavelength of λ = 1.54 Å. The sample detector distance was fixed at 538 mm. The water cluster size was calculated by $d=\raisebox{1ex}{$2\mathrm{\pi }$}\!\left/ \!\raisebox{-1ex}{$q$}\right.$ on the basis of Bragg’s law (51). All the samples of N117 and sulfone-based hydrocarbon membranes were equilibrated in DI water for 24 hours at 25°C before SAXS testing.

PALS

PALS was conducted using a fast-fast coincidence system. A ²²Na source was used with an activity of ~10 μCi. All the membrane samples were immersed in DI water for 24 hours to ensure hydration. After removing surface water, each sample was placed in a custom-designed container and sealed with a Kapton film to form a small chamber. Two identical sample pieces were sandwiched around the ²²Na positron source to ensure uniform positron injection. All spectra were subjected to source correction and analyzed by using the LT program to decompose the positron lifetimes and corresponding intensities. The free volume size in the polymer matrix was determined by the following equation on the basis of the Tao-Eldrup model.

$${\mathrm{\tau }}_{\mathrm{o}-\text{Ps}}=\frac{1}{2}{[1-\frac{R}{R+∆R}+\frac{1}{2\mathrm{\pi }}\text{sin}\left(\frac{2\mathrm{\pi }R}{R+∆R}\right)]}^{-1}$$ (5)

where τ_o-Ps is the o-Ps lifetime in the membrane, R is the radius of the free volume pore, and ΔR = 0.16 nm is the thickness of the homogeneous electron layer overlapping with the o-Ps wave function.

Differential scanning calorimetry

DSC was performed to characterize the amount of freezing water (loosely bound water and free water) in ion-transport channels. The membrane samples were equilibrated in DI water for 24 hours in advance and then sealed in an aluminum pan after quickly wiping off the surface water. During the testing, the samples were cooled from 25° to −60°C and heated to 25°C with the scan rate of 10°C min⁻¹. The weight of the membrane sample was ~4 mg.

Nuclear magnetic resonance

¹H NMR proton spin-spin relaxation time (T₂) measurement was conducted by a MesoMR23-060H-1 with the proton resonance frequency of 22 MHz at room temperature. The samples were immersed in DI water for 24 hours, and the surface water was quickly wiped off before measurements. For ¹H NMR cryoporometery measurements, the samples were soaked in perfluorooctane to prevent water evaporation. The temperature range was −60° to 5°C, and the data were collected every 2°C with a holding time of 5 min.

Zeta potential

The zeta potential of membrane samples was determined using streaming potential measurement conducted with a SurPASS 3 system using a 1.0 mM KCl solution as the standard electrolyte across a pH range of 7 to 10.

Ionic conductivity and permeability

The resistances were measured by performing electrochemical impedance spectroscopy of a static cell assembled on the sample membranes. The frequency range was from 0.2 MHz to 0.1 Hz. The ASR and ionic conductivity (σ) of the membrane were calculated by equations as follows

$$\text{ASR}=A\times (R_{\mathrm{w}}-R_{\mathrm{o}})$$ (6)

$$\mathrm{\sigma }=\frac{L}{A\times (R_{\mathrm{w}}-R_{\mathrm{o}})}$$ (7)

where A and L are the active area and thickness of the membrane, and R_w and R_o are the resistances of the cell with or without a membrane, respectively. The membrane samples were immersed in testing solution for 24 hours in advance.

Ionic permeability was tested using an H-cell sandwiching the membrane sample. For polysulfide ions, the feed side was 1 M K₂S₂ or K₂S₄ in a 1 M KOH solution, while the permeate side was filled with 1 M KOH. For ferrocyanide and ferricyanide ions, the feed and permeate sides were 0.5 M K₄[Fe(CN)₆] or K₃[Fe(CN)₆] in 1 M KCl and 1 M KCl, respectively. The volume of solution was fixed at 8 ml in both sides of half cells. The absorbance of different ions was collected by an ultraviolet-visible spectrophotometer (ASL SEC2020) under room temperature. The permeability coefficient (D) was calculated using the following equation

$$V_{\mathrm{p}}\frac{d{c}_{\mathrm{p}}\left(t\right)}{\mathit{dt}}=\frac{\mathit{AD}}{L}[c_{\mathrm{f}}-c_{\mathrm{p}}\left(t\right)]$$ (8)

where $c_{\mathrm{f}}$ and $c_{\mathrm{p}}\left(t\right)$ are the concentrations of ions in the feed and permeate sides, respectively; $V_{\mathrm{p}}$ is the volume of solution in the permeate side; A is the effective area of ~0.78 cm², and L is the thickness of the membrane sample. The ionic selectivity was defined as the ratio of the ionic conductivity to ionic permeability of the membrane.

