Mesocrystal TiO 2 ﬁ lms: in situ topotactic transformation and application in dye-sensitised solar cells †

Thin ﬁ lm ceramics and semiconductors play an important role in energy- and environment-related areas such as photovoltaics, energy storage and water puri ﬁ cation. The morphology and structure of materials signi ﬁ cantly a ﬀ ect their properties and performance in applications. Mesocrystal materials with a hierarchical structure and designable overall shape possess not only the properties from nanosized building-blocks but also collective functions from the crystallographically ordered assembly, meeting the criteria of high performance candidates in various applications. In this study, a facile and versatile method was developed to prepare mesocrystal ﬁ lms by simply making a printable paste of the topotactic precursor, followed by in situ topotactic transformation of printed ﬁ lms. Using TiO 2 as the model material, mesocrystal TiO 2 ﬁ lms made from NH 4 TiOF 3 paste possess high surface area, crystallographic orientation of anatase nanoparticles and overall large particle size, performing well in dye-sensitised solar cells (DSSCs) as either single-layer photoanodes or additional scattering layers.


Introduction
Thanks to its elemental abundance, nontoxicity, low cost, high chemical stability and unique physicochemical properties, titanium dioxide (TiO 2 ) has been intensively investigated in energy-and environment-related areas such as photocatalysis, hydrogen generation, Li-ion batteries, and solar cells. [1][2][3][4][5] To satisfy the criteria for different applications, there has been a lot of effort on tuning the size, crystallinity, shape and architecture of TiO 2 because these parameters can signicantly affect its properties and performance. A very characteristic example is as photoanode for dye-sensitized solar cells (DSSCs). [6][7][8][9] A good photoanode requires specically high dye loading capacity, efficient electron collection and transport and also a pronounced light-scattering ability. In most of the DSSCs studies, mesoporous anatase TiO 2 nanoparticle lms made from commercially available TiO 2 paste (Ti-Nanoxide T/SP, Solaronix) are widely used. The highly dispersed nanoparticles (15-20 nm) possess a high surface area leading to efficient dye loading and high transparency ensuring favourable light penetration. 10,11 However, the negligible light scattering in nanoparticle lms leads to a short pathway for incident light, limiting the photon absorption by the adsorbed dye molecules. To solve this problem, an additional scattering layer composed of large TiO 2 particles is generally applied on top of the transparent active layer in many reports to improve the light harvesting in DSSCs. 12 Another method is to develop hierarchicalstructured TiO 2 as photoanode such as mesoporous TiO 2 microspheres, 13 TiO 2 nanoowers 14 and hierarchical TiO 2 nanotube arrays. 15 It is also not surprising that the combination of the above two strategies, using TiO 2 with hierarchical structure as additional scattering layer has been shown to signicantly enhance the light harvesting in DSSCs. 16,17 As a recently-developed hierarchical structure, mesocrystals, consisting of nanocrystals with common crystallographic orientation, have drawn a lot of attention since they were rst proposed by Cölfen in 2005 as the metastable intermediates in non-classical crystallization processes. 18 Over the past 16 years, mesocrystal structures were found or realized in a broad range of materials including biominerals and mimetics (e.g. sea urchin spines, 19 vaterite CaCO 3 (ref. 20)), metal oxides (e.g. ZnO, 21 TiO 2 , 22 Fe 2 O 3 , 23 Co 3 O 4 (ref. 24)), metal sulde (e.g. ZnS 25 ), etc., and have been demonstrated to bring the material not only properties associated with individual nanoparticles but also collective properties and functions. TiO 2 mesocrystals (mcTiO 2 ) have been investigated in many applications and showed excellent performance, considering their advanced properties from the unique structure. [26][27][28][29] However, in most existing work, not only for mcTiO 2 but also for other mesocrystals, lengthy hydrothermal or solvothermal treatments were always adopted to enable the 'bottom-up' nanocrystal oriented assembly. 26,27,30,31 To make mesocrystals, another 'top-down' strategy via topotactic conversion has also been explored in some cases. [32][33][34] This solid-to-solid transformation is based on crystallographic similarity between target product and the precursor, which is NH 4 TiOF 3 in the case of anatase TiO 2 . 22 Compared with a 'bottom-up' strategy, topotactic conversion uses a solid precursor as template, making it easier to design, control and predict the morphology and structure of the product. Considering the demands of TiO 2 lms in many applications, such as photoanodes for solar cells, electrodes for energy storage and immobilised photocatalysts for easier reuse, a general and simple method to make functional mcTiO 2 lms will broaden the application of mcTiO 2 in energy-and environment-related areas and bring the specic properties from the unique structure of mesocrystals to those applications.
