Contents lists available at ScienceDirect

Journal of Membrane Science

journal homepage: www.elsevier.com/locate/memsci

Ceramic tubular nanofiltration membranes with tunable performances byatomic layer deposition and calcination

He Chen, Xiaojuan Jia, Mingjie Wei, Yong Wang⁎

State Key Laboratory of Materials-Oriented Chemical Engineering, Jiangsu National Synergetic Innovation Center for Advanced Materials, and College ofChemical Engineering, Nanjing Tech University, Nanjing 210009, PR China

A R T I C L E I N F O

Keywords:Atomic layer deposition (ALD)Ceramic membranesNanofiltrationTubular membranes

A B S T R A C T

Ceramic nanofiltration (NF) membranes are of particular significance for molecular separations under harshconditions. However, they are usually manufactured by the sol-gel process which frequently suffers from lowefficiency and poor control in the membrane properties. Herein we demonstrate an efficient and morecontrollable strategy to produce ceramic tubular NF membranes based on atomic layer deposition (ALD).Tubular ceramic membranes with pore size of ~5 nm are used as the substrates, on which titanium alkoxide(titanicone) is ALD-deposited. Subsequent calcination in air degrades the organic moieties in titanicone, thusconverting the dense layer of titanicone into a microporous layer of TiO2. This microporous TiO2 layer serves asa thin separation layer delivering the NF size-sieving function. The thickness of the TiO2 layer can be readilytuned by changing the ALD cycle numbers, and correspondingly the original substrate membranes areprogressively tightened with rising ALD cycles, and the membrane with 300 cycles exhibits a molecular-weight-cut-off (MWCO) of ~680 Da and water permeability of 30 L m−2 h−1 bar−1. Such a water permeability is higherthan many other ceramic tubular membranes with similar MWCOs because of the ultrathin nature of themicroporous TiO2 layer established by titanicone deposition and calcination.

1. Introduction

Nanofiltration (NF) has become increasingly important in people'sdaily life and also in industry for their expanding applications invarious fields, for instance, water softening and pharmaceutic synthesisand purification. There are both polymeric and ceramic NF mem-branes, and the latter are very much desired and sometimes evenirreplaceable in applications where high temperature, high pressure, oraggressive solvents are involved because of their superiority in thermal,chemical, and mechanical stability compared to their polymericcounterparts [1–4]. Ceramic NF membranes are typically producedthrough the sol-gel process, and very tight membranes with molecularweights cut off (MWCO) as low as < 1000 Da have been successfullyfabricated through this technique [5–8]. However, the water perme-ability of thus-produced NF membranes is often less than10 L m−2 h−1 bar−1, severely limiting their productivity in their appli-cations [9–11]. Recently, a TiO2-doped ZrO2 NF membrane with aMWCO of ~500 Da and water permeability of 35–40 L m−2 h−1 bar−1

was fabricated by using a modified colloidal sol-gel process [12].However, the sol-gel method for the preparation of NF membranes isrelatively tedious and less efficient as multiple rounds of time-

consuming coating and drying operations are required to form a tightselective layer on the macroporous substrates. Moreover, due to thenature of the sol-gel method, the thickness of the produced layer ofeach coating round is typically larger than a few hundreds ofnanometers. As a result, the thickness of the coating layers andconsequently the MWCOs cannot be tuned in a precise and progressiveway.

In contrast to the liquid phase sol-gel process, atomic layerdeposition (ALD) based on self-limiting reactions taking place on thegas-solid interfaces has emerged to be a highly precise technique forthe deposition of metal oxides on various substrates. ALD is capable ofdepositing conformal and uniform layers with a predictable control inthe thickness of the deposition layers at the sub-angstrom level [13]. Byleveraging the extreme precision in the thickness control as well as thehigh infiltrability of the vaporized precursors in ALD, we demonstratedthe capability and flexibility of ALD of metal oxides along the pore wallsin the progressively tuning pore sizes and surface properties andsubsequently the separation properties of a number of porous mem-branes [14–20]. For instance, Al2O3 has been successfully depositedinto ceramic microfiltration (MF) membranes to tune the pore size aswell as permeation of the membranes [17]. The performance of the

http://dx.doi.org/10.1016/j.memsci.2017.01.020Received 2 November 2016; Received in revised form 17 December 2016; Accepted 13 January 2017

⁎ Corresponding author.E-mail address: yongwang@njtech.edu.cn (Y. Wang).

Journal of Membrane Science 528 (2017) 95–102

Available online 14 January 20170376-7388/ © 2017 Elsevier B.V. All rights reserved.

MARK

http://www.sciencedirect.com/science/journal/03767388http://www.elsevier.com/locate/memscihttp://dx.doi.org/10.1016/j.memsci.2017.01.020http://dx.doi.org/10.1016/j.memsci.2017.01.020http://dx.doi.org/10.1016/j.memsci.2017.01.020http://crossmark.crossref.org/dialog/?doi=10.1016/j.memsci.2017.01.020&domain=pdf

membranes was changed gradually with the number of ALD cycles asthe water permeability was decreased while the retention rate wasincreased with cycle numbers. The original MF membranes were finallyturned to be ultrafiltration (UF) membranes with the increase of theALD cycles. In these previous works, because the substrate membranescontained pores typically larger than 30 nm vaporized ALD precursorspenetrated into deep interior of the membranes, producing coatinglayers not only on the membrane surfaces but also along the poresbeneath the surfaces. This strategy may also lead to NF membranesprovided we keep increasing ALD cycle numbers to significantly reducethe effective pore sizes down to a few nanometers. However, it requiresthousands of ALD cycles to shrink the original MF membranes into theNF category, which would be very tedious, time-consuming, and evenworse, the porosity of the produced NF membranes would be sig-nificantly reduced, thus consequently giving a low permeability.

