Fuel cells with an operational range of –20 °C to 200 °C enabled by phosphoric acid-doped intrinsically ultramicroporous membranes

Conventional proton exchange membrane fuel cells (PEMFCs) operate within narrow temperature ranges. Typically, they are run at either 80‒90 °C using fully humidified perfluorosulfonic acid membranes, or at 140‒180 °C using non-humidified phosphoric acid (PA)-doped membranes, to avoid water condensation-induced PA leaching. However, the ability to function over a broader range of temperature and humidity could simplify heat and water management, thus reducing costs. Here we present PA-doped intrinsically ultramicroporous membranes constructed from rigid, high free volume, Tröger’s base-derived polymers, which allow operation from −20 to 200 °C. Membranes with an average ultramicropore radius of 3.3 Å show a syphoning effect that allows high retention of PA even under highly humidified conditions and present more than three orders of magnitude higher proton conductivity retention than conventional dense PA-doped polybenzimidazole membranes. The resulting PA-doped PEMFCs display 95% peak power density retention after 150 start-up/shut-down cycles at 15 °C and can accomplish over 100 cycles, even at −20 °C. Most proton exchange membrane fuel cells are designed to operate within a temperature range of a few tens of degrees, but functioning in a broader range of conditions could be advantageous. Here the authors use ultramicroporous, phosphoric acid-doped membranes that allow fuel cell operation from −20 °C to 200 °C.

P roton exchange membrane fuel cells (PEMFCs) have been applied in numerous portable and stationary applications due to their high power density and minimal pollution 1 . The two conventional types of hydrogen PEMFCs, low-temperature PEMFCs using perfluorosulfonic acid (PFSA) polyelectrolytes (for example, Nafion) and high-temperature PEMFCs using phosphoric acid (PA)-doped polyelectrolytes (for example, polybenzimidazole (PBI/PA)), each have their advantages and limitations. PFSA-based PEMFCs have been commercialized for vehicles, as they have excellent proton conductivity under fully humidified conditions at relatively low temperatures (~80 °C). As relative humidity (RH) decreases, sulfonated proton exchange membranes (PEMs) become much less conductive and eventually dehydrate. In commercial PEMFC usage for vehicles using conventional PEMs, water and heat management, such as the need for large radiators to dissipate waste heat, are required 2,3 . In contrast, acid-base-type PEMFCs operate at temperatures above 140 °C and typically do not require humidification or heat management systems. A higher operating temperature enhances electrocatalyst reactivity and increases tolerance to CO or H 2 S contaminants in hydrogen inlet streams (>1% CO at 150 °C) [4][5][6][7] . Although PEMFCs based on PBI/PA membranes can operate reliably at 160 °C without additional humidification for over 27,000 hours, the water-soluble PA doped into the membrane leaches out when humidity is present under cold start-up or/and frequent start-up/shut-down cycles, which limits the fuel cell function below 140 °C (refs. [8][9][10][11][12][13][14]. Thus, broadening the operation temperature and RH windows and further achieving subzero start-up capabilities become the critical challenges for commercialization of high-temperature (HT) PEMFC systems [15][16][17][18] .
At present, common strategies to improve cold start-up performance depend on engineering methods and external assistance to manage water during shut-down and start-up, such as using a three-dimensional (3D) fine mesh cathode flow channel, introducing gas purging procedures, installing an additional internal/external heating device or including an alternative hydrogen pump 16,[18][19][20] . However, improving PEM proton conductivity below 0 °C, which would seem to be a straightforward strategy to solve cold start-up challenges, has rarely been applied in low-temperature PEMFCs. Since PA-doped PEMs do not rely on water to transport protons, it is reasonable to hypothesize that PA-doped PEMFCs could start-up and operate below 0 °C if a solution to the intractable problem of how to prevent PA leaching under low-temperature conditions is found.
The key to PA leaching with water absorption/desorption in HT PEMs is believed to be because of weaker interactions of PA in the PBI matrix. To address this, the introduction of stronger basicity to form quaternary ammonium-biphosphate ion pairs (QAPOH) has resulted in strong PA interactions in HT PEMs [21][22][23][24] . This strategy of strong ion pair interactions decreased PEM PA loss in the presence of water, and the resulting PEMFCs exhibited stable performance at 80 °C, which was a remarkably lower temperature than conventional PA-doped PBI PEMFCs. However, these QAPOH-based PEMFCs have not been explored to operate at temperatures lower than 80 °C, nor at subzero temperatures, without suffering the loss of PA.
Here, we demonstrate that HT PEMFCs based on intrinsically microporous polymers strongly mitigate PA leaching, which is one of the major limitations of conventional PBI-based fuel cells.

Fuel cells with an operational range of -20 °C to 200 °C enabled by phosphoric acid-doped intrinsically ultramicroporous membranes
In the tuneable subnanometre-level intrinsic microporosity of polymers derived from Tröger's base (TB), a V-shaped bridged bicyclic diamine [25][26][27][28][29][30][31][32] , both the syphoning effect of microporosity and the acid-base interactions with PA are proposed to synergistically support high PA retention and efficient proton conduction in PEMFCs. These improved properties enable PEMFC operation under a wide range of conditions at temperatures from -20 °C to 200 °C without external humidification, allowing multiple start-up and shut-down cycles, which is a substantial improvement over existing technologies.

