Synthesis of Silica Membranes by Chemical Vapor Deposition Using a Dimethyldimethoxysilane Precursor

membranes

Article Synthesis of Silica Membranes by Chemical Vapor Deposition Using a Dimethyldimethoxysilane Precursor

S. Ted Oyama 1,2,3,* , Haruki Aono 2, Atsushi Takagaki 2,4 , Takashi Sugawara 2 and Ryuji Kikuchi 2 1 College of Chemical Engineering, Fuzhou University, Fuzhou 350116, China 2 Department of Chemical Systems Engineering, The University of Tokyo, 7-3-1 Hongo, Bunkyo-ku, Tokyo 113-8656, Japan; [email protected] (H.A.); [email protected] (A.T.); [email protected] (T.S.); [email protected] (R.K.) 3 Department of Chemical Engineering, Virginia Tech, Blacksburg, VA 24061, USA 4 Present address: Department of Applied Chemistry, Faculty of Engineering, Kyushu University, 744 Motooka, Nishi-ku, Fukuoka 819-0395, Japan * Correspondence: [email protected]; Tel.: +81-3-3727-1254

Received: 19 February 2020; Accepted: 15 March 2020; Published: 22 March 2020

Abstract: Silica-based membranes prepared by chemical vapor deposition of tetraethylorthosilicate (TEOS) on γ-alumina overlayers are known to be effective for hydrogen separation and are attractive for membrane reactor applications for hydrogen-producing reactions. In this study, the synthesis of the membranes was improved by simplifying the deposition of the intermediate γ-alumina layers and by using the precursor, dimethyldimethoxysilane (DMDMOS). In the placement of the γ-alumina layers, earlier work in our laboratory employed four to five dipping-calcining cycles of boehmite sol precursors to produce high H2 selectivities, but this took considerable time. In the present study, only two cycles were needed, even for a macro-porous support, through the use of finer boehmite precursor particle sizes. Using the simplified fabrication process, silica-alumina 7 2 1 1 composite membranes with H2 permeance > 10− mol m− s− Pa− and H2/N2 selectivity >100 were successfully synthesized. In addition, the use of the silica precursor, DMDMOS, further improved the H2 permeance without compromising the H2/N2 selectivity. Pure DMDMOS membranes proved to be unstable against hydrothermal conditions, but the addition of aluminum tri-sec-butoxide (ATSB) improved the stability just like for conventional TEOS membranes.

Keywords: silica-alumina membrane; dimethyldimethoxysilane (DMDMOS); hydrothermal stability; chemical vapor deposition; gamma-alumina intermediate layers; hydrogen helium separation

1. Introduction Silica-based membranes have been known to be eﬀective for selective hydrogen permeation for more than 30 years, and are potentially less expensive than palladium-based membranes. The subject of silica membranes has been covered in recent reviews [1–3] and several recent papers [4–8]. Notable past work includes the original papers [9,10], the use of sol-gel techniques [11,12], chemical vapor deposition (CVD) [13,14] and plasma methods [15,16], the employment of diverse CVD precursors [17–21] and compositions [22–24], the determination of porosity [25], and the elucidation of the mechanism of permeation [26–28]. Important precedents are the use of dimethyldimethoxysilane (DMDMOS) [17] and other methyl-substituted siloxanes [29,30]. The separation mechanism of small species like He, H2, and Ne in silica membranes prepared by chemical vapor deposition of alkoxide substrates is based on molecular hopping between solubility

Membranes 2020, 10, 50; doi:10.3390/membranes10030050 www.mdpi.com/journal/membranes Membranes 2020, 10, 50 2 of 21 sites [31]. The sites in the silica network are approximately 0.3 nm, allowing only small gas molecules like He, H2, or Ne to enter them [31,32]. The only way for larger gas molecules to permeate is to pass through pores or defects such as pinholes or cracks. Therefore, the avoidance of defects is important for high H2 selectivity, and this will be one of the topics covered in this study. Permeance in membranes with larger pores may occur by a modified gas translational mechanism [33]. Usually, the sol-gel cycle in the synthesis of intermediate layers must be repeated several times to ensure a defect-free membrane [34]. However, too many repetitions not only require longer preparation time with extra energy consumption, but also increase the γ-alumina thickness, thereby posing the risk of cracking. Particularly, a previous study [35] achieved very high H2 permeance and selectivity by applying four to five dipping-calcining cycles to form a graded γ-alumina intermediate layer on a macro-porous support, but this was time-consuming and undesirable for reproducibility and cost reasons. It is desirable to reduce the required number of dipping-calcining cycles for a specific membrane support, and this will be one of the topics covered in this study. Scanning electron microscopy (SEM) is used to image samples at intermediate stages of synthesis so as to provide insight on the difference between successful and failed membranes. In addition, the H2 permeance is further improved by using a novel silica precursor compound, dimethyldimethoxysilane [17], which makes the silica network slightly looser. A potential use of silica membranes is in membrane reactors [36–40], so a significant concern is the stability under thermal or hydrothermal conditions [41–43]. Particularly, exposure to water vapor at an elevated temperature quickly leads to deterioration of the membrane performance. In general, the H2 permeance will decrease due to the densification of the silica network, and the permeance of larger gas molecules will increase due to pinhole generation caused by the degradation of γ-alumina, when it is used as a support. This was confirmed by Reference [17], where hydrothermal stability tests were conducted on both the inside and outside of the membrane tube. The densification of the silica is believed to occur due to the hydrolysis of Si-O-Si linkages and the formation of Si-OH hydroxyl groups, and the degradation of the γ-alumina is believed to occur due to sintering and pore enlargement. To improve the stability of the silica layer metal elements such as Al [44,45] or Ti [46] can be incorporated into the Si network. They are thought to impart higher tolerance against the attack of the H2O molecules to the Si-O-Si bond. For example, in the case of a silica-alumina composite 7 2 1 1 membrane [44], the initial H2 permeance was 2–3 10− mol m− s− Pa− with a H2/CH4 selectivity × 7 2 1 1 of 940 at 600 ◦C, and the H2 permeance was maintained above 10− mol m− s− Pa− even after exposure to 16 mol% steam for 500 h at 600 ◦C. In the case of a silica-titania composite membrane [46], the initial H permeance was 3 10 7 mol m 2 s 1 Pa 1 and was reduced by only 30% after 130 h of 2 × − − − − exposure in 75 mol% H2O at 650 ◦C, though the H2/CH4 selectivity was at moderate level (40–60). The above-mentioned membranes are much more stable than pure silica membranes, where a reduction of more than 90% of the H2 permeance is typical [47,48]. In this study, silica-alumina composite membranes are employed for hydrothermal stability. The stability of γ-alumina can also be improved by metal-doping. Silica membranes fabricated by counter diffusion chemical vapor deposition (CVD) of tetramethoxyorthosilicate (TMOS)/O2 on pure or Ni-doped γ-alumina showed hydrothermal stability at 500 ◦C with a steam/N2 ratio of 3 [49]. The authors pointed out that for the membrane with pure γ-alumina, the N2 permeance fluctuated during the hydrothermal exposure and significantly increased from the initial value, but by doping 5 mol% Ni into the γ-alumina, the problem was minimized. The actual stabilization mechanism, however, was not clarified in the paper. Subsequent work [41] discussed the membrane degradation mechanism in detail, and attributed the fluctuation of the N2 permeance to the combination of γ-alumina pore enlargement, pore plugging by grain growth, and subsequent grain rearrangement. To achieve a high H2 permeance without affecting the H2 selectivity against the target compound to be separated, it is important to strictly control the pore size of the silica network. Silica membranes fabricated by counter diffusion CVD of TMOS, and one-sided diffusion CVD of phenyltrimethoxysilane (PTMS) and dimethoxydiphenylsilane (DMDPS) have been investigated [50]. The use of DMDPS Membranes 2020, 10, 50 3 of 21 Membranes 2020, 10, x FOR PEER REVIEW 3 of 21

selectivity of 6800 at 300 °C, which is comparable6 to 2the 1performance1 of palladium membranes. resulted in a H2 permeance of the order of 10− mol m− s− Pa− and a H2/SF6 selectivity of 6800 at Subsequent work [51,52] using this membrane confirmed that the H2/toluene separation ability was 300 ◦C, which is comparable to the performance of palladium membranes. Subsequent work [51,52] stable for more than 1000 h. It should be noted, however, that the H2/N2 selectivity was lower than using this membrane conﬁrmed that the H2/toluene separation ability was stable for more than 1000 h. ~100 due to an open porous silica network (0.3–0.47 nm [51]) compared to the TEOS- or TMOS- It should be noted, however, that the H2/N2 selectivity was lower than ~100 due to an open porous derived networks (~0.3 nm). Several studies have focused on silica membranes with other functional silica network (0.3–0.47 nm [51]) compared to the TEOS- or TMOS-derived networks (~0.3 nm). Several groups like vinyltriethoxysilane [53]. Noteworthy are recent studies that describe high-performance studies have focused on silica membranes with other functional groups like vinyltriethoxysilane [53]. silica membranes [54–56]. Noteworthy are recent studies that describe high-performance silica membranes [54–56]. Silica membranes prepared by counter diffusion CVD have also been studied using a number of Silica membranes prepared by counter diﬀusion CVD have also been studied using a number silica precursors [57]. Among them, the use of DMDMOS resulted in a superior H2 permeance of 9.0 of silica precursors [57]. Among them, the use of DMDMOS resulted in a superior H permeance of −7 −2 −1 −1 2 × 10 mol7 m s Pa2 1and H1 2/N2 selectivity of 920 at 500 °C. However, the paper [17] did not report 9.0 10− mol m− s− Pa− and H2/N2 selectivity of 920 at 500 ◦C. However, the paper [17] did not detailed× information about long-term stability or the gas permeation mechanism of the new report detailed information about long-term stability or the gas permeation mechanism of the new membrane, so these will be covered in this study. membrane, so these will be covered in this study.