Alternatively, ALD can be performed on substrate membranes withsmall pores (e.g. less than 20 nm). In this case, deposition inside themembrane will also take place but can be limited to a much lessthickness because of the pronounced diffusion resistance of the smallpores to ALD precursors with appropriately controlled depositionparameters. Additional dense layers would form on top of the substratemembranes with extending ALD cycles. Considering that the typicalprecursor, water, serving as the oxidant, can be replaced by alcohols, toincorporate organic moieties into the deposited layers which can besubsequently cavitated by removing these organic moieties, the as-deposited dense layers can be converted to microporous ones whichmay serve as selective layers exhibiting the NF size-discriminatingfunction. This modified ALD process using alcohols rather than wateras the oxidizing precursor is also known as molecular layer deposition(MLD) [21], however, the deposition mechanism has no difference withthe typical ALD process. The metal alkoxide films formed through thereaction of metal precursors and diols or triols are defined as“metalcones” [22], and correspondingly aluminum alkoxide and tita-nium alkoxide are known as alucone and titanicone, respectively.Recently, Liang et al. demonstrated that micropores can be obtainedon alucone-deposited silica particles after burning off or etching awaythe organic moieties in the metalcones [23]. The pore sizes werearound 1 nm, which is exactly falling in the range of pores of NFmembranes. Therefore, it is expected that ALD deposition of metal-cones on suitable porous substrates followed by cavitation to producecontinuous microporous ultrathin selective layers may open a newavenue for the highly controllable and efficient fabrication of inorganicNF membranes.

The right substrate membranes are vital for the production of high-quality NF membranes through this method of metalcone depositionand subsequent cavitation. The material, the pore size, and theconfiguration of the substrate membranes should all be carefullyconsidered from the aspect of ALD processing as well as the real-worldapplications of the produced NF membranes. Very recently, thin filmsof microporous TiO2 obtained by calcining titanicone exhibited ex-cellent retentions to salts and dyes through the NF mechanism [24].Unfortunately, anodic aluminum oxide (AAO) chips were used assubstrates in that work, which significantly hampers the possibility ofthese composite membranes to be used in real applications because ofthe poor mechanical stability and high cost of AAO supports. Moreover,long precursor pulse, exposure, and purge durations of a few minuteswere employed in this work, and the titanicone deposition predomi-nantly took place into deep interior of the AAO supports.

In the present work, tubular ceramic ultrafiltration (UF) mem-branes with the pore size of ~5 nm are chosen as the substrate. Thischoice is made because the tubular configuration is most extensivelyused in industry for ceramic membranes on one hand, and membraneswith pores significantly larger than 5 nm would need too many ALDcycles to form an additional depositing layer on the membrane surfaceon the other. On these tubular ceramic UF membranes, we depositdense titanicone layers with varied thickness by changing the deposi-

tion cycles. Short precursor pulse and exposure durations are used tolimit the deposition to take place in the near-surface region, thusforming a more asymmetric structure within short operation durations.The deposited tubular membranes are subsequently calcined to convertthe dense titanicone layers to microporous TiO2 layers. Thus producedcomposite structures exhibit excellent NF performances with tunableMWCOs simply by changing the ALD cycle numbers.

2. Experimental

2.1. Materials

Asymmetric tubular ceramic membranes consisting of α-aluminasupporting layers and TiO2 inner separation layers (Fraunhofer-Institut für Keramische Technologien und Systeme IKTS, Germany)were used as substrates. The tubular membranes have an externaldiameter of 10 mm and a tube wall thickness of 1.7 mm, and theiraverage pore size is 5 nm according to the manufacturer. TiCl4(99.99%, Metalorganic Center, Nanjing University), as one of thecheapest Ti sources, was used as metal source, while ethylene glycol(EG, 99.5%, TCI Shanghai) was served as organic source for the ALDprocess. N2 with ultra-high purity of 99.999% (Tianhong Gas, Nanjing)was used as the purge and carrier gas during the ALD process. Siliconwafers (SY100W-01, Seyang Electronics) were cleaned with ethanoland deionized water and dried in nitrogen gas before using as controlsubstrates. Polyethylene glycols (PEG, Aladdin, Shanghai, China) withvarious molecular weights (600, 1500, 4000 and 10,000 Da) were usedto determine the MWCOs of different membrane samples.