Polymer intrinsic microporosity
Four high molecular weight TB-based polymers (that is, dimethyldiphenylmethane (DMDPM-TB), dimethylbiphenyl (DMBP-TB), trimethylphenylindan (TMPI-TB) and triptycene (Trip-TB) were prepared (Fig. 1a, Supplementary Fig. 1 and Supplementary Table  1). The four polymers were designed to elicit variations in the intrinsic micropore size and distribution through differences in steric hindrance, rotational flexibility and packing efficiency of the polymer chains 33,34 . Microporosity characterization employed positron annihilation lifetime spectroscopy (PALS), which is based on lifetime measurements of γ-ray-generated positronium located in material free volumes 35 . The peak-normalized PALS spectra of the four TB films covering a broad channel number from around 1,000 to 2,400 were analysed by both PATFIT and CONTIN programs (Fig. 1b, Supplementary Fig. 2 and Supplementary Table 2) 36,37 . As shown in Table 1, the average radius (R 3 ) of the spherical pores correlating to the triplet positronium (o-Ps) third mean lifetime (τ 3 ), their corresponding relative intensities (I 3 ) and the corresponding relative fractional free volume (FFV 3 ) within the films were determined by semi-empirical equations using the finite-term lifetime analysis PATFIT program 37 . The three DMDPM-TB, DMBP-TB and TMPI-TB membranes exhibited one kind of ultra-microcavity, with R 3 < 3.5 Å, while the Trip-TB had two o-Ps components indicating two kinds of microcavities (τ 3 /I 3 /R 3 /FFV 3 and τ 4 /I 4 /R 4 /FFV 4 correlating to the third and fourth mean lifetime, respectively), which is probably due to the internal free-volume elements of the triptycene structures disrupting the polymer chain packing. The small-sized microcavities in Trip-TB have an average radius of 2.8 Å (R 3 ), which is dominant by the annihilation intensity, although the FFV 3 is less than half of FFV 4 . Among the four TB-based films, DMBP-TB and Trip-TB have the largest overall FFV, and a 3D modelled structure of DMBP-TB in an amorphous cell is also provided in Fig. 1c.
We were also curious about the microporous structures of TB-based polymers after doping with PA. However, the PALS technique cannot be applied to analyse PA-doped membranes due to contamination of the 22 Na source by PA. Therefore, we used the apparent Brunauer-Emmett-Teller (BET) surface areas (S BET ) from N 2 adsorption/desorption isotherms collected at 77 K to characterize the microporosity both before and after PA doping (Table 1 and Supplementary Fig. 3; see Supplementary Table 3 for detailed PA-swelling properties). However, it is noteworthy that gas sorption measurements cannot completely elucidate the critical ultramicropore size distribution responsible for molecular sieving 38 . DMDPM-TB, which has the most polymer chain flexibility amongst the four TB-based polymers, was soluble in 85% PA, so that its BET measurement after PA doping could not be performed and the membrane was unsuitable for PA-doped PEM applications. For the DMBP-TB and TMPI-TB membranes, the values of S BET decreased to almost 0 after PA doping, suggesting that the micropores were filled with PA. More interestingly, the Trip-TB showed a reduction in S BET surface area from 868 to 275 m 2 g −1 , suggesting that 32% of the microporous volume was unoccupied. This is in good agreement with the PALS analysis, as the FFV of the smaller porosity (R 3 = 2.8 Å, FFV 3 ) was 31% of the overall FFV of Trip-TB (Table 1). This probably indicates that PA could more readily occupy bigger pores R 4 (4.4 Å), but was less able to occupy the smaller pores R 3 (2.8 Å) due to backbone rigidity. Besides microporosity, we found that the rigid bicyclic tertiary amine centres in TB-based membranes appear to be critical motifs for PA absorption, as we observed little PA uptake for the prototypical intrinsic microporous membrane PIM-1 (R = 4.1 and 2.7 Å, PA uptake = 5%; Supplementary Fig. 4) 25 .

Syphoning effect of ultramicroporous membranes
The membranes were treated with either 85 wt% PA (Fig. 2a) or pure PA (Fig. 2b) and analysed by 31 P NMR to confirm the syphoning effect and illustrate the impact of water. In both cases, m-PBI/ PA without intrinsic micropores showed a broad signal around 0 ppm, indicating the presence of bulk free PA. All TB-based membranes showed more upfield 31 P NMR chemical shifts of 31 P NMR, and the most upfield chemical shifts were observed for DMBP-TB/ PA membranes. The more upfield chemical shifts compared with free PA are probably the result of the ring current shielding effect from the delocalized aromatic ring electron systems in these polymers [39][40][41][42][43][44] . The pore-size dependent negative 31 P NMR chemical shift trend is consistent with a recent report of PA distribution in porous electrocatalysts 44 , further confirming the location of PA molecules in ultramicropores. Comparing Fig. 2a,b, the absence of water made the 31 P NMR chemical shifts move more upfield, which is in contrast to the hydrogen-bonding model, in which the hydrogen bonding between water and PA leads to an upfield shift in 31 P NMR 45,46 .