In this work, we report a facile method to prepare functional mcTiO 2 lms by making a printable paste of the NH 4 TiOF 3 followed by in situ thermal topotactic transformation of printed NH 4 TiOF 3 lms, extending an approach that has received little prior attention. 35 The NH 4 TiOF 3 precursor is prepared through an HF free route at room temperature and mixed with organic binders to form a viscous paste. The topotactic conversion occurs simultaneously with the thermal removal of organic components aer lm printing, signicantly simplifying the manufacturing steps of mcTiO 2 lms . The obtained mcTiO 2 lm was characterized by X-ray Diffraction (XRD), Scanning Electron Microscopy (SEM), Transmission Electron Microscopy (TEM), UV-Visible (UV-Vis) spectroscopy, Attenuated Total Reection-Fourier Transmission Infrared (ATR-FTIR) analysis and tested in the example application of DSSCs to investigate the functions brought by the structural features. The mcTiO 2 lms were applied as either single-layer photoanode or additional scattering layer for DSSCs using the commercial dye LEG4 as the photo-sensitizer. Compared with commercial transparent TiO 2 (Ti-Nanoxide T/SP, Solaronix), diffusing TiO 2 (Ti-Nanoxide D/SP, Solaronix) and reective TiO 2 (Ti-Nanoxide R/SP, Solaronix), mcTiO 2 prepared from NH 4 TiOF 3 paste possesses not only properties from individual nanoparticles but also collective functions, leading to satisfactory performance in DSSCs and promising prospects for other potential applications.

Experimental
Synthesis of NH 4 TiOF 3 NH 4 TiOF 3 powder was prepared from a room temperature hydrolysis reaction of (NH 4 ) 2 TiF 6 with addition of ammonia solution, which is modied from a reported method. 28 Typically, 3.959 g (NH 4 ) 2 TiF 6 (Acros Organics) was dissolved in 45 ml of H 2 O, and then the solution was mixed with 5 ml 3 M ammonia solution under magnetic stirring. Aer stirring the solution for 5 min and letting it stand for 6 h at room temperature, the precipitate was collected by centrifugation and washed sequentially with water and ethanol, and dried in ambient conditions.

Preparation of NH 4 TiOF 3 paste
The preparation of NH 4 TiOF 3 paste follows a reported analogous method for TiO 2 paste with a few modications. 36 Before making the paste, 1 g of two kinds of ethyl cellulose (EC) powders, i.e. EC (10 cP, Merck) and EC (30-70 cP, Merck) were dissolved in anhydrous ethanol respectively and stirred overnight at room temperature to get 10 g of EC stock solutions (10 wt%). They were labelled as EC-10 and EC-30 according to different viscosities. To prepare the paste, 0.278 g NH 4 TiOF 3 was mixed with 0.65 g anhydrous terpineol solvent, 0.45 g EC-10 stock solution, 0.35 g EC-30 stock solution and 0.8 ml anhydrous ethanol. The amount of NH 4 TiOF 3 was calculated to satisfy the Ti concentration as $10.8 wt%, the same as commercial titania pastes from Solaronix. Aer stirring overnight, the mixture was sonicated for 2 min and stirred at 13 500 rpm for 1 min using an ultra-terrax mixer. The sonication and stirring procedure was repeated 3 times. Ethanol and water were then removed by rotary-evaporation at 40 C.