2.2. Titanicone ALD and calcination

The membrane samples were placed in the chamber of the ALDreactor (Savannah S100, Cambridge NanoTech) and were heated at100 °C for 10 min before the start of deposition in vacuum (~1 Torr).TiCl4 was maintained at room temperature, while EG was heated to80 °C. TiCl4 and EG vapor were pulsed into the reactor alternatively.For one titanicone ALD cycle, each precursor was pulsed into thechamber for 0.03 s, then held for 5 s, and subsequently the chamberwas purged for 10 s with a steady N2 flow rate of 20 sccm. The ceramicmembranes as well as silicon wafers were deposited for 100, 200, 300,400 and 500 cycles, respectively. The deposited samples were calcinedin air at 400 °C for 2 h with a heating rate of 1 °C/min in a hightemperature furnace (KSL-1100X-s, MTI Corporation).

2.3. Characterizations

Fourier Transformation Infrared (FTIR) spectra of the ceramicmembranes with different treatments and the bare membrane werecollected using a Nicolet 8700 FTIR spectrometer in the mode ofattenuated total reflection (ATR). Surface and cross-sectional morphol-ogies of the ceramic membranes with and without titanicone depositionand calcination were examined by a field emission scanning electronmicroscopy (FESEM, Hitachi S-4800) with an operating voltage of5 kV. A thin layer of Pd/Au alloy was sputtering-coated onto the surfaceof the samples before SEM observations to avoid electron charging.

X-ray diffraction patterns of the membrane samples were per-formed on a Rigaku MiniFlex 600 X-ray diffractometer with a Cu Kα X-ray source. XRD measurements were operated in the reflection modewith Cu-Kα radiation (40 kV, 15 mA) and a diffracted beam mono-chromator, using a step scan mode with a step of 0.02°. A spectroscopicellipsometer (Complete EASEM-2000U, J.A. Woollam) with an inci-dence angle of 70° was used to determine the thickness and refractiveindex of the TiO2 films formed on silicon wafers after titaniconedeposition and calcination.

H. Chen et al. Journal of Membrane Science 528 (2017) 95–102

96

2.4. Filtration tests

The water permeability and the retention of PEG were measured ina homemade apparatus at room temperature under the pressure of0.5 MPa. The retention for PEG was studied by filtration experimentsusing PEG mixtures with a total concentration of 3 g/L. The PEGretention rates of the membranes were determined by measuring thePEG concentrations of both feed and permeate side. The PEG retentionrate can be calculated through the following Eq. (1):

R C C(%) = (1 − / ) × 100p f (1)

where R is the PEG retention rate of the membrane, Cp and Cf is theconcentration of PEG solution in the permeate and feed solution,respectively. The PEG solutions were analyzed by gel permeationchromatography (GPC). GPC of PEG was performed withUltrahydrogel® columns using a Waters Isocratic HLPC 1515 pumpand a Waters 2414 refractive index detector. The molecular weight ofthe PEG corresponding to a 90% retention level was taken as theMWCO of the membrane.

3. Results and discussion

3.1. The fabrication process and the growth rate of titanicone andTiO2

Fig. 1 presents the schematic diagram of the preparation of ceramictubular NF membranes by ALD and subsequent calcination. The baremembrane has a layered structure including a supporting layer as wellas a separation layer (Fig. 1a). The precursors are alternativelyinfiltrated into the supporting layer as well as separation layer of thetubular membrane and are chemisorbed on the surface of particulatesconstituent of the supporting and separation layers. The secondprecursor (EG) reacted with the first precursor (TiCl4) previouslysorbed on the particulate surface following the surface reactions below:

-OH*+TiCl4→-O-TiCl3*+HCl (g)

-O-TiCl3*+3HO-C2H4-OH→-O-Ti-(O–C2H4-OH)3*+3HCl (g)

Titanicone is thus conformally coated along the particulates in theinitial stage, and subsequently a dense layer of titanicone is formed onthe top surface of the separation layer with rising ALD cycles (Fig. 1b).The thickness of this dense layer can be tuned simply by varying thedeposition cycles. Then the titanicone-deposited ceramic membranesare calcined at 400 °C for 2 h in air, so that organic components intitanicone are removed to generate microporous TiO2 layer, which willserve as the new selective layer delivering the NF function (Fig. 1c). Inthis way, it is expected that the permeation performance of the originalUF membranes could be tightened to be NF membranes with tunableMWCOs.

To investigate the deposition mechanism and the growth rate oftitanicone, ellipsometry was used to monitor the change of thethickness of titanicone deposited on silicon wafers before and aftercalcination with varied deposition cycles. As shown in Fig. 2, there is alinear correlation between the thickness of the deposited titaniconelayer and the cycle numbers, implying that the growth of titanicone isindeed following the ALD mechanism which produces constant thick-ness of deposited materials as a result of self-limiting reactions

between the two precursors. Furthermore, the as-fitted growth rate oftitanicone can be determined to be ~1.5 Å/cycle. Such a growth rate oftitanicone is much lower than that reported in other works (~4 Å/cycle)[21]. This should be ascribed to that they used a much longer pulsetime for each precursor (5 s for TiCl4 and 8.5 s for EG) whereas we usedonly 0.03 s for each precursor which allows much less precursors to besorbed on the surface, producing less amount of titanicone each cycle.There is a significant reduction in the thickness of the depositedtitanicone layers after calcination in air because of the elimination oforganic moieties in titanicone. After calcination the linear relationshipbetween the thickness of the deposited layers and the cycle numbers ismaintained, and correspondingly the growth rate of TiO2 layer can bedetermined to be ~0.4 Å/cycle, indicating that calcination does notweaken the integrity of the original titanicone layers. By comparing thethickness of the titanicone layers before and after calcination, we canestimate the shrinking ratio of the titanicone layer during calcination tobe 73%, which is similar to the reported value (77%) in other work[21].