Despite m-PBI/PA having the highest water uptake (Supplementary Table 3), it shows the most downfield chemical shift. Thus, it is reasonable to conclude that the effect of ultramicropore shielding is more profound than hydrogen bonding for influencing 31 P NMR chemical shifts. The reduced average distance between the micropore wall and PA in DMBP-TB membranes indicates the strongest spatial interactions with PA. Besides solid-state 31 P NMR to study the PA-saturated samples, we also adjusted the acid doping level (ADL) to investigate acidbase interactions by solution-state 31 P NMR analysis in DMSO-d 6 ( Supplementary Fig. 6). When the tertiary amine in TB is protonated by PA to form N-H + ⋯H 2 PO 4 − interactions, the phosphorous signal would have a downfield shift due to the electron-withdrawing effect 23 . As a result, when ADL = 2, PA fully protonated the TB units and downfield chemical shifts were observed compared with free PA, except for Trip-TB, which might be a result of greater steric hindrance and a more rigid polymer backbone. Both Mulliken and natural population analysis suggest that the DMBP-TB unit has the strongest Lewis basicity (Supplementary Fig. 5 and Supplementary Data 1), which is consistent with its highest 31 P NMR chemical shift with 2 equivalents of PA ( Supplementary Fig. 6). Interestingly, when the samples were saturated by 85 wt% PA, the solution-state 31 P NMR showed the same negative chemical shift trend as the solid-state 31 P NMR analysis ( Fig. 2a and Supplementary Fig. 6). Thus, it is hypothesized that both the capillary force in spatially confined micropores and the acid-base interactions play critical roles in retaining PA when water is present 31,32 . Therefore, both the microporosity evaluation and the 31 P NMR analysis suggest that a large fractional free volume (FFV 3 ) resulting from appropriate micropores (R 3 ≈ 3.3 Å) is beneficial for PA absorption and retention.
To further confirm the syphoning effect of microporosity to PA, the apparent BET surface areas of the TB membranes were obtained after soaking the TB/PA membranes three times in   distilled water (each time for 12 h) at 40 °C to remove the doped PA. As shown in Table 1, the apparent BET surface areas of DMBP-TB and TMPI-TB membranes were only 16 and 15 m² g −1 respectively after the PA removal process. These values are close to those of the pristine PA-doped DMBP-TB and TMPI-TB membranes. In contrast, the Trip-TB membrane, having the larger pore size, showed obvious BET recovery after PA removal. These results indicate that the intrinsic micropores having an average ultramicropore radius of 3.3-3.4 Å showed the strongest syphoning interaction with PA. Thus, the DMBP-TB membrane displayed the largest ADL retention value of 72.5%, which is much better than that of the m-PBI and Trip-TB membranes, as shown in Table 1.

Proton conductivity and Pa retention
The proton conductivities of PA-doped membranes were measured in-plane over the temperature range of -30 °C to 180 °C without humidification (Fig. 3a). DMBP-TB exhibited the highest conductivity due to its highest PA uptake, while TMPI-TB/PA and Trip-TB/PA membranes also showed higher or equivalent proton conductivities than m-PBI/PA, despite their lower PA uptake and ADL (as shown in Table 1). This suggests that the intrinsically microporous structures in TB-based membranes probably facilitate proton transport. The highest proton conductivity of 159 mS cm -1 was achieved at 180 °C for DMBP-TB membranes. More importantly, we observed that the low-temperature proton conductivities (from -30 °C to 0 °C) of TB-based PEMs were superior, and these properties are highly relevant to low-temperature fuel cells and cold start-up 15,47 . We propose that the PA occupying the intrinsic micropores is responsible for higher proton conductivities at lower temperatures. In the case of the DMBP-TB/PA membrane, the phenomenon was pronounced and showed one order of magnitude higher proton conductivity (10 mS cm -1 ) than m-PBI/PA (less than 1 mS cm -1 ) (Fig. 3a) at -30 °C. Additionally, we observed a similar trend for through-plane conductivities, which provides a more appropriate estimation of membrane performance in practical fuel cells ( Supplementary Fig. 7) 48 . As shown in Supplementary Fig. 8, both in-plane and through-plane conductivities fit linearly within the Arrhenius plot from room temperature to 180 °C. , Changes in in-plane proton conductivity at 80 °C measured at different RH by ramping sequentially from 5% to 90% with ~10% RH increments for PA-doped PEMs. After 90% RH, the RH was reduced to 5% RH, and equilibrated for 14 h before the next cycle began. c, Changes in in-plane proton conductivity at 80 °C in the 5th RH cycle. d-f, Conductivity and PA loss for PA-doped membranes as a function of time at 40 °C/60% RH (d), 80 °C/40% RH (e) and 160 °C/0% RH (f). In d-f, solid symbols refer to conductivity retention (the left axis) and hollow symbols refer to PA loss (the right axis). Error bars in a and c represent the standard deviation for three separate measurements.