Preparation of TiO 2 photoanodes
Fluorine-doped tin oxide (FTO)-coated glass substrates (TEC7, Merck) were cut into pieces of size 3 cm Â 4 cm (for 4 cells), and sonicated in 2 vol% Hellmanex detergent solution for 30 min, DI water for 15 min and rinsed with ethanol. Then they were dried with hot air and treated by UV/ozone for 20 min. Prior to screen printing, FTO substrates were pre-treated by a 40 mM TiCl 4 aqueous solution at 70 C for 30 min, followed by a water rinse and a heat treatment at 500 C for 30 min. For the single layer DSSCs, the NH 4 TiOF 3 paste or commercial paste (nanoparticle only paste Ti-Nanoxide T/SP, mixed titania particle paste Ti-Nanoxide D/SP, and large titania particle only paste Ti-Nanoxide R/SP) was screen-printed on top of TiCl 4 treated FTO substrates through a 90 polyester screen with 4 circles (D ¼ 6 mm). The lms were sintered at 500 C for 2 hours using different temperature ramps. The obtained TiO 2 lms were named as F (mesocrystal TiO 2 from NH 4 TiOF 3 paste), T (transparent TiO 2 from Ti-Nanoxide T/SP), D (diffusing TiO 2 from Ti-Nanoxide D/SP), and R (reective TiO 2 from Ti-Nanoxide R/SP). A numeric suffix will be used to note the heating rate. For example, F-2 means 1 layer of TiO 2 lm made from NH 4 TiOF 3 paste with a heating rate of 2 C min À1 . For the two-layers cells, Ti-Nanoxide T/SP was used to print the transparent active TiO 2 layer and sintered at 500 C for 15 min. Additional scattering layers were prepared by depositing screen-printed NH 4 TiOF 3 paste or commercial paste (Ti-Nanoxide D/SP or Ti-Nanoxide R/ SP) on top of the transparent TiO 2 layer, followed by sintering at 500 C for 2 hours with a heating rate of 2 C min À1 . The obtained lms were named as TF, TD, and TR respectively.

Fabrication of dye-sensitized solar cell
The TiO 2 lms were pre-heated at 120 C for 30 minutes before being immersed in a dye bath containing 0.2 mM LEG4 (Dyenamo, Fig. S1 †) in acetonitrile/tert-butanol (1 : 1, v/v). Aer sensitization for 20 hours, the electrodes were rinsed with acetonitrile and dried in air. Counter electrodes were prepared by doctor-blading platinum precursor paste (Solaronix Platisol-T) on FTO substrates followed by heating in air at 450 C for 15 min. The working electrode and counter electrode were assembled together into a sandwich type cell and sealed with a hot-melt Surlyn lm. An I À /I 3 À electrolyte solution was injected through a pre-drilled hole in the counter electrode by vacuum back lling and the hole was then sealed with Surlyn cover and a microscope coverslip. The electrolyte was composed of 0.1 M LiI, 0.05 M I 2 , 0.6 M 1,2-dimethyl-3-propylimidazolium iodide (DMPII), 0.5 M 4-tert-butylpyridine (tBP) in acetonitrile. Fig. S2 † shows digital photos of the example DSSC devices.
Characterization XRD was carried out on a Bruker D2 PHASER using CuKa radiation. UV-Vis absorption/diffuse reectance spectra (DRS) were obtained using a JASCO V-670 UV/Vis/NIR spectrophotometer. Infrared (IR) spectra from 4000 to 500 cm À1 were recorded on a PerkinElmer Spectrum 65 ATR-IR spectrometer.
To prepare lm samples for XRD, UV-Vis DRS and IR characterizations, NH 4 TiOF 3 paste or commercial TiO 2 pastes were doctor-bladed on glass slide with the thickness controlled by 1 layer of Scotch tape, followed by a drying process at 120 C for 10 min and a sintering process if needed. Surface and crosssectional SEM images of samples were collected using a ZEISS SIGMA Field Emission SEM, operated in SE2 mode with a 10 kV accelerating voltage. To prepare specimens for cross-section analysis, the substrates were scored before screen-printing. The substrates were broken along the score aer preparation of lms and adhered on a stub with the cross section facing up. TEM, Selected-Area Electron Diffraction (SAED) and Electron Dispersive Spectroscopy (EDS) were carried out using a FEI Titan Themis microscope. Photocurrent density-voltage (J-V) curves were recorded on an Autolab potentiostat (Metrohm) with class AAA SLB300A solar simulator (Sciencetech) as the light source. The light intensity was calibrated to AM1.5G using a silicon reference cell. The active area of the solar cell was xed with a black metal mask with a circular aperture of 0.0707 cm 2 (d ¼ 3 mm). The frequency-modulated electrochemical impedance spectroscopy (EIS) was recorded with a similar setup to that of the J-V measurements. The frequency range was set to 1 MHz to 0.05 Hz with an AC voltage amplitude of 10 mV. The measurements were performed under white light emitting diode (LED) illumination with adjustable intensity. The plots were tted using the ZView (Scribner Associates) soware. Incident photon-to-current conversion efficiency (IPCE) was recorded with a photo-spectrometer setup (Bentham PVE300) by illuminating the solar cell with a modulated monochromatic light (xenon and quartz halogen lamps) through 1.85 mm slit. The incident light was calibrated with a reference silicon photodiode and the spectral resolution was set to 5 nm.