3.2. Changes in compositions of the membranes during depositionand calcination

The deposited and the calcined membranes were characterized byFTIR, so as to investigate the surface chemistry of the treated ceramicmembranes. As can be clearly seen from Fig. 3, no obvious changescould be observed after deposition with 100 cycles. While a new peakemerged at around 1100 cm−1 after deposition with 500 cycles whichcan be attributed to the stretching vibration of C-O bonds in theprecursor EG, indicating the successful deposition of titanicone on thesurface of the membranes. It is worth noting that the peak at~1100 cm−1 disappeared after calcination, proving that the organiccomponents were removed during the calcination process and titani-cone was converted to TiO2.

Furthermore, XRD was used to confirm the successful formation ofTiO2 layer on the surface of the ceramic membrane, and the diffractionpeaks of the ceramic membranes before and after the titaniconedeposition and calcination were compared. As shown in Fig. 4, the

Fig. 1. Schematic diagram for the fabrication process of ceramic tubular NF membranes.

Fig. 2. The thickness of titanicone layers deposited on silicon wafers with different cyclenumbers and the thickness of the corresponding TiO2 layers after calcination.

H. Chen et al. Journal of Membrane Science 528 (2017) 95–102

97

bare membrane has very strong peaks originated from α-Al2O3 [25]and anatase phase of TiO2 [26], corresponding to the fact that thesupport layer and the separation layer were α-Al2O3 and anatase TiO2,respectively. After deposition with 400 cycles and calcination, theintensities of the diffraction peaks of both α-Al2O3 and anatase TiO2dramatically decreased, suggesting that the top layer of the pristinemembranes was somehow covered. The peak intensity further de-creased after deposition with 500 cycles and calcination, suggestingthat the thickness of the TiO2 layer was increased with the ALD cyclenumbers. We note that neither did the peak intensity of anatase TiO2enhance nor did any new peak appear after the calcination process.Moreover, we performed the same test on silicon wafers with identicaltreatments for comparison, and no peaks could be observed either, sowe can conclude that the newly formed TiO2 layer was amorphous anddid not show any crystalline peaks [27]. However, both α-Al2O3 andanatase TiO2 still can be detected after 500 deposition cycles andcalcination. This was ascribed to the probing depth of XRD, whichcould reach several micrometers of the samples [28] while the newlyformed TiO2 layer was merely less than 30 nm, therefore the peakintensity was not strong enough to shield characteristic diffractionpeaks of the pristine layer of the membrane.

3.3. Morphology evolution of the membranes during deposition andcalcination

To reveal the surface evolution of the membranes after titaniconedeposition and calcination, SEM was used to characterize the surfacemorphologies of the bare and the calcined membranes. As can be seenfrom Fig. 5a, the bare membrane exhibits a surface morphology ofaccumulated fine particulates as a result of its sol-gel fabricationprocess, and there are many tiny gaps between the adjacent particu-lates, which are defined as membrane pores with the diameter of a fewnanometers. The surface of the membrane became dense after 200 or500 deposition cycles and no pores could be observed. Since the growthrate of ALD titanicone is 1.5 Å/cycle and the pore size of bare ceramicmembranes is around 5 nm, after 20 or more deposition cycles, thepores would be fully filled after deposition for only 20 cycles whichproduce a coating layer with a thickness of ~3 nm. With more than 20cycles an extra thin layer would form on the surface of the separationlayer of the membrane surfaces. The surface of membranes after ALDfor 200 or 500 cycles is therefore dense and smooth, as can be seenfrom Fig. 5b and c. Because there is a significant reduction in thicknessafter calcination as we discussed earlier the thickness of the depositionlayer formed on membrane surface with 100 cycles drastically shrinksand the pores are unveiled after calcination (Fig. 5d). However, the sizeof the particulates on the surface becomes larger as a result of thecoating of TiO2 along the original particulates. For the membranesample with 500 ALD cycles, the thickness of the deposition layer islarge enough (can be observed in Fig. 6b), so that the layer aftercalcination still covers the original membrane surface (Fig. 5f). Forthose with 200 ALD cycles, the original membrane surface is partiallycovered, leaving behind some narrow pores (Fig. 5e).

We further investigated the evolution of the cross-sectional mor-phology of the tubular membranes with titanicone deposition andcalcination. As can be seen clearly from Fig. 6a, the bare membraneexhibits a layered structure consisting of a thick supporting layer and athin separation layer with a thickness of ~1.5 µm. After 500 depositioncycles, an additional dense and very thin layer with a thickness of 70–80 nm (highlighted in the square in Fig. 6b) can be observed on top ofthe original separation layer of the tubular membrane, which isbelieved to be the titanicone layer. After calcination in air, this thinlayer disappeared since the titanicone deposition layer shrank aftercalcination and the titanicone is turned to TiO2. It is also worth notingthat neither the deposition nor the calcination process had evidentinfluence on the cross-sectional morphology of the supporting layer ofceramic membranes as shown in Fig. 6a–c. In contrast to a relativelylong precursor pulse and exposure durations of several minutes used inother works [21,24,29], a relatively short pulsing and exposureduration (0.03 s and 5 s, respectively) was deliberately used in ourwork, which does not allow the precursors to diffuse through the tightselective layer to reach the supporting layer, limiting the depositionpredominantly happening in the selective layer of the tubular mem-brane. Such a deposition limited to the near-surface region is essentialin the preservation of permeation performance of the ceramic mem-branes after this modification process. Moreover, these shorteneddurations will significantly reduce the time required for the membranemodification by ALD, and also will consume much less precursors.

