1. Introduction

Titania (TiO$_{2})$ has been extensively investigated as an n-type semiconductor performing as a photo-anode for photo-electro-chemical (PEC) water splitting due to its chemical resistance,environmental suitability and low cost. Another main advantage of TiO$_{2}$ is the suitable band edge positions,since a semiconductor for water splitting needs to have a minimum energy gap of 1.23 eV along with conduction and valence bands overlapping the oxidation of H$_{2}$O and reduction of H$^{+}$ respectively. The main limitations of stoichiometric TiO$_{2}$ are its large band gap (3.2 eV) and high electron-hole recombination due to the short life of the excited electron. A variety of TiO$_{2}$ structures have been investigated,for example carbon-doped TiO$_{2}$^[1] or Sr-doped TiO$_{2}$^[2],hydrogen treated TiO$_{2}$^[3],TiO$_{2}$ nanostructures^{[3, 4, 5]} and TiO$_{2}$ heterostructures^{[2, 6]} in order to overcome the limitations and achieve improved performance for water-splitting applications.

Previous work on Ti-suboxides has shown that it is possible to control the number oxygen vacancies and its electrical conductivity from 10$^{-9}$ S/m (TiO$_{2})$ to 10$^{4}$ S/m (Magnli phases)^[7]. In this paper,we investigate the photo-electro-chemical properties of a novel functionally graded structure,whereby fine scale fibres are formed with a gradually increasing density of oxygen vacancies along its length. Figure 1 shows a schematic representation of a functionally graded fibre structure with one end consisting of TiO$_{2}$ with oxygen vacancies introduced by carbo-thermal reduction and the other edge is chemically reduced to a larger extent and therefore consists primarily of Magnli phases (Ti$_{n}$O$_{2n-1}$,3 $<$ $n$ $<$ 10). The heavily reduced,and electrically conductive,end is then bonded with Ag wires using Al paste; which has been shown to form anohmic electrical contact with the Magnli phase region^[7]. The photocurrent density efficiency was determined and the semiconductor-electrolyte interface was studied.

2. Experimental

2.1 Materials preparation and fibre characterization

The TiO$_{2}$ fibres were produced using a thermoplastic extrusion process. Titanium dioxide powder (PI-KEM,99.5%,0.3 $\mu$m particle size,specific surface area 7.49 m$^{2}$/g) was pre-coated with three monolayers of stearic acid (93661,FlukaChemie AG,Switzerland). The stearic acid was solved in toluene and mixed with the ceramic powder in a jar mill with zirconia milling media for 12 h. The toluene was dried out using a rotary evaporator (Rotavapor R-134,B\"{u}chiLabortechnik AG,Switzerland). The pre-mixed powder was blended with polyethylene binder (1700MN18C Lacqtene PEBD,Arkema Group,Cedex,France) using a torque rheometer (HAAKE PolyLab Mixer,Rheomix 600,Thermo Scientific,Karlsruhe,Germany). For the two step mixing a temperature of 150 C (1st step) and 120 C (2nd step) was used. After mixing a thermoplastic homogeneous feedstock with 54 vol.% of TiO$_{2}$ powder was achieved. This feedstock was used for thermoplastic extrusion of fibres with a diameter of 500 $\mu$m using a capillary rheometer (RH7-2,Malvern,Herrenberg,Germany) at a temperature of 120 C. A special ceramic die design (Empa,Switzerland) with an orifice of 500 $\mu$m was used to produce the fibres. The fibres were extruded with a ram speed of 0.5~mm/s and a pressure of 13 MPa^[8]. The `green' TiO$_{2}$ fibres were cut on a conveyor belt into 170 mm long lengths. The `green-body' fibres were then sintered at 1300 C for 1.5 h in chamber furnace (UAF,LENTON,UK) with a dwell stage at 500 C in order to burn out the binder. This sintering pattern was selected after optimization in order to achieve high density samples ($>$ 97% of the theoretical) and avoid significant grain growth. After sintering,the diameter of the TiO$_{2}$ fibres was 440 $\mu$m.

