TiO2 photoanodes with exposed {0 1 0} facets grown by aerosol-assisted chemical vapor deposition of a titanium oxo/alkoxy cluster

Photoelectrochemical water splitting is a promising technology for the development of solar fuels. Titanium dioxide (TiO2) is one of the most studied metal oxides in this field as a photoanode. Achieving its full potential requires controlling its morphology and crystallinity and especially the exposure of its most active crystal facets. Herein, we present the formation of nanostructured TiO2 photoanodes with anatase phase and high exposure of the {0 1 0} facet, the most active TiO2 phase and facet. TiO2 photoanodes were prepared from a Ti7O4(OEt)20 titanium oxo/alkoxy cluster solution using aerosol assisted chemical vapor deposition. Characterization techniques such as SEM and TEM reveal that these TiO2 photoanodes consist of morphologies resembling the crystals of gypsum, sand and water found in nature, also known as desert roses. Furthermore, TEM and XRD analysis also reveals that the metastable anatase TiO2 phase is maintained up to 1000 C and exceeds the typical anatase-to-rutile phase-transition temperature of 500–750 C, a feature that could be exploited in the smart ceramics industry. Photoelectrochemical measurements show that these desert-rose TiO2 photoanodes achieve excellent photocurrent densities with an incident photon-to-current efficiency of 100% at 350 nm and a faradaic efficiency for oxygen evolution of 90%.


Introduction
A key approach to reduce global warming is to change and decarbonize the current energy portfolio, highly based on fossil fuels, to a more sustainable one. 1 The abundant solar energy reaching the Earth's surface (1.3 Â 10 5 TW year À1 ) provides a clean alternative and can be used to produce clean hydrogen from water via photoelectrochemical (PEC) water splitting. Among the different light harvesting materials used as photoanodes in PEC cells, TiO 2 is the most studied material owing to its good properties such as chemical and thermal stability, low cost, electronic properties and long durability. 2,3 Moreover, TiO 2 also nds applications in the decomposition of organic pollutants, photovoltaics, self-cleaning coatings, electrochromic display devices, Li-ion batteries and biomedical devices. 2,4 Nevertheless, it suffers from a few disadvantages, such as the large band-gap and fast recombination of electrons and holes, which can limit its practical application, especially in PEC devices. 5 An approach to overcome some of these limitations is by designing nanostructured TiO 2 crystals with most active facets exposed, since they can offer more available active surface area for the charge transfer process at the photocatalyst-electrolyte interface. 6 Under equilibrium conditions, anatase TiO 2 crystals typically grow with a majority of {1 0 1} facets exposed that have one of the lowest surface energy (0.44 J m À2 ) and poor PEC or photocatalytic activity. 7 In this regard, there is a great scientic interest in growing anatase TiO 2 crystals with high energy facets exposed, such as {0 1 0} and {0 0 1}, which are known to be the most active ones for photocatalytic or PEC applications, especially the {0 1 0} facet. [8][9][10][11] Nowadays, the hydrothermal method is the most employed method for the fabrication of nanostructured TiO 2 photoanodes and a wide range of different morphologies have been achieved so far, such as nanotubes, 12 nanorods, 13 nanowires, 14 nanobelts 15 and even ower-like nanostructures. 16 Chemical vapor deposition (CVD) is an alternative method for the preparation of nanostructured TiO 2 lms. This method allows the fabrication of robust lms with a relatively low processing cost, facilitating the scaling up. 17 Different variants of CVD have been used for TiO 2 growth, such as aerosol-assisted CVD or metal-organic CVD (MOCVD). For instance, Gardecka et al. successfully synthesized nanostructured and dendritic TiO 2 photoanodes using MOCVD and titanium tetraisopropoxide as the precursor. 18 Other morphologies such as cauliower-like structures, needle-like structures and compact domes with pyramidal features (doped with W) have also been successfully grown by AACVD using titanium isopropoxide and titanium ethoxide as TiO 2 precursors. [19][20][21][22][23] In this publication, we present the rst formation of nanostructured anatase TiO 2 having the appearance of crystals of gypsum, sand and water, typically known as "desert roses", with a high exposure of one of the most photocatalytically-active TiO 2 {0 1 0} facet. A similar morphology had only been produced before with rutile TiO 2 phase using a hydrothermal method and MoO 3 to stabilize {0 0 1} surfaces, 24 but never with anatase TiO 2 that is the most photocatalytically-active phase. The "desert rose"-like anatase TiO 2 is grown by AACVD on different substrates using a sophisticated but inexpensive precursora titanium oxo/alkoxy cluster, also called cage, with formula Ti 7 O 4 (OEt) 20 . Upon deposition, the lms are covered in carbon residue, but posterior calcination in air reveals the TiO 2 {0 1 0} facet is predominant on the surface of the rose petals. When deposited on a conductive transparent support and tested in PEC cells for water oxidation, the resulting desert-rose TiO 2 photoanodes exhibit high photocurrents and stability and 100% incident-to-photon efficiency (IPCE) performance at 350 nm. Therefore, the results herein presented reveal new strategies for the design and fabrication of nanostructured TiO 2 photoanodes using the AACVD technology and metal oxo/alkoxy clusters.