3.4. Separation performances of the membranes

The PWF of the bare and modified membranes were presented inFig. 7. The bare membrane has an initial PWF of ~100 L m−2 h−1 bar−1.With 100 deposition cycles and calcination, the PWF of the membranedropped to ~80 L m−2 h−1 bar−1. As revealed by SEM examinations,most of the pristine pores in the ceramic membrane were blocked bythe newly formed TiO2 layer after 100 deposition cycles and calcina-tion; nevertheless, the as-prepared layer was microporous and hydro-philic, which led to a relatively slight decrease in PWF. With deposition

Fig. 3. FTIR spectra of ceramic membranes subjected to different treatment.

Fig. 4. XRD patterns of the bare and the calcined membranes with different cycles.Peaks originated from α-Al2O3 and anatase phase TiO2 are highlighted with circles andtriangles, respectively.

H. Chen et al. Journal of Membrane Science 528 (2017) 95–102

98

for 200 cycles and calcination, the PWF of the membrane decreasedsharply to ~45 L m−2 h−1 bar−1. Such a significant drop of the permea-tion should be ascribed to that the pristine layer was gradually coveredby the microporous TiO2 layer, which was in good agreement with theSEM and XRD results. After 300 or more deposition cycles andcalcination, the PWF of the membrane further declined and remainedat a steady value of ~30 L m−2 h−1 bar−1. Therefore, after 300 or moredeposition cycles and subsequent calcination the newly formed micro-porous TiO2 layer was serving as the separation layer.

The permeability of a porous membrane is mainly determined bythe pore size, porosity, and the thickness of the selective layer. In thepresent work, the pore size is dependent on the type of the organicgroups in the precursors and is considered to remain unchanged as weconstantly used EG as the organic precursors. The thickness of theselective layer also changed very slightly as deposition with 100 morecycles and subsequent calcination only increased the thickness by~4 nm. We then investigate the change of porosity of the selectivelayers with cycle numbers.

The porosity of the microporous TiO2 formed on the tubularmembranes cannot be directly measured. Alternatively, we measuredthe porosities of the TiO2 layers formed on the silicon wafers whichshould be comparable to that formed on the tubular membranes asthey were deposited under identical conditions. The porosity of porousmaterials can be estimated by Eq. (2) [30]:

n n ϕ n ϕ= (1 − ) +f s pore air pore2 2 2 (2)

where nf, ns, and nair represent the refractive index of the porousmembrane, the material consisting of the membrane, and air, respec-tively. Φpore is the porosity of the membrane. In our case, ns and nairare known as 2.52 and 1, respectively, and nf can be determined byellipsometry.

As can be seen from Fig. 8, with 50 deposition cycles andcalcination, the porosity of the membrane was relatively high at~45%. This should be ascribed to that the silicon wafer has a thinbare SiO2 layer on the surface, which had non-negligible effects on thereflective index measurements of TiO2 layer since the TiO2 layer wasonly ~2 nm in thickness. The refractive index of the SiO2 layer wasdifferent from that of microporous TiO2 layer, which would affect thecalculation results of porosity inevitably. After 100 deposition cycles

and calcination, the porosity of the membrane dropped to ~35%, whichshould be attributed to that the newly formed TiO2 layer was ~4 nmand the influence of the bare SiO2 layer was weakened. The porosity ofthe membrane slightly decreased when the deposition cycles wasincreased to 200, which should also be ascribed to the furtherweakening influence of bare SiO2 layer. While with 300 or moredeposition cycles and calcination, the porosity of the membrane barelychanged and finally stabilized at ~30%. Therefore, because of the slightor no change in pore size, porosity, and thickness of the selective layer,the calcined membranes with 300 or more ALD cycles exhibit un-changed permeability.

We investigated the retention performances of the calcined mem-branes with various ALD cycles by testing their MWCOs. PEGs withvarious molecular weights were used as the probe molecules. The PEG-retention curves, as a function of molecular weight, of the bare and thecalcined membranes were presented in Fig. 9. The bare membraneexhibits a MWCO of ~7200 Da. MWCO of the calcined membraneswith a cycle number of 100 and 200 was reduced to ~5500 Da and~2600 Da, respectively. MWCO was further lowered down to ~680 Daby calcining the membrane deposited with 300 cycles. Further increasein deposition cycles exceeding 300 did not tighten the membrane anylonger. The calcined membranes with 400 and 500 cycles showedsimilar MWCOs to the membrane prepared with 300 cycles. Such atrend of changes in MWCOs with cycle numbers is similar to that ofwater permeability. With ALD cycles lower than 300 corresponding tothe initial stage of deposition and after calcination the coating ofmicroporous TiO2 predominantly took place around the particulatesconsisting of the original selective layer of the substrate membrane,thus reducing the effective pore size and consequently reduced thepermeability and MWCOs. With 300 or more cycle numbers themicroporous TiO2 layer was formed on the top surface of the substratemembranes to become the new selective layer. Consequently, thepermeability and retention remained mostly unchanged due to thevery limited change in pore size, porosity, and thickness of this newlayer as we discussed above.