2. Materials and Methods

2.1. Materials The membranes used in this research consisted of three layers: a commercial α-alumina support (Noritake Co.) with with nomina nominall pore pore size size of of 60 60 nm, a γ-alumina intermediate layer deposited by sequential dip-coatingdip-coating/calcining,/calcining, and and a topmosta topmost silica silica layer layer placed placed by chemical by chemical vapor depositionvapor deposition (CVD). Before(CVD). silicaBefore deposition, silica deposition, two membranes two membranes were synthesized were synthesized in parallel in under parallel the sameunder conditionsthe same (Figureconditions1). Then,(Figure one 1). membrane Then, one membrane was used for was CVD used and for the CVD other and membrane the other membrane was used forwas scanning used for electronscanning microscopy electron microscopy (SEM) analysis (SEM) to have analysis identical to substrateshave identical for clarifying substrates the relationship for clarifying between the therelationship surface or between cross-sectional the surface morphology or cross-sectio and thenal gas morphology separation and ability. the gas separation ability.

Figure 1. Membrane fabrication procedures. CVD-Chemical vapor deposition, SEM-Scanning electronFigure 1. microscopy. Membrane fabrication procedures. CVD-Chemical vapor deposition, SEM-Scanning electron microscopy.

Membranes 2020, 10, 50 4 of 21

2.2. Membrane Fabrication

2.2.1. Preparation of the Dipping Sols The γ-alumina layers were prepared by sol-gel coating of boehmite sols, which were synthesized through the hydrolysis of aluminum isopropoxide (Aldrich, >98%) and subsequent peptization by nitric acid, as reported previously [58]. The boehmite particle sizes were controlled by the time of hydrolysis and the amount of nitric acid. The actual dipping solutions were prepared by mixing the boehmite sols with a polyvinyl alcohol (PVA, Polysciences, M.W. = ~78,000) solution to minimize the risk of cracking by slightly raising the sol viscosity. The obtained concentration of the sol and PVA were 0.15 wt.% and 0.35 wt.%, respectively.

2.2.2. Preparation of the Intermediate Layers First, a commercial α-alumina tube (i.d. = 4 mm, o.d. = 6 mm, length = 30 mm) with 60 nm pores (Noritake Corporation) was connected to non-porous alumina tubes (i.d. = 4 mm, o.d. = 6 mm, length = 200 mm) at both ends with ceramic joints, which were made by applying a glass paste and firing at 1000 ◦C. Second, the tube was dipped into the dipping solution for 10 s with its outer surface wrapped with Teflon tape. Third, the tube was dried in a laminar flow enclosure with high efficiency particulate air filters for 4 h, and was calcined at 650 ◦C for 3 h. Here, a relatively slow heating/cooling rate of 1.5 ◦C/min was applied in order to minimize the risk of cracking due to different thermal expansion coefficients between the support and the intermediate layer. The dipping-calcining cycle was repeated as necessary.

2.2.3. Preparation of the Topmost Silica Layers The topmost silica layers were synthesized by chemical vapor deposition (CVD), which placed a thin (20~30 nm) silica layer on top of the γ-alumina substrate by the thermal decomposition of the silica precursor compound. First, a conventional silica precursor, tetraethyl orthosilicate (TEOS) (Aldrich, 98%), was used to concentrate on the optimization of the intermediate layer. The schematic diagram of the CVD apparatus is shown in Figure2a. The 6 mm diameter support tube was coaxially fixed inside the stainless reactor using machined Swagelok fittings with Teflon ferrules. The assembly was placed in an electric furnace and was heated to 650 ◦C at a heating rate of 1.5 ◦C/min. The bubbler temperature for the TEOS delivery was kept at 25 ◦C, which gave a TEOS vapor pressure of 250 Pa (obtained from the following Antoine equation).

B log P = A , A = 4.17312, B = 1561.277, C = 67.572, P[bar], T[K] (1) 10 − T + C −

To improve the hydrothermal stability, a silica-alumina composite membrane was fabricated using a second bubbler to deliver aluminum tri-sec-butoxide (ATSB) as a secondary component in addition to the TEOS (Figure1a). In this case, the bubbler temperature for the ATSB delivery was set at 96 ◦C so that the Al/Si ratio would be 0.03. This value was determined by considering the tradeoﬀ between the hydrothermal stability and the H2 selectivity obtained in a previous study [44]. The two carrier gases were premixed with a dilution Ar gas before introduction to the inside of the support tube. The lines leading to the inner tube were heated by a ribbon heater to prevent condensation of the precursor compounds. A balance argon gas was also introduced to the outer shell in order to maintain pressure balance between inside and outside of the support tube and minimize the loss of H2 permeance due to silica deposition within the pores of the γ-alumina. After the CVD process was ﬁnished, the assembly was purged with the balance and dilution Ar gases for about 15 min to sweep out any unreacted precursor compounds or reaction byproducts. To further improve the H2 permeance without compromising the H2 selectivity, a novel silica precursor reported in the literature, dimethyldimethoxysilane (DMDMOS) [17], was also examined. The reported H permeance and H /N selectivity were 9.0 10 7 mol m 2 s 1 Pa 1 and 920, respectively, 2 2 2 × − − − − Membranes 2020, 10, 50 5 of 21 but the authors did not describe any detailed information about the gas permeation mechanism or the long-termMembranes stability 2020, 10, of x FOR the PEER membrane. REVIEW Therefore, these will be investigated in this study. 5 of 21

(a)

(b)

(c)

FigureFigure 2. Experimental 2. Experimental setup setup for for several several typestypes of chemical vapor vapor deposition deposition (CVD). (CVD). (a) One-sided (a) One-sided diﬀusiondiffusion CVD CVD of tetraethylorthosilicate of tetraethylorthosilicate (TEOS) (TEOS)/aluminum/aluminum tri-sec-butoxide tri-sec-butoxide (ATSB), (ATSB), (b) counter-di(b) counter-ﬀusion

CVDdiffusion of dimethyldimethoxysilane CVD of dimethyldimethoxysilane (DMDMOS)/O 2(DMDMOS)/O,(c) one-side2 di, ﬀ(cusion) one-side CVD ofdiffusion DMDMOS CVD/ATSB of /O2.

DMDMOS/ATSB/O2.

Membranes 2020, 10, 50 6 of 21

To synthesize DMDMOS membranes, both counter-diffusion CVD (Figure2b) and one-sided diffusion CVD (Figure2c) were applied, whereas the previous study [ 17] dealt with only counter-diffusion CVD. Particularly, the one-sided diffusion CVD made it easier to add the stream of ATSB to improve the hydrothermal stability. In both cases, the CVD temperature was set at 500 ◦C in order to prevent support cracking due to the large reaction heat between DMDMOS and O2 in accordance with the previous report [17].

2.3. Evaluation of the As-Synthesized Membranes

2.3.1. SEM Analysis The surface and cross-sectional microstructures of the membranes were studied using ﬁeld emission scanning electron microscopy (FESEM, S-900, Tokyo, Japan). Samples were obtained by mechanically slicing the tubular membranes with a diamond saw and subsequently chopping the sliced discs with a cutter. The samples were sputtered with platinum for 15 s before the measurements with SEM. Through the observation, the following points were examined (Table1).

Table 1. Characterization of Membranes by Scanning Electron Microscopy.

Samples before CVD Samples after CVD Particle size of γ-alumina · Porosity Smoothness Surface · · Smoothness Presence of pinholes · · Presence of pinholes · Thickness of the γ-alumina layer Thickness of the silica layer · · Presence of γ-alumina particle Presence of silica inﬁltration Cross-section · · inﬁltration into α-alumina into γ-alumina · · Presence of cracks Presence of cracks · ·

2.3.2. Single Gas Permeation Test The CVD process was interrupted at various deposition times, and gas permeation tests were carried out at 650 ◦C using the same apparatus. The sample gases (H2, He, Ne, N2, CO2, Ar) were introduced into the inside of the membrane tube, the bottom end of which was closed. Then, the gas permeation rates from the outer shell were measured either by a ﬂow meter or a gas chromatograph. Permeances were calculated using the following equation:

F P = (2) A∆p

2 1 1 1 where P is the permeance (mol m− s− Pa− ), F is the molar flow rate (mol s− ), A is the permeation area (m2), and ∆p is the pressure difference across the membrane (Pa). The permeation area, A, was calculated by: πL(r1 r2) A = − (3) ln r1 r2 where L is the effective membrane length, r1 is the outer diameter, and r2 is the inner diameter. The ideal selectivity was defined as the ratio of the single gas permeances:

Pi αij = (4) Pj

2 1 1 where αij is the selectivity of gas i versus j, Pi is the permeance of gas i (mol m− s− Pa− ), and Pj 2 1 1 9 2 1 1 is the permeance of gas j (mol m− s− Pa− ). For permeances higher than 10− mol m− s− Pa− , Membranes 2020, 10, 50 7 of 21