Results & discussion NH 4 TiOF 3 powder samples were synthesized based on a reported method, modied to have an increased reaction volume. 28 The XRD pattern in Fig. 1a matched well with the powder diffraction pattern of reported NH 4 TiOF 3 (CCDC 1880514) simulated using Mercury soware, 37,38 indicating that phase-pure NH 4 TiOF 3 was successfully obtained. Fig. 1b shows the SEM image of the NH 4 TiOF 3 particles with a diameter range of 200-350 nm, in line with the size range for optimum visible light Mie scattering. 39 NH 4 TiOF 3 powders were mixed with organic binders to prepare printable pastes. Doctor-bladed lms with thickness controlled by 1 layer of Scotch tape were sintered at 500 C for 2 hours with different heating rate, namely 1 C min À1 , 2 C min À1 , 5 C min À1 and 10 C min À1 . Fig. S3 † shows a digital photo of the doctor-bladed lm on a piece of glass slide (1.8 cm Â 2.6 cm). No cracking or ake-off was observed aer the sintering procedure, indicating a good mechanical adhesion of the printed lm. XRD patterns of the obtained lms are shown in Fig. 2a, conrming that all samples were pure phase anatase (ICDD 86-1156). The crystallite domain size of each sample was calculated from the full width at half maxima (FWHM) of (101) diffraction peak (2-theta ¼ 25.3 ) using the Scherrer equation, and assuming negligible instrumental broadening (Table 1). In DSSCs, the crystallinity of TiO 2 is associated with the trap states distribution of the photoanode, which will have inuence on the charge recombination rate thereby affecting the power conversion efficiency (PCE). 40 Sample F-2 under the heating rate of 2 C min À1 shows the largest average crystal size among all the samples, which may lead to a relatively-lower charge  recombination rate and higher photovoltages when applied in DSSCs. However, larger crystallite size may also result in smaller surface area therefore a lower dye coverage and current density. The performance of TiO 2 in DSSCs requires an optimal balance between two parameters.
To better evaluate the properties and performance of mcTiO 2 made from NH 4 TiOF 3 paste, three commercial available pastes were selected as references, namely nano-particle only paste Ti-Nanoxide T/SP, mixed titania particle paste Ti-Nanoxide D/SP, and large titania particle only paste Ti-Nanoxide R/SP. More details of commercial pastes and the comparison with NH 4 -TiOF 3 paste are listed in Table S1. † TiO 2 lms from commercial available pastes were also made by doctor-blading and sintering at 500 C for 2 hours with a heating rate of 2 C min À1 . XRD patterns of obtained reference TiO 2 (T-2, D-2 and R-2) are given in Fig. S4. † The crystallite domain size for these samples were also calculated from (101) diffraction peak and collected in Table 1, which shows a consistent order with the particle sizes. Compared with the crystallite size of the reference TiO 2 , mesocrystal F-2 sample is in the middle of the order, which may offer the possibility of a balanced crystallinity and surface area.
Also, the approximate 20 nm size of the crystallites has been generally regarded as optimal for the conventional I À /I 3 À electrolyte-based DSSCs. 41 SEM images of mesocrystal TiO 2 annealed under different heating rate are given in Fig. 2b, c and S5. † It is clear to see that aer sintering at 500 C, all samples retained the overall shapes and sizes of the original NH 4 TiOF 3 powders, with 200-350 nm particle range which is benecial for visible light scattering. The high-resolution images indicate that the cubic particles are built of smaller nanoparticles, consistent with the crystallinedomain size determined by the Scherrer equation. The most typical feature of a topotactic conversion is that the product keeps the overall shape and size of precursor but alters the crystal structure to a mesocrystal form, which means the building blocks of products from topotactic transformation should be crystallographically ordered. 22 This can be proven by TEM. Fig. 3 shows the TEM images of a TiO 2 sample sintered at 500 C for 2 hours with a heating rate of 2 C min À1 . The TEM image of a typical particle shows a rhombus shape and the porous structure (Fig. 3a). The corresponding SAED pattern of the circled area shows a typical mesocrystal diffraction pattern (single-crystal like with minor distortion), which can be indexed to the [11À1] zone of anatase. 42 From higher magnication in Fig. 3b, square facets of nanoparticles can be clearly recognised, corresponding to exposed {001} facets of truncated bipyramidal anatase (see inset of Fig. 3b). The size of nanoparticles is consistent with the crystalline domain size calculated from XRD results. The high resolution TEM image (Fig. 3c) shows planes with lattice spacing of 0.36 nm, corresponding to the (101) and (011) planes of the anatase phase, and the 82 angle between (101) and (011) matches the theoretical value calculated from the lattice constants of anatase (ICDD 86-1156). The EDS analysis ( Fig. 3d-f) shows dominant Ti (44.4%) and O (54.5%) elements and homogeneous distribution. The negligible concentration of N (1.09%) and F (0.0147%) (see Fig. S6 and Table S2 †) suggests the efficient removal of N and F aer the sintering progress. Collectively, it can be conrmed that anatase TiO 2 mesocrystals were successfully obtained.