Since PA loss under humidified conditions is one of the major limitations for conventional m-PBI/PA PEMFCs, we herein used two test protocols to compare TB-based membranes and m-PBI to elucidate the impact of ultramicroporous structures on PA retention properties. First, RH cycling experiments from 5% to 90% RH at 80 °C were performed sequentially (Fig. 3b,c) 21 . The samples were equilibrated at 5% RH for 14 h before the beginning of the next RH cycle, and each RH cycle took a total of 34 h. The proton conductivities of PA-doped PEMs gradually increased with RH in all the runs 40 and m-PBI/PA achieved the highest proton conductivity at 90% RH (576 mS cm −1 ) on the first run. However, when the second cycle started from 5% RH, an 80% reduction in proton conductivity for the m-PBI/PA membrane was observed, and the ADL value decreased from 10.5 to 4.6. In subsequent runs, under low RH conditions, the conductivity of the m-PBI/PA membrane was reduced by nearly three orders of magnitude due to excessive loss of PA, which is consistent with previous findings 21 . In contrast, for the DMBP-TB/PA membrane with ultramicroporosity, the proton conductivity progressively decreased, reaching 60% during the first five RH cycles. For the subsequent sixth and seventh runs, the conductivity exhibited little change and stabilized, implying almost no PA loss occurred during these two RH cycles. The ADL of the DMBP-TB/PA membrane stabilized at about 7.5, that is, only a 36% loss of the original PA content. This is slightly better than the QAPOH reported by Kim and coworkers (40% PA loss after five cycles under similar conditions) 21 , which, to date, is one of the best PEMs having the highest PA retention properties. It is noteworthy that TMPI-TB/PA could not survive after the RH cycling experiments because it dissolved, and Trip-TB/PA showed almost no proton conductivity under low humidity (RH < 20%) and only 14.1% remaining PA after the first RH cycle. These results indicate that suitable backbone rigidity, functional group Lewis basicity and ultramicroporosity are all crucial aspects that determine the membrane PA retention properties.
Holding the PA-doped membranes at specific temperatures and RH conditions over a long period of time is another approach to evaluate PA retention properties 21 . We selected conditions of 40 °C/60% RH and 80 °C/40% RH to determine the extent of low-temperature water condensation-induced PA loss, and conditions of 160 °C/0% RH to study PA evaporation. As shown in Fig. 3d-f, the PA-doped intrinsically microporous TB PEMs showed higher PA retention than the conventional m-PBI/PA membrane, no matter how harsh the conditions were. Amongst the TB/PA membranes, DMBP-TB with an appropriate ultramicropore size and the largest FFV 3 exhibited the highest PA retention. For example, under 160 °C/0% RH conditions for 85 h, the loss of PA was only 10% for DMBP-TB/PA, while it was about 15% and 25% for Trip-TB/PA and m-PBI/PA membranes, respectively (Fig. 3f). After 150 h exposure in 80 °C/40% RH, the PA content of m-PBI/PA was only 60% of the original value, while the DMBP-TB/PA membrane still retained at least 89% of the PA (Fig. 3e). At lower temperature and higher RH conditions (40 °C/60% RH), the PA loss of m-PBI/PA membranes became more serious. Both the proton conductivity and PA content of m-PBI/PA dropped by over 50% after the 100-hour evaluation period, while those of DMBP-TB decreased by less than 20% and those of TMPI-TB/PA and Trip-TB/PA membranes decreased by 35-45% (Fig. 3d). The substantially improved performance of DMBP-TB/PA, which is comparable to QAPOH 21 , further verifies that the ultramicroporous TB/PA membranes with subnanometre cavities of appropriate micropore sizes (R 3 ) can mitigate PA loss and enable operation over a full range of temperatures with enhanced water tolerance.

H 2 /O 2 fuel cell performance
Membrane electrode assemblies (MEAs) using PA-doped TB and m-PBI/PA membranes were prepared with carbon-supported Pt catalysts and polytetrafluoroethylene (PTFE) emulsion as the binder. We first evaluated different membranes at 160 °C without external humidification or backpressure, which are typical conditions employed for evaluating PA-doped membranes 49 . As shown in Fig. 4a, although the O 2 reduction reaction (ORR) activity is still lower in the kinetic region, probably due to Pt catalyst poisoning resulting from phosphate adsorption and/or the utilization of non-proton-conductive PTFE binder 50 , the DMBP-TB/PA-based MEA achieved the best fuel cell performance, with a peak power density of 815 mW cm −2 at 160 °C, which is more than twice that of the MEA prepared from m-PBI/PA membrane (397 mW cm −2 , see Supplementary Table 4 for a comparison of representative high-temperature PEMFC performance reported in the literature). Similar behaviour was also observed for the Trip-TB/PA MEA, which demonstrated better fuel cell performance than m-PBI/PA, despite its lower PA uptake. The m-PBI/PA MEA showed the lowest performance, due to its highest cell high-frequency resistance (HFR), which increased from 224 mΩ cm 2 to 254 mΩ cm 2 during the test (Fig. 4a). The trend in peak power densities of these MEAs is consistent with their membrane proton conductivities, as shown in Fig. 3a and Supplementary Fig. 7.
