Design aspects of all atomic layer deposited TiO₂–Fe₂O₃ scaffold-absorber photoanodes for water splitting

Abstract

Iron and titanium oxides have attracted substantial attention in photoelectrochemical water splitting applications. However, both materials suffer from intrinsic limitations that constrain the final device performance. In order to overcome the limitations of the two materials alone, their combination has been proposed as a solution to the problems. Here we report on the fabrication of an atomic layer deposited (ALD) Fe₂O₃ coating on porous ALD-TiO₂. Our results show that successful implementation requires complete mixing of the TiO₂ and Fe₂O₃ layers via annealing resulting in the formation of a photoactive iron titanium oxide on the surface. Moreover, we found that incomplete mixing leads to crystallization of Fe₂O₃ to hematite that is detrimental to the photoelectrochemical performance. IPCE and transient photocurrent measurements performed using UV and visible light excitation confirmed that the iron titanium oxide extends the photocurrent generation to the visible range. These measurements were complemented by transient absorption spectroscopy (TAS), which revealed a new band absent in pristine hematite or anatase TiO₂ that we assign to charge transfer within the structure. Taken together, these results provide design guidelines to be considered when aiming to combine TiO₂ and Fe₂O₃ for photoelectrochemical applications.

---

1 Introduction

Photoelectrochemical hydrogen production from water could provide a clean and inexhaustible source of energy if successfully realized on a large scale.¹ For this reason, it has attracted wide interest among researchers ever since the pioneering reports on the topic were published.^2,3 Efficient photoelectrochemical water splitting has high demands on the material properties, which have proven to be difficult task to combine all in only one semiconductor. Therefore, a solution to the problem has been sought from material combinations.

One strategy to combine two materials is to divide the functions of light absorption and charge transport between an absorber and a high surface area scaffold similarly to dye-sensitized solar cells.⁴ The scaffolding approach (also known as the host-scaffold/guest-absorber approach) was first demonstrated in water splitting by Sivula et al. where a 60 nm thick hematite layer was deposited on nanostructured WO₃.⁴ Later, demonstrations of a 25 nm thick hematite layer on TiS₂ nanonets, a 5 nm thick layer on ZnO nanorods,⁵ and a few nanometer thick hematite coating on porous ZnO⁶ have been reported. The suitability of TiO₂ as a scaffold for hematite has been studied previously despite the unfavorable band alignment for electron injection. Successful combining of the good electron transport properties of TiO₂ and light absorption of iron oxide holds great promise for efficient solar water splitting with safe, abundant and stable materials.

Successful implementation of hematite on TiO₂ has been reported to require an underlayer (Nb₂O₅),⁷ while at the same time successful realization without an underlayer has also been reported.⁸ Moreover, annealing the TiO₂ with Fe₂O₃ to create a thin Fe₂TiO₅ layer on the surface has been proposed to be the mechanism to reach high photocurrents.⁹ Therefore, the varying experimental results implies that the question of how to combine TiO₂ and Fe₂O₃ in the scaffold/absorber configuration requires further clarification. Furthermore, a recent review tells that the working principles of the Fe₂O₃–TiO₂ heterojunction devices still remain unclear and suggests further investigation on this promising topic.¹⁰

The aim of this paper is to establish whether TiO₂ is a suitable scaffold for Fe₂O₃ in the scaffold/absorber configuration. To answer this question, we applied 100, 200 and 400 atomic layer deposition (ALD) cycles of Fe₂O₃ on porous TiO₂. The thicknesses correspond nominally to 5, 10 and 20 nanometers. The composite materials were characterized by scanning electron microscopy (SEM), transmission electron microscopy (TEM), X-ray photoelectron spectroscopy (XPS), Raman spectroscopy, X-ray diffraction (XRD), and they were tested as anodes in a photoelectrochemical (PEC) water splitting cell. The steady-state PEC experiments were complemented by transient absorption (TAS) and transient photocurrent (TPC) measurements to gain insight into the water oxidation dynamics.