ATR-FTIR analysis was used to provide more information on the topotactic conversion process. Samples scratched from NH 4 TiOF 3 doctor-bladed lm which was only dried at 120 C for 10 min and mcTiO 2 lm aer sintering at 500 C for 2 hours were measured, together with powder form NH 4 TiOF 3 and ethyl cellulose as references (Fig. 4). Before sintering at 500 C, NH 4 TiOF 3 sample (Fig. 4c) showed absorption peaks of both Table 1 Comparison of the crystal sizes, dye uptake capabilities and photovoltaic properties of single-layer TiO 2 photoanodes. All the samples were sintered at 500 C for 2 hours with a heating rate indicated by the numeric suffix a  NH 4 TiOF 3 and organic binders. The bands at 3196 cm À1 , 3082 cm À1 and 1416 cm À1 are attributed to the n 3 , n 1 (or n 2 + n 4 ) and n 4 modes of NH 4 + respectively. 32 The disappearance of all these peaks in mcTiO 2 sample (Fig. 4d) indicates the removal of ammonium aer the topotactic conversion. Due to the hydroscopic nature of NH 4 TiOF 3 , deformation band of adsorbed water at 3275 cm À1 appeared in Fig. 4c, which was, however, not detected in the mcTiO 2 sample. Other typical absorption peaks of NH 4 TiOF 3 were observed between 1000 cm À1 to 500 cm À1 . The band at 870 cm À1 is ascribed to n (Ti]O) (terminal oxygen) or Ti-O-Ti antisymmetric stretching. 43 The band at 752 cm À1 corresponds to a combination of lattice mode of Ti-O and Ti-F bands, which is a typical feature of oxouorotitanates. 44 The bands at 588 cm À1 and 555 cm À1 were due to Ti-F and Ti-O respectively. 32,44 Aer calcination, the product mcTiO 2 displayed only a broad stretching vibrational Ti-O-Ti absorption peak in this range, indicating the decomposition of Ti-F aer topotactic conversion. The vibrations between 3000 cm À1 and 2840 cm À1 are associated with C-H stretching of either ethyl cellulose or terpineol added as organic binders, 45,46 which are not observed in mcTiO 2 sample, indicating the combustion of organic additives aer the topotactic conversion. Another indicative absorption peak at 1050 cm À1 , which may be from -C-O-Cstretching of the pyranose ring in ethyl cellulose, 46 is also not found in the spectrum of mcTiO 2 sample. In short, the NH 4 -TiOF 3 lm made from NH 4 TiOF 3 paste undergoes a topotactic transformation to form mesocrystal TiO 2 and a simultaneous removal of organic binders. A schematic illustration of this in situ topotactic transformation process is given in Fig. 2e. More generally, this strategy could be transferrable to other functional ceramics as long as the thermal topotactic transformation can be applied.
To evaluate the photovoltaic performance of anatase mcTiO 2 , the obtained anatase mcTiO 2 lms were rstly tested as single-layer photoanodes. Considering the low thickness of the single-layer TiO 2 obtained from screen-printing (4-6 mm, Fig. 2b), LEG4 dye with high extinction coefficient was selected as the sensitizer. 47 To investigate the light scattering effect and dye-loading capacity of single-layer mcTiO 2 lms, doctor-bladed lms were tested by UV-Vis DRS before and aer LEG4 dye adsorption. According to the UV-Vis DRS spectrum in Fig. 5a, before absorbing dye molecules, bare anatase mcTiO 2 (sample F-2) has the highest reectance compared with all samples from commercial pastes, indicating the highest light scattering ability. This is because of the optimal particle size of anatase mcTiO 2 inherited from the precursor NH 4 TiOF 3 aer the topotactic conversion. Aer dye loading (Fig. 5b), the reectance in the visible range of all the samples drastically decreased due to the light absorption by dye molecules. The lower decrease from R-2 sample indicates a poor dye uptake capability. Quantied dye coverage on different TiO 2 lms is collected in Table 1. It is shown that all mcTiO 2 lms made from NH 4 TiOF 3 paste possess comparable dye uptake capability to that from Solaronix T and Solaronix D and much higher dye loading than Solaronix R, conrming the high surface area arising from individual nanoparticles in mcTiO 2 .