On the basis of the MWCO, the membranes prepared by titaniconedeposition with a cycle number larger than 300 followed by calcinationcan be considered to be NF membranes with the pore size lying in therange of micropores [31]. That is to say, by depositing of titanicone for

Fig. 5. Surface SEM images of the bare ceramic membrane (a), the titanicone-deposited membrane with a cycle number of 200 (b) and 500 (c), the titanicone-deposited and calcinedceramic membrane with a cycle number of 100, 200, and 500 (d–f). (a)–(f) have the same magnification and the scale bar is shown in (f).

H. Chen et al. Journal of Membrane Science 528 (2017) 95–102

99

about 200–300 cycles, the initial UF membranes were converted to NFmembranes with tightening size-discriminating properties. The poresize of the membrane prepared by ALD for 300 cycles and subsequentcalcination can be estimated to be approximately 1.3 nm by fitting itsMWCO into Eq. (3) [32]:

r g= 0.1673 × (MW( /mol))0.557 (3)

where r is the molecular radius (Å), and MW is the molecular weight(Da).

By considering both the water permeability and MWCO of the

Fig. 6. The cross-sectional SEM images of the bare (a) and the deposited membranesbefore (b) and after (c) calcination at a deposition cycle number of 500. The bottom-leftand bottom-right panels in (a–c) show the enlarged morphology of the supporting layerand the selective layer of the membrane, respectively. The up and bottom panels in (a–c)have the same magnifications, respectively and the scale bars are shown in (c).

Fig. 7. Permeability of the bare membrane and the calcined membranes with varieddeposition cycles.

Fig. 8. Calculated porosity of porous TiO2 layer on silicon wafers with varied depositioncycles.

Fig. 9. PEG retention of the bare ceramic membrane and the calcined membranes withvaried deposition cycles.

H. Chen et al. Journal of Membrane Science 528 (2017) 95–102

100

membrane prepared by titanicone deposition for 300 cycles andsubsequent calcination, its NF performance is outstanding comparedto other NF membranes prepared by different methods. As shown inTable 1, our membrane exhibits larger water permeability than manyother ceramic tubular NF membranes with similar MWCOs.Considering also the high efficiency of the fabrication process andprecise control in membrane structure and properties this ALD andsubsequent calcination strategy is expected to be extended to produceother NF membranes (or even UF membranes with small pore sizes) ofdifferent materials with tunable properties.

4. Conclusions

In summary, we demonstrate a facile and effective method tofabricate ceramic tubular NF membranes. ALD is employed to deposittitanicone on the substrate membranes with a pore size of ~5 nm andTiCl4 and ethylene glycol are used as the precursors. We deliberatelyuse short pulse and exposure durations of both ALD precursors to limitthe deposition mainly taking place in the near-surface region of thesubstrate and an additional dense layer is formed after larger ALDcycles. This dense deposition layer is subsequently converted to amicroporous TiO2 layer by burning off the organic components, asrevealed by FTIR, XRD, and SEM characterizations. The separationproperties of the ceramic membranes can be tuned simply by varyingthe ALD cycles. When the cycle number is increased up to 300 themembranes exhibit progressively reduced water permeability andincreased retention rates. The water permeability and MWCO maintainlargely unchanged with further increasing cycle numbers beyond 300.The original substrate membrane gives a water permeability of~100 L m−2 h−1 bar−1 and MWCO of ~7200 Da, which are reduced to45 L m−2 h−1 bar−1 and ~2600 Da, respectively after deposition with200 cycles and calcination. The calcined membrane with 300 ALDcycles is tightened to the NF category and exhibits MWCO of ~680 Daand water permeability as high as 30 L m−2 h−1 bar−1 which is higherthan many other tubular NF membranes prepared by the sol-gelprocess. Such a deposition of metalcone and subsequent calcinationstrategy is expected to be used to prepare NF membranes with differentmaterials and the NF performances may be further tuned by usingprecursors with differently sized organic groups.

Acknowledgments

Financial support from the National Basic Research Program ofChina (2015CB655301), the Natural Science Research Program of theJiangsu Higher Education Institutions (13KJA430005), the NaturalScience Foundation of Jiangsu Province (BK20150063), and theProject of Priority Academic Program Development of JiangsuHigher Education Institutions (PAPD) is gratefully acknowledged.

References

[1] S. Condom, A. Larbot, S.A. Younssi, M. Persin, Use of ultra-and nanofiltration

ceramic membranes for desalination, Desalination 168 (2004) 207–213.[2] S. Benfer, U. Popp, H. Richter, C. Siewert, G. Tomandl, Development and

characterization of ceramic nanofiltration membranes, Sep. Purif. Technol. 22(2001) 231–237.

[3] A. Kermanpur, E. Ghassemali, S. Salemizadeh, Synthesis and characterisation ofmicroporous titania membranes by dip-coating of anodised alumina substratesusing sol–gel method, J. Alloy. Compd. 461 (2008) 331–335.