Our results show that an annealing step which creates a thin FeTi-oxide layer on the surface is necessary for successful device operation. Importantly, we find that crystallization of ALD-Fe₂O₃ to hematite will lead to reduced photocurrent once hematite begins to dominate the surface. The growth of hematite can be avoided by using a low number of ALD cycles. For our device architecture, we discover that 100 ALD-cycles is sufficient to increase the absorption of the photoanode without significant hematite crystal growth.

The results presented herein provide new insight into the fabrication of porous TiO₂/Fe₂O₃ photoanodes with the scaffold/absorber technique by clarifying the surface science of the material composition from the perspective of solar water splitting. While this study was carried out by the means of ALD the here presented phenomena and the design principles derived from them are applicable to other fabrication techniques as well.

2 Results and discussion

The TiO₂ scaffolds used in this work were fabricated according to a previously published method of ALD on cellulose substrates.¹¹ The resulting weblike TiO₂ structure was selected to this study for its large pore size allowing easy penetration of ALD precursors. SEM and TEM images of the TiO₂ scaffolds coated with 100, 200 and 400 cycles of ALD-Fe₂O₃ after annealing are presented in Fig. 1 (see Fig. S1† for a zoomed-out image). TEM images show the hollow tubular structure of the TiO₂ scaffold on top of which the ALD-Fe₂O₃ grow. Based on the SEM images, the application of 100 and 200 iron oxide deposition cycles was found to form individual islands which cover the TiO₂ scaffolds throughout. The islands are present already prior to annealing. Increasing the cycle number resulted in increased island sizes that grow large enough to coalesce together.

Fig. 1 (a–c) SEM images of a TiO₂ scaffold coated with 100, 200, and 400 cycles of ALD-Fe₂O₃, respectively. Images (d–f) are TEM images of the corresponding samples.

X-ray photoelectron spectroscopy was implemented to further characterize the surface of the structure. Table 1 shows the relative elemental concentrations of Ti, Fe and O for the TiO₂ sample with 100 cycles of ALD-Fe₂O₃ coating. Upon annealing at 680 °C the Ti/Fe ratio increases from 1.16 to 2.03, which indicates diffusion of Ti towards the surface. Furthermore, annealing increases the Fe 2p and Ti 2p peak separation by 0.3 eV (Ti 2p peak shifts to lower binding energy), which can be assigned to the formation of a mixed FeTi-oxide. In the studied samples, titanium was found to have oxidation state Ti⁴⁺ and iron Fe³⁺ suggesting the formation of a Fe(III)Ti(IV) oxide. A similar peak shift has been previously observed and associated to the formation of amorphous Fe_xTi_1−xO_y, where a thin TiO₂ coating was deposited over Fe₂O₃ nanorods.¹² The XPS spectra for Fe 2p and Ti 2p energy regions are given in the ESI (Fig. S2†). No Sn was detected on the surface, while the diffusion of tin originating from the underlying FTO has been reported to occur upon annealing at high temperatures.¹³

Table 1 Relative surface concentrations and core level binding energies (BE) of Ti, Fe and O for TiO₂ scaffold coated with 100 cycles of Fe₂O₃. The BE scale was calibrated according to the C 1s C–C/H component at 285.0 eV. The relative surface concentrations of all the detected elements including adventitious C and traces of N, Ca, Na, Mg and Si are shown in the ESI (Table S1)

Sample	Concentration, at% (XPS peak position, eV)
	Fe 2p3/2	Ti 2p3/2	O 1s	Ti/Fe	Δ(Fe 2p3/2–Ti 2p3/2)
As-deposited	11.69 (710.62)	13.53 (458.84)	74.78 (530.07)	1.16	(251.78)
Annealed	7.44 (710.53)	15.07 (458.43)	77.49 (529.78)	2.03	(252.10)

---

X-ray diffraction (XRD) was used to identify the crystal structures. As XRD is a bulk characterization method, all samples indicated crystalline SnO₂ from the substrate and anatase TiO₂. In the as-deposited condition, the existence of hematite was visible for the sample with 400 ALD cycles, as indicated by its highest intensity {104} peak at 33.3° (see Fig. S3† for the diffractogram). The hematite signal was not detected for the 100 or 200 cycle samples suggesting an amorphous structure. Even though ALD metal oxide films obtained at high temperatures are usually crystalline as-deposited, ultra-thin films (<5 nm) regularly remain amorphous until a critical film thickness value is reached.¹⁴ Upon annealing, the high intensity peak of hematite at 33.3° is intensified for the 400 cycles sample which suggests increased degree of crystallinity in the hematite layers compared to the non-annealed ones. Hematite was found in the annealed 100 and 200 cycle samples only in trace amounts.