Owing to the higher proton conductivity for DMBP-TB/PA membranes even at -30 °C, we then decided to extend the MEA operating temperature range without external humidification or backpressure. Surprisingly, the DMBP-TB/PA-based MEA can easily start up as low as -20 °C and can operate at a peak power density of 83 mW cm -2 (Fig. 4b,c), while the m-PBI/PA-based MEA failed to do so due to its considerably lower proton conductivity ( Supplementary Fig. 9). As the operating temperature increased from -20 °C to 160 °C, the peak power density of the DMBP-TB MEA continued to rise and the HFR decreased from 5.0 Ω cm 2 to 0.23 Ω cm 2 , although no performance improvement and HFR reduction were observed when further increasing the cell temperature to 200 °C. The MEA derived from DMBP-TB/PA showed over 200 mW cm -2 of peak power density when operating at 40 °C, while the m-PBI/PA-based MEA showed less than 100 mW cm -2 under the same conditions (Fig. 4d). As shown in Supplementary Fig. 10, the H 2 /air cell performance at 160 °C, 80 °C and 40 °C showed the same trend. Therefore, the DMBP-TB/PA-based MEA as a single set-up can operate smoothly from -20 °C to 200 °C, which is one of the broadest PEMFC operating temperature ranges achieved so far.
Low-temperature durability tests for DMBP-TB/PA-and m-PBI/ PA-based MEAs were performed at 40 °C with H 2 /O 2 and without external humidification (Fig. 5a). The cell voltage of the m-PBI/PA MEA decreased from 0.4 V to 0.18 V within only 4 hours due to the dramatic loss of PA ( Supplementary Fig. 11). After the durability test, the membrane was wiped to remove any leached PA and the residual ADL was determined by acid-base titration and found to be reduced to 3.0, which corresponds to only 30% PA retention. In contrast, the DMBP-TB/PA MEA showed no obvious cell voltage degradation and only 5% PA loss over a period of 150 hours.
On the basis of these results, we then evaluated the durability of DMBP-TB/PA MEAs using a more demanding accelerated stress test (AST) to investigate the PA leaching under low-temperature fuel cell operating conditions 10,51 . The AST consists of shut-down and start-up cycling, as well as a step-V procedure from 1 V to 0.1 V with a scan rate of 25 mV min −1 without backpressure or external humidification (Fig. 5b,c). For the DMBP-TB/PA MEAs, the AST could be performed for over 150 cycles at both 40 °C and 15 °C, while the m-PBI/PA failed after only two cycles at 40 °C ( Supplementary Figs. 12-14). Exceptional stability was observed for the DMBP-TB/PA MEA at 15 °C, which started with 199.2 mW cm -2 peak power density and an ADL of 10.7, and after 150 aggressive start-up/shut-down cycles ended with 189 mW cm -2 peak power density (95% retention) and an ADL of 9.2 (86% retention) (Fig. 5b). Even at -20 °C, the AST could be carried out for more than 100 cycles, as shown in Fig. 5c and Supplementary Fig. 14. This good performance for PA-doped, non-humidified MEAs at low temperatures further supports the enhanced PA capillary retention and cold start-up capability of the intrinsically ultramicroporous TB membranes.

Conclusion
In summary, four TB-based polymers with tuneable intrinsic ultramicropores were synthesized as PA-doped PEMs. Fuel cells based on intrinsic microporous TB/PA PEMs were shown to be more tolerant towards water condensation owing to the delocalized effect and acid-base interactions, enabling higher PA retention and MEAs operating at much lower temperatures than conventional PA-doped m-PBI PEMFCs. More specifically, the DMBP-TB/PA membranes, which had optimized ultramicroporous structure (R ≈ 3.3 Å) and large free volume, exhibited impressive proton conductivity and PA retention compared with other TB/PA and m-PBI/PA membranes. A peak power density of 815 mW cm -2 was achieved for the DMBP-TB/ PA MEA at 160 °C, which was twice that of the m-PBI/PA MEA. Moreover, the substantially improved properties of DMBP-TB/PA enabled the MEA to operate across a broad temperature range from -20 °C to 200 °C, although the cell performance in the low-current region may be further improved by inhibiting phosphate anion adsorption on the Pt catalyst surface. The low-temperature performance of DMBP-TB/PA MEAs was remarkably better than m-PBI/ PA MEAs, and also outperformed PFSA-based PEMFCs, due to the non-humidified conditions and easy subzero start-up and operation. The ultramicroporous TB-based membranes studied herein not only provide new strategies to solve low-temperature PEMFC operation and cold start-up issues, but also break through the classical definition of low-temperature and high-temperature PEMFCs. Especially for different electrolytes, such as proton conductive ionic liquids 52,53 , the pore size and distribution and the functionality of the microporous structures (acidophilicity and absorption for electrolyte) can be carefully tuned to ensure electrolyte retention and to enhance cell performance. Thus, we believe this approach has the potential to expand PEMFC application opportunities under a broad range of conditions.

Synthesis of TB polymers.