[4] J.M. Skluzacek, M.I. Tejedor, M.A. Anderson, NaCl rejection by an inorganicnanofiltration membrane in relation to its central pore potential, J. Membr. Sci.289 (2007) 32–39.

[5] Q. Xu, M.A. Anderson, Sol–gel route to synthesis of microporous ceramicmembranes: preparation and characterization of microporous TiO2 and ZrO2xerogels, J. Am. Ceram. Soc. 77 (1994) 1939–1945.

[6] R. Kreiter, M.D. Rietkerk, B.C. Bonekamp, H.M. van Veen, V.G. Kessler, J.F. Vente,Sol–gel routes for microporous zirconia and titania membranes, J. Sol.-Gel Sci.Technol. 48 (2008) 203–211.

[7] B. Topuz, M. Çiftçioğlu, Sol–gel derived mesoporous and microporous aluminamembranes, J. Sol.-Gel Sci. Technol. 56 (2010) 287–299.

[8] X. Chen, W. Zhang, Y. Lin, Y. Cai, M. Qiu, Y. Fan, Preparation of high-flux γ-alumina nanofiltration membranes by using a modified sol–gel method,Microporous Mesoporous Mater. 214 (2015) 195–203.

[9] A. Larbot, S. Alami-Younssi, M. Persin, J. Sarrazin, L. Cot, Preparation of a γ-alumina nanofiltration membrane, J. Membr. Sci. 97 (1994) 167–173.

[10] P. Baticle, C. Kiefer, N. Lakhchaf, A. Larbot, O. Leclerc, M. Persin, J. Sarrazin, Saltfiltration on gamma alumina nanofiltration membranes fired at two differenttemperatures, J. Membr. Sci. 135 (1997) 1–8.

[11] J. Schaep, C. Vandecasteele, B. Peeters, J. Luyten, C. Dotremont, D. Roels,Characteristics and retention properties of a mesoporous γ-Al2O3 membrane fornanofiltration, J. Membr. Sci. 163 (1999) 229–237.

[12] Y. Lu, T. Chen, X. Chen, M. Qiu, Y. Fan, Fabrication of TiO2-doped ZrO2nanofiltration membranes by using a modified colloidal sol-gel process and itsapplication in simulative radioactive effluent, J. Membr. Sci. 514 (2016) 476–486.

[13] S.M. George, Atomic layer deposition: an overview, Chem. Rev. 110 (2009)111–131.

[14] Q. Wang, X. Wang, Z. Wang, J. Huang, Y. Wang, PVDF membranes withsimultaneously enhanced permeability and selectivity by breaking the tradeoffeffect via atomic layer deposition of TiO2, J. Membr. Sci. 442 (2013) 57–64.

[15] F. Li, L. Li, X. Liao, Y. Wang, Precise pore size tuning and surface modifications ofpolymeric membranes using the atomic layer deposition technique, J. Membr. Sci.385 (2011) 1–9.

[16] Q. Xu, Y. Yang, X. Wang, Z. Wang, W. Jin, J. Huang, Y. Wang, Atomic layerdeposition of alumina on porous polytetrafluoroethylene membranes for enhancedhydrophilicity and separation performances, J. Membr. Sci. 415 (2012) 435–443.

[17] F. Li, Y. Yang, Y. Fan, W. Xing, Y. Wang, Modification of ceramic membranes forpore structure tailoring: the atomic layer deposition route, J. Membr. Sci. 397(2012) 17–23.

[18] Q. Xu, Y. Yang, J. Yang, X. Wang, Z. Wang, Y. Wang, Plasma activation of porouspolytetrafluoroethylene membranes for superior hydrophilicity and separationperformances via atomic layer deposition of TiO2, J. Membr. Sci. 443 (2013)62–68.

[19] Q. Xu, J. Yang, J. Dai, Y. Yang, X. Chen, Y. Wang, Hydrophilization of porouspolypropylene membranes by atomic layer deposition of TiO2 for simultaneouslyimproved permeability and selectivity, J. Membr. Sci. 448 (2013) 215–222.

[20] H. Chen, L. Kong, Y. Wang, Enhancing the hydrophilicity and water permeability ofpolypropylene membranes by nitric acid activation and metal oxide deposition, J.Membr. Sci. 487 (2015) 109–116.

[21] A.I. Abdulagatov, R.A. Hall, J.L. Sutherland, B.H. Lee, A.S. Cavanagh, S.M. George,Molecular layer deposition of titanicone films using TiCl4 and ethylene glycol orglycerol: growth and properties, Chem. Mater. 24 (2012) 2854–2863.

[22] S.M. George, B.H. Lee, B. Yoon, A.I. Abdulagatov, R.A. Hall, Metalcones: hybridorganic–inorganic films fabricated using atomic and molecular layer depositiontechniques, J. Nanosci. Nanotechnol. 11 (2011) 7948–7955.

[23] X. Liang, M. Yu, J. Li, Y.-B. Jiang, A.W. Weimer, Ultra-thin microporous–mesoporous metal oxide films prepared by molecular layer deposition (MLD),Chem. Commun. (2009) 7140–7142.