Raman spectroscopy was used as a complementary experiment to study the annealed samples in more detail. The TiO₂ scaffold shows typical anatase signal peaks at 144, 197, 395, 515 and 636 cm⁻¹.¹⁵ The hematite peaks¹⁶ at 244, 291, 297, 410, 499, 610 cm⁻¹ become clearly visible only after the coating thickness increases up to 400 cycles (Fig. 2). For the 100 and 200 cycle coatings the hematite signal is weak indicating that only a minor amount of ALD-Fe₂O₃ crystallized to hematite being in line with the XRD results. The LO (longitudinal optical mode) peak at 660 cm⁻¹, which was observed only for the 400 cycle coated sample, has been assigned to disorder within the hematite lattice.¹⁷ The appearance of this peak has been also previously observed to appear for high temperature (650 °C) annealed Fe₂O₃–TiO₂ systems and assigned to a reaction between hematite and TiO₂.¹² This peak is not detected for the 100 and 200 cycle samples as the hematite signal itself is barely visible.

Fig. 2 Raman spectra of TiO₂ scaffold with varying thickness of ALD-Fe₂O₃ coating. H: hematite, A: anatase, LO: longitudinal optical mode.

Fig. 3 shows IV-curves for TiO₂ scaffold coated with ALD-Fe₂O₃ along with the bare weblike TiO₂ scaffold for comparison. The IV-curves are measured after annealing at 680 °C, which was found to be prerequisite for obtaining reasonable photocurrent from the ALD-Fe₂O₃ coated samples (see Fig S4† for the IV-curves of as-deposited samples). The TiO₂-web shows an IV-curve with photocurrent in the same order of magnitude that has been previously reported for cellulose templated porous ALD-TiO₂.¹⁸ This allows us to conclude that the web worked as expected and provided a suitable platform for the ALD-Fe₂O₃ coatings.

Fig. 3 IV-curves of the studied samples.

For the ALD-Fe₂O₃ samples we found out that the 100 and 200 cycle coatings gave significantly higher photocurrent than the thicker 400 cycle coating. Based on the results obtained by Raman spectroscopy and XRD we can see that the drop of the photocurrent is caused by the crystallization of ALD-Fe₂O₃ into hematite (see Fig. 2 & 3). This is in accordance with the observation noted previously by Sivula et al.⁴ and Stefik et al.⁷ that hematite overlayers on titania results in no photocurrent. This indeed is expected as hematite has a conduction band edge below that of titania, creating an injection barrier for the electrons photogenerated in the hematite layer.^7,19 The photoelectrochemical performance of our materials closely matches those published earlier for Fe_x–TiO₂ photoanodes prepared via sol–gel.²⁰

The XPS analysis indicated the formation of an FeTi-oxide layer on the surface upon annealing. When the ALD-Fe₂O₃ cycle number is kept low, in our case 200 or below, the mixing of ALD-Fe₂O₃ with Ti during annealing is complete resulting in FeTi-oxide on the surface. This explains why the 100 & 200 cycle coated samples overperform the 400 cycle coated sample, where part of the ALD-Fe₂O₃ film crystallizes as hematite. Previously, a Fe₂TiO₅ (pseudobrookite) layer has been synthesized on TiO₂ nanotubes by high temperature annealing of electrodeposited Fe₂O₃.⁹ More frequently, however, the Fe₂TiO₅ has been designed to inject electrons to hematite^12,21–23 in contrast to what we report here.