Membranes 2020, 10, x FOR PEER REVIEW 7 of 21 an electronic mass flow meter was used for the measurement, and for lower permeances, a Shimazu 8A gasto the chromatograph outer shell in order was employed. to sweep the In thepermeated latter case, molecules He gas atto athe known detector flow of rate the gas was chromatograph. also introduced toThen, the outerthe gas shell flow in rate order was to sweepcalculated the permeated from the He molecules flow rate to theand detector sample ofgas the concentration, gas chromatograph. which Then,was obtained the gas flowfrom rate the wasaverage calculated peak area from and the the He ca flowlibration rate and curve. sample The gas selectivity concentration, was calculated which was as obtainedthe ratio of from the thesingle average gas permeances peak area and of H the2 to calibration N2, CO2, or curve. Ar. The selectivity was calculated as the ratio of the single gas permeances of H2 to N2, CO2, or Ar. 2.3.3. Hydrothermal Stability Test 2.3.3. Hydrothermal Stability Test As was described in the introduction (Section 1), both the topmost selective layer and its support layerAs must was be described substantially in the stable introduction under the (Section operatin1), bothg conditions; the topmost otherwise, selective it layer is impossible and its support to use layerthe membrane must be substantially in practical applications. stable under theHowever, operating there conditions; are limited otherwise, numbers it of is impossiblepublications to dealing use the membranewith hydrothermal in practical stability applications. tests However,on both theresides areof limitedthe membrane. numbers ofTherefore, publications in dealingthis study, with hydrothermal stability tests tests targeting on both sides both ofthe the silica membrane. and the γ Therefore,-alumina layers in this were study, conducted hydrothermal at 650 stability°C, up to tests 200 targetingh. both the silica and the γ-alumina layers were conducted at 650 ◦C, up to 200 h. 1 1 1 First, an Ar flow flow at 6.6 μµmol s −1 (flow(flow rates rates in in μµmolmol s− s1− cancan be be converted converted into into ml ml (NTP) (NTP) min min−1 by− bymultiplying multiplying by 1.5) by 1.5)was was passed passed through through a heated a heated bubbler bubbler containing containing distilled distilled water water and was and then was thenintroduced introduced to the to inside the inside of the of themembrane membrane tube tube to todirectly directly contact contact the the silica silica layer. layer. The The bubblerbubbler temperature was setset at 56 ◦°CC so that the Ar stream would contain 16 mol% water vapor. At the same 1 time, another ArAr flowflow atat 11.611.6 µμmolmol ss−−1 waswas introduced introduced to the outside of the membrane tube. Both sides of the membrane were kept at at atmosphericmospheric pressure. The hydrotherm hydrothermalal treatment was interrupted at several exposure times,times, andand thethe permeancespermeances ofof HH22,, He, and N2 werewere measured to monitor the changes in thethe membrane membrane performance. performance. To To make make the permeancethe permeance measurements, measurements, the humidified the humidified Ar was changedAr was tochanged a pure Arto fora pure about Ar 20 for min about to dry 20 the min membranes. to dry the The membranes. wet Ar flow The was wet resumed Ar flow immediately was resumed after theimmediately permeance after measurements. the permeance The cyclemeasurements. was repeated Th severale cycle timeswas repeated until the membraneseveral times performance until the becamemembrane almost performance unchanged. became almost unchanged. 1 Next, the configurationconfiguration waswas altered.altered. AnAn ArAr flowflow atat 11.611.6 µμmolmol ss−−1 containingcontaining 16 16 mol% mol% water water vapor was introduced to the outer shell to directly contact the γ-alumina layer, while another dry Ar stream 1 at 6.6 µμmolmol ss−−1 was introduced to the inner tube. The ga gass permeances were measured after several exposure times just like the previous case. Figure 33 summarizessummarizes thethe experimentalexperimental proceduresprocedures forfor thethe above-described hydrothermal stabil stabilityity tests. As As a a note, the lines for the delivery of humidifiedhumidified Ar were maintainedmaintained aboveabove 100100 ◦°CC usingusing ribbonribbon heatersheaters toto preventprevent condensationcondensation ofof waterwater vapor.vapor.

Figure 3. Hydrothermal stability test for both the inside and outside of the membrane tube. MFC-Mass ﬂowFigure controller. 3. Hydrothermal stability test for both the inside and outside of the membrane tube. MFC- Mass flow controller.

Membranes 2020, 10, 50 8 of 21

3. Results and Discussion

3.1. Morphology and Structure of the Membranes Studies with scanning electron microscopy (SEM) revealed a progression of structure in the different membranes (Figure4). The surface image of the α-alumina support with nominal pore size of 60 nm (Figure4a) treated once with a 200 nm sol shows that although a certain amount of γ-alumina was deposited on the α-alumina, there were large areas of uncovered α-alumina particles (Figure4b). The image of the α-alumina support treated by a 200 nm sol twice shows that the α-alumina was totally covered by γ-alumina (Figure4c). These results mean that at least two dipping-calcining cycles are required to eliminate large defects if large sols are used first. In addition, the support had a coarse surface and relatively large pores (>5 nm) (Figure4d), so additional treatment with a smaller sol was necessary, thus bringing the total number of dipping-calcining cycles to three. The γ-alumina surface prepared by dip-coating the α-alumina support with a 40 nm sol (Figure4e) and 80 and 40 nm sols (Figure4f) shows the successful formation of smooth and uniform γ-alumina layers. In addition, their cross-sectional images show that the thickness of the γ-alumina layer was about 1 µm for the former (Figure4g) and about 2 µm for the latter (Figure4h). These results suggest that macro-porous α-alumina can be completely covered by γ-alumina even if the boehmite particle size is smaller than, or similar to, the support pore size. This has not been reported before. To further investigate the microstructural differences between these two samples, pictures at higher magnification were taken (Figure5). Careful observation reveals a slightly higher porosity for the sample with one treatment with 40 nm particles (Figure5a) than that with successive 80 and 40 nm particles (Figure5b). Then, the particle stacking density is slightly lower for the former. However, it was difficult to confirm a significant difference even with the nearly maximum resolution of SEM. Earlier studies [54–56] have demonstrated the usefulness of nanopermporometry in quantitating the pore size distribution of intermediate layers. Unfortunately, this was not undertaken in the present study. However, the microscopy does indicate that when only one intermediate layer was applied, the surface was only partly covered. Although there were no clear differences between the samples before CVD, significant differences were found for the samples after CVD. For the sample with one layer of 40 nm sol, the formation of a silica layer was not complete and there were many visible pinhole defects (Figure5c), with those at the center of the picture being especially conspicuous. For the sample with successive layers of 80 and 40 nm sols, the surface was formed from continuous rounded structures, which is characteristic of a CVD-derived silica layer (Figure5d) [ 59]. In addition, its cross-sectional image revealed that the thickness of the silica layer was about 30–40 nm (Figure5e), which was similar to that in the previous study. These results clearly show that a smooth and uniform silica layer was successfully formed, although the dipping-calcining cycle was performed only twice on the macro-porous support. As a summary of all the above findings, at least two dipping-calcining cycles are necessary in order to obtain a pinhole-free membrane. MembranesMembranes 20202020,, 1010,, x 50 FOR PEER REVIEW 9 9of of 21 21

2 µm 2 µm 2 µm

(a) Surface of the α-alumina (b) Surface of the 60 nm α- (c) Surface of the 60 nm α- support with average pore size alumina support treated once alumina support treated twice of 60 nm with a 200 nm sol with a 200 nm sol (low magnification)

200 nm 200 nm 200 nm

(d) Surface of the 60 nm α- (e) Surface of the 60 nm α- (f) Surface of the 60 nm α- alumina support treated twice alumina support treated once alumina support treated with a with a 200 nm sol (high with a 40 nm sol (low 80 nm and a 40 nm sol (low magnification) magnification) magnification)

1 µm 2 µm

(g) Cross-section of the 60 nm (h) Cross-section of the 60 nm α-alumina support treated α-alumina support once with a 40 nm sol sequentially treated with an 80 nm sol and a 40 nm sol

FigureFigure 4. Scanning 4. Scanning electron electron micr microscopyoscopy (SEM) (SEM) images images (a–h of) of the the membranes membranes at at low low magniﬁcation. magnification.

MembranesMembranes2020 2020, 10, ,10 50, x FOR PEER REVIEW 10 of10 21 of 21

50 nm 50 nm

(a) Surface of the 60 nm α-alumina support treated (b) Surface of the 60 nm α-alumina support once with a 40 nm sol (high magnification) sequentially treated with an 80 nm sol and a 40 nm sol (high magnification)

200 nm 200 nm

30-40 nm

200 nm

(e) Cross-section of the silica layer deposited on the γ-alumina layer

Figure 5. SEM images of the membranes at high magniﬁcation. Figure 5. SEM images (a–e) of the membranes at high magnification.

Membranes 2020, 10, 50 11 of 21

3.2. Gas Permeation Properties of the Membranes Figure5 shows the changes in gas permeances during CVD for various types of membranes fabricated in this research. At the beginning of the study, a conventional silica precursor, TEOS, was applied to concentrate on the optimization of the intermediate layer. For the membrane with one treatment with a 40 nm sol, the H permeance at 650 C was 1.2 10 7 mol m 2 s 1 Pa 1 and the H /N 2 ◦ × − − − − 2 2 selectivity was 26 after 120 min of CVD (Figure6a). This low selectivity is resonant with the defective surface structure revealed by SEM (Figure5c). On the other hand, for the membrane successively treated with 80 nm and 40 nm sols, the H permeance was the same (1.2 10 7 mol m 2 s 1 Pa 1), 2 × − − − − but the H2/N2 selectivity was 330 after 90 min of CVD (Figure6b). This result clearly demonstrated that a defect-free membrane (Figure5d) can be synthesized with only two dipping-calcining cycles in spite of the use of a macro-porous 60 nm α-alumina support. Previous studies had used supports with substrate pores of 5 nm [28,35,44]. Therefore, the repetition number of the dipping-calcining cycle was fixed at two for the subsequent studies. Figure6c shows the results for one of the successfully prepared membranes by the dual-element CVD of TEOS and ATSB. After 105 min of deposition, the H2 permeance and the H2/N2 selectivity at 650 C were 2.5 10 7 mol m 2 s 1 Pa 1 and 980, respectively. The original purpose of the addition ◦ × − − − − of ATSB was to improve the hydrothermal stability (described later), but the H2 permeance was also improved compared to the TEOS-derived membranes. To further improve the H2 permeance, the silica precursor was changed from TEOS to DMDMOS. Figure6d shows the results for the membrane prepared by counter di ffusion CVD of DMDMOS and O2. In this case, the membrane fabrication was mostly finished within the first 15 min. After that, the H2 permeance became almost constant, whereas the N2 permeance still decreased slowly, probably because of the modification of some small remaining pores. It is evident that thermal decomposition of DMDMOS was almost negligible, and it actually needed the presence of O2 to form a SiO2 film. 7 After 60 min of CVD, the H2 permeance and the H2/N2 selectivity at 500 ◦C were 3.3 10− mol 2 1 1 × m− s− Pa− and 82, respectively. Considering the fact that the measurement was taken at 500 ◦C, the H2 permeance was already much higher than that for a TEOS-derived membrane, but it was not as high as the literature value (9.0 10 7 mol m 2 s 1 Pa 1)[17]. Figure6e shows the results for × − − − − the membrane prepared by one-sided diffusion CVD of DMDMOS and O2. After 45 min of CVD, the H permeance and the H /N selectivity at 500 C were 3.8 10 7 mol m 2 s 1 Pa 1 and 90, 2 2 2 ◦ × − − − − respectively. Finally, Figure6f shows the results of the membrane prepared by one-sided di ffusion CVD of DMDMOS/ATSB/O2. After 45 min of CVD, the H2 permeance and the H2/N2 selectivity at 500 C were 4.0 10 7 mol m 2 s 1 Pa 1 and 130, respectively. ◦ × − − − − Overall, the H2 permeances of the DMDMOS-derived membranes were not as high as that reported in the literature (9.0 10 7 mol m 2 s 1 Pa 1)[17], probably due to the lack of optimization of the CVD × − − − − parameters. However, the permeance measurements for several gas species (Figure7a) revealed that the DMDMOS-derived membrane may have a looser silica network compared to the TEOS-derived membranes. The first evidence for this inference is that the He/H2 permeance ratio for the fresh DMDMOS membrane was as low as 1.2. The value was much lower than those for TEOS-derived membranes (usually >2.0), suggesting that the average pore size of the DMDMOS-derived silica network was significantly larger than the kinetic diameters of H2 (0.289 nm) or He (0.26 nm). In fact, the H2 permeance was higher than that for a TEOS-derived membrane, but the He permeance was lower. Membranes 2020,, 10,, 50x FOR PEER REVIEW 1212 of of 21