Materials
Titanium(IV) ethoxide Ti(OEt) 4 , anhydrous toluene ($99.9%) and ethanol (<0.0003% water) were provided by Sigma Aldrich. Aluminoborosilicate glass (ABS) coated with a uorine-doped tin oxide (FTO) transparent conductive layer (8 U sq À1 ) was provided by Solaronix SA, Switzerland. These FTO-ABS substrates withstand 800 C heating in air, with no deterioration of the FTO conductivity. 25 They were cleaned by ultrasonication in a 2% aqueous Hellmanex III solution, deionized water, acetone and isopropyl alcohol (each step for 3 min), followed by rinsing in deionized water and compressed-air drying and an oxygen plasma treatment for 20 min to enhance surface energy. Alumina substrates (100 mm Â 100 mm Â 1 mm) for high-temperature studies were provided by Almath and quartz substrates (25 mm Â 12 mm) for time-resolved microwave conductivity measurements were provided by H. Baumbach & Co. Ltd.

Synthesis of Ti 7 O 4 (OEt) 20
Ti 7 O 4 (OEt) 20 titanium oxo/ethoxy cluster was synthesized by a controlled hydrolysis in toluene as Eslava et al. previously described. 26 Briey, 0.34 mL of deionized water and 5.0 mL of anhydrous ethanol were added dropwise to a solution containing 7.0 mL of Ti(OEt) 4 in anhydrous toluene (15 mL) under argon atmosphere. Aer overnight stirring, evaporation of the solvent resulted in the formation of a white/yellowish crystalline solid precipitate of Ti 7 O 4 (OEt) 20 .

Preparation of TiO 2 -Rose and TiO 2 photoanodes
Photoanodes were prepared using AACVD. The aerosol droplets were generated using a TSI Model 3076 Constant Output Atomiser, a 0.05 M solution of Ti 7 O 4 (OEt) 20 in toluene, and nitrogen as a carrier gas at a constant ow rate of 1.5 L min À1 . Depositions were carried out onto FTO-ABS, quartz or alumina substrates placed horizontally inside a tube furnace. Deposition times of 0.5, 1, 1.5 and 2 h at 500 C and deposition temperatures of 400, 500, 600 and 700 C for 1 h were performed to assess the growth mechanism and optimization. The optimal deposition conditions for PEC performance were found to be 500 C and 1 h. At the end of the deposition, the substrate was le to cool down under nitrogen ow. The obtained lms were further annealed in air at a heating rate of 10 C min À1 up to 800 C, kept at this temperature for 2 h, and then le to cool down in air. The obtained photoanodes at 500 C with 1 h deposition conditions were denoted as TiO 2 -Rose-AD (asdeposited) and TiO 2 -Rose-800 (annealed). For comparison, TiO 2 photoanodes were prepared following the same methodology but using 0.05 M Ti(OEt) 4 (same precursor molar concentration) and 0.35 M Ti(OEt) 4 (same Ti molar concentration) toluene solutions instead. The resultant photoanodes were accordingly denoted as TiO 2 -0.05M-AD, TiO 2 -0.35M-AD, TiO 2 -0.05M-800 and TiO 2 -0.35M-800.

Characterization
Unit cell calculations were performed at 150 K in a RIGAKU SuperNova manufactured by Agilent Technologies. Fieldemission scanning electron microscopy micrographs (FE-SEM) were acquired using a JEOL FESEM6301F instrument. X-ray photoelectron spectroscopy (XPS) was performed using a Thermo Fisher Scientic K-alpha + spectrometer using a microfocused monochromatic Al X-ray source (72 W). C 1s peak was used for internal charge correction. X-ray diffraction (XRD) patterns were collected from 10 to 80 (2q) Bragg-Brentano with a Bruker AXS D8 Advance using Cu Ka (0.154 nm) radiation, 0.023 (2q) steps and a total integration time of 960 s. The rutile TiO 2 fraction in the lms was calculated using the following equation: 27 where I anatase is the measured intensity of the anatase (1 0 1) diffraction plane and I rutile is the measured intensity of the rutile (1 1 0) diffraction plane. The amount of anatase in the lm was the remaining fraction (X anatase ¼ 1 À X rutile ), since no other phases were observed. The coherent diffraction domain size was calculated using the Scherrer equation at the (1 0 1) anatase TiO 2 diffraction. 28 Preferred crystal orientation in the lm was evaluated by calculating texture coefficients (TC (h k l) ) using the Harris method and a powder diffraction standard for anatase (ICDD-JCPDS 75-1537). 