[24] Z. Song, M. Fathizadeh, Y. Huang, K.H. Chu, Y. Yoon, L. Wang, W.L. Xu, M. Yu,TiO2 nanofiltration membranes prepared by molecular layer deposition for waterpurification, J. Membr. Sci. 510 (2016) 72–78.

[25] J. Lee, J.-I. Lee, H.-I. Lee, Methodologies to observe and characterize interfacialreaction products in (Al2O3)p/Al and SiCp/Al composites-using SEM, XRD, TEM,Scr. Mater. 35 (1996) 721–726.

[26] K. Thamaphat, P. Limsuwan, B. Ngotawornchai, Phase characterization of TiO2powder by XRD and TEM, Kasetsart J. (Nat. Sci.) 42 (2008) 357–361.

[27] H. Zhang, M. Finnegan, J.F. Banfield, Preparing single-phase nanocrystallineanatase from amorphous titania with particle sizes tailored by temperature, NanoLett. 1 (2001) 81–85.

[28] T. Hirata, E. Asari, M. Kitajima, Infrared and Raman spectroscopic studies of ZrO2polymorphs doped with Y2O3 or CeO2, J. Solid State Chem. 110 (1994) 201–207.

[29] S.-M. Lee, E. Pippel, U. Gösele, C. Dresbach, Y. Qin, C.V. Chandran, T. Bräuniger,G. Hause, M. Knez, Greatly increased toughness of infiltrated spider silk, Science324 (2009) 488–492.

[30] T.C. Choy, Effective medium theory: principles and applications, Oxford UniversityPress, Oxford, 2015.

[31] J. Sekulić, J.E. ten Elshof, D.A. Blank, A microporous titania membrane fornanofiltration and pervaporation, Adv. Mater. 16 (2004) 1546–1550.

[32] P. Puhlfürß, A. Voigt, R. Weber, M. Morbé, Microporous TiO2 membranes with a

Table 1Water permeability and MWCOs of ceramic tubular NF membranes reported inliterature.

Water permeability (L m−2 h−1 bar−1) MWCO (Da) Ref.

7.8 1000 [33]13 750 [34]20 500–600 [35]20 1000 [36]20 1000 [8]26 990 [2]35–40 ~500 [12]30 680 This work

H. Chen et al. Journal of Membrane Science 528 (2017) 95–102

101

http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref1http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref1http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref2http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref2http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref2http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref3http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref3http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref3http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref4http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref4http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref4http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref5http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref5http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref5http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref6http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref6http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref6http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref7http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref7http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref8http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref8http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref8http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref9http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref9http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref10http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref10http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref10http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref11http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref11http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref11http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref12http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref12http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref12http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref13http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref13http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref14http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref14http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref14http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref15http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref15http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref15http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref16http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref16http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref16http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref17http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref17http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref17http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref18http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref18http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref18http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref18http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref19http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref19http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref19http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref20http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref20http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref20http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref21http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref21http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref21http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref22http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref22http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref22http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref23http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref23http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref23http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref24http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref24http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref24http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref25http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref25http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref25http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref26http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref26http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref27http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref27http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref27http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref28http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref28http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref29http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref29http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref29http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref30http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref30http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref31http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref31http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref32

cut off < 500 Da, J. Membr. Sci. 174 (2000) 123–133.[33] I. Sentana, R.D. Puche, E. Sentana, D. Prats, Reduction of chlorination byproducts

in surface water using ceramic nanofiltration membranes, Desalination 277 (2011)147–155.

[34] X. Da, X. Chen, B. Sun, J. Wen, M. Qiu, Y. Fan, Preparation of zirconiananofiltration membranes through an aqueous sol–gel process modified by glycerolfor the treatment of wastewater with high salinity, J. Membr. Sci. 504 (2016)

29–39.[35] T. Van Gestel, C. Vandecasteele, A. Buekenhoudt, C. Dotremont, J. Luyten,

R. Leysen, B. Van, der Bruggen, G. Maes, Salt retention in nanofiltration withmultilayer ceramic TiO2 membranes, J. Membr. Sci. 209 (2002) 379–389.

[36] M. Ebrahimi, D. Willershausen, K.S. Ashaghi, L. Engel, L. Placido, P. Mund,P. Bolduan, P. Czermak, Investigations on the use of different ceramic membranesfor efficient oil-field produced water treatment, Desalination 250 (2010) 991–996.

H. Chen et al. Journal of Membrane Science 528 (2017) 95–102

102

http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref32http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref33http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref33http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref33http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref34http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref34http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref34http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref34http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref35http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref35http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref35http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref36http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref36http://refhub.elsevier.com/S0376-7388(16)32115-9/sbref36

Ceramic tubular nanofiltration membranes with tunable performances by atomic layer deposition and calcinationIntroductionExperimentalMaterialsTitanicone ALD and calcinationCharacterizationsFiltration tests

Results and discussionThe fabrication process and the growth rate of titanicone and TiO2Changes in compositions of the membranes during deposition and calcinationMorphology evolution of the membranes during deposition and calcinationSeparation performances of the membranes

ConclusionsAcknowledgmentsReferences