For this study, four Tröger's base polymers (that is, DMDPM-TB, DMBP-TB, TMPI-TB and Trip-TB) were prepared through acid-promoted polycondensation reactions between commercially available aromatic diamine species and dimethoxymethane according to the literature procedures 33,34 . A mixture of aromatic diamine (20.0 mmol) and dimethoxymethane (100 mmol) was added to a 250 ml three-necked round-bottomed flask equipped with mechanical stirring, followed by cooling to 0 °C in an ice-water bath. TFA (800 mmol) was added dropwise through a constant-pressure funnel at 0 °C over a 4 h period, then the reaction mixture was continuously stirred at 0 °C for 24 h. After warming to ambient temperature, the reaction was allowed stirred for another 96 h. The resulting solution was precipitated in aqueous ammonium hydroxide solution, and the precipitate was filtered, followed by washing with excess deionized water. The dried residue was dissolved in chloroform and precipitated in methanol, and this purification process was repeated twice. Finally, the polymer was obtained after being thoroughly dried in a vacuum oven at 50 °C for 12 h. The NMR spectra of the resultant TB polymers matched well with previous reports, and their weight average molecular weight (M w ), number-average molecular weight (M n ) and polydispersity (Đ) were determined using a gel permeation chromatography instrument equipped with a Waters 1515 isocratic HPLC pump and Waters 2414 refractive index detector (Supplementary Table 1). Polystyrene standards were used for calibration and the flow rate of the mobile phase (HPLC-grade tetrahydrofuran) was 0.6 ml min -1 .
Membrane preparation. The TB polymers were dissolved in CHCl 3 (10 wt%) at room temperature. The degassed solution was cast on a glass plate and dried at room temperature for 24 h. The obtained membrane was immersed in deionized water to peel off the plate and dried in a vacuum oven at 50 °C for 12 h. All the undoped membranes had a thickness of about 80 μm.
The m-PBI M w ≈ 50 kDa, inherent viscosity of 0.88 dl g −1 , 0.5 g per 10 ml 96% H 2 SO 4 at 30 °C), prepared according to a literature procedure 54 , was dissolved in NMP to give a 5 wt% solution after filtering. Then, the solution was cast on a glass substrate and the solvent was evaporated at 80 °C for 12 h in a closed oven. The obtained membrane was peeled off the substrate, and then boiled in deionized water for 12 h to remove trace amounts of NMP. Finally, the m-PBI membranes were obtained with a thickness of about 60 μm, after drying in a vacuum oven at 120 °C.
PA doping. The PA-doped membranes were obtained by immersing TB membranes in PA solution at 20 °C for 0.5-8 h until their weight reached a constant value, followed by wiping with filter papers and drying in a vacuum oven at 60 °C for 2 h. All the membranes increased in thickness after PA doping. The thickness of PA-doped TB membranes was about 160 μm, while that of PA-doped m-PBI was about 100 μm. The PA-doping level and the volume expansion measurements of the membranes were carried out in triplicate, independently, with different pieces of membranes, to examine the reproducibility. The PA uptake and the PA-doping level were determined using acid-base titration 55 . In an ice-water bath, 0.100 M NaOH solution was used to titrate PA-doped PEM samples (approximately 1 cm × 4 cm), using methyl orange as an indicator. After being neutralized, the samples were weighed (W dry ) after being thoroughly washed with deionized water and dried in a vacuum oven at 100 °C for 4 h. The PA uptake was calculated from the following equation: where V NaOH (litres) is the volume of NaOH used, C NaOH (mol l −1 ) is the molar concentration of NaOH and Equiv mol represents the equivalent mole of titrant for PA (in this case Equiv mol = 1), W dry (g) is the dry polymer weight and 98.0 (g mol −1 ) is the molecular weight of PA. The PA-doping level per repeat unit (ADL) was calculated on the basis of the following equation: where M W (g mol −1 ) is the molecular weight of the polymer repeat unit. The thickness, the width and the length of the membranes were measured before and after PA doping. Then, the swelling ratio (SR) was calculated using the following equations: where T dry , L dry , and T wet , L wet , V wet are the thickness, length and volume of the membranes before and after PA doping, respectively.
PA removal procedure. The PA-saturated membranes with ADL determined were immersed in distilled water at 40 °C for 12 h, and then the same procedure was repeated three times. The remaining ADL values were determined again by acidbase titration using methyl orange as an indicator. After durability testing, the membrane was wiped with filter paper to remove the leached PA. Then, the content of residual acid (ADL) in the membrane was determined again by acid-base titration using methyl orange as an indicator.
Microporosity characterization. PALS was employed to determine the microcavity size and free volume in the membranes. A positron can form a neutral atom called a positronium (Ps) after picking up an electron 35 . A Ps has two states: the para-Ps (p-Ps) and the ortho-Ps (o-Ps). An o-Ps atom picks up an electron to annihilate into two photons during the collisions with the walls of free-volume cavities. In polymers, the lifetime for o-Ps pick-off annihilation usually ranges from 1 ns to 10 ns, which is determined by the size of the free-volume cavities, and its intensity is proportional to the fraction of the free-volume cavities.
The tests were performed as follows: a 22 Na positron source (~10 μCi) was sandwiched between two identical membrane samples with dimensions of 1.5 mm × 10 mm × 10 mm. Within a few picoseconds, a 1.28 MeV γ-ray was emitted by the 22 Na nucleus simultaneously with a positron. The time delay between the emission of the γ-ray (1.28 MeV) and one of the 0.511 MeV annihilation photons determines the positron lifetime. A fast-fast coincidence system served to conduct the lifetime measurements with a time resolution of ∼220 ps and a channel width of 12.6 ps. The lifetime spectra were analysed using the data-processing programs PATFIT and CONTIN.