29 Raman spectroscopy was carried out on a Renishaw inVia system using a 532 nm diode-pumped solid-state laser (DPSS) manufactured by Cobolt. The laser beam was focused onto the sample using a 50Â long distance objective. Thermogravimetric analysis (TGA) was performed using a Setaram Setsys Evolution 16 TGA-DTA-DSC equipment for TiO 2 powders (a few mg of TiO 2 was scratched from the FTO). TGA of Ti 7 O 4 (OEt) 20 and Ti(OEt) 4 precursors were performed in a glove box under argon atmosphere using a Perki-nElmer TGA 4000 apparatus. High-resolution transmission electron microscopy (HRTEM) micrographs of lms were obtained using a JEOL JEM-2100Plus microscope. For the sample preparation, a few milligrams of lm was scratched and dispersed in ethanol followed by TEM grid loading. Timeresolved microwave conductivity measurements (TRMC) were carried out using a set up and procedure previously described in literature. [30][31][32] During measurements, a change in the microwave power reected by the cavity upon excitation with a 3 ns pulse laser was monitored. For these experiments, measurements were performed using a wavelength tunable optical parametric oscillator (OPO) coupled to a diode-pumped Qswitched Nd:YAG 3 ns pulse laser at wavelengths of 350, 650 and 1200 nm. A dielectric constant of 41 was used for the calculation of the TRMC signal. 33 Ultraviolet-visible (UV-Vis) absorption spectra were collected in an Agilent Cary 100 diffuse reectance UV-Vis spectrophotometer. UV-Vis trans-ectance measurements were collected in a Lambda 950 spectrometer (Perkin Elmer) with an integrating sphere (150 mm InGaAs) and mounting the sample in the center.

PEC measurements
PEC performance of photoanodes was evaluated using a Com-pactStat. potentiostat (Ivium Technologies). Photocurrents were measured under simulated sunlight (AM 1.5G, 100 mW cm À2 ) from a ltered 300 W xenon lamp source (Lot Quantum Design) or under UV illumination (365 nm, 3.6 mW cm À2 ) from a ModuLight IM3412 LED light (Ivium Technologies). PEC cells were prepared with a three-electrode conguration with Pt as the counter electrode, a silver chloride (Ag/AgCl/3.5 M KCl) reference electrode and as-prepared photoanodes as the working electrode. 1 M aqueous KOH (pH ¼ 13.7) was used as the electrolyte solution. Illumination was directed towards the back of the FTO-ABS working electrode and a mask was placed on top of the photoelectrode to dene the illuminated area. Photocurrent-time curves were measured at an applied bias of 1.23 V vs. the reversible hydrogen electrode (V RHE ). Photocurrent-potential curves were recorded at a scan rate of 20 mV s À1 . The measured Ag/AgCl potentials (E Ag/AgCl ) were converted to RHE potentials ðE RHE Þ and vice versa using the Nernst equation. PEC impedance spectroscopy (PEIS) was carried out under simulated sunlight (AM 1.5G, 100 mW cm À2 ) at the light open circuit potential (OCP) of the cell, at a frequency range of 10 5 -0.1 Hz with an amplitude of 10 mV. EIS measurements at different potentials were also performed under dark conditions to obtain Mott-Schottky plots. These measurements were carried out at a xed frequency of 500 and 1000 Hz, based on the following equation: 34 where C is the semiconductor depletion layer capacitance, N d the electron carrier density, e the elemental charge value, 3 0 the permittivity of the vacuum, 3 the relative permittivity of the semiconductor, U s the applied potential, U Fb the at band potential, and [K B T/e] a temperature-dependent correction term. The electron carrier density (N D ) was obtained from Mott-Schottky plots using the following equation: Where 3 ¼ 41 for anatase TiO 2 and [d(1/C 2 )/d(U s )] À1 is the inverse of the slope obtained from Mott-Schottky plot. Incident photon-to-current efficiency (IPCE) measurements were calculated using the same Xe light source and a triple grating Czerny-Turner monochromator. 35 The intensity of monochromatic light was measured at the working electrode position with a SEL033/U photodetector (International Light Technologies). Oxygen (O 2 ) measurements were conducted using a Pyroscience FireStingO 2 bre-optic oxygen meter combined with a TROXROB10 oxygen probe, together with a TDIP temperature sensor to give automatic compensation for minor uctuation in the PEC cell temperature.