Han et al.²⁴ noted on the importance of not covering the whole TiO₂ substrate with hematite in their system where the photoactive material was the TiO₂. Our system is significantly different as the photoactive material is the mixed FeTi-oxide. To prove that the improved efficiency stems from the increased spectral response, originating from the mixed oxide, incident photon-to-current-efficiency (IPCE) was measured. As expected, the bare TiO₂ gives rise to photocurrent only with UV excitation, and no signal with visible light excitation due to the large band gap of the material. However, upon coating of the TiO₂ electrode with ALD-Fe₂O₃ followed by annealing, the spectral response is observed to extend into the visible range up to 600 nm. The IPCE of the hematite containing sample (400 cycles) is reduced compared to the 100 & 200 cycle samples, because hematite cannot contribute to the photocurrent due to the conduction band misalignment. The IPCE spectra are given in the ESI† along with UV-Vis absorption spectra of the materials Fig. S5a and b.†

Transient photocurrent (TPC) decays, measured with UV (355 nm) and visible light (410 nm) laser excitation, complement the steady-state IPCE measurements. The TPC experiment measures the decay of the photocurrent signal created by the laser excitation. The signal is a sum of two decay pathways: the electron–hole recombination and the electron extraction by the external circuit.^25,26Fig. 4 shows TPC decays for bare TiO₂ and FeTi-oxide normalized relative to the TiO₂ 355 nm signal. All the presented decays have similar kinetics, which imply that the aforementioned processes are not significantly altered by the coating within the resolution of the experiment. This similarity of the decay curves is evidence that the FeTi-oxide layer is able to inject electrons to the TiO₂ network through which the charges diffuse to the FTO contact.

Fig. 4 Photocurrent decays of the ALD-Fe₂O₃ sensitized (100 cycles, annealed) TiO₂ electrodes and the reference TiO₂ scaffold. Decays are normalized relative to the TiO₂ 355 nm signal.

The charge carrier recombination dynamics were studied in the ps–ns timescale using transient absorption spectroscopy of annealed planar 200 cycles ALD-Fe₂O₃ samples deposited on FTO and TiO₂ coated FTO. The 200 cycle sample was selected for the study to achieve high enough absorbance on the planar samples. The excitation wavelength was set to 410 nm to selectively excite only the ALD-Fe₂O₃ overlayer. In this time-scale the transient absorption is composed of a superposition of the absorptions of both photoexcited holes and electrons.

The negative transient absorption band below 450 nm is due to the bleaching of the ground state absorption. The strong positive absorption band centered at 570–580 nm in the bare hematite sample has previously been assigned to the absorption of photoexcited electrons.²⁷ However, recent investigations of hematite photoanodes under applied bias voltages have revealed a positive absorption at the same wavelength that is formed due to high valent iron-oxo species.^28,29 We have previously observed that a bias dependent bleaching of this absorption occurs due to electron transfer into oxidised surface states.³⁰ Thus, we conclude that this absorption band is most probably due to photoexcited holes in the hematite bulk. The decay of the photoexcited electrons is a multiexponential process dominated by recombination and electron trapping.³⁰ The decay of the TAS signal was best fit with three exponential functions with lifetimes of 0.29, 17, and 320 ps (Fig. S6a–c & Table S2†).

The transient absorption spectrum of the ALD-Fe₂O₃ sample deposited on top of TiO₂ and annealed at 680 °C shows the formation of a new broad positive absorption band between 450–680 nm (see Fig. 5b). Neither bare hematite or TiO₂ (Fig. S6d†) show a positive transient absorption below 535 nm. Additionally, the globally fitted three exponential decay of the TAS signal now has lifetimes of 0.26, 4.7, and 320 ps. The fastest and slowest components of the decay match those for the bare hematite sample, whereas the middle component is now significantly faster. We determine this new positive absorption band to indicate that electrons have been transferred to the TiO₂ layer. The transfer of electrons between the layers explains the faster decay component, and the positive transient absorption spanning from 450 to 680 nm is assigned to the free electrons transferred to the TiO₂ layer.³¹ Interestingly, the ground state bleach observed for the bare hematite sample below 450 nm is not visible for the sample deposited on top of TiO₂. This can be due to the transfer of conduction band electrons away from the FeTi-oxide layer leading to the ground state bleach disappearing, further supporting our determination of the electron transfer occurring through the mixed oxide layer, as indicated by the TPC results. Typically, electron transfer between unmixed layers should take place in the picoseconds to nanoseconds timescale. The fact that the change in the TAS spectrum occurs immediately after excitation suggests towards the proposed mixing of the two metal oxide layers.