To produce the functionally graded fibres the TiO$_{2}$ sintered fibres were reduced using a carbo-thermal process performed in a tubular furnace (LTF,LENTON,UK) with carbon black powder as reducing agent that was in contact with one end of the 10 mm long fibres. The reduction process was performed at 1300 C for 1 h under constant argon flow. The manufacturing and characterization of homogeneous TiO$_{2}$ and Magnli phases fibres has been presented in detail in previous work^[7] that was focusing on structural and electrical characterization. In the present work these homogenous fibres (length 10 mm,$\phi$ 0.44 mm) are also tested to compare with the functionally graded fibres. Figure 2 shows a schematic diagram of the process developed to manufacture the functionally graded titania fibres. To characterize the structure of the graded fibres and determine the phases present at both extremes of the fibre lengths,X-ray diffraction (XRD),Bruker D8-Advance using Cu-K$\alpha$ wavelengths,and scanning electron microscopy (SEM,JEOL JSM6480LV) were used.

2.2 Characterization of photocurrents

The photocurrent of TiO$_{2}$,Magnli phases and functional graded fibres was studied as a function of the intensity of the UV light using a novel LED reaction cell. The incident photon-to-photocurrent efficiency (IPCE) was determined according to the equation:

${\rm IPCE}(\%) = \frac{i_{\rm ph}hc}{\lambda P e}\times 100,$

(1)

where $i_{\rm ph}$ is the photocurrent density,$h$ is Planck's constant,$c$ velocity of light,$P$ the light power density,$\lambda$ is the irradiation wavelength,and $e$ is the elemental charge. PEC measurements on individual fibres were performed in a three electrode electrochemical system using a saturated calomel electrode (SCE) reference electrode and a platinum wire as a counter electrode in 1 M KOH electrolyte.

The reaction cell was equipped with an LED board with 36 LEDs controlled in 12 individual channels of three LEDs each ( Figure 3). The arrangement of LEDs was designed to provide an even illumination within the reaction cell. The LED board is an insulated metal substrate (IMS) printed circuit board (PCB) for better thermal management and each LED generates 800~mW of optical power at full electric power; this corresponds to $\sim$1.9 kW/m$^{2}$ of optical power at a distance of 100 mm from the LEDs. In this work,tests were conducted on single fibres using 40% of the power,i.e. at 760 W/m$^{2}$. There is a $\pm$10% variation in optical power between LEDs,as well as $\pm$2 nm variation in dominant wavelength^[9]. In order to achieve an improved control of the optical power delivered to the fibres,an electrically erasable programmable read-only memory (EEPROM) chip was applied to the LED light source. The EEPROM holds the information about the intensity of each channel at different electrical power levels and the information on dominant wavelength offset from the desired wavelength; in this case 368 nm. Using this information the controller can adjust the input power to all channels to provide the correct optical power. Since the LEDs are heat sensitive,a temperature sensor was placed on the PCB of the UV light source and the sensor was used as an overheating protection. The LEDs were controlled with a pulse width modulated (PWM) signal generated by the LED driver circuitry. These LEDs are therefore cycled between being fully on and fully off and the duty cycle of the PWM signal determines the average intensity. UV light of 368 nm was chosen since previous work on TiO$_{2}$ has reported that doping or oxygen vacancies have the greatest affect in this wavelength region^{[10, 11, 12]}.

The reactor used was a glass container with double walls with a space between the reactor vessel walls for a coolant to keep the temperature of the fibre in the reactor at a stable temperature. Linear sweeps of potential were collected electrode using an Ivium Technologies potentiostat. The interface of the semiconductor and the electrolyte was also studied using a Solartron SI 1250 coupled with a SI 1286 dielectric interface.