Structural characterization
Titanium oxo/ethoxy cluster Ti 7 O 4 (OEt) 20 ( Fig. 1) was rstly reported in 1967 by K. Watenpaugh and C. N. Caughlan as one of the rst hydrolysis products of Ti(OEt) 4 in dry ethanol bubbled with partially-dried air. 38 In this work, we successfully prepared it in gram scale following a controlled hydrolysis of Ti(OEt) 4  The deposition time during AACVD process was studied. Fig. 3 shows SEM cross-sectional micrographs of desert rose TiO 2 photoanodes obtained at four different deposition times and all annealed at 800 C. The thickness of the lms increased from 0.77 (s ¼ 0.05) mm for 0.5 h to 2.6 (s ¼ 0.20), 2.8 (s ¼ 0.07), and 2.5 (s ¼ 0.26) mm for 1, 1.5 and 2 h, respectively (s stands for std deviation). Desert-rose owers grow perpendicular to the FTO substrate with plate-like petals emerging from the stem of the ower and achieving a good coverage of the support. Interestingly, aer 1 h of deposition, the thickness of the lms remains practically constant between 2.6 and 2.8 mm, although some random owers grow as a second layer (see some roses in the background of the micrographs in Fig. 3c and d). This growth is conrmed by top-view SEM micrographs of the same photoanodes ( Fig. S1 †). A homogeneous rst layer of similar-size roses is achieved at 0.5 and 1 h deposition time, but excessive time leads to some secondary larger owers above the rst layer. A deposition of 1 h was found to be optimal. The deposition temperature was also studied at 400, 600 and 700 C for 1 h. Films did not grow at 400 C but the higher deposition temperatures were successful (Fig. S2 † SEM micrographs). Finer nanostructures were observed at higher temperatures, which are typical when precursor decomposition and/or chemical reactions mostly occur in the vapor phase, followed by surface adsorption and heterogeneous reactions. 40 No plate-like "petal" morphologies were obtained at different temperatures, so 500 C was conrmed to be optimal, together with 1 h deposition time.
Following work was carried out using lms deposited at these   conditions. This optimization based on morphology was further conrmed by PEC measurements (results not shown).
HR-TEM micrographs of both TiO 2 -Rose-AD and TiO 2 -Rose-800 lm fragments are shown in Fig. 4. First, it is noteworthy to highlight the presence of a thin amorphous carbon layer at the crystallite interface of TiO 2 particles for TiO 2 -Rose-AD (Fig. 4a-c). Aer annealing in air, TiO 2 -Rose-800 shows no distinguished amorphous carbon layer (Fig. 4d-f).  (Fig. 4h). This assignment of facets is shown on a SEM micrograph in Fig. 4i.
The composition and chemical state of lm surfaces were evaluated using XPS analysis (Table S1 †). TiO 2 -Rose-AD possess a large amount of carbon on the surface, 55.7 at%, that agrees well with the black color appearance, HR-TEM micrographs and attributed deposited carbon. Conversely, TiO 2 -Rose-800 just shows 18.1 at% C, assigned to volatile organic compounds deposited during storage of samples. TGA in air on TiO 2 -Rose-AD sample (Fig. S3 †) showed one single predominant step at $400 C, indicating the temperature at which carbon deposits burn off in air.
Ti 2p high resolution XPS spectra are shown in Fig. 5a and corresponding binding energies listed in Table S2. † In both TiO 2 -Rose-AD and TiO 2 -Rose-800, the two characteristic peaks of Ti +4 in anatase TiO 2 attributed to Ti 2p 1/2 and Ti 2p 3/2 are observed. 43 Interestingly, a shi towards higher binding energies is observed for TiO 2 -Rose-AD, indicating the possibility of  Ti-O-C bonds in the lm. 44 O 1s high resolution XPS spectra are shown in Fig. 5b and corresponding binding energies listed in Table S3. † Three peaks at different binding energies are observed. The main peak at lower binding energies corresponds to crystal lattice O-Ti 4+ in the TiO 2 lattice structure, whereas the smaller peaks at slightly higher energies are attributed to hydroxyl groups or adsorbed water on the surface. 45 As observed for Ti 2p, a shi towards higher binding energies is observed for TiO 2 -Rose-AD, suggesting that crystal lattice oxygen is also attached to a non-anatase element, supporting the hypothesis of Ti-O-C bonds in the as-deposited samples. 44 Finally, C 1s XPS spectra are shown in Fig. 5c. The main peak at 284.8 eV corresponds to C-C bonds whereas the smaller peaks at higher binding energies are assigned to different carbon environments, such as C-OH and C-O-C. 46,47 Interestingly, no additional peaks at $281.5 eV corresponding to Ti-C are observed for TiO 2 -Rose-AD. Therefore, only C-O-Ti bonds are conrmed on TiO 2 -Rose-AD lms. 44,48 The smaller peaks at $293 and $295 eV mainly observed in TiO 2 -Rose-800 correspond to K 2p 3/2 and K 2p 1/2 , respectively, impurities from the KOH electrolyte used during PEC measurements. 49 11,50,51 The crystal structure of anatase TiO 2 highlighting the crystal planes is shown in Fig. S4. † The preferred orientation of TiO 2 -Rose agrees well with SEM (Fig. 2) and HRTEM (Fig. 4) micrographs where plate-like sheets and rectangular shape particles are observed, respectively. The diffraction planes together with the morphological analysis carried out in HRTEM micrographs conrm that the TiO 2 -Rose lms are mainly exposed of {0 1 0} anatase TiO 2 facets with some regions of {1 0 1} and {0 0 1} facets. 