The single-layer DSSCs were characterized by measuring J-V behaviour under 1 sun illumination. The results are given in Fig. 6 and also summarized in Table 1. Fig. 6a shows the comparison among DSSCs with anatase mcTiO 2 as photoanodes with different temperature ramp during the mcTiO 2 lm annealing process. In general, the performance of DSSCs under different conditions varies in a very small range in terms of all   parameters. Among all the heating rate, the samples annealed with a ramp of 2 C min À1 gave relatively higher PCE, resulting from the optimal balance between the open circuit voltage (V OC ) and the short circuit current density (J SC ), despite that samples annealed with a ramp of 5 C min À1 possessed higher dye loading and correspondingly higher J SC . Films made from commercial pastes were also tested in DSSC devices to compare with the performance of mcTiO 2 as single-layer photoanodes (Fig. 6b). Compared with Solaronix T and D, mcTiO 2 -based DSSCs F-2 showed comparable V OC and ll factor (FF) but a bit lower J SC . As a result, the PCE of F-2 (4.57 AE 0.17%) was lower than T-2 and D-2 by 0.24% and 0.47% respectively. The slight drop in photocurrent is attributed to the big particle size. Even though mcTiO 2 can adsorb comparable amount of dye molecules to nanoparticle based lms as discussed above, the excellent light scattering and reection may lead to the waste of backward scattered light (see Fig. S7 †), therefore reducing overall light harvesting. This is a drawback of mcTiO 2 when applied in single-layer photoanodes, but can be utilized as a benet when it is applied as the additional scattering layer, which will be discussed in the next section. Solaronix R was also used as the reference TiO 2 . The DSSCs based on large particle TiO 2 (R-2) showed the highest V OC among all devices. However, this is with the sacrice of J SC due to the lowest dye loading, which is not favourable for single-layer photoanodes. The mcTiO 2 particles have similar overall particle size to R but higher porosity due to the mesocrystal structure. As a result, mcTiO 2 DSSCs F-2 offer similar V OC , but much higher J SC than devices based on R, resulting in a better performance as singlelayer photoanodes.
To better understand the interfacial charge transfer mechanism of mcTiO 2 based single-layer DSSCs, EIS measurements were conducted under white LED illumination at open circuit conditions. By varying the light intensity, the V OC can be tuned and hence the quasi-Fermi level in the photoanode. During all measurements, a 10 mV perturbation was applied in the 0.05 to 1 MHz frequency range. Fig. 7a and b display examples of Nyquist plots for DSSCs based on different single-layer photoanodes under a bias of 0.7 V. There are two semicircles in each plot. The rst one in the high-frequency region denotes the electron transfer at the Pt/electrolyte interface, and the other semicircle represents the electron recombination at the TiO 2 / dye/electrolyte interface. A transmission line equivalent circuit model 48 shown in the inset of Fig. 7b was used to extract electron transfer parameters of cells from all Nyquist plots under a series of applied bias potentials. Examples of tted results from plots under 0.7 V are shown in Table S3. † The chemical capacitance, C m , provides information on the electronic structure of TiO 2 in DSSCs as 49 with N t the total number of trap states below the conduction band, L the lm thickness, p the porosity of the lm, q the elementary charge, k B the Boltzmann constant and (E redox À E c ) the energy difference between redox potential of electrolyte and conduction band of TiO 2 . Fig. 8a shows the typical exponential increase of C m with the increasing applied potential and according to eqn (1), the exponential trap distribution parameter a can be extracted from the slope. Different a values for cells with different single-layer photoanodes, as shown in Table  S3, † reect their different depth of trap distributions. 50 The highest a of cell F indicates the shallowest distribution of trap states and relatively smaller proportion of deep traps, which would be benecial for charge collection via the trap-detrap process. 