Fig. 5 Transient absorption spectra of (a) hematite and (b) FeTi-oxide. Timescale is linear until 1 ps and logarithmic for longer delay times.

3 Conclusions

We have presented a case study on the combination of ALD-TiO₂ and thin ALD-Fe₂O₃ coating for water splitting purposes in a scaffold/absorber configuration. Our results indicate that the condition for the approach is to induce mixing of the Fe₂O₃ and TiO₂ layers at high temperature, resulting in photoactive Fe(III)Ti(IV) oxide on the surface. Furthermore, growth of hematite crystals when a large number of ALD-Fe₂O₃ cycles was applied resulted in decreased device performance. With 100 and 200 ALD-Fe₂O₃ cycles crystallization to hematite was avoided, and an increase in the photocurrent over TiO₂ was achieved. We consider these results as a indication that ALD-TiO₂ and ALD-Fe₂O₃ is a suitable pair for a host-scaffold/guest-absorber approach when appropriate post-treatments are applied. The tunability of ALD offers vast possibilities for optimization of both the scaffold and coating thicknesses. We propose that follow-up studies should look into applying the same approach on transparent porous TiO₂ structures, possibly fabricated by the means of nanocelluloses, with even smaller ALD-Fe₂O₃ cycle number to reveal the full potential of the approach.

4 Experimental

4.1 TiO₂ thin film preparation

FTO substrates (TEC 7/2.2, 6–8 ohm sq⁻¹, Solems) were cleaned with detergent and water, and subsequently sonicated in water (Milli-Q), acetone and 2-propanol for 15 min in each solvent. A titanium dioxide compact layer was prepared on the cleaned substrates by spin coating from a titanium isopropoxide precursor solution.³² On these substrates, the porous TiO₂ structures were built by coating a cellulose scaffold with 30 nm of TiO₂ using atomic layer deposition. The precursor for the ALD-TiO₂ was tetrakis(dimethylamido)titanium(IV). After ALD the samples were annealed at 450 °C to crystallize the TiO₂ and remove the cellulose from the structure. The details of the method, including the ALD procedure, have been published previously.¹¹

Planar samples for transient absorption measurements were prepared by coating FTO substrates with 30 nm of TiO₂ using the same ALD procedure as for the cellulose substrates, and annealed at 450 °C. When used as a reference for the ALD-Fe₂O₃ coated samples, same post-annealings were applied to all.

4.2 Atomic layer deposition of Fe₂O₃

Hematite thin films were deposited in an F-120 ALD reactor (ASM Microchemistry, Finland) by applying the FeCl₃ + H₂O-process at 400 °C.³³ At this deposition temperature, the growth rate of the hematite thin films on planar substrates is 0.5 Å per cycle as determined via energy dispersive X-ray spectroscopy (EDX) measurements. The EDX spectra were collected using an Oxford INCA 350 microanalysis system connected to a Hitachi S-4800 field emission scanning electron microscope (FESEM). The film thicknesses were calculated from the EDX spectra using the GMRfilm software³⁴ and assuming a bulk density value of 5.3 g cm⁻³.

4.3 Annealing

The high temperature treated samples were annealed at 680 °C for 1 hour using a 20 °C min⁻¹ heating rate. After annealing the samples were taken out of the oven to room temperature while still hot and let cool down.

4.4 Scanning electron microscopy

Field emission scanning electron microscopy images (FESEM) images were taken with a Carl Zeiss Ultra 55 instrument. Acceleration voltages varied from 1 to 5 kV.

4.5 Transmission electron microscopy

Samples for the transmission electron microscopy study of the structure were prepared by scraping a small amount of the sample in ethanol and casting a drop of the ethanol on a carbon-coated copper grid. After evaporation of ethanol at room temperature, the samples where studied with JEOL TEM 2010 with an acceleration voltage of 200 kV.