11,41,52,53 None of the diffraction planes shown in Fig. 6a are indexed to rutile TiO 2 phase despite annealing at 800 C, above the typical anatase-to-rutile phase-transformation temperature, which is 600 C for powders and expected to be slightly higher for substrate-constrained lms (ca. 750 C). 54,55 To further investigate the maximum temperature at which the metastable but more photocatalytically active anatase phase is preserved, the Ti 7 O 4 (OEt) 20 precursor was deposited by AACVD on top of alumina substrate at 500 C and further annealed in air for 2 h at 900-1200 C. The use of a different support did not affect the nal morphology of TiO 2 (Fig. S5 †). The XRD patterns are shown in Fig. S6 † and the percentage of each phase vs. temperature in Fig. 6b. TiO 2 -Rose shows presence of pure anatase TiO 2 up to 900 C and a gradual transformation to rutile phase for temperatures of 1000 C and above (Fig. 6b). A high anatase percentage (22%) is obtained in the lm at 1000 C air annealing. These results reveal that these lms achieved using AACVD would offer advantages when used as functional coatings on smart tiles with antibacterial and self-cleaning properties. The ceramics of tiles require temperatures above 900 C for their preparation, which limits their coating to rutile phase which is less photocatalytically active than anatase. 56 High-temperature-stable anatase TiO 2 is typically achieved by doping TiO 2 with metal and non-metal ions co-doping (combining both metal and non-metal ions) and by enriching with oxygen, which strengthens Ti-O-Ti bonds and thus delays the transformation to rutile phase. 56 Recently, high-temperature anatase TiO 2 photoanodes have also been synthesized by anodization of titanium foils followed by a solvothermal treatment,  keeping stable anatase phase up to 900 C. 57 The authors attributed the anatase high temperature stability to a phonon connement effect, typically observed in anatase TiO 2 with small crystallite sizes ($30 nm). 58,59 Since neither doping treatment nor oxygen enrichment modications were undertaken to our TiO 2 samples, coherent crystal domain size calculations and Raman analysis were carried out to further investigate the possibility of this phonon connement effect on our lms. 57,59 Table 2 shows the calculated anatase coherent crystal domain size of TiO 2 -Rose photoanodes annealed in air at different temperatures (patterns in Fig. S7 †). The largest domain size is found in the as-deposited sample, 46.6 nm, most likely due to the presence of interstitial carbon in the anatase TiO 2 lattice structure. Substitutional doping of C +4 with Ti +4 can be discarded owing to the large difference between their ionic radius, being 16 and 61 pm for C +4 and Ti +4 , respectively. 60 At annealing temperatures ranging from 600 to 800 C, the domain size is signicantly smaller, from 28 to 36 nm, and in the range where phonon connement can occur. 58, 59 We attribute this reduction in size to the removal of interstitial carbon present in as-deposited samples. Moreover, the presence of amorphous carbon in the anatase TiO 2 grain boundaries must also have limited the TiO 2 domain sizes. 60,61 Amorphous carbon in the interface of TiO 2 crystals was conrmed in HRTEM micrographs of TiO 2 -Rose-AD (inset of Fig. 4e and f) and by XPS analysis, where Ti-O-C bonds were observed. Above 900 C, grain boundary restrictions disappear and anatase TiO 2 domain sizes grow up to $50 nm owing to sintering of crystals. 62,63 This crystal domain growth is accompanied by a transformation from anatase to rutile (Fig. 6b).
Raman spectroscopy was carried out to further conrm the possibility of phonon connement. As previously reported in literature a shi of the E g Raman mode at 144 cm À1 of anatase TiO 2 towards lower wavenumber supports the phonon connement model of high-temperature anatase TiO 2 . 57,59 Fig. S8 † shows the Raman spectra of TiO 2 -Rose annealed at different temperatures (600 to 900 C) conrming the shi towards lower wavenumber values when annealing temperature is increased. This Raman shi along with the calculated anatase coherent crystal domain size around 30 nm supports that hightemperature anatase TiO 2 may be achieved through phonon connement effects, in addition to substrate constraint effects. Fig. S9 † shows the Raman spectra of TiO 2 -Rose lms. In all cases, only Raman bands ascribed to tetragonal anatase TiO 2 are observed, in agreement with XRD patterns. Particularly, the sharp bands at ca. 144, 198, 400, 520 and 640 cm À1 correspond to E g , E g , B 1g , B 1g and E g Raman vibration modes of anatase TiO 2 , respectively. 64 Two additional bands at ca. 1340 and 1590 cm À1 appear for TiO 2 -Rose-AD only, assigned to D and G bands of graphitic carbon structures (Fig. S9b †). 65 These results agree well with XPS and HRTEM, where Ti-O-C bonds and amorphous carbon layers were observed for TiO 2 -Rose-AD.
UV-Vis spectroscopy measurements for TiO 2 -Rose-800 and TiO 2 -Rose-AD are shown in Fig. S10. † As expected, a clear absorption edge at $400 nm is observed for TiO 2 -Rose-800, whereas lower transectance values at higher wavelengths with no clear absorption edge is observed for TiO 2 -Rose-AD owing to carbon coverage.