-5 3 -5 3 -1 -1 10 10 10 10 Pa Pa -1 H

-1 -6

-6 H s

10 2 s

10 2 /N -2 /N -2 2 2 2

2 10 10 -7 selectivity / - -7 selectivity / - 10 H 10 H 2 2 N N -8 2 10-8 2 10 H /N 1 H /N 1 2 2 10 2 2 10 -9 10-9 10

-10 0 -10 0 Permeancem / mol Permeance / mol m 10 10 10 10 0 30 60 90 120 0 306090 Deposition time / min Deposition time / min

(a) Change in gas permeances with deposition time (b) Change in gas permeances with deposition time for a membrane prepared by one-sided diffusion for a membrane prepared by one-sided diffusion CVD of pure TEOS CVD of pure TEOS (dip-coating: 40 nm sol only) (dip-coating: 80 nm sol and 40 nm sol)

-5 3 -5 3 -1 -1 10 10 10 10 Pa Pa -1 -1 H -6 H -6 s s H 10 2 10 2 2 /N /N -2 -2 2 2 2 10 2 10 selectivity / - / selectivity -7 - / selectivity -7 10 H 10 H /N 2 2 2 N -8 2 -8 10 H /N 1 10 1 2 2 10 10 N 10-9 10-9 2

-10 0 -10 0 Permeance / mol m mol / Permeance Permeance / mol m mol / Permeance 10 10 10 10 0 153045607590105 015304560

Deposition time / min Deposition time / min

(c) Change in gas permeances with deposition time (d) Change in gas permeances with deposition time for a membrane prepared by one-sided diffusion for a membrane prepared by counter diffusion CVD CVD of TEOS/ATSB of DMDMOS/O2 (dip-coating: 80 nm sol and 40 nm sol) (dip-coating: 80 nm sol and 40 nm sol)

-5 3 -5 3 -1 -1 10 10 10 10 Pa Pa -1

-6 H -1

-6 H s

10 2 s /N

10 2 -2 /N -2 2

2 10 2 selectivity / - 10 2 -7 -7 selectivity / - H 10 H 10 2 2 N N -8 2 -8 2 10 H /N 1 10 H /N 1 2 2 2 2 10 10 -9 10-9 10

-10 0 -10 0 Permeance /mol m Permeance / mol m / mol Permeance 10 10 10 10 0153045 0153045 Deposition time / min Deposition time / min

(e) Change in gas permeances with deposition time (f) Change in gas permeances with deposition time for a membrane prepared by one-sided diffusion for a membrane prepared by one-sided diffusion

CVD of DMDMOS/O2 CVD of DMDMOS/ATSB/O2 (dip-coating: 80 nm sol and 40 nm sol)

FigureFigure 6. Change 6. Change in gas in gas permeances permeances with with deposition deposition time time for forvarious various types types of membranes: of membranes. (a–f).

The data for the didifferentﬀerent gasesgases (Figure(Figure7 7a)a) can can be be analyzed analyzed by by a a normalized normalized Knudsen-based Knudsen-based permeance method [[20].20]. The The method method is derived from thethe gasgas translationtranslation model.model. First, the gas permeances of the various gases with didifferentﬀerent kinetickinetic diametersdiameters were convertedconverted to normalizednormalized Knudsen permeances using Equation (5). The obtained results were then plotted as a function of

Membranes 2020, 10, x FOR PEER REVIEW 13 of 21 Membranes 2020, 10, 50 13 of 21 molecular size (Figure 7b). Finally, Equation (6) was used to fit the experimental data, using dp as fittingKnudsen parameter. permeances using Equation (5). The obtained results were then plotted as a function of molecular size (Figure7b). Finally, Equation (6) was used to fit the experimental data, using d p as 𝑃 fitting parameter. 𝑓 = 𝑀 (5) 𝑃 Pi f = q 𝑀 (5) MHe PHe 𝑑Mi 1− 𝑑3 𝑓 = 1 di (6) − 𝑑dp f =1− (6) 𝑑 3 1 dHe − dp In these equations, where f represents the ratio of the permeance of the i-th component (𝑃) to that predictedIn these equations, from a reference where componentf represents (He) the ratiobased of on the the permeance Knudsen diffusion of the i-th mechanism, component di ( isPi )the to kineticthat predicted diameter from of gas a reference i (nm), d componentHe is the kinetic (He) diameter based on of the helium Knudsen and didp ffisusion the estimated mechanism, membranedi is the kinetic diameter of gas𝑀i (nm), dHe is the kinetic diameter of helium and dp is the estimated membrane 𝑃 q pore size (nm). M 𝑀 is the permeance of the i-th component predicted from the He pore size (nm). P He is the permeance of the i-th component predicted from the He permeance He Mi permeance under the Knudsen diffusion mechanism. MHe and Mi are the molecular weight of helium under the Knudsen diffusion mechanism. MHe and Mi are the molecular weight of helium and gas i, and gas i, respectively1 (g mol-1). The analysis gives a pore size of 0.39 nm (Figure 7b). A comparable respectively (g mol− ). The analysis gives a pore size of 0.39 nm (Figure7b). A comparable pure SiO 2 puremembrane SiO2 membrane derived from derived TEOS from gives TEOS a pore gives size a pore of 0.34 size nm of [0.3453] tonm 0.37 [53] nm to [0.3760], nm confirming [60], confirming that the thatstructure the structure of the DMDMOS-derived of the DMDMOS-derived membrane membrane is looser. is looser.

(a) (b) -1 1.0 10-6 He H

Pa 2

-1 0.8 s

-7 e -2

10 c

Ne n 0.6 CO a

-8 2 Ar e

10 m N2 r 0.4 d = 0.39 nm e p -9 p 10 SF6 0.2

-10 0.0

10 Normalized Knudsen-based 0.2 0.3 0.4 0.5 0.6 0.20 0.25 0.30 0.35 0.40 Permeance / mol m mol / Permeance Kinetic diameter / nm Kinetic diameter /nm

Figure 7. (a) GasGas permeancespermeances for for a a membrane membrane synthesized synthesized by by counter-di counter-diffusionﬀusion CVD CVD of DMDMOS of DMDMOS and b andO2 measured O2 measured at 650 at ◦650C (before °C (before hydrothermal hydrothermal treatment), treatment), ( ) determination(b) determination of pore of pore size. size.

The second evidence for the more open structure of the DMDMOS-derived membrane over a pure TEOS-derived membrane is that the permeance order of CO2, Ar, and N2 followed the order of pure TEOS-derived membrane is that the permeance order of CO2, Ar, and N2 followed the order of their molecular size rather than their molecular we weight.ight. This This behavior behavior can can hardly hardly be be observed for for a TEOS-derived membrane, membrane, where where larger larger gas gas molecule moleculess permeate permeate through through defects defects in the in the membrane membrane by theby theKnudsen Knudsen mechanism, mechanism, thereby thereby resulting resulting in ina apermeance permeance order order dominated dominated by by their their molecular weight. The molecular sieving eﬀect for CO2, Ar, and N2 strongly suggests that there actually were weight. The molecular sieving effect for CO2, Ar, and N2 strongly suggests that there actually were some micropores present which allowed their passag passage.e. Furthermore, Furthermore, although the permeances of the 9 2 1 1 above three gas species were on the order of 10− −9mol m−−2 s−−1 Pa−1 , the permeance of SF6 was on the above three gas species were on the order of 10 mol m s Pa , the permeance of SF6 was on the order of 10 10 mol m 2 s 1 Pa 1. This indicated that the membrane had few defects such as pinholes or order of 10−−10 mol m−−2 s−1− Pa−1.− This indicated that the membrane had few defects such as pinholes or cracks, and that CO2, Ar, and N2 permeated through the membrane mainly by the molecular sieving cracks, and that CO2, Ar, and N2 permeated through the membrane mainly by the molecular sieving mechanism, and the contribution of the Knudsen permeationpermeation was notnot important.important. Considering the above ﬁndings, there would be a chance to further improve the H2 permeance of the DMDMOS above findings, there would be a chance to further improve the H2 permeance of the DMDMOS membranes byby minimizing minimizing the the membrane membrane thickness thickness through through the optimization the optimization of the CVD of parameters.the CVD parameters.

3.3. Hydrothermal Stability of the Membranes

Membranes 2020, 10, 50 14 of 21

Membranes 2020, 10, x FOR PEER REVIEW 14 of 21 3.3. Hydrothermal Stability of the Membranes First, a long-term hydrothermal stability test was conducted for both the inside and outside of First, a long-term hydrothermal stability test was conducted for both the inside and outside of the the membrane tube using a membrane prepared from TEOS and ATSB. The result is shown in Figure membrane tube using a membrane prepared from TEOS and ATSB. The result is shown in Figure8. 8. When the inner tube was exposed to 16 mol% water vapor, the H2 permeance was rapidly lost for When the inner tube was exposed to 16 mol% water vapor, the H permeance was rapidly lost for the first 12 h, and then became somewhat stabilized after 96 h of exposure2 . The tendency was quite the ﬁrst 12 h, and then became somewhat stabilized after 96 h of exposure. The tendency was quite normal, and the loss of permeance was much smaller than that for pure silica membranes. The N2 normal, and the loss of permeance was much smaller than that for pure silica membranes. The N permeance remained almost unchanged. Next, the configuration of the apparatus was altered and2 permeance remained almost unchanged. Next, the conﬁguration of the apparatus was altered and the outer shell was exposed to 16 mol% water vapor. The original purpose of this treatment was to the outer shell was exposed to 16 mol% water vapor. The original purpose of this treatment was to examine the degree of pinhole enlargement caused by the deterioration of γ-alumina. examine the degree of pinhole enlargement caused by the deterioration of γ-alumina.