4.6 X-ray photoelectron spectroscopy

XPS measurements were performed utilizing a non-monochromatized Al K_α X-ray source (hν = 1486.6 eV) and an Argus hemispherical electron spectrometer (Omicron Nanotechnology GmbH) installed in a multifunctional UHV system with a base pressure below 1 × 10⁻¹⁰ mbar.³⁵ The surface elemental concentrations and chemical states of compounds were identified by analyzing high-resolution XPS spectra using CasaXPS software (Version 2.3.16 PR 1.6).³⁶ The binding energy scale was calibrated according to the C 1s C–C/H component at 285.0 eV. After a Shirley-type background subtraction, the peaks were least-squares fitted with a combination of symmetric Gaussian–Lorentzian line shapes. The relative atomic concentrations were calculated using Scofield photoionization cross sections³⁷ and an experimentally measured transmission function of the Argus analyser.

4.7 X-ray diffraction

The crystal structures were defined by XRD (Panalytical Empyrean multipurpose diffractometer) with Cu K_α radiation (λ = 1.5405 Å) and a cathode voltage and current of 45 kV and 40 mA, respectively. The samples were scanned over 2θ = 20–90° with a step size of 0.018° and step duration of 19 s per step. The references were ICOD 00-046-1088, ICOD 01-070-6826/ICSD 98-020-2243 and ICSD 98-008-8418 for SnO₂, TiO₂ and α-Fe₂O₃, respectively.

4.8 Raman spectroscopy

Raman spectra were collected in the backscattering geometry using a confocal Raman microscope equipped with a 532 nm laser source and a 100× objective (NT-MDT Ntegra). The laser power was adjusted to 10 mW at the sample surface using an ND filter. The acquisition time for each measurement was 600 s.

4.9 IV-curves and IPCE

The IV-curves were measured using a three electrode setup having the sample as the working electrode, a platinum counter electrode and a Ag/AgCl reference electrode. 0.1 M NaOH was used as the electrolyte contained in a 1 cm × 5 cm quartz cuvette. The IV-curves were recorded with Agilent E5272A source measure unit under AM1.5G 100 mW cm⁻² (1 Sun) illumination using Sciencetech SS150 – AAA Solar simulator as the light source. The samples were illuminated through the glass substrate. The light intensity was set with a calibrated reference solar cell from Oriel instruments (model 91150V). The photocurrents were scaled relative to the active area of the samples that was typically a 50 mm² square. IPCE was measured with the same setup filtering the light trough band pass filters (FKB-VIS-40 from Thorlabs, FWHM 40 nm) that were placed in front of the sample.

4.10 UV-Vis absorption

UV-Vis absorption was measured with Shimadzu UV-3600 spectrophotometer. The absorbance was recorded from FTO substrates having a 30 nm ALD-TiO₂ layer on top of which ALD-Fe₂O₃ (100, 200 or 400 cycles) was deposited, and then subsequently annealed at 680 °C. The reference hematite film was prepared directly on the FTO substrate without the TiO₂ layer.

4.11 Transient photocurrent measurement

The TPC decays were measured with the same three electrode set up described above keeping the working electrode at a potential of 0.5 V vs. Ag/AgCl. A 50 ohm resistor was connected in series with the sample. The sample was excited with laser pulses (355 nm & 410 nm) and the resulting voltage drop over the resistor was measured as a function of time with an oscilloscope (Tektronix TDS 5032B).

4.12 Transient absorption spectroscopy

The transient absorption studies were performed using the femtosecond pump-probe method. The primary laser pulses were obtained with a Ti:Sapphire laser (Libra F, Coherent Inc., 800 nm, 100 fs pulse width, repetition rate 1 kHz). Approximately 90% of the fundamental beam was directed to an optical parametric amplifier (Topas C, Light Conversion Ltd.) to produce the excitation pump pulses at 410 nm with approximately 1 mm beam diameter at the sample, attenuated to 0.65–0.8 mW with neutral density filters with 500 Hz repetition rate. The white light probe pulses were obtained directing the remaining fundamental laser beam energy through a motorized translational stage to a Ti:sapphire crystal for white continuum generation. The measurement system (ExciPro, CDP systems) was equipped with a Si CCD detector. The absorbance changes were averaged 2500 times for each delay time, and the whole measurement sequence was performed four times for each sample to minimize variations caused by possible excitation power fluctuations.

