The synthesis of white 6FDA-TFDB polymer was shown in Scheme S1. Loading freshly-made pristine 6FDA-TFDB membranes in the chamber (Supplementary Fig. 1), trimethyl aluminum (TMA) and water were applied on membrane samples as vapor phase precursors for Al2O3 deposition26–27, and a short exposure time of 8s was adopted for precise control of TMA deposition on membrane surfaces and within micropores (Fig. 1). The resulting white and transparent membranes were designated as 6FDA-TFDB-x, where x refers the number of cycles of the ALD process (4–12). Afterward, the 6FDA-TFDB-x polymer membrane was pyrolyzed in a controlled environment to create CMS membranes. The pyrolysis temperature was controlled via a three-zone tube furnace at 550 oC under Ar purging. Details of the experimental method related to the fabrication of CMS membranes are described in the Supporting Information.
The physicochemical properties of membranes were first characterized. The TGA trace of 6FDA-TFDB and 6FDA-TFDB-x precursor membranes are shown in Supplementary Fig. 2. 6FDA-TFDB and 6FDA-TFDB-x precursor membranes exhibit similar trends of weight loss, suggesting the trace Al in the polymer membranes has insignificant effects in the membrane thermal stability. In addition, the surface and cross-sectional morphologies of the 6FDA-TFDB and 6FDA-TFDB-x precursor membranes were examined using SEM. As shown in Supplementary Fig. 3, the precursor membranes demonstrate rather homogeneous morphological structures. The Al distribution on the surface and in cross-sections of 6FDA-TFDB-x CMS membranes was analyzed by Al and F elemental EDS mapping (Supplementary Fig. 4). Apparently, the aluminum content on the surface of membrane increases with different cycles (highlighted as red spots in Supplementary Fig. 4a-4d, yellow spots refer the overlap of red and green). As expected, the EDS mapping results of the cross-sectional specimen (Supplementary Fig. 4e-4h) indicate that the content of aluminum increases from 4 to 7 cycles (cross-sectional specimen shown in Supplementary Fig. 5). On the other hand, the concentration of fluorine increases with the number of cycles from 7 to 12 cycles, which was due to the in situ formation of Al-F complex during the ALD and pyrolysis process28.
Structural characterization of the 6FDA-TFDB-x CMS membranes
The elements contents of polymer precursor membrane samples were further studied by X-ray photoelectron spectrometer (XPS) (Fig. 2 and Supplementary Fig. 6–7). XPS results in Fig. 2a and Supplementary Fig. 6 demonstrate that the content of F-Al in 6FDA-TFDB precursor membranes gradually increases with cycles at the peak of 685.2 eV. However, the F-C content decreases from 4 to12 cycles at the peak of 688.1 eV. It is hypothesized that the highly reactive Al(CH3)3 defluorinate the polymer precursor, leading to the reduction of F-C content. Such an Al-induced defluorination phenomenon of F-containing polymers was also observed in literature29–30. Furthermore, as the deconvoluted F1s peaks shown in Supplementary Fig. 6, the degree of F-Al area is found to increase from 39–74% with the cycles increasing from 4 to 12, corroborating the formation of F-Al bonds in the ALD process. On the other hand, as shown in Fig. 2b, there is a sharp decrease of C1s signal (~ 293.1 eV) corresponding to the CF3 group when increasing cycles from 4 to 12, indicating that CF3 bonds reduce with more ALD cycles, consistent with the trends shown in Fig. 2a and Supplementary Fig. 6. In addition, XPS of Al2p in 6FDA-TFDB-x precursors was conducted with results shown in Supplementary Fig. 7. The binding energy of 74.9 eV is assigned to Al-O, suggesting the formation of aluminum oxide in the membranes. Moreover, the binding energy of 76.5 eV in the Al2p spectra corresponds to Al-F with intensities of signals gradually increasing with the cycles31–32. Such a trend not only agrees with the result in Fig. 2a but also provides further evidence of Al-F formation upon ALD treatment.