-6 -1 10 4 He/H Pa

-1 H

2 2

-7 permeance ratio / - s 3

-2 10

10-8 ↑↓ 2 Exposure Exposure from inside from outside -9 10 N 1 2

10-10 0 Permeance / mol m / mol Permeance 0 50 100 150 200 Exposure time / h

Figure 8. Results of the long-term hydrothermal stability tests to both the inside and outside of the membraneFigure 8. Results tube. of the long-term hydrothermal stability tests to both the inside and outside of the membrane tube. However, unlike the case of membranes prepared by counter-diffusion CVD [41,49], there was no However, unlike the case of membranes prepared by counter-diffusion CVD [41,49], there was remarkable change in the N2 permeance. This was probably due to the structural difference between theno remarkable membranes preparedchange in by the counter-di N2 permeance.ffusion CVDThis andwas one-sided probably diduffusione to the CVD. structural In the former difference case, thebetween silica the film membranes is generally prepared deposited by within counter-diff the poresusion of γCVD-alumina, and one-sided thus the continuitydiffusion CVD. of the In silica the mightformer be case, susceptive the silica to film the is structural generally change deposited of γ -alumina.within the Inpores the latterof γ-alumina, case, a thin thus silica the continuity “layer” is depositedof the silica on might top of be the susceptiveγ-alumina tolayer, the structural so pinholes change can hardlyof γ-alumina. be created In the even latter if the case, average a thin Kelvin silica “layer” is deposited on top of the γ-alumina layer, so pinholes can hardly be created even if the pore diameter of γ-alumina increases by the hydrothermal exposure. Rather than a change of the N2 average Kelvin pore diameter of γ-alumina increases by the hydrothermal exposure. Rather than a permeance, a sharp decrease of the H2 permeance was observed again for the first 12 h of exposure. change of the N2 permeance, a sharp decrease of the H2 permeance was observed again for the first This was probably because of the attack of the H2O molecules on both sides of the silica layer, as shown in12 Figureh of exposure.9. This was probably because of the attack of the H2O molecules on both sides of the silica layer, as shown in Figure 9. Although the reported kinetic diameter of H2O has some variation in the literature (0.265 nm [32], 0.2995 nm and 0.314 nm [61]), the H2O permeance is expected to be more than two magnitudes lower than the H2 permeance according to the literature [57]. Therefore, it is likely that only the silica network near the surface underwent densification during the first hydrothermal treatment to the inner tube. Then, the silica network near the intersection between the silica and γ-alumina underwent densification during the treatment of the outer shell. Another remarkable finding was that the He/H2 permeance ratio tended to increase by exposure to hydrothermal conditions, which has also been pointed out in the literature [41]. This is also indicative of the densification of the silica layer. The kinetic diameters of He and H2 are 0.26 and 0.289 nm respectively, whereas the pore size of the TEOS-derived membrane right after synthesis is roughly estimated to be 0.3 nm. As the densification proceeds, the pore size and the H2 molecular size will become extremely close to each other, but there still will be a relatively

Membranes 2020, 10, 50 15 of 21

Membraneslarge diﬀ 2020erence, 10, betweenx FOR PEER the REVIEW poresize and the He molecular size. This is why the reduction in15 the of He21 permeance was not as sharp as that in the H2 permeance.

Figure 9. Proposed mechanism for the sharp decrease of the H2 permeance, which was observed twice. Figure 9. Proposed mechanism for the sharp decrease of the H2 permeance, which was observed twiceNext,. a hydrothermal stability test was conducted for a pure silica membrane synthesized from DMDMOS, and the result is shown in Figure 10. It was revealed that, although the DMDMOS-derived membraneAlthough had the better reported initial kinetic performance diameter than of H a2 TEOS-derivedO has some variation membrane, in the it hadliterature extremely (0.265 poor nm [32],hydrothermal 0.2995 nmstability and 0.314 and nm could [61]), not themaintain H2O permeance its superior is expected performance to be more over than a long two period magnitudes of time. 7 2 1 1 lowerThe H 2thanpermeance the H2 permeance before the hydrothermalaccording to the exposure literature was [57]. 3.2 Ther10−efore,mol mit −is likelys− Pa that− at only 650 ◦theC, andsilica it × 8 2 1 1 networkwas reduced near by the about surface 75%, underwent and that after densification 192 h of exposure during wasthe 7.8first 10hydrothermal− mol m− s treatment− Pa− . Another to the × innerremarkable tube. Then, ﬁnding thewas silica that network the N 2nearpermeance the intersection also underwent between the a very silica large and drop γ-alumina fromthe underwent order of 9 2 1 1 10 2 1 1 densification10− mol m− durings− Pa− theto treatment 10− mol mof− thes− outerPa− shbyell. the Another hydrothermal remarkable exposure. finding This was strongly that the suggests He/H2 permeancethatMembranes the micropores 2020 ratio, 10, xtended FOR through PEER to REVIEWincrease which N by2 could exposure permeate to hydrothermal had shrunk into conditions, much smaller which micropores has also16 been of that 21 pointedno longer out allowed in the theliterature permeation [41]. This of N 2is. also indicative of the densification of the silica layer. The kinetic diameters of He and H2 are 0.26 and 0.289 nm respectively, whereas the pore size of the TEOS- -6 -1 derived membrane right10 after synthesis is roughly estimated to be 0.3 nm.4 AsHe/H the densification proceeds, the pore size and the H2 molecular size will become extremely close to each other, but there Pa -1

still will be a relatively large difference between the pore size and theHe He molecular2 size. This is why

-7 permeance ratio / - s 2 3 the reduction in the-2 He permeance10 was not as sharp as that in the H Hpermeance. 2 Next, a hydrothermal stability test was conducted for a pure silicaHe/H membrane synthesized from 2 DMDMOS, and the result is-8 shown in Figure 10. It was revealed that, although the DMDMOS-derived membrane had better initial10 performance than a TEOS-derived membrane,2 it had extremely poor hydrothermal stability and could not maintain its superior performance over a long period of time. −7 −2 −1 −1 The H2 permeance before10 the-9 hydrothermal exposure was 3.2 × 10 mol m 1 s Pa at 650 °C, and it was reduced by about 75%, and that after 192 h of exposure was 7.8 × 10−8 mol m−2 s−1 Pa−1. Another N remarkable finding was that the N2 permeance also underwent a very2 large drop from the order of 10−9 mol m−2 s−1 Pa−1 to 1010−10-10 mol m−2 s−1 Pa−1 by the hydrothermal exposure. This0 strongly suggests that Permeance / mol m / mol Permeance the micropores through which0 N2 could 50 permeate 100 had shrunk 150 into 200 much smaller micropores that no longer allowed the permeation of N2. Exposure time / h

Figure 10. Results of the hydrothermal stability test for a pure silica membrane synthesized fromFigure DMDMOS. 10. Results of the hydrothermal stability test for a pure silica membrane synthesized from DMDMOS.

To obtain a deeper understanding of the gas permeation mechanism of this new membrane, the temperature dependence of the gas permeances after the hydrothermal treatment was investigated and the result is shown in Figure 11. The gas permeation behavior after the densification was quite similar to that for a TEOS-derived membrane, probably because the average pore size of the silica layer became about 0.3 nm or smaller by the hydrothermal exposure. It is remarkable that, although the permeance of CO2 was higher than that of N2 before the exposure, their permeance order became reversed after the exposure, indicating that the permeation mechanism changed from molecular sieve to Knudsen. In any case, a membrane with H2 permeance on the order of 10−8 mol m−2 s−1 Pa−1 is not attractive for membrane reactor applications as a minimum permeance of 10−7 mol m−2 s−1 Pa−1 is needed [40], and therefore, improvement was necessary.

-6 -1 10 Pa -1 -7 He s

-2 10 H 2 Ne 10-8

10-9 N 2 CO 2 10-10 Permeance / mol m Permeance / mol 0 100 200 300 400 500 600 700 Temperature / ℃

Figure 11. Temperature dependence of the gas permeances after the hydrothermal treatment for a pure silica membrane synthesized from DMDMOS.

Then, a hydrothermal stability test was conducted for a silica-alumina composite membrane synthesized by the one-sided diffusion CVD of DMDMOS/ATSB/O2, and the result is shown in Figure 12. It can be seen that the hydrothermal stability of the membrane was significantly improved. The initial H2 permeance was 4.0 × 10−7 mol m−2 s−1 Pa−1, and that after 94 h of exposure became 2.5 × 10−7

Membranes 2020, 10, x FOR PEER REVIEW 16 of 21

-6 -1 10 4 He/H Pa -1

He 2

-7 permeance ratio / - s 3 -2 10 H 2 He/H 2 10-8 2

10-9 1 N 2 10-10 0 Permeance / mol m / mol Permeance 0 50 100 150 200 Exposure time / h

MembranesFigure2020 10., 10 Results, 50 of the hydrothermal stability test for a pure silica membrane synthesized from16 of 21 DMDMOS.