Conflicts of interest

There are no conflicts to declare.

Acknowledgements

M. L. and T. I. thank Academy of Finland (Centre of Excellence in Atomic Layer Deposition), and EU-FP7 Grant “G-PHOTOCAT” (Grant No. 309636). H. A. thanks the Academy of Finland (grant nos 141481, 286713, 310359, and 309920) and the Jenny and Antti Wihuri Foundation. K. L. thanks the Academy of Finland (grant nos 141481 and 286713). A. H. thanks the Finnish Foundation for Technology Promotion. A. H. and T. I. wish to acknowledge the late Dr Leonid Khriachtchev for his assistance in the Raman measurements.

References

1. T. Hisatomi, J. Kubota and K. Domen, Chem. Soc. Rev., 2014, 43, 7520–7535.
2. A. Fujishima and K. Honda, Nature, 1972, 238, 37.
3. P. J. Boddy, J. Electrochem. Soc., 1968, 115, 199.
4. K. Sivula, F. L. Formal and M. Grätzel, Chem. Mater., 2009, 21, 2862–2867.
5. Y.-K. Hsu, Y.-C. Chen and Y.-G. Lin, ACS Appl. Mater. Interfaces, 2015, 7, 14157–14162.
6. D. Barreca, G. Carraro, A. Gasparotto, C. Maccato, T. Altantzis, C. Sada, K. Kaunisto, T.-P. Ruoko and S. Bals, Adv. Mater. Interfaces, 2017, 1700161.
7. M. Stefik, M. Cornuz, N. Mathews, T. Hisatomi, S. Mhaisalkar and M. Grätzel, Nano Lett., 2012, 12, 5431–5435.
8. L. Liu, Mater. Lett., 2015, 159, 284–288.
9. Q. Liu, J. He, T. Yao, Z. Sun, W. Cheng, S. He, Y. Xie, Y. Peng, H. Cheng, Y. Sun, Y. Jiang, F. Hu, Z. Xie, W. Yan, Z. Pan, Z. Wu and S. Wei, Nat. Commun., 2014, 5, 5122.
10. S. Kment, F. Riboni, S. Pausova, L. Wang, L. Wang, H. Han, Z. Hubicka, J. Krysa, P. Schmuki and R. Zboril, Chem. Soc. Rev., 2017, 46, 3716–3769.
11. A. Hiltunen, K. Lahtonen, J. Saari, A. Ojanperä, E. Sarlin, H. Wondraczek, A. Efimov, K. Kaunisto, P. Vivo, C. Maccato, D. Barreca, P. Fardim, N. Tkachenko, M. Valden and H. Lemmetyinen, ChemPhysChem, 2017, 18, 64–71.
12. X. Li, P. S. Bassi, P. P. Boix, Y. Fang and L. H. Wong, ACS Appl. Mater. Interfaces, 2015, 7, 16960–16966.
13. A. Annamalai, P. S. Shinde, A. Subramanian, J. Y. Kim, J. H. Kim, S. H. Choi, J. S. Lee and J. S. Jang, J. Mater. Chem. A, 2015, 3, 5007–5013.
14. V. Miikkulainen, M. Leskelä, M. Ritala and R. Puurunen, J. Appl. Phys., 2013, 113, 021301.
15. B. Lafuente, R. T. Downs, H. Yang and N. Stone, The power of databases: the RRUFF project, in Highlights in Mineralogical Crystallography, ed. T. Armbruster and R. M. Danisi, W. De Gruyter, Berlin, Germany, 2015, pp. 1–30.
16. B. Lafuente, R. T. Downs, H. Yang and N. Stone, The power of databases: the RRUFF project, in Highlights in Mineralogical Crystallography, T. Armbruster and R. M. Danisi, W. De Gruyter, Berlin, Germany, 2015, pp. 1–30.
17. A. M. Jubb and H. C. Allen, ACS Appl. Mater. Interfaces, 2010, 2, 2804–2812.