For an n-type semiconductor electrode at open circuit,the Fermi level is typically higher than the redox potential of the electrolyte^[13] and hence there is an upward bending of the band edges. At positive potentials the holes move towards the electrode-electrolyte interface and electrons move to the interior of the semiconductor,where the Magnli phases are highly conductive and prevent electron-hole recombination. The gradual increase of the conductivity across the length of fibres improves the interconnectivity and the collection of the photo generated electrons at the base of the fibres^[14]. At a certain potential the Fermi level is at the same energy level as the solution redox potential and therefore there is no band bending. This potential is referred to as the flat band potential ($E_{\rm fb})$ which can be calculated based on the Mott-Schottky relationship by measuring the apparent capacitance as a function of the applied potential (Equation (2)). The donor density can be calculated from the gradient ($n)$ of the (1/$C^{2}$ versus $E$) curve and $E_{\rm fb}$ determined by extrapolation to the condition,$C$ $=$ 0.

$\frac{1}{C^2}=\frac{2(E-E_{\rm fb}-kT/e)}{N_{\rm D} \varepsilon \varepsilon_0 e A^2},$

(2)

$N_{\rm D}=\frac{2}{\varepsilon \varepsilon_0 A^2 e n},$

(3)

where $N_{\rm D}$ is the carrier density,$k$ the Boltzmann constant,$T$ the temperature,$\varepsilon$ the dielectric constant of the anodic film,$\varepsilon_{0}$ the permittivity of free space,$e$ the charge of an electron,and $A$ the electrode surface. There are two capacitance values to be considered,the space charge region and the double layer. Since the space charge capacitance is much smaller than the double layer capacitance^[15] the contribution of the space charge capacitance to the total capacitance is considered negligible. The equivalent circuit used for the modelling of the electrochemical impedance spectroscopy measurements is an $R$ ($R$-$C$) circuit^[16].

3. Results and discussion

The morphology and structure of the functionally graded fibres were characterized by SEM and XRD. Figures 4(a) and 4(b) present SEM images of lightly reduced TiO$_{2-x}$ fibre end (i.e. furthest away from carbon source) and Figures 4(c) and 4(d) show the heavily reduced Magnli phase end of the functionally graded fibre.

Using the linear intercept method,the measured grain size was smaller at the TiO$_{2-x}$ end of the functionally graded fibre. The grain sizes are 5.9 and 8.3 $\mu$m for the TiO$_{2-x}$ and the Magnli phases end respectively. At the Magnli phase end of the fibres oxygen diffusion is more intense and this can lead to grain boundary migration that increases the grain size^[17]. After sintering and reduction the diameter is 440 $\mu$m across the whole length of the fibres. XRD spectra for the functionally graded fibres are presented in Figure 5 and show the different phases formed at each end of the fibres. The XRD spectrum of the TiO$_{2-x}$ fibre end shows the presence of rutile TiO$_{2}$ and titanium suboxides while the XRD spectrum of the Magnli phase fibre end shows the presence of titanium suboxides whereas the TiO$_{2}$ peaks are no longer present.

To investigate the effect of the oxygen vacancies on the electrode-electrolyte interface electrochemical impedance measurements on single Magnli phases and functionally graded fibres were also conducted. Measurements conducted on single TiO$_{2}$ fibres were subject to noise due to their low electrical conductivity (10$^{-9}$ S/m) and accurate fitting of the impedance measurements was therefore not possible. Figures 6(a) and 6(b) show the Mott-Schottky plots for the Magnli phase and functionally graded fibres. Both curves exhibit positive slopes suggesting the n-type semiconductor behaviour of both materials.