To obtain a a deeper deeper understanding understanding of of the the gas gas permeation permeation mechanism mechanism of this of this new new membrane, membrane, the temperaturethe temperature dependence dependence of the of the gas gas permeances permeances after after the the hydrothermal hydrothermal treatment treatment was was investigated investigated and the result is shown in Figure 1111.. The gas permeation behavior after the densiﬁcationdensification was quite similar toto thatthat forfor a a TEOS-derived TEOS-derived membrane, membrane, probably probab becausely because the the average average pore pore size ofsize the of silica the silica layer layerbecame became about about 0.3 nm 0.3 or nm smaller or smaller by the by hydrothermal the hydrothermal exposure. exposure. It is It remarkable is remarkable that, that, although although the thepermeance permeance of COof CO2 was2 was higher higher than than that that of of N N2 before2 before the the exposure, exposure,their their permeancepermeance order became reversed after the exposure, exposure, indica indicatingting that the the permeation permeation mechanism mechanism changed changed from from molecular molecular sieve sieve 8 2 1 1 to Knudsen. In anyany case,case, aa membranemembrane with with H H2 2permeance permeance on on the the order order of of 10 10− −mol8 mol m m− −2s −s−1 PaPa−−1 isis not not 7 2 1 1 attractive for membranemembrane reactorreactor applicationsapplications asas aa minimum minimum permeance permeance of of 10 10− −7mol mol m m− −2s s−−1 PaPa−−1 is needed [40], [40], and therefore, improvement was necessary.

-6 -1 10 Pa -1 -7 He s

-2 10 H 2 Ne 10-8

10-9 N 2 CO 2 10-10 Permeance / mol m Permeance / mol 0 100 200 300 400 500 600 700 Temperature / ℃

Figure 11. Temperature dependence of the gas permeances after the hydrothermal treatment for a pure Figuresilica membrane 11. Temperature synthesized dependence from DMDMOS. of the gas permeances after the hydrothermal treatment for a pure silica membrane synthesized from DMDMOS. Then, a hydrothermal stability test was conducted for a silica-alumina composite membrane synthesizedThen, a byhydrothermal the one-sided stability diﬀusion test was CVD conducted of DMDMOS for /aATSB silica-alumina/O2, and thecomposite result is membrane shown in synthesizedFigureMembranes 12 2020. It by can, 10 the, x be FOR one-sided seen PEER that REVIEW thediffusion hydrothermal CVD of DMDMOS/ATSB/O stability of the membrane2, and the was result signiﬁcantly is shown improved. in 17Figure of 21 12.The It initial can be H seenpermeance that the hydr was 4.0othermal10 7 stabilitymol m 2ofs the1 Pa membrane1, and that was after significantly 94 h of exposure improved. became The 2 × − − − − initial2.5mol m10 H−2 27 spermeance−mol1 Pa− m1. The2 s was 1reductionPa 4.01. The× 10 was− reduction7 mol only m− 238% wass−1 Pa in only−1 ,this and 38% case. that in thisafterThis case. 94new h This ofmembrane exposure new membrane wasbecame used was2.5 for × used 10the−7 × − − − − formembrane the membrane reactor reactorstudies studies on the ethane on the ethanedehydrogenation dehydrogenation reaction reaction in a forthcoming in a forthcoming study. study.

-6 -1 10 4 He/H He Pa

-1 H

2 2

-7 / - ratio permeance s 3

-2 10

-8 10 He/H 2 2

-9 10 N 1 2

10-10 0 Permeance / m mol 0 25 50 75 100 Exposure time / h Figure 12.12.Result Result of theof hydrothermalthe hydrothermal stability stability test for test a silica-alumina for a silica-alumina composite membranecomposite synthesizedmembrane

bysynthesized one-side dibyﬀ usionone-side CVD diffusion of DMDMOS CVD of/ATSB DMDMOS/ATSB/O/O2. 2.

In order to make a comparison, surface SEM pictures after the long-term hydrothermal stability tests were taken for both a defect-free membrane and a deteriorated membrane. Figure 13a shows the surface of a membrane which still had H2/N2 selectivity above 100. Although the silica surface shows a granular morphology compared to that of a fresh surface (Figure 5d), the γ-alumina was still completely covered by a continuous silica layer. On the other hand, Figure 13b shows the surface of a different membrane which had H2/N2 selectivity of around only 10 after the hydrothermal exposure. There are many clearly visible pinhole defects, through which larger gas molecules such as N2 or CO2 could permeate by Knudsen diffusion.

200 nm 200 nm

(a) Surface of the topmost Si-Al composite layer (b) Surface of the topmost Si-Al composite layer after after hydrothermal treatment (membrane with final hydrothermal treatment (different membrane with H2/N2 selectivity > 100) final H2/N2 selectivity ~ 10)

Figure 13. SEM images of the membrane surface (a,b) after the long-term hydrothermal treatment.

Figure 14 shows a summary of the performance of a total number of 16 membranes fabricated in this study, the intermediate layers of which were synthesized by the sequential treatment with an 80 and a 40 nm sol. Even though the synthesis steps were simplified compared to a previous study

Membranes 2020, 10, x FOR PEER REVIEW 17 of 21 mol m−2 s−1 Pa−1. The reduction was only 38% in this case. This new membrane was used for the membrane reactor studies on the ethane dehydrogenation reaction in a forthcoming study.

-6 -1 10 4 He/H He Pa

-1 H

2 2

-7 / - ratio permeance s 3

-2 10

-8 10 He/H 2 2

-9 10 N 1 2

10-10 0 Permeance / m mol 0 25 50 75 100 Exposure time / h MembranesFigure2020 12., 10 ,Result 50 of the hydrothermal stability test for a silica-alumina composite membrane17 of 21

synthesized by one-side diffusion CVD of DMDMOS/ATSB/O2. In order to make a comparison, surface SEM pictures after the long-term hydrothermal stability In order to make a comparison, surface SEM pictures after the long-term hydrothermal stability tests were taken for both a defect-free membrane and a deteriorated membrane. Figure 13a shows tests were taken for both a defect-free membrane and a deteriorated membrane. Figure 13a shows the surface of a membrane which still had H2/N2 selectivity above 100. Although the silica surface the surface of a membrane which still had H2/N2 selectivity above 100. Although the silica surface shows a granular morphology compared to that of a fresh surface (Figure5d), the γ-alumina was still shows a granular morphology compared to that of a fresh surface (Figure 5d), the γ-alumina was still completely covered by a continuous silica layer. On the other hand, Figure 13b shows the surface of a completely covered by a continuous silica layer. On the other hand, Figure 13b shows the surface of diﬀerent membrane which had H2/N2 selectivity of around only 10 after the hydrothermal exposure. a different membrane which had H2/N2 selectivity of around only 10 after the hydrothermal exposure. There are many clearly visible pinhole defects, through which larger gas molecules such as N2 or CO2 There are many clearly visible pinhole defects, through which larger gas molecules such as N2 or CO2 could permeate by Knudsen diﬀusion. could permeate by Knudsen diffusion.

200 nm 200 nm

FigureFigure 13. SEM 13. SEM images images of the of membrane the membrane surface surface (a,b) after after the the long-term long-term hydrothermal hydrothermal treatment. treatment.

Figure 14 showsshows aa summarysummary ofof the the performance performance of of a a total total number number of of 16 16 membranes membranes fabricated fabricated in thisin this study, study, the the intermediate intermediate layers layers of whichof which were were synthesized synthesized by by the the sequential sequential treatment treatment with with an an 80 and80 and a 40 a nm40 nm sol. sol. Even Even though though the the synthesis synthesis steps steps were were simplified simplified compared compared to a previousto a previous study study [35], Si-Al composite membranes with H permeance > 1.5 10 7 mol m 2 s 1 Pa 1 and H /N selectivity > 2 × − − − − 2 2 100 have been successfully fabricated many times, demonstrating the level of reproducibility of the membrane fabrication process used in this study. There are many steps that enter into the synthesis, including joining of the support to the dense alumina tubes, the preparation of the sols, the dip-coating procedure, and the CVD process, and it is difficult to control all the steps. Nevertheless, the results are significant as the values of H permeance > 1.5 10 7 mol m 2 s 1 Pa 1 and H /N selectivity 2 × − − − − 2 2 > 100 noted above were shown to be requirements for a commercial process [36]. This involved the study of the model A B + C reaction, the dihydrogen-ation of ethane (C H C H + H ) over → 2 6 → 2 4 2 a 5 wt% Cr/ZSM-5 catalyst in a conventional packed-bed reactor (PBR) and in a membrane reactor (MR) fitted with hydrogen-selective silica membranes. Specifically, it was demonstrated that for the 7 6 2 1 1 all-important permeance range of 10− and 10− mol m− s− Pa− , there was an enhancement in yield as H2 selectivity increased from 10 to 20 and 30, but beyond 100, there was little improvement even with a selectivity of 1000 [36]. Thus, a selectivity of 100 is sufficient for most applications involving membrane reactors. It should be noted, however, that the membrane performance (especially the H2 permeance) tended to be reduced after the long-term hydrothermal treatment, and it was not easy to maintain the above-mentioned performance over a long period of time. It was also remarkable that, when the Membranes 2020, 10, x FOR PEER REVIEW 18 of 21

[35], Si-Al composite membranes with H2 permeance > 1.5 × 10−7 mol m−2 s−1 Pa−1 and H2/N2 selectivity > 100 have been successfully fabricated many times, demonstrating the level of reproducibility of the membrane fabrication process used in this study. There are many steps that enter into the synthesis, including joining of the support to the dense alumina tubes, the preparation of the sols, the dip- coating procedure, and the CVD process, and it is difficult to control all the steps. Nevertheless, the results are significant as the values of H2 permeance > 1.5 × 10−7 mol m−2 s−1 Pa−1 and H2/N2 selectivity > 100 noted above were shown to be requirements for a commercial process [36]. This involved the

study of the model A → B + C reaction, the dihydrogen-ation of ethane (C2H6 → C2H4 + H2) over a 5 wt% Cr/ZSM-5 catalyst in a conventional packed-bed reactor (PBR) and in a membrane reactor (MR) fitted with hydrogen-selective silica membranes. Specifically, it was demonstrated that for the all-important permeance range of 10−7 and 10−6 mol m−2 s−1 Pa−1, there was an enhancement in yield as H2 selectivity increased from 10 to 20 and 30, but beyond 100, there was little improvement even with a selectivity of 1000 [36]. Thus, a selectivity of 100 is sufficient for most applications involving membrane reactors. MembranesIt should2020, 10, be 50 noted, however, that the membrane performance (especially the H2 permeance)18 of 21 tended to be reduced after the long-term hydrothermal treatment, and it was not easy to maintain the above-mentioned performance over a long period of time. It was also remarkable that, when the initial H2/N2 selectivity was lower than 100, the membrane performance tended to deteriorate by the initial H2/N2 selectivity was lower than 100, the membrane performance tended to deteriorate by the hydrothermal treatment, probably because of fatal pinhole enlargement. This provides a direction for hydrothermal treatment, probably because of fatal pinhole enlargement. This provides a direction for improvement, that is the number of defects should be reduced at the outset. improvement, that is the number of defects should be reduced at the outset. 104 Pure Si membranes 103 (right after synthesis) Pure Si membrane (after hydrothermal treatment) 102

selectivity / - selectivity / Si-Al composite membranes 2 1 (right after synthesis) /N