The average radius (R) of the spherical pores correlating to the o-Ps lifetime (τ), their relative intensities (I) and the overall relative FFV within the membranes were determined by the proposed semi-empirical equations as given below: where R represents the radius of the cavities, ΔR was obtained by empirical calibration as a constant (0.1656 nm) representing the electron layer thickness and C is a proportionality material-dependent constant (about 0.018, which is obtained through molecular simulation). A Micromeritics ASAP 2020 instrument was used to measure N 2 adsorption/ desorption of all TB polymers. Before the gas sorption analysis, undoped samples were degassed at 393 K for 12-16 h under vacuum, while PA-doped samples were degassed at 333 K for 2 h. The equilibration time was extended to 180 s to achieve the adsorption equilibrium (in the time interval, the consecutive pressure value agrees within 1.3 × 10 −4 bar). BET N 2 adsorption isotherms were applied to calculate specific surface areas.
Molecular simulation. The molecular dynamics (MD) simulation was constructed by the Forcite module in the Materials Studio software package (Accelrys). In one cubic simulation box, a polymer chain of 10 monomer units was used as template chain for the adjacent initial packing with the Amorphous Cell module. In every packing model four polymer chains were grown one after the other under periodic boundary conditions at 308 K and at an initial density of 0.1 g cm −3 . The density of the model was obtained and the final equilibration of the packing models was carried out with the following sequence of simulation steps: (1) 50 ps NVT-MD simulations at 600 K (a stimulated annealing), (2) 100 ps NVT-MD simulations at 308 K (back to target temperature), (3) 20 ps NpT-MD at 308 K and 10 bar with a time step of 0.1 fs (for the first time, volume fluctuations are allowed in the system), followed by (4) 280 ps NpT-MD at 308 K and 1.0 bar with a time step of 1.0 fs and (5) a long continuous NpT-MD simulation with the same conditions over 1 ns.
Calculated Mulliken and natural population analysis charges of N atoms for TB units. Geometry optimizations, vibrational frequencies and thermal energy corrections were performed with the B3LYP-D3 functional in conjunction with a mixed basis set of 6-31G(d) for all atoms in the gas phase. The universal solvation model (SMD) was employed to account for the effects of water solution 56 . All the optimized structures were confirmed by frequency calculations to be either minima or transition states at the same level of theory. To obtain more accurate electronic energies, single-point energy calculations were performed at the SMD B3LYP-D3/6-311+g** levels with the B3LYP-D3/6-31g(d) optimized structures. Structures were generated using CYLview 57 . All calculations were carried out with Gaussian 09 packages 58 . The detailed calculated Cartesian coordinates for the four TB-based units are included in Supplementary Data 1.

Proton conductivity.
A BioLogic VSP-300, FR impedance/gain phase analyser with two-point, in-plane membrane geometry using impedance spectroscopy at frequencies ranging from 1.0 MHz to 10.0 Hz was employed to measure the proton conductivity (σ) of the PEMs. The dimensions of the sample membranes were approximately 1 cm × 4 cm and were seated between two Pt-coated electrodes of a cell. An environmental chamber was used to place the testing cells, with a controlled atmosphere of temperature, pressure and humidity.
The through-plane conductivity cell was constructed using a Pt-plated stainless steel electrolytic tank (E200, provided by Tianjin Aida Hengsheng Technology Development). Pt electrodes were connected to the IT5102 Internal Resistance Test Device (ITECH) by means of pins. Before conducting the measurements, the cell was calibrated by measuring the impedance of Nafion 211 membranes from DuPont, which were soaked in deionized water for 24 h before use 48 . Membranes were cut into a circle with a diameter of 1.5 cm for the through-plane impedance measurement. The membrane sample was sandwiched between two Pt electrodes, and the cell was tightened with screws. The cell was put into the environmental chamber with a controlled temperature to collect impedance data. From the following equation, the values of σ in mS cm −1 for both cases were calculated: where L is the gap between electrodes for in-plane measurements and the thickness of the membrane for through-plane measurements; A is the membrane cross-sectional area in the case of the in-plane set-up and the area of the electrodes in the case of the through-plane set-up; R is the bulk membrane resistance.

RH cycling experiments.
The RH experiments were run sequentially by ramping up from 5% RH to 90% RH, with ~10% RH increments in an environmental chamber and with an equilibration period of two hours at each given RH. The samples were equilibrated at 5% RH for 14 h before the beginning of the next RH cycle, which was 34 h for each RH cycle. The in-plane conductivities of PEMs at 80 °C were measured every two hours and three times for each given RH, and the average was taken as the conductivity at a given RH.
Conductivity stability. The PEM strips (approximately 1 cm × 4 cm) were placed into an environmental chamber at 80 °C under 40% RH (or 180 °C with 0% RH). The changes of in-plane proton conductivity were monitored (σ t ). The proton conductivity retention (R c ) of the membranes was determined using the following equation: where σ i is the initial proton conductivity.