51,52 This could have beneted from the orientation of cubic nanoparticles in the mesocrystal structure, since many studies have demonstrated the strong dependence of the trap states distribution on the structural order of the charge carrier transport lms. 53,54 In addition to the morphology, the crystallinity of TiO 2 also impacts the density of states (DOS) in terms of both bulk and surface trap concentrations, 55 which is denoted as N t in eqn (2). According to Fig. 8a, the C m values of the cells follow the order of T > D $ F > R, consistent with that expected from their order of crystallinity estimated from crystallite domain size, implying that the effect from crystallinity makes the main contribution to the DOS of photoanodes. The recombination resistance (R rec ) at the TiO 2 /dye/ electrolyte interface was plotted versus applied potential in Fig. 8b. The order of the R rec for all TiO 2 follows R > D $ F > T, consistent with the order of their V OC under 1 sun illumination. The potential dependence of R rec follows the Buttler-Volmer relationship as 49 where l is the reorganization energy of acceptor species, and b is the transfer coefficient, k r the rate constant accounting for recombination kinetics, N s the total number of surface states contributing to the recombination. By use of eqn (3), the b for Fig. 7 (a) Nyquist plots for single-layer DSSCs measured at 0.7 V under white light LED illumination. (b) Magnified version of (a). Inset of (b) equivalent circuit for the electrochemical impedance fitting. R s : series resistance; R 1 : resistance at the Pt/electrolyte interface; CPE1: capacitance at the Pt/electrolyte interface, derived from a constant phase element; DX1: transmission line element for the porous electrode (type 11: Bisquert #2); CPE2: chemical capacitance at the triple contact FTO/TiO 2 /electrolyte interface, derived from a constant phase element.
each TiO 2 was also obtained and shown in Table S3. † The highest b for cell F implies its stronger potential dependence of R rec . It has been reported that the b value tells the degree of the impact from surface states on the charge recombination. 56 If b ¼ 1, the charge recombination solely happens from the conduction band of TiO 2 . When b is less than 1, which is normally observed in DSSCs, lower b indicates more contribution from the surfaces states to the charge recombination. Based on this, a more ideal charge recombination occurs in cell F (higher b), which is likely because of its shallower distribution of trap states as discussed above.
In order to determine the effect of recombination kinetics on the cell performance, it is also useful to plot R rec versus total electron density (n), which can be obtained from C m by the following equation 57 As shown in Fig. S8, † cell F shows slightly lower R rec at a xed n compared with cell D, suggesting slightly faster recombination at the surface of mcTiO 2 in cell F. Also taking D as the reference, R showed similar kinetics and T showed the slowest. By comparing R rec versus potential and R rec versus n plots, one can observe an obvious shi of R and T, which means recombination kinetics is not the determinant of R rec at xed potential. One possible reason is that their different crystallinity and surface states (denoted by N s ) make more contributions to the value of R 0 therefore R rec . The smallest N s for sample R leads to the highest R rec among all TiO 2 , while the highest N s for sample T due to the highest surface area drops off its R rec in potential plots. This is also consistent with the conclusion based on the C m analysis. For cell F, considering the similar surface states between cell F and cell D, the slightly faster recombination kinetics results in slightly lower R rec at high applied potential. However, at lower applied potential, R rec of cell F becomes higher compared with cell D, which can be attributed to its shallower trap distribution. From eqn (2), the position of conduction band of TiO 2 also alters the value of R 0 therefore R rec when different TiO 2 is applied as photoanodes. Even though CB edge shis of TiO 2 with exposed {001} facets have been detected and reported in some published reports, 58 the impact of this on R rec is not signicant here by comparing cell F and cell D. Electron lifetime calculated from s n ¼ R rec Â C m was also plotted versus applied potential (Fig. 8c) and total electron concentration (Fig. S8 †). The lifetime follows the same trends as R rec in both plots, further supporting the discussion based on R rec results.