The flat band potential was determined to be 1.92 V versus SCE for the Magnli phase fibre whereas for the functionally graded fibre it was -2.33 V versus SCE. The Mangli phases exhibit 2-D electronic conductivity and therefore the number of available carries is not as high in this region compared to the TiO$_{2-x}$ fibre end where the oxygen vacancies introduced by partial reduction increase the carrier density. By determining the gradient ($n)$ of the Mott-Schottky plots in Figure 6 and using Equation (3) the carrier density was determined; for the homogenous Magnli phase fibre the carrier density is 4.16 $\times$ 10$^{14}$ cm$^{-3}$ and for the functionally graded fibre is 5.72 $\times$ 10$^{17}$ cm$^{-3}$. Oxygen vacancies are known to be shallow donors for the rutile phase of TiO$_{2}$ and therefore increase the carrier density^[18]. In the case of the Magnli phases the structure and the conduction mechanism changes and thus the oxygen vacancies do not increase the carrier density^[19]. The EIS measurements ( Figure 6) that showed the better performance (more negative flat band potential) of the functionally graded fibres are confirmed from the photocurrent measurements that are shown in Figures 7(a)-7(c) which compare the photocurrent densities of measurements of TiO$_{2}$,Magnli phase and functionally graded fibres. The photocurrent density produced from the functionally graded fibre is much higher (0.786 mA/cm$^{2})$ than both the TiO$_{2}$ (0.032 mA/cm$^{2})$ and Magnli (0.065 mA/cm$^{2})$. The IPCE was also calculated as shown in Table 1 at $-0.241$ V versus SCE (0 V versus RHE). The TiO$_{2}$ fibre has the lowest efficiency,where the Magnli phases fibre has an increased performance but is not as high as the functionally graded fibre,which is in agreement with the measured carrier density. In addition to the higher carrier density the functionally graded fibres have improved interconnectivity due to the gradually increasing oxygen vacancies that enhance the electron-hole separation.

The calculated IPCE for the functionally graded fibres at 368 nm is higher than the efficiency reported in other work on modified TiO$_{2}$^{[1, 5, 11]} and therefore shows the promise of functionally graded fibres as part of a fibre array for water splitting applications. Further work on the graded fibres will explore the measurement of the photocurrent under visible light,the determination of the efficiency and testing of larger scale fibre arrays.

4. Conclusions

This paper has investigated work to enhance the performance of TiO$_{2}$ as a photo anode for water splitting by forming functionally graded and fine scale fibres that could be used as an array of active fibres. The fibres have been formed by carbo-thermal reduction on extruded TiO$_{2}$ fibres and the low-cost extrusion process is of interest to readily form large scale arrays of such structures. XRD shows that the fibre end furthest from the carbon source during the reduction process consists mostly of rutile TiO$_{2}$ along with titanium suboxides (TiO$_{2-x})$ due the oxygen vacancies introduced by carbo-thermal reduction. This fibre end is placed in contact with the electrolyte and acts to absorb the UV light. The fibre end that is heavily reduced and in direct contact with the carbon source during the reduction process consists of Magnli phases that exhibit electrical conductivity close to metallic (10$^{4}$ S/m). This end of fibre is used to encourage electron-hole separation. As shown from the EIS data,the level of oxygen vacancies along the fibre length affect the flat band potential that moves to negative values (at both ends) and are therefore more beneficial of water splitting applications. In addition,the TiO$_{2-x}$ fibre end increases the carrier density due to the increased shallow donors. The photocurrent density of the functionally graded fibres was significantly higher compared to homogeneous TiO$_{2}$ and Magnli phases fibres reaching an efficiency of 34.8% and demonstrates the potential of functionally graded fine scale TiO$_{2}$ fibres for water splitting applications.

Acknowledgements

The research leading to the materials development and characterisation was from the European Union's Seventh Framework Programme (FP7/2007-2013)/ERC Grant Agreement no. 320963 on Novel Energy Materials,Engineering,Science and Integrated Systems (NEMESIS). Reactor development was supported by The European Union under FP7 Project 309846,"Photocatalytic Materials for the Destruction of Recalcitrant Organic Industrial Waste - PCATDES.''