2 10

H Si-Al composite membranes (after hydrothermal treatment) 100 10-8 10-7 10-6 H permeance / mol m-2 s-1 Pa-1 2 FigureFigure 14. 14.Summary Summary of of the the performance performance of of a totala total number number of 16of membranes16 membranes synthesized synthesized from from TEOS TEOS or TEOSor TEOS+ ATSB. + ATSB. 4. Conclusions 4. Conclusions 7 2 1 1 Silica-alumina composite membranes with H2 permeance > 1.5 10− mol m− s− Pa− and Silica-alumina composite membranes with H2 permeance > 1.5 ×× 10−7 mol m−2 s−1 Pa−1 and H2/N2 H /N selectivity > 100 have been successfully fabricated repeatedly, even though the membrane selectivity2 2 > 100 have been successfully fabricated repeatedly, even though the membrane fabrication fabrication steps were simpliﬁed compared to previous studies. In addition, the use of a novel silica steps were simplified compared to previous studies. In addition, the use of a novel silica precursor, precursor, dimethyl dimethoxy silane (DMDMOS), further improved the H2 permeance without dimethyl dimethoxy silane (DMDMOS), further improved the H2 permeance without compromising compromising the H2 selectivity. Even though the pure silica membranes synthesized from DMDMOS the H2 selectivity. Even though the pure silica membranes synthesized from DMDMOS had poor had poor hydrothermal stability just like the TEOS-derived membranes, the addition of aluminum hydrothermal stability just like the TEOS-derived membranes, the addition of aluminum tri-sec- tri-sec-butoxide (ATSB) rendered the membrane much more stable. butoxide (ATSB) rendered the membrane much more stable. Author Contributions: H.A. carried out the research and wrote the initial draft, A.T. and R.K. helped to direct the researchAuthor andContributions: analyze the H.A. results, carried T.S. helpedout the withresearch the scanningand wrote electron the initial microscopy, draft, A.T. S.T.O. and R.K. planned helped the to work, direct obtainedthe research thefunding, and analyze and the edited results, the manuscript.T.S. helped with All authorsthe scanning have electron read and microscopy, agreed to the S.T.O. published planned version the work, of theobtained manuscript. the funding, and edited the manuscript.

Funding:Funding:This This work work was was supported supported by by the the Japan Japan Science Science and and Technology Technology Agency Agency under under the the CREST CREST program, program, Grant Number JPMJCR16P2. Grant Number JPMJCR16P2. Conﬂicts of Interest: The authors declare no conﬂict of interest. Conflicts of Interest: The authors declare no conflict of interest. References

1. Gallucci, F.; Fernandez, E.; Corengia, P.; van Sint Annaland, M. Review Recent advances on membranes and membrane reactors for hydrogen production. Chem. Eng. Sci. 2013, 92, 40–66. [CrossRef] 2. Khatib, S.J.; Oyama, S.T. Review of silica membranes for hydrogen separation prepared by CVD. Sep. Purif. Technol. 2013, 111, 20–42. [CrossRef] 3. Khatib, S.J.; Oyama, S.T.; de Souza, K.R.; Noronha, F.B. Review of silica membranes for hydrogen separation prepared by chemical vapor deposition. In Inorganic, Polymeric and Composite Membranes, Structure, Function and Correlations, Membrane Science and Technology Series; Oyama, S.T., Stagg-Williams, S.M., Eds.; Elsevier Science: Amsterdam, The Netherlands, 2011; Volume 14, p. 25. 4. Ten Hove, M.; Luiten-Olieman, M.W.J.; Huiskes, C.; Nijmeijer, A.; Winnubst, L. Hydrothermal stability of silica, hybrid silica and Zr-doped hybrid silica membranes. Sep. Purif. Technol. 2017, 189, 48–53. [CrossRef]

5. Akamatsu, K.; Tago, T.; Seshimo, M.; Nakao, S.-I. Long-term stable H2 production from methylcyclohexane using a membrane reactor with a dimethoxydiphenylsilane-derived silica membrane prepared via chemical vapor deposition. Ind. Eng. Chem. Res. 2015, 54, 3996–4000. [CrossRef] Membranes 2020, 10, 50 19 of 21

6. Nagasawa, H.; Minamizawa, T.; Kanezashi, M.; Yoshioka, T.; Tsuru, T. Microporous organosilica membranes for gas separation prepared via PECVD using diﬀerent O/Si ratio precursors. J. Membr. Sci. 2015, 489, 11–19. [CrossRef] 7. Gao, X.; Diniz da Costa, J.C.; Bhatia, S.K. Adsorption and transport of gases in a supported microporous silica membrane. J. Membr. Sci. 2015, 460, 46–61. [CrossRef] 8. Yin, Y.; Xua, T.; Shen, X.; Wu, H.; Jiang, Z. Fabrication of chitosan/zwitterion functionalized titania–silica hybrid membranes with improved proton conductivity. J. Membr. Sci. 2014, 469, 355–363. [CrossRef] 9. Okubo, T.; Inoue, H. Introduction of speciﬁc gas reactivity to porous glass membranes by treatment with tetraethoxysilane. J. Membr. Sci. 1989, 42, 109–119. [CrossRef]

10. Gavalas, G.R.; Megiris, C.E.; Nam, S.W. Deposition of H2-permselective SiO2 films. Chem. Eng. Sci. 1989, 44, 1829–1839. [CrossRef] 11. De Lange, R.S.A.; Keizer, K.; Burggraaf, A.J. Analysis and theory of gas transport in microporous sol-gel derived ceramic membranes. J. Membr. Sci. 1995, 104, 81–91. [CrossRef] 12. Okubo, T.; Haruta, K.; Kusakabe, K.; Morooka, S.; Anzai, H.; Akiyama, S. Preparation of sol-gel derived thin membrane on a porous ceramic hollow filter by the filtration technique. J. Membr. Sci. 1991, 59, 73–83. [CrossRef]

13. Nam, S.W.; Ha, H.Y.; Yoon, S.P.; Han, J.; Lim, T.H.; Oh, I.H.; Hong, S.A. Hydrogen-permselective TiO2/SiO2 membranes formed by chemical vapor deposition. Korean Membr. J. 2001, 3, 69–78. 14. Megiris, C.E.; Glezer, J.H.E. Permeation of silicon dioxide ﬁlms by low-pressure chemical vapor deposition on dense and porous alumina substrates. Chem. Eng. Sci. 1992, 47, 3925–3933. [CrossRef] 15. Ngamou, P.; Overbeek, J.; Kreiter, R.; van Veen, H.; Vente, J.; Wienk, I.; Cuperus, P.; Creatore, M. Plasma deposited hybrid silica membranes with a controlled retention of organic bridges. J. Mater. Chem. A 2013, 1, 5567–5576. [CrossRef] 16. Walkiewicz-Pietrzykowska, A.; Cotrino, J.; Gonzalez-Elipe, A.R. Deposition of thin ﬁlms of SiOxCyH in a surfatron microwave plasma reactor with hexamethyldisiloxane as precursor. Chem. Vap. Depos. 2005, 11, 317–323. [CrossRef]

17. Nomura, M. Stability of H2-permselective SiO2 ﬁlms formed by chemical vapor deposition. Membrane 2007, 32, 340–346. [CrossRef] 18. Zhang, X.; Yamada, H.; Saito, T.; Kai, T.; Murakami, K.; Nakashima, M.; Ohshita, J.; Akamatsu, K.; Nakao, S. Development of hydrogen-selective triphenylmethoxysilane-derived silica membranes with tailored pore size by chemical vapor deposition. J. Membr. Sci. 2016, 499, 28–35. [CrossRef] 19. Niimi, T.; Nagasawa, H.; Kanezashi, M.; Yoshioka, T.; Ito, K.; Tsuru, T. Preparation of BTESE-derived organosilica membranes for catalytic membrane reactors of methylcyclohexane dehydrogenation. J. Membr. Sci. 2014, 455, 375–383. [CrossRef] 20. Lee, H.R.; Kanezashi, M.; Shimomura, Y.; Yoshioka, T.; Tsuru, T. Evaluation and fabrication of pore-size-tuned silica membranes with tetraethoxydimethyl disiloxane for gas separation. AIChE J. 2011, 57, 2755–2765. [CrossRef] 21. Yamaguchi, T.; Ying, X.; Tokimasa, Y.; Nair, B.N.; Sugawara, T.; Nakao, S.I. Reaction control of

tetraethylorthosilicate (TEOS)/O3 and tetramethylorthosilicate (TMOS)/O3 counter diﬀusuin chemical vapour deposition for preparation of molecular-sieve membranes. Phys. Chem. Chem. Phys. 2000, 2, 4465–4475. [CrossRef] 22. Kageyama, N.; Sugawara, T.; Takagaki, A.; Kikuchi, R.; Oyama, S.T. Overcoming pressure drop losses in membrane reactors by semi-batch operation. Sep. Purif. Tech. 2017, 185, 175–185. [CrossRef] 23. Hekkink, J.H.A.; de Lange, R.S.A.; Ten Hoeve, A.A.; Blankenvoorde, P.J.A.M.; Keizer, K.; Burggraaf, A.J.