Photoelectrochemical characterization
PEC performance of TiO 2 -Rose lms on FTO-ABS substrates was evaluated. No PEC activity was observed for TiO 2 -Rose-AD when irradiated (Fig. S11 †), which we ascribe to the carbon residues coverage and consequent light shielding. However, TiO 2 -Rose-800 shows high PEC response a photocurrent plateau of $0.67 mA cm À2 with simulated sunlight (Fig. 7a) and 3.0 mA cm À2 with UV light (365 nm, 3.6 mW cm À2 , Fig. 7b). Photostability measurements for 2 days including some recovery periods in the dark are shown in Fig. 7c. Aer 24 h of continuous light irradiation, 70% of the total photocurrent is still maintained and a 6 h period in the dark recovers 15% of original photocurrent. The photocurrent decrease is assigned to photocorrosion with photogenerated electrons and holes trapped in the structure. 66 Actually, during the irradiation time the TiO 2 changed color from white to brownish, attributed to the reduction of Ti 4+ to Ti 3+ by trapped photogenerated electrons (Fig. 7c inset). 66 During the recovery period in the dark, the TiO 2 becomes white again, due to the back oxidation to Ti 4+ with atmospheric oxygen, recovering some PEC activity.
IPCE values for TiO 2 -Rose-800 start to increase from 400 nm and reach a remarkable 100% at 350 nm (Fig. 7d). Below 350 nm wavelengths IPCE values decrease due to FTO-ABS substrate light absorption, as conrmed by transmittance measurements on FTO-ABS substrates (Fig. S12 †). The use of frontillumination avoids such decrease at wavelengths below 350 nm, but maximum IPCE values are then 67% due to a longer electron path where more electron-hole recombination can occur (Fig. S13 †). 67,68 Integrating the product of the IPCE curve (Fig. 7d) and the photon intensity in the AM 1.5G solar spectrum results in a photocurrent density value of 0.9 mA cm À2 at 1.23 V RHE , which is slightly higher than the 0.67 mA cm À2 obtained in J-V and Jtime curves at 1.23 V RHE (Fig. 7a and d). This variation is attributed to a spectral mismatch between the simulated sunlight (ltered Xe source) used in the J-V and J-time measurements and the real AM 1.5G solar spectrum used in the IPCE integration. 69 These high IPCE and integrated photocurrent values further conrm the excellent performance of these rose-like shaped photoanodes prepared using Ti 7 O 4 (OEt) 20 oxo clusters. To further understand why the as-deposited dark TiO 2 samples (TiO 2 -Rose-AD) show no PEC activity, as compared to those post annealed at 800 C in air (TiO 2 -Rose-800), we investigated the charge carrier dynamics (i.e., mobility and lifetime) of these samples deposited onto quartz substrates, by TRMC. This technique probes the generation and decay of mobile charges upon pulsed irradiation at various wavelengths (350, 650 and 1200 nm). Fig. 8a shows the microwave conductance transients of TiO 2 -Rose-800 and TiO 2 -Rose-AD aer a 3 ns laser pulse of 350 nm with a photon ux of 3.97 Â 10 13 photons pulse À1 cm À2 , in which we are probing the charge dynamics for excitation energies above the band gap of anatase TiO 2 (3.2 eV). It has been previously reported that the TRMC signal from TiO 2 is predominately a measure of electron mobility and lifetime, since holes are rapidly trapped. 30,70 A strong initial TRMC signal (4Sm) for TiO 2 -Rose-800 (3.40 Â 10 À2 cm 2 V À1 s À1 ) indicates higher electron mobilities compared to the moderate signal of TiO 2 -Rose-AD (6.64 Â 10 À3 cm 2 V À1 s À1 ) at equivalent photon ux. Interestingly, the TRMC signal decays for the two samples are different. As shown in Fig. S14a, † the TRMC signal for the TiO 2 -Rose-800 sample can be tted with a combination of an exponential decay (<100 ns) with a time constant s of 13 ns and a power law decay (>100 ns) with a decay exponent of $0.5. The exponential decay is assigned to band-to-band recombination pathways, while the power law decay can be attributed to traplimited bimolecular recombination mechanism. [71][72][73][74] In contrast, the TRMC signal for the TiO 2 -Rose-AD sample can be tted with only a power law decay with a decay exponent of $0.5 (Fig. S14b †); only trap-limited bimolecular recombination occurs in this sample. This behavior is consistent with the relatively constant mobility and similar decay kinetics at various light intensities (Fig. S15 †).