On the basis of the above characterization and analysis, it can be seen that mcTiO 2 possesses similar dye uptake capability to commercial nanoparticle TiO 2 due to the high surface area from nanosized building-blocks, excellent light scattering behaviour similar to commercial large-size particles because of the overall mesocrystal size, and reasonable electron transport ability and recombination resistance, resulting from the porous mesocrystal structure and the crystallographic order of nanoparticles. Such a combination of properties in one material makes TiO 2 mesocrystals promising photoanodes for DSSCs and also prospective candidates in other applications. However, the high reectance may limit its performance in single-layer photoelectrodes. To make the best use of all the properties from the specic structure, mcTiO 2 was also tested as  a scattering layer on top of transparent nanoparticle TiO 2 lms made from Solaronix T. In this conguration, as shown in the schematic illustration (Fig. S9 †), the portion of light which passes through the transparent layer will be scattered for greater path length following Mie's scattering theory. 39 As a result, optimal light harvesting can be reached when both layers possess excellent dye coverage. Compared with commercial reective TiO 2 in T+R , mcTiO 2 in T + F has similar light scattering (see Fig. S9 †) but higher dye loading in the scattering layer, leading to better utilization of forward scattered light therefore higher J SC as shown in Fig. 9a and Table 2. Compared with commercial diffuse TiO 2 in T + D, mcTiO 2 in T + F has higher light reection. But the presence of the transparent layer avoids waste of the backward scattered light. As a consequence of the optimised light harvesting, T + F showed similar current to T + D. This was consistent with the recorded incident photon-to-current conversion efficiency (IPCE) spectra and integrated J SC from IPCE spectra (Fig. 9b), showing similar IPCE from T + F and T + D and lower maximum with T + R.
The average V OC of DSSCs is 0.70 AE 0.01 V, 0.69 AE 0.01 V, and 0.72 AE 0.01 V for T + F, T + D and T + R, respectively. This is consistent with their inverse order of chemical capacitance and the order of recombination resistance, as shown in Fig. 9c and d. Exponential trap distribution parameter a for three twolayers photoanodes was also calculated based on Fig. 9c and listed in Table S4. † It is not surprising that they have similar a values, therefore similar trap distribution, since commercial nanoparticle TiO 2 (T) was used as main working layer in all three photoanodes. The additional scattering layer modies the electronic structure of the three photoanodes to varying degrees. In terms of C m , three photoelectrodes follow the order T + D > T + F > T + R. As discussed above, the lowest C m of T + R may come from the lowest total number of trap states, which is because of the high crystallinity of R. For T + F and T + D, considering their similar a and close crystallinity of mcTiO 2 and diffractive TiO 2 (D), the lower C m of T + F can be ascribed to the positive shi of CB edge according to eqn (2). This interpretation can be corroborated by R rec analysis. From Fig. S10, † T + F has the lowest R rec at xed total electron concentration, indicating the recombination is faster on the surface of T + F, consistent with the observations in single-layer DSSCs. T + D and T + R have similar recombination kinetics. The smallest surface area of R leads to the lowest N s therefore the highest R rec at xed potential, in line with the highest V OC of T + R. The N s of T + F and T + D are regarded to be similar because of their similar surface area estimated from dye uptake and similar trap distribution depth from the same a value. In this case, there should be some impacts from other aspects to offset the fast recombination kinetics of T + F in order to endow it with comparable R rec at xed applied potential and slightly higher V OC under 1 sun illumination compared with T + D. According to eqn (4), the aforementioned positive CB edge shi of mcTiO 2 in T + F may explain the results.
The long-term stability of DSSCs using different TiO 2 as scattering layers was evaluated by recording the PCE of cells kept in dark and ambient condition. The results are given in Fig. S11. † Aer 92 days, the cell T + F still showed 96.8% of the initial efficiency, as good as the cells based on all commercial TiO 2 photoanodes.

Conclusions
In summary, we have demonstrated a facile and versatile method to prepare mesocrystal lms by making the printable paste of the topotactic precursor followed by in situ topotactic transformation of printed lms. In this work we prove that this technique works very well for anatase TiO 2 . The obtained mcTiO 2 from printed NH 4 TiOF 3 lms possesses typical structural features of mesocrystals, which brings properties associated with individual nanoparticles, advantages from the crystallographic orientation of nanoparticles and collective functions of the mesocrystal. In a specic application, DSSCs, we applied the mcTiO 2 as single-layer photoanodes, in which the properties from the structural features were discussed and compared with commercial TiO 2 in detail; and also as scattering layer to make better use of all the structural properties. Compared with commercial scattering TiO 2 , mcTiO 2 showed a better performance as the scattering layer owing to its high dye loading capacity, superior visible light scattering and reasonable charge transport. We believe that the mcTiO 2 lms from in situ topotactic transformation could also make the best use of the structural features in other applications using porous TiO 2 . Furthermore, this method simplies manufacturing processes for mesocrystal lms and opens up general opportunities for other functional ceramic lms.

Author contributions
BL designed, carried out and analysed most of the practical work and draed the manuscript. AVO carried out conrmatory PCE measurements and the IPCE measurements. NR directly supervised most of the work and both NR and AI inputted into manuscript preparation and results interpretation.