CMS membrane samples were also characterized using XPS and the spectra exhibit distinct peaks resulting from the presence of F1s, N1s, Al2p, and C1s as shown in Fig. 2c-2d and Supplementary Fig. 8–9. As expected, the F1s XPS spectrum of untreated 6FDA-TFDB CMS membranes shown F-C bonds at the peak of 688.1 eV, which, however, does not exhibit F-Al bonds since ALD treatment was not applied in the process of CMS fabrication (Fig. 2c). Interestingly, upon ALD treatment, F-C bonds disappear, consistent with our previous study that the ALD tends to defluorinate the precursor, essentially leading to the loss of F-C groups. Moreover, the content of F-Al as indicated by the peak at 688.1 eV increases with cycles to increase the content of aluminum in CMS membranes. On the other hand, the content of pyridinic N in CMS has been found to play a critical role in gas transport properties, and the decrease of pyridinic N reduces gas permeabilities33. As Fig. 2d shows, the N1s spectra clearly display the presence of both pyridinic N and pyrrolic N, corresponding to the peak at 398.0-398.8 eV and 400.3-400.8 eV, respectively. The intensities of pyridinic N and pyrrolic N peaks decrease with increased cycles. Based on the deconvoluted N1s peaks shown in Supplementary Fig. 8, the areas of pyridinic N of 6FDA-TFDB and 6FDA-TFDB-x (x = 4, 7, 10, 12) CMS membranes are estimated to be 39.4%, 39.1%, 36.3%, 33.1% and 26.8%, respectively. The increase of cycles to reduce the contents of pyridinic N, indicating the ALD treatment affects the formation of pyridinic N during pyrolysis. In addition, Al2p XPS of 6FDA-TFDB-x CMS membranes show that Al-F content of membranes increases with the cycles (Supplementary Fig. 9), as more Al2O3 were converted into Al-F complex28. Raman was also conducted on CMS membranes with the D and G peaks shown in Raman spectra (Supplementary Fig. 10). The smaller the ID/IG ratio, the lower the defect density in the material and the higher the graphite-like regularity13,20. As the ALD increase from 4 to 12 cycles, the ratio of ID/IG gradually decreases, indicating an increased degree of graphitization and a reduced level of defects in the amorphous CMS structure.
The evolution of chemical structure of precursors upon ALD process is further probed by 19F NMR, ATR-FTIR and TGA-FTIR. As illustrated in 19F NMR spectra of 6FDA-TFDB and 6FDA-TFDB-x precursors (Fig. 3a) and CMS of 6FDA-TFDB-x, aromatic and alkane C-F3 in TFDB and 6FDA moieties are clearly detected. More importantly, an apparent new peak at ~ 57.2 ppm emerges, suggesting the formation of Al-F bonds. 19F NMR spectra of 6FDA-TFDB-x CMS membranes show that the chemical shifts of Al-F shift from − 109.51 ppm to -126.38 ppm due to terminal F atoms are highly shielded with the ALD increasing (Fig. 3b). In addition, as shown by ATR-FTIR in Fig. 3c, the peak of 1256 cm-1 is ascribed to C-F3 stretching vibration of polymers34. Upon ALD treatment, detectable C-F3 shifts from 1256 cm-1 to 1250 cm-1 were found in 6FDA-TFDB-x precursor membranes, which ascribe to the increased interaction between F and Al with cycles since the heavy mass Al reduces the vibration frequency of C-F3 stretching and rocking. As TGA-FTTR results of 6FDA-TFDB-x membranes shown in Fig. 3d-3f, major off-gases including CO2, CO and CF3H were released during the pyrolysis, consistent with the literature reported elsewhere20. Moreover, the amount of CF3H gas tends to decrease when the ALD cycles increase from 4 to 12, as indicated by the TGA-FTIR spectra data, likely due to partial -CF3 groups transformed into AlF3, consistent with the EDS mapping results in Supplementary Fig. 4. CMS membrane samples were also tested using ATR-FTIR with results shown in Supplementary Fig. 11b, which display broad shoulders of absorbance peaks at ∼890–1145 cm− 1, corresponding to Al-F stretching vibrations35–36. The occurrence of absorbance peaks of Al-F3 in a broad range of vibrational frequencies again confirms the formation of Al-F bonds37.
To further illustrate the critical role of Al-F bonds formed in situ during ALD treatment, we prepared CMS membranes by mixing precursor polymers with 2% and 5% Al2O3 particles, following by pyrolysis with same conditions as ALD treated samples. XPS (Supplementary Fig. 12–15) was performed on both precursors and CMS membranes. The Al2p XPS and F1s XPS results do not reveal the F-Al bond in the 6FDA-TFDB/Al2O3 precursor or CMS membranes (Supplementary Fig. 12–13), since the peaks of binding energy completely differ from the Al-F bond as demonstrated in samples with ALD treatment (Fig. 3a, Supplementary Figs. 7 and 8). In addition, C1s XPS results show that the C-F bond was not impacted by the addition of Al2O3 in precursors as the intensities of binding energy remain similar for polymer and Al2O3-doped polymer membranes (Supplementary Fig. 14), different from membranes with ALD cycles (Fig. 2b). In a sharp contrast to the case of ALD-treated membranes, the N1s spectra of 6FDA-TFDB/Al2O3 CMS membranes show the same ratio of pyridinic N and pyrrolic N while such ratio in 6FDA-TFDB CMS membranes changes significantly with different ALD cycles (Supplementary Figs. 8 and 15). The above characterization proves that Al-F bond cannot be established by simply mixing Al2O3 with -CF3 containing polymers, reflecting the crucial role of ALD treatment in creating Al-F bonds in membranes.