TRMC measurements were also performed at longer wavelengths of 650 and 1250 nm. Since these wavelength energies are lower than the bandgap of TiO 2 , these measurements effectively probe the photogenerated charges that can reside within the bandgap. Expectedly, no TRMC signal for TiO 2 -Rose-800 sample was observed at these excitation wavelengths. However, a clear TRMC transient signal for TiO 2 -Rose-AD (see Fig. 8b) was observed for both 650 and 1200 nm excitation wavelengths. The mobility slightly decreases with increasing wavelength (5.78 Â 10 À3 and 3.61 Â 10 À3 cm 2 V À1 s À1 for the 650 and 1200 nm excitation, respectively), but the decay still follow the same power law mechanism (see Fig. S16 †). We attribute this to the carbon impurities embedded in the unannealed samples that introduce localized electron trapping states, delaying the electron and hole recombination, but also minimalizing charge mobility. 75 These localized states sit at energetic positions deep within the band gap of anatase TiO 2 . Therefore, photogenerated charge carriers in the carbon doped TiO x C y (TiO 2 -Rose-AD) samples will not sit below the water oxidation potential, nor have a high enough photovoltage to achieve photoactivity for water splitting. This explains the absence of photocurrent from this sample (Fig. S11 †).   9 shows a schematic representation of the processes occurring in the two different samples. The main difference between the two is in the absence of band-to-band recombination for the TiO 2 -Rose-AD sample. This suggests that carriers, even aer excitation beyond the bandgap, rapidly decay into the trap states, which later do not contribute to any photocurrent. Our observations are in agreement with surface photovoltage (SPV) measurements of carbon-doped titania which elucidated deep-isolated and catalytically-poor trap states. 76,77 EIS measurements under simulated sunlight were carried out to understand charge transfer processes in the photoanodes. Fig. 10a shows Nyquist plots along with the equivalent circuit used, which include a R 1 /CPE 1 pair which describes the semiconductor resistance and capacitance at the depletion layer and a second R 2 /CPE 2 pair for the resistance and capacitance of the semiconductor at the interface between the electrolyte and photoanode (Helmholtz layer). 78 Based on the obtained tted results, TiO 2 -Rose-800 photoanode has resistance values of 35.6 U for R 1 and 821 U for R 2 . The small resistance values suggest a better separation efficiency and faster transfer rate for photogenerated electrons and holes at the electrode/electrolyte interface. 78,79 This good charge-transfer properties agree well with J-V curves, where high photocurrent performances are obtained for TiO 2 -Rose-800.
EIS measurements in the dark were carried out to characterize the intrinsic properties of the photoanodes, such as carrier densities (N D ) and at-band potentials. The EIS data were acquired in the form of Mott-Schottky plots at 500 and 1000 Hz, as shown in Fig. 10b. Plots indicate that the sample possess a positive slope, typical of n-type semiconductors. 34 The at-band potentials, obtained from the X-axis intercept of the Mott-Schottky plots, shows a very small variation of 0.02 V between the two frequencies, which indicates a very low frequency dispersion and true measured value for the at-band potential. 80 A at-band potential of 0.12 V RHE and an electron carrier density of 6.43 Â 10 18 cm À3 were calculated at 500 Hz. Similar at-band potential values have been previously reported in literature for nanostructured TiO 2 photoanodes. 34 O 2 evolution measurements at 1.23 V RHE were carried out for TiO 2 -Rose-800 and results are shown in Fig. 11. Fig. S17 † shows the intensity-time curve obtained during the measurements. The amount of O 2 gas evolved was accumulated inside the cell, and thus O 2 content increased over time. The calculated faradaic efficiency for TiO 2 -Rose-800 photoanodes is $90% at the end of the oxygen measurement, providing further evidence that these photoanodes have high activity for oxygen evolution.

Precursor dependence
TiO 2 photoanodes using Ti(OEt) 4 as a starting precursor at two different concentrations have also been prepared for comparison. Fig. S18 † shows SEM micrographs of both TiO 2 photoanodes. Randomly distributed particles of irregular shape are observed for the 0.35 M deposition (TiO 2 -0.35M-800) while plate-like particles are observed for the 0.05 M deposition (TiO 2 -0.05M-800). As expected, thicker lms were obtained when using a 0.35 M solution of Ti(OEt) 4 . Importantly, the desert-rose morphology is not achieved in any, which indicates that the desert rose morphology is unique to the use of Ti 7 O 4 (OEt) 20  precursor. Their optical properties were compared. Fig. S19 † shows the Kubelka-Munk function F(R) of diffuse reectance UV-Vis spectra, related to the absorption coefficient (a). 35 TiO 2 -Rose-800 shows the highest absorption coefficient at 350 nm, followed by TiO 2 -0.35M-800 and TiO 2 -0.05M-800. Therefore, the spectra indicate that the desert rose like lms absorb the most light, ascribed to their higher lm density and thickness. Their XRD patterns are shown in Fig. S20 † showing anatase phase in all the cases. Importantly, the preferred orientation observed for TiO 2 -Rose-800 towards the (2 0 0) diffraction plane is no longer observed for TiO 2 -0.05M-800 and TiO 2 -0.35M-800 lms, indicating again that the desert rose morphology with {0 1 0} facets exposed parallel to the FTO-ABS substrate is unique to the use of Ti 7 O 4 (OEt) 20 clusters. Finally, PEC and IPCE performances are shown in Fig. S21-S22. † TiO 2 -Rose-800 exhibits the highest PEC performance at all applied voltages, but specially at low voltages, indicating a better potential onset which we assign to its unique morphology. Furthermore, TiO 2 -Rose-800 also has the highest IPCE performance at 1.23 V RHE (Fig. S21b †).
