Low temperature sintering of binder-containing TiO2/metal peroxide pastes for dye-sensitized solar cells

Nano-structured metal oxide ﬁ lms are key components of dye-sensitized (DSC) solar cells. Scaling such devices requires lower temperature processing to enable cheaper substrates to be used. In this context, we report a new and scalable method to sinter binder-containing metal oxide pastes to make DSC photo-electrodes at lower temperatures. Metal peroxide powders (CaO 2 , MgO 2 , or ZnO 2 ) were added to terpineol-based P25 pastes containing ethyl cellulose binder or to commercial TiO 2 paste (DSL18NR-T). Thermal analysis shows that binder decomposition occurs at 300 (cid:1) C instead of the standard 450 (cid:1) C for a TiO 2 -only paste and suggests that the metal peroxides act as combustion promoters releasing heat and oxygen within the ﬁ lm while heating. The data show that this heat and oxygen release coincide best with binder combustion for ZnO 2 and DSC device tests show that adding ZnO 2 to TiO 2 pastes produces the best performances a ﬀ ording h ¼ 7.5% for small devices (0.26 cm 2 ) and h ¼ 5.7% at 300 (cid:1) C or 450 (cid:1) C for DSL18NR-T/ZnO 2 for larger (1 cm 2 ) devices. To the best of our knowledge, the performance of the (0.26 cm 2 ) cells is comparable to the highest e ﬃ ciency devices reported for DSCs fabricated using low temperature methods. The device e ﬃ ciency is most strongly linked with J sc ; BET and dye sorption measurements suggest that J sc is linked with the metal oxide surface area and dye loading. The latter is linked to the availability of surface sorption sites for dye molecules which is strongly negatively a ﬀ ected by any residual organic binder which resulted from incomplete combustion.

Low temperature sintering of binder-containing TiO 2 /metal peroxide pastes for dye-sensitized solar cells † Peter J. Holliman, * ab Dhiyaa K. Muslem, a Eurig W. Jones, a Arthur Connell, a Matthew L. Davies, a Cecile Charbonneau, b Matthew J. Carnie b and David A. Worsley b Nano-structured metal oxide films are key components of dye-sensitized (DSC) solar cells. Scaling such devices requires lower temperature processing to enable cheaper substrates to be used. In this context, we report a new and scalable method to sinter binder-containing metal oxide pastes to make DSC photo-electrodes at lower temperatures. Metal peroxide powders (CaO 2 , MgO 2 , or ZnO 2 ) were added to terpineol-based P25 pastes containing ethyl cellulose binder or to commercial TiO 2 paste (DSL18NR-T).
Thermal analysis shows that binder decomposition occurs at 300 C instead of the standard 450 C for a TiO 2 -only paste and suggests that the metal peroxides act as combustion promoters releasing heat and oxygen within the film while heating. The data show that this heat and oxygen release coincide best with binder combustion for ZnO 2 and DSC device tests show that adding ZnO 2 to TiO 2 pastes produces the best performances affording h ¼ 7.5% for small devices (0.26 cm 2 ) and h ¼ 5.7% at 300 C or 450 C for DSL18NR-T/ZnO 2 for larger (1 cm 2 ) devices. To the best of our knowledge, the performance of the (0.26 cm 2 ) cells is comparable to the highest efficiency devices reported for DSCs fabricated using low temperature methods. The device efficiency is most strongly linked with J sc ; BET and dye sorption measurements suggest that J sc is linked with the metal oxide surface area and dye loading. The latter is linked to the availability of surface sorption sites for dye molecules which is strongly negatively affected by any residual organic binder which resulted from incomplete combustion.