18. Z. Li, C. Yao, Y. Yu, Z. Cai and X. Wang, Adv. Mater., 2014, 26, 2262–2267.
19. M. Grätzel, Nature, 2011, 414, 338–344.
20. E. Courtin, G. Baldinozzi, M. T. Sougrati, L. Stievano, C. Sanchez and C. Laberty-Robert, J. Mater. Chem. A, 2014, 2, 6567–6577.
21. L. Wang, N. T. Nguyen, X. Huang, P. Schmuki and Y. Bi, Adv. Funct. Mater., 2017, 27, 1703527.
22. J. Deng, X. Lv, J. Liu, H. Zhang, K. Nie, C. Hong, J. Wang, X. Sun, J. Zhong and S.-T. Lee, ACS Nano, 2015, 9, 5348–5356.
23. D. Monllor-Satoca, M. Bartsch, C. Fabrega, A. Genc, S. Reinhard, T. Andreu, J. Arbiol, M. Niederberger and J. R. Morante, Energy Environ. Sci., 2015, 8, 3242–3254.
24. H. Han, F. Riboni, F. Karlicky, S. Kment, A. Goswami, P. Sudhagar, J. Yoo, L. Wang, O. Tomanec, M. Petr, O. Haderka, C. Terashima, A. Fujishima, P. Schmuki and R. Zboril, Nanoscale, 2017, 9, 134–142.
25. B. C. O'Regan and F. Lenzmann, J. Phys. Chem. B, 2004, 108, 4342–4350.
26. S. R. Pendlebury, A. J. Cowan, M. Barroso, K. Sivula, J. Ye, M. Grätzel, D. R. Klug, J. Tang and J. R. Durrant, Energy Environ. Sci., 2012, 5, 6304–6312.
27. S. Sorenson, E. Driscoll, S. Haghighat and J. M. Dawlaty, J. Phys. Chem. C, 2014, 118, 23621–23626.
28. B. Klahr and T. Hamann, J. Phys. Chem. C, 2014, 118, 10393–10399.
29. O. Zandi, A. R. Schon, H. Hajibabaei and T. W. Hamann, Chem. Mater., 2016, 28, 765–771.
30. T.-P. Ruoko, K. Kaunisto, M. Bärtsch, J. Pohjola, A. Hiltunen, M. Niederberger, N. V. Tkachenko and H. Lemmetyinen, J. Phys. Chem. Lett., 2015, 6, 2859–2864.
31. T. Yoshihara, R. Katoh, A. Furube, Y. Tamaki, M. Murai, K. Hara, S. Murata, H. Arakawa and M. Tachiya, J. Phys. Chem. B, 2004, 108, 3817–3823.
32. S. Pathak, A. Sepe, A. Sadhanala, F. Deschler, A. Haghighirad, N. Sakai, K. C. Goedel, S. D. Stranks, N. Noel, M. Price, S. Hüttner, N. A. Hawkins, R. H. Friend, U. Steiner and H. J. Snaith, ACS Nano, 2015, 9, 2311–2320.
33. J. A. Klug, N. G. Becker, S. C. Riha, A. B. F. Martinson, J. W. Elam, M. J. Pellin and T. Proslier, J. Mater. Chem. A, 2013, 1, 11607–11613.
34. R. A. Waldo, Microbeam Anal., 1988, 23, 310.
35. K. Lahtonen, M. Lampimäki, P. Jussila, M. Hirsimäki and M. Valden, Rev. Sci. Instrum., 2006, 77, 083901.
36. N. Fairley, CasaXPS: Spectrum Processing Software for XPS, AES and SIMS, Casa Software Ltd., Cheshire, UK, 2009, version 2.3.16 PR 1.6, http://www.casaxps.com/.
37. J. H. Scofield, J. Electron Spectrosc. Relat. Phenom., 1976, 8, 129–137.

---

Footnote

† Electronic supplementary information (ESI) available: Additional SEM image, XPS data, XRD diffractograms, IV-curves, IPCE, and TAS spectra. See DOI: 10.1039/c8se00252e

---