The signicant PEC performance of TiO 2 -Rose-800, which reaches 100% IPCE, is mainly attributed to the desert-rose morphology and exposure of {0 1 0} facets at multiple layers. Comparison to TiO 2 -0.05M-800 and TiO 2 -0.35M-800, which lacks {0 1 0} exposure supports this assignment as well as related literature on the activity of different TiO 2 facets. 8 As revealed by Pan et al., anatase crystals with larger exposure of {0 1 0} facets possess the highest photocatalytic activity owing to the combination of 100% coordinated Ti 5c atoms on the surface and a more favorable CB position. 8 The combination of these two factors allows all photogenerated electrons be efficiently transferred via surface Ti 5c atoms, reducing the probability of electron-hole recombination and thus improving the PEC performance. 8 Moreover, as revealed in the cross-sectional SEM micrographs of TiO 2 -Rose-800, each rose grows perpendicularly to the FTO-ABS substrate, leaving a small gap between each ower for the electrolyte to permeate while exposing multiple layers of TiO 2 {0 1 0} facets per substrate area. This desert-rose characteristic morphology, with multiple plate-like sheets growing from the stem of the ower, also contribute to a superior light scattering and absorption in comparison to more irregular morphologies such as the ones found for TiO 2 -0.05M-800 and TiO 2 -0.35M-800 samples. 4 It is believed that the formation of this specic desert-rose morphology arises from the different chemical structure of Ti 7 O 4 (OEt) 20 in comparison to Ti(OEt) 4 when used as AACVD precursor. For instance, Ti 7 O 4 (OEt) 20 precursor has a condensation degree of 0.57 (O/Ti ¼ 4/7) unlike 0 (nul) in Ti(OEt) 4 . It consists of seven TiO 6 octahedra units that form a titanium oxo core which is surrounded by a large number of ethoxide groups, whereas the Ti(OEt) 4 lacks such titanium oxo core. 39 Such a different chemical structure might result in a completely different decomposition path during the AACVD deposition. To conrm the different decomposition, we carried out TGA analysis of Ti 7 O 4 (OEt) 20 . Fig. S23a † shows that the thermal  decomposition of Ti 7 O 4 (OEt) 20 occurs in three main steps at 224, 277 and 331 C giving rise to a mass of 27.4% for the decomposition residue (TiO 2 ). The biggest weight loss (51.1%) from ca. 100 to 225 C corresponds mainly to decomposition of alkoxy ligands and formation of fragments containing titanium oxide species. 81 In contrast, the TGA pattern of Ti(OEt) 4 shows that decomposition occurs in one single step at $160 C (Fig. S23b †).

Conclusions
We have demonstrated the growth of nanostructured TiO 2 lms having a morphology like the crystals of gypsum, sand and water found in nature, known as desert roses. This morphology was successfully grown by AACVD of Ti 7 O 4 (OEt) 20 with N 2 as a carrier. The desert-rose TiO 2 consists of plate-like particles with preferential exposure of {0 1 0} anatase TiO 2 facets, further conrmed by analyzing lattice fringes on the surface. Roses grow perpendicular to substrates such as FTO or alumina, offering an excellent substrate coverage with multiple layers of plate-like TiO 2 per substrate area. In addition, desert-rose TiO 2 show a high preservation of the metastable anatase phase despite annealing in air at very high temperatures, even 1000 C. Rutile phase only appears above 900 C, unlike typical 600 C threshold. Such feature could be exploited in smart tiles with antibacterial and self-cleaning properties, whose ceramics require high temperature preparation. When desert-rose TiO 2 lms are deposited on FTO-ABS substrates and annealed in air, they offer excellent photoelectrochemical performance as photoanodes for oxygen evolution in aqueous electrolytes. Photocurrent plateaus of $0.67 mA cm À2 under simulated sunlight (100 mW cm À2 ) or 3.0 mA cm À2 under 365 nm UV light (3.6 mW cm À2 ) are achieved as well as an IPCE of $100% at 350 nm. Such remarkable performance is attributed to an excellent morphology, preferential exposure of {0 1 0} TiO 2 facets and, upon calcination, the minimization of surface states that would otherwise trap photoinduced charge carriers. On balance, we have extended the use of metal oxo/ alkoxy clusters to AACVD for functional coatings, discovering a novel and simple strategy to obtain a faceted semiconductor without the use of dopants. These results will trigger research in using metal oxo clusters in the preparation of efficient, nanostructured and stable photoelectrodes, as well as, other possible components of energy devices.

Conflicts of interest
There are no conicts to declare.