Characterization and permeation properties of binary SiO2-TiO2 and SiO2-Al2O3 modiﬁed gamma-alumina membranes. Key Eng. Mater. 1991, 61, 375–378. 24. Yoshida, K.; Hirano, Y.; Fujii, H.; Tsuru, T.; Asaeda, M. Hydrothermal stability and performance of silica-zirconia membranes for hydrogen separation in hydrothermal conditions. J. Chem. Eng. Jpn. 2001, 34, 523–531. [CrossRef] 25. Kanezashi, M.; Yada, K.; Yoshioka, T.; Tsuru, T. Organic-inorganic hybrid silica membranes with controlled silica network size: Preparation and gas permeation characteristics. J. Membr. Sci. 2010, 348, 310–318. [CrossRef] Membranes 2020, 10, 50 20 of 21

26. Li, H.; Schygulla, U.; Hoffmann, J.; Nichoff, P.; Haas-Santo, K.; Dittmeyer, R. Experimental and modeling study of gas transport through composite ceramic membranes. Chem. Eng. Sci. 2014, 108, 94–102. [CrossRef] 27. Oyama, S.T.; Lee, D.; Hacarlioglu, P.; Saraf, R.F. Theory of hydrogen permeability in nonporous silica membranes. J. Membr. Sci. 2004, 244, 45–53. [CrossRef] 28. Gu, Y.; Oyama, S.T. Ultrathin, hydrogen-selective silica membranes deposited on alumina-graded structures prepared from size-controlled boehmite sols. J. Membr. Sci. 2007, 306, 216–227. [CrossRef] 29. Mise, Y.; Ahn, S.-J.; Takagaki, A.; Kikuchi, R.; Oyama, S.T. Fabrication and evaluation of trimethylmethoxysilane (TMMOS)-derived membranes for gas separation. Membranes 2019, 9, 123. [CrossRef] 30. Kato, H.; Lundin, S.-T.B.; Ahn, S.-J.; Takagaki, A.; Kikuchi, R.; Oyama, S.T. Gas separation silica membranes prepared by chemical vapor deposition of methyl-substituted silanes. Membranes 2019, 9, 144. [CrossRef] 31. Oyama, S.T.; Yamada, M.; Sugawara, T.; Takagaki, A.; Kikuchi, R. Review on mechanisms of gas permeation through inorganic membranes. J. Jpn. Pet. Inst. 2011, 54, 298–309. [CrossRef] 32. Igi, R.; Kanezashi, M.; Yoshioka, T.; Fujisaki, S.; Iwamoto, Y.; Tsuru, T. Permeation properties of hydrogen and water vapor through porous silica membranes at high temperatures. AIChE J. 2011, 57, 618–629. 33. Kanezashi, M.; Sasaki, T.; Tawarayama, H.; Nagasawa, H.; Yoshioka, T.; Ito, K.; Tsuru, T. Experimental and theoretical study on small gas permeation properties through amorphous silica membranes fabricated at different temperatures. J. Phys. Chem. 2014, 118, 20323–20331. [CrossRef] 34. Okubo, T.; Watanabe, M.; Kusakabe, K.; Morooka, S. Preparation of γ-alumina thin membrane by sol-gel processing and its characterization by gas permeation. J. Mater. Sci. 1990, 25, 4822–4827. [CrossRef] 35. Gu, Y.; Oyama, S.T. High molecular permeance in a poreless ceramic membrane. Adv. Mater. 2007, 19, 1636–1640. [CrossRef] 36. Ahn, S.J.; Yun, G.N.; Takagaki, A.; Kikuchi, R.; Oyama, S.T. Effects of pressure contact time, permeance, and selectivity in membrane reactors: The case of the dehydrogenation of ethane. Sep. Purif. Technol. 2018, 194, 197–206. [CrossRef] 37. Ma, L.-C.; Castro-Dominguez, B.; Kazantzis, N.K.; Ma, Y.H. A cost assessment study for a large-scale water gas shift catalytic membrane reactor module in the presence of uncertainty. Sep. Purif. Technol. 2016, 166, 205–212. [CrossRef] 38. Meng, L.; Yu, X.; Niimi, T.; Nagasawa, H.; Kanezashi, M.; Yoshioka, T.; Tsuru, T. Methylcyclohexane dehydrogenation for hydrogen production via a bimodal catalytic membrane reactor. AIChE J. 2015, 61, 1628–1638. [CrossRef] 39. Araki, S.; Miyanishi, H.; Yano, H.; Tanaka, S.; Miyake, Y. Water gas shift reaction in a membrane reactor using a high hydrogen permselective silica membrane. Sep. Sci. Technol. 2013, 48, 76–83. [CrossRef] 40. Lim, H.; Gu, Y.; Oyama, S.T. Studies of the effect of pressure and hydrogen permeance on the ethanol steam reforming reaction with palladium- and silica-based membranes. J. Membr. Sci. 2012, 396, 119–127. [CrossRef]

41. Nagano, T.; Sato, K. Degradation mechanism of an H2-permselective amorphous silica membrane. J. Mater. Sci. 2014, 49, 4115–5120. [CrossRef] 42. Nomura, M.; Aida, H.; Gopalakrishnan, S.; Sugawara, T.; Nakao, S.I.; Yamazaki, S.; Inada, T.; Iwamoto, Y. Steam stability of a silica membrane prepared by counter diﬀusion chemical vapor deposition. Desalination 2006, 193, 1–7. [CrossRef] 43. Lim, H.; Gu, Y.; Oyama, S.T. Reaction of primary and secondary products in a membrane reactor: Studies of ethanol steam reforming with a silica-alumina composite membrane. J. Membr. Sci. 2010, 351, 149–159. [CrossRef] 44. Kageyama, N.; Takagaki, A.; Sugawara, T.; Kikuchi, R.; Oyama, S.T. Synthesis and characterization of a silica-alumina composite membrane and its application in a membrane reactor. Sep. Purif. Technol. 2018, 195, 437–445. [CrossRef] 45. Gu, Y.; Hacarlioglu, P.; Oyama, S.T. Hydrothermally stable silica–alumina composite membranes for hydrogen separation. J. Membr. Sci. 2008, 310, 28–37. [CrossRef] 46. Gu, Y.; Oyama, S.T. Permeation properties and hydrothermal stability of silica-titania membranes supported on porous alumina substrates. J. Membr. Sci. 2009, 345, 267–275. [CrossRef] 47. Sea, B.K.; Soewito, E.; Watanabe, M.; Kusakabe, K.; Morooka, S.; Kim, S.S. Hydrogen recovery from a

H2–H2O–HBr mixture utilizing silica-based membranes at elevated temperatures. 1. Preparation of H2O- and H2-selective membranes. Ind. End. Chem. Res. 1998, 37, 2502–2508. [CrossRef] Membranes 2020, 10, 50 21 of 21

48. Akamatsu, K.; Nakane, M.; Sugawara, T.; Nakao, S.I. Performance under thermal and hydrothermal condition of amorphous silica membrane prepared by chemical vapor deposition. AIChE J. 2009, 55, 2197–2203. [CrossRef]

49. Nagano, T.; Sato, K.; Saito, T.; Takahashi, S. Hydrothermal stability of mesoporous Ni-doped γ-Al2O3. J. Ceram. Soc. Jpn. 2009, 117, 832–835. [CrossRef] 50. Ohta, Y.; Akamatsu, K.; Sugawara, T.; Miyoshi, A.; Nakao, S. Development of pore-size-controlled silica membranes for gas separation by chemical vapor deposition. J. Membr. Sci. 2008, 315, 93–99. [CrossRef]

51. Seshimo, M.; Saito, T.; Akamatsu, K.; Segawa, A.; Nakao, S. Inﬂuence of toluene vapor on the H2-selective performance of dimethoxy diphenylsilane-derived silica membranes prepared by the chemical vapor deposition method. J. Membr. Sci. 2012, 415–416, 51–56. [CrossRef]

52. Seshimo, M.; Akamatsu, K.; Furuta, S.; Nakao, S. H2 purification durability of dimethoxy diphenylsilane-derived silica membranes with H2-toluene mixtures. Ind. Eng. Chem. Res. 2013, 52, 17257–17262. [CrossRef] 53. Ahn, S.-J.; Yun, G.-N.; Takagaki, A.; Kikuchi, R.; Oyama, S.T. Synthesis and characterization of hydrogen selective silica membrane prepared by chemical vapor deposition of vinyltriethoxysilane. J. Membr. Sci. 2018, 550, 1–8. [CrossRef] 54. Nishida, R.; Tago, T.; Saitoh, T.; Seshimo, M.; Nakao, S.-I. Development of CVD silica membranes having high hydrogen permeance and steam durability and a membrane reactor for a water gas shift reaction. Membranes 2019, 9, 140. [CrossRef][PubMed] 55. Sawamura, K.-I.; Okamoto, S.; Todokoro, Y. Development of mass production technology of highly permeable nano-porous supports for silica-based separation membranes. Membranes 2019, 9, 103. [CrossRef][PubMed] 56. Ishii, K.; Ikeda, A.; Takeuchi, T.; Yoshiura, J.; Nomura, M. Silica-based RO membranes for separation of acidic solution. Membranes 2019, 9, 94. [CrossRef][PubMed] 57. Nomura, M.; Nishi, Y.; Sakanishi, T.; Utsumi, K.; Nakamura, R. Preparation of thin LiSiO membranes by using a CVD method. Energy Procedia 2013, 37, 1012–1019. [CrossRef] 58. Oyama, S.T.; Hacarlioglu, P. The boundary between simple and complex descriptions of membrane reactors: The transition between 1-D and 2-D analysis. J. Membr. Sci 2009, 337, 188–199. [CrossRef] 59. Oyama, S.T.; Lee, D.; Sugiyama, S.; Fukui, K.; Iwasawa, Y. Characterization of a highly selective hydrogen permeable silica membrane. J. Mater. Sci. 2001, 36, 5213–5223. [CrossRef] 60. Belhaj Messaoud, S.; Takagaki, A.; Sugawara, T.; Kikuchi, R.; Oyama, S.T. Alkylamine-silica hybrid membranes for carbon dioxide/methane separation. J. Membr. Sci. 2015, 477, 161–171. [CrossRef] 61. Kanezashi, M.; Sasaki, T.; Tawarayama, H.; Yoshioka, T.; Tsuru, T. Hydrogen permeation properties and hydrothermal stability of sol-gel-derived amorphous silica membranes fabricated at high temperatures. J. Am. Ceram. Soc. 2013, 96, 2950–2957. [CrossRef]

© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).