The gas sorption and pore sizes of CMS membranes were analyzed with N2 adsorption experiments. As shown in Fig. 4a, the N2 adsorption capacity drops as the number of cycles increases, owing to the pore-blocking effect of ALD treatments with added cycles. As a result, the N2 derived BET surface area decreases from 949.8 to 847.6 m2 g-1 for the 6FDA-TFDB and 6FDA-TFDB-12 CMS membranes, respectively (Supplementary Table 1). The pore size distributions of CMS membranes were further examined based on non-local density functional theory with results illustrated in Fig. 4b, showing that the 6FDA-TFDB CMS membrane possesses an average pore size of 6 to 11 Å. In contrast, after 4–12 cycles, the average pore size decreases to the range of 5.5 to 8.0 Å. The 6FDA-TFDB-x CMS membranes show a clearly reduced pore size compared to the pristine 6FDA-TFDB CMS membrane, suggesting that ALD treatment is expected to enhance molecular sieving properties of CMS membranes by narrowing the pore sizes and creating ultra-micropores in CMS matrix as will be discussed later.
Pure and mixed CO2/CH4 gas separation performance
To study effects of ALD on membrane separation performance, the gas permeability and selectivity of both precursors and CMS membranes derived from 6FDA-TFDB and 6FDA-TFDB-x were tested with permeation results summarized in supplementary Table 1–2, Fig. 4 and Supplementary Fig. 16. The results of polymeric precursors in Fig. 4c and 4d show that the gas permeability decreases continuously while the selectivity generally decreases with the increase of cycles. ALD deposition layer near the surface of membranes creates additional gas transport resistance, thereby reducing the gas permeabilities, which, however, does not improve molecular sieving properties of polymeric membranes. Surprisingly, 6FDA-TFDB-x CMS membranes show pronounced enhancement of gas selectivities upon ALD treatment. Similar to polymeric precursors, the more cycles and the lower gas permeabilities of CMS membranes. The drop of permeability results from the additional layer by ALD deposition and consistent with the decrease of content of pyridinic N as shown in Fig. 2d, which also agrees with the finding reported in the literature33. On the other hand, the effect of ALD on gas selectivity is relatively complicated. In the beginning, the gas selectivity of 6FDA-TFDB-x CMS membranes is amplified by introducing cycles from 4 to 7 with a drop of gas permeability. For instance, 6FDA-TFDB-7 shows a CO2/CH4 selectivity of 58.4 with a CO2 permeability of 7767 Barrer, compared with a CO2/CH4 selectivity of 30.2 and a CO2 permeability of 12653 Barrer in the case of untreated 6FDA-TFDB CMS membranes.
By plotting with upper bounds, the overall performance of 6FDA-TFDB-4 and 6FDA-TFDB-7 CMS surpass the latest upper bounds for CO2/CH4 and O2/N2 (Fig. 4e-5f). In fact, the CO2/CH4 selectivity of 6FDA-TFDB-7 CMS membranes is the highest among reported 550 oC pyrolyzed CMS membranes derived from polymers including Matrimid, 6FDA-BPDA/DAM, 6FDA-mPDA/DABA, 6FDA/DETDA-DABA, 6FDA/1,5-ND:ODA,TB, PIM, and PIM-PI, as shown in supplementary Table 4. When further increasing the number of cycles to 10 and 12, the gas selectivity of CMS membranes tends to decrease but remains higher than untreated membranes. In any case, the above results suggest that ALD strategy is an effective strategy in developing CMS membranes with high permeability and selectivity and there exists an optimal cycle for preparing the most selective CMS materials. Presumably, the existence of Al-F complex plays the key role in affecting CMS membrane separation performance, generated from two possible paths: 1) defluorination of polymer precursors with Al(CH3)3 during ALD treatment in the absence of pyrolysis; 2) reaction of Al2O3 with HF released during pyrolysis starting at a temperature of ~ 450 oC38. Al-F complex is expected to interact strongly with strands including -C = O39, C-N40 and C-F moieties by Vander Waals force during pyrolysis, which likely promotes an efficient packing of strands and reduces the pore sizes, essentially improving the gas selectivity. However, when excessive cycles applied, the pore size reduces to a level where the fast gas permeability (i.e. H2, CO2) drops more rapidly than the slow gas (i.e. N2, CH4), leading to a decrease of gas selectivity.