For hydroxide ion migration, the feed and permeate sides were 1 M KOH and 1 M KCl aqueous solutions, respectively. The change of pH value was monitored by a Eutech PC700 Meter under room temperature.

Chemical stability of the membrane

The K-type membrane samples were stored in the electrolyte solution for 40 days at 30°C (62, 63), followed by washing out of the electrolyte with DI water and drying at 80°C for 24 hours. Fourier transform infrared spectroscopy (Thermo Fisher Scientific) was performed to assess the chemical stability of the membrane at the range of 500 to 4000 cm⁻¹. TGA was conducted by Mettler TGA2 from 30° to 800°C to confirm the thermal stability of the membrane.

Cell assembly and testing

The flow cell consisted of a membrane, pretreated graphite felts (geometric area of 2 cm by 2 cm), poly(ether ether ketone) frames, graphite plates with flow fields, and current collectors. For the S-Fe RFB, the posolyte was prepared using 0.5 M K₄[Fe(CN)₆] with 1 M KCl as the supporting electrolyte, while the negolyte was prepared using 1 M K₂S₄ in a 1 M KOH solution with 50 mM FMN-Na as the molecular catalyst (11). All the electrolytes were dissolved in deoxygenated DI water under argon protection, and the volume of solution was 10 ml. The flow cell was tested in an atmospheric environment with the flow rate of 50 ml min⁻¹. For flow cell testing of pristine membrane (without surface coatings), a glass fiber was sandwiched between the membrane and the positive electrode.

Charge-discharge measurements were collected by the LAND Battery Testing System. The cut voltage was set as 1.3 V (10 to 30 mA cm⁻²), 1.4 V (40 mA cm⁻²), or 1.5 V (50 mA cm⁻²) for the charging procedure and 0.2 V for the discharging procedure. Long-cycle testing was conducted with a current density of 20 mA cm⁻² and cut voltage of 1.3 to 0.2 V. After 3000 hours, the cut voltage was changed to 1.2 to 0.2 V to avoid the side reaction. The volume of electrolytes of the 100 cm² cell stack with two cells in series was 200 ml. The stack was precycled for 10 cycles and tested at 20 mA cm⁻² with the flow rate of 200 ml min⁻¹. OCV measurements were conducted with a PSFB static cell (fig. S5A) under room temperature.

For the “EE recovery” process (64), the original posolyte and negolyte were collected and stored in the glove box. The electrolytes were replaced with 10 ml of 1 M K₂S₂ in a 1 M KOH aqueous solution on both sides and circulated by peristaltic pumps for 6 hours. The polysulfide aqueous solution can redissolve the insulating solid sulfur, which forms on the posolyte-side electrode, as shown in the following equation

$${\mathrm{S}}_n^{2-}+\mathrm{S}\to {\mathrm{S}}_{n+1}^{2-}(n=1 \text{to} 7)$$ (9)

Afterward, the electrodes were flushed with DI water to remove the electrolyte. During the process, the cell was not disassembled. The cycling test using the original electrolytes continued to measure after the “EE recovery” process.

For the polysulfide-iodide RFB, the posolyte was prepared using 1 M KI with 1 M KCl as the supporting electrolyte. The battery charging process was controlled with a cutoff voltage of 1.6 V or a capacity limit of 89 mA·hour (50% SOC of 10 ml of 1 M KI in 1 M KCl). The discharge process was terminated at a cutoff voltage of 0.2 V.

Computational methods

The Gaussian version 09 software package (65) was used to optimize monomers and hydrated ions and calculate the binding energy between the monomers with/without a sulfonic acid group and the water molecule at the B3LYP/6-311+G** level, along with Grimme’s DFT-D3 empirical dispersion correction (66). The true local or global minima of these structures were ensured by the absence of imaginary vibrational frequencies. The open-source software Multiwfn was used to analyze the wave function of the corresponding structure to obtain the electrostatic potential isosurface (67), which was displayed through VMD software (68).

For the kinetic diffusion behavior of K⁺ ions on the monomer surface, the climbing image nudged elastic band method was used to calculate diffusion energy barriers on the basis of the Vienna ab initio simulation package (69). Ion-electron interactions were assessed using the projector augmented wave method. The exchange-correlation functional was elucidated through the generalized gradient approximation based on Perdew-Burke-Ernzerhof (70). All computations used a plane-wave cutoff energy of 450 eV, ensuring a convergence criterion for atomic forces of 0.02 eV/Å and setting an energy convergence tolerance at 10⁻⁵ eV. Grimme’s DFT-D3 correction was implemented to consider long-range van der Waals interactions. To prevent interactions among periodic images, a vacuum layer exceeding 10 Å was established in the a-direction.

Supplementary Materials

This PDF file includes:

Figs. S1 to S37

Tables S1 to S4

References