O'Regan and Grätzel's breakthrough in dye-sensitised solar cell (DSC) technology 1 showed that sintering pre-made TiO 2 nanocrystals 2-4 at 450-600 C signicantly increases the photo-anode surface area, dye loading and short circuit current. This has led to considerable interest in DSC devices as promising candidates for large scale, low cost PVs because they should be manufacturable using printing and roll-to-roll processing using abundant and non-toxic raw materials. 5,6 For DSC devices, the mechanical and electrical connectivity between the TiO 2 particles are key to long-term device function as charge must travel through the photo-electrode to the current collecting working electrode. 7 Typically this is achieved by printing a colloidal suspension of well-dispersed, crystalline TiO 2 nanoparticles where the paste rheology is controlled by the addition of a long-chain organic polymer (the binder). The binder enables crack-free lms to be printed with variable but controlled lm thicknesses at a range of length scales. Without the binder, it is impossible to print large enough photo-electrodes to scale the technology. This means that currently printed lms must be sintered at >450 C to completely combust the organic binder to free up dye sorption sites on the surface of the TiO 2 particles and to enable the formation of robust interparticle connections but low enough to minimise any reduction in the TiO 2 surface area. However, this limits the working electrode substrate to FTO-coated glass or Ti foil 8 which are inexible and heavy or expensive, 9 respectively. By comparison, cheaper, more lightweight and exible substrates which are suitable for large-scale roll-to-roll manufacture (e.g. metal foils, TCO-coated plastic) require lower processing temperatures; typically 300 C for metal foils and 150 C for plastics. 10 However, there are no reports in the literature to date of methods of sintering binder-containing pastes at low temperature.
Instead, although low temperature sintering of photoanodes for DSCs has been widely studied, reports have centred on using binder-free pastes. However, as discussed previously, binder-free pastes are signicantly limited for larger scale applications because of the difficulties in controlling printability and lm consistency over wider areas (e.g. lm thickness, and inter-particle adhesion which can lead to cracking and substrate adhesion which can lead to delamination). Binderfree approaches include spin coating, 11 sol gel, 12 hydrothermal treatment 13-15 chemical sintering, 16,17 pressure sintering, [18][19][20] electrophoretic deposition of TiO 2 nanoparticles, 21 varying TiO 2 particles 22 and the use of electron beam showers. 23 Variations in radiant heating have also been studied including laser treatment, 24 microwave irradiation 25 and post-sintering O 2 plasma 26 aimed at removing residual organic matter from the metal oxide surface. Whilst these approaches have shown promise at the laboratory scale, producing consistent, large-scale meso-porous lms for DSC device manufacturing will require the presence of a binder in the TiO 2 colloid. Hence, the study of low temperature approaches to sintering binder-containing metal oxide pastes remains a key challenge for DSC manufacturing.
In this context, this paper reports a new approach to lower sintering temperatures of binder-containing metal oxide pastes to manufacture photo-electrodes for DSCs. We believe this is the rst report using solid peroxides as combustion promoters. This approach enables accurate control of paste rheology by retaining an organic polymer binder in the metal oxide pastes which is essential to produce coherent, large scale lms with consistent thickness. Our combined experimental approach is to screen larger lms (3 Â 2 cm) to examine lm coherence and colour, to understand the fundamental thermal chemistry of screen printing pastes and how this inuences sintering and then to make larger DSC devices than are typically reported (1 cm 2 ) to test the effects of sintering on performance. Finally, we have studied the metal oxide area and dye loading to investigate the links between the short circuit current and sintering treatment.

Experimental
Paste and device manufacture P25 pastes were prepared by thoroughly mixing P25 (16 g, Degussa) with terpineol (64.9 g, Fluka) and ethyl cellulose (8 g, Fluka Product# 46070). Peroxide containing pastes were prepared by adding calcium peroxide, magnesium peroxide or zinc peroxide (10% w/w versus TiO 2 , all Aldrich) and thoroughly mixing for 1 h. Commercial DSL18NR-T paste (Dyesol) was either used as purchased and metal peroxides were added as described above.