In order to decouple the effect of Al-F from Al2O3 on gas separation performance, 2% and 5% Al2O3 and AlF3 were blended with 6FDA-TFDB separately to prepare CMS membranes. Mixed matrix membranes made of Al2O3 or AlF3 with 6FDA-TFDB are opaque, different from transparent 6FDA-TFDB-x samples with the ALD process (Supplementary Fig. 17). Furthermore, the gas separation performance of those membranes was tested with results shown in supplementary Table 2. As a typical pore former, the addition of Al2O3 improves the gas permeability with a simultaneous loss of gas selectivity. However, in the case of AlF3 mixed matrix membranes, AlF3 tends to reduce gas permeability and increase gas selectivity in a strong agreement with aforementioned discussions. For instance, 5 wt% AlF3 6FDA-TFDB CMS membranes display an increase of CO2/CH4 selectivity from 30.2 to 42.6, and O2/N2 selectivity from 4.9 to 5.6, compared with untreated 6FDA-TFDB CMS, respectively. These experiments further corroborate our findings on the crucial role of the Al-F bond in determining CMS structures and molecular sieving properties.
The plasticization response and ageing behavior of 6FDA-TFDB-x CMS membranes
The gas separation performance of 6FDA-TFDB and 6FDA-TFDB-7 CMS membranes under high operation pressures was explored using an equimolar binary mixture of CO2/CH4. As shown in Fig. 5, the CMS membranes demonstrate excellent CO2/CH4 mixed-gas separation performance at elevated pressures. The CO2/CH4 mixed-gas selectivity reaches as high as 55.6 at a feed pressure of 400 psi. Moreover, the CO2/CH4 selectivity of 6FDA-TFDB and 6FDA-TFDB-7 CMS membranes maintain nearly unchanged comparing with the pure gas permeation results (Fig. 5a). Such mixed gas separation performance places the CMS membranes well above the mixed-gas trade-off curve for CO2/CH4 (Fig. 5b).
A key obstacle for most CMS membranes lies in their lack of long-term stability41–42. To investigate the durability of 6FDA-TFDB and 6FDA-TFDB-7 CMS membranes over time, we performed an aging study as long as 480 h, during which the upstream and downstream of membrane samples were pulling vacuum with results shown in Fig. 5 and supplementary Table 3. Similar to other reported CMS, the untreated 6FDA-TFDB CMS membrane shows a typical physical aging phenomenon and the CO2 permeability drops over 70.3% from 12653 to 3764 Barrer during the first 120 h (Fig. 5c-5d). The aging tends to slow down after 120 h but there is still a drop of CO2 permeability over 30% during aging from 120→480 h. In a sharp contrast, the ALD treated 6FDA-TFDB-7 CMS membranes demonstrate significantly enhanced aging resistance. The CO2 permeability decreases only 13.6% from 7767 to 6709 Barrer during the first 120 h aging, much lower than the case of untreated CMS. After aging for 120 h, the CO2 permeability of 6FDA-TFDB-7 decrease 10.3%. Combing the two stages of aging, the ALD treated CMS samples only show a loss of CO2 permeability less than 13.6%, considerably lower than that in the case of untreated CMS (i.e. 70.3%). On the other hand, both CMS samples exhibit pronounced increases of gas selectivity after aging. Note that the CO2/CH4 selectivity of 6FDA-TFDB-7 CMS membrane is as high as 93.9 after aging 480 h. We hypothesize that the existence of the F-Al complex in the 6FDA-TFDB-7 CMS membranes results in rigidified CMS matrix, thereby suppressing the pore collapse of membranes and retaining high gas permeability during aging.
Outlook
Our work has demonstrated a facile approach of finely modulating CMS pore structure and separation performance. Various 6FDA-based CMS polyimide precursors could be using the strategy to obtain the controlled aperture CMS membranes. More broadly, the concept and ALD approach can also be extended to other important CMS precursors such as PIM-based or TB precursors to create a new family of advanced CMS membranes. The design principle described in this work offers an efficient and facile method for precisely manipulating CMS pore size with potential for gas separations such as CO2 removal and H2 purification.