PA retention properties. The stability of PA in the membranes was determined in an environmental chamber with constant temperature and humidity for several hours. The weight of the membrane (W i ) was measured at different times and the liquid water on the surface of the membrane was blotted with a filter paper and the membrane was dried in a vacuum oven at 60 °C for 2 h before weighing. The stability of PA was calculated using the weight loss ratio (R w ) of the acid in the sample, as depicted in the following equation: where W a is the weight of the membrane after PA doping and W dry is the weight of the membrane before PA doping. 31 P NMR analysis. 31 P NMR spectra were recorded on a Bruker DPX-400 spectrometer at room temperature in DMSO-d 6 and were referenced to H 3 PO 4 (external standard) at 0 ppm. Two equivalents of PA-doped samples were obtained through adding 2 equiv. 85% PA direct to the polymer powder. After 30 minutes of ultrasonication and heating for 2 hours at 80 °C, the mixtures were dissolved in DMSO-d 6 . The PA-saturated membranes were directly dissolved in DMSO-d 6 for 31 P NMR spectroscopic evaluation. Solid-state 31 P NMR spectra of the PA-saturated membranes were acquired on a Bruker Avance 600 MHz Wide Bore spectrometer (14.1 T) using a 4 mm HXY probe with a ZrO 2 rotor, DR mode, lambda/2 and range coil. The magic angle spinning (MAS) rate was 9 kHz and chemical shifts were referenced relative to (NH 4 ) 2 HPO 4 (1.00 ppm).
To further reduce the effect of different amounts of water uptake in the membranes on 31 P NMR spectra, solid-state 31 P NMR spectra of mixtures of polymer powders and solid PA were measured. A certain amount of solid PA was well ground with polymer powder in a glovebox under an inert atmosphere of Ar. The amount of solid PA added to the polymer powders was equivalent to the molar PA absorbed in the membranes. 31 P MAS NMR of the mixture was carried out on a Bruker AVANCE III 600 spectrometer at a resonance frequency of 242.9 MHz using a 4 mm HX double-resonance MAS probe at a sample spinning rate of 12 kHz. Single-pulse 31 P MAS NMR experiments with 1 H decoupling were performed with a 90° pulse width of 3.2 μs and a 180 s recycle delay, and a 1 H decoupling strength of 80 kHz and 160 scans. The chemical shift of 31 P NMR was externally referenced relative to (NH 4 ) 2 HPO 4 (1.00 ppm).
Single-cell performance. Single cells with an active area of 4.0 cm 2 were used to measure the initial fuel cell performance of m-PBI/PA and TB/PA membranes. The MEAs for fuel cell testing were prepared by the catalyst-coated substrate method and the catalyst layer was composed of 0.50 mg cm −2 Pt (40 wt%) and 0.30 mg cm −2 PTFE loadings (the weight ratio of Pt/C to PTFE was 4 to 1), respectively. For all the tested cells, both the cathode and the anode used the same gas diffusion electrodes (GDEs). The catalyst ink was prepared by mixing PTFE emulsion (60%) and Pt/C (40 wt%) in water:isopropanol (1:6 by weight) for 30 minutes using a magnetic stirrer, followed by ultrasonication for 30 minutes, to ensure good dispersion. The GDE was prepared by painting the catalyst ink onto the carbon paper with gas diffusion layer (HCP120, HESEN), and then the obtained GDE was heated at 350 °C for 30 min in an N 2 atmosphere in a tubular furnace. To fabricate the MEA, the membrane sample was sandwiched between two GDEs and then hot-pressed at 100 °C under a pressure of 0.5 MPa. A test station (Smart 2-WonATech) was employed to study the initial fuel cell performance with no humidification and backpressure. The anode and cathode were fed with dry H 2 and O 2 at 200 sccm, respectively. The MEAs were activated at a constant voltage of 0.3 A until the current became stable. Then, steady-state polarization curves were recorded by polarizing the cell voltage from 1.0 V to 0.15 V in steps of 0.05 V, and holding the voltage for 30 s at each point. The high-frequency impedance (HFR) measurement at 1 kHz was conducted when the cell reached steady state at various cell voltages using a BioLogic VSP-300 potentiostat. The stability of PEMs was evaluated by AST. Using a constant temperature (that is, -20 °C, 15 °C and 40 °C), shut-down and start-up cycling AST were performed without backpressure or external humidification. After a step-V procedure from 1 V to 0.1 V with a scan rate of 25 mV min −1 , the cell was powered off, and then the fuel intake valves were closed. Restarting the cell after 5 minutes, opening the fuel intake valves and stabilizing the cell for 5 minutes were carried out to start another step-V cycle.
For the H 2 /air cell test, the area of the MEA was 1.00 cm 2 . A VersaStat 3 potentiostat was used to measure the electrochemical impedance spectra, and cyclic voltammograms and cell performance were recorded with no humidification and backpressure. The flow of the dry H 2 and O 2 were 30 sccm and 100 sccm, respectively. The same procedures as for the H 2 /O 2 cell test were used to activate the MEAs and record the polarization curves.

Data availability
Source data are provided with this paper. The authors declare that the data supporting the findings of this study are available within the paper and Supplementary information. Further data beyond the immediate results presented here are available from the corresponding authors on reasonable.

Code availability
This study did not generate any datasets.