Photo-electrodes were prepared by doctor blading the metal oxide paste onto FTO-coated glass (TEC15, NSG) and sintering in air at either 450 C, 300 C or 150 C for 30 min. Larger lms were prepared (3 Â 2 cm) to study lm coherence over a larger area. DSC devices were prepared using 1 cm 2 lms (minimum 3 replicates per treatment) to study lm coherence and reproducibility. Selected lms were then immersed in 40 mM TiCl 4 -THF 2(aq) at 70 C for 30 min before rinsing with water and air drying for 10 min. The lms were then re-sintered at the same temperature used during the rst sintering process so that no lm experienced a temperature greater than its initial sintering temperature at any point during processing.
Counter electrodes were prepared by spreading the PT1 paste (Dyesol) onto TEC 8 glass (NSG) and heating to 400 C in air for 30 min. The photo and counter electrodes were then sealed together using a Surlyn gasket at 120 C followed by fast dyeing with 0.3 mM N719 (Dyesol) in acetonitrile-tert-butanol (1 : 1 v/v) for 10 min as described previously. 27 The electrolyte was 0.8 M 1-methyl-3-propyl imidazolium iodide, 0.05 M tert-butyl ammonium iodide, 0.05 M I 2 , 0.3 M benzimidazole and 0.05 M guanidinium thiocyanate in acetonitrile.

Device characterization
Current-voltage characteristics were studied using an ABET Solar Simulator with a Xe arc lamp and a Keithley 2400 at 100 mW cm À2 or 1 Sun between 0 and 1 V. The spectral response was measured from 300-800 nm on a QEÂ10 Quantum Efficiency Measurement System in DC mode at a resolution of 10 nm. Lamps were calibrated to 1 Sun (100 mW cm À2 ) using a certied (Oriel 91150V) mono-crystalline Si reference cell traceable to the National Renewable Energy Laboratory (NREL).
Photocurrent and photovoltage transients were measured as described in ref. 28. The white bias light was provided by a BRIDGELUX 9000 lumen LED array (Farnell) whilst the pulse light was provided by a bank of four OSLON PowerCluster green LED arrays (RS). The pulse intensity was chosen to make sure DV < 10 mV above V OC . A pulse length of 250 ms was utilized and was generated via a fast MOSFET transistor controlled by a National Instruments USB-6251 data acquisition board (DAQ) and WaveMetrics IGOR Pro soware. Voltages were measured directly using the DAQ whilst currents were measured via a voltage drop against a 1 U resistor.
XRD was carried out using a PANalytical diffractometer at 45 kV and 35 mA between 20 and 60 2q using Ni-ltered Cu-K a1 radiation (l ¼ 1.5405 A). Thermal gravimetric analysis and differential scanning calorimetry were performed using a SDTQ600 TGA/DSC (TA Instruments Ltd). The pastes were either run as prepared or were pre-dried at 100-110 C for 2 h to evaporate the solvent to focus the analysis on binder combustion. The samples were ramped at 10 C min À1 between room temperature and 600 C under owing air. The surface area of pre-sintered doctor-bladed lms was determined using BET (Brunauer-Emmett-Teller) isotherms at À196 C using a Micromeritics Gemini III 2375. Electron microscopy and energy dispersive X-ray elemental analyses were performed on a Field Emission Gun-Scanning Electron Microscope (FEG-SEM) Hitachi S-4800 (12 keV, 10 mA) equipped with an Oxford instruments X-Max (50 mm 2 window) detector.
N719 dye sorption was studied using sintered P25 and P25/ ZnO 2 lms either by directly immersing the lm in N719 dye solution for 24 h or by scraping sintered lms off a glass substrate and immersing the collected powder into N719 dye solution for 24 h. N719 uptake was quantied by rst desorbing the dye using 0.1 M NaOH (aq) -ethanol (1 : 1 v/v) followed by calculation of the concentration of the desorbed solution using UV-visible spectroscopy at 512 nm in conjunction with a calibration graph (see ESI Fig. 1 †) obtained by using six N719 standards from 0 to 1.0 mM (11.82 mM À1 cm À1 ). 29 The equilibrium N719 adsorption capacity of N719 was measured for freely dispersed P25, P25/ZnO 2 powder at seven N719 concentrations (25-500 mg L À1 ) at 22, 40 and 50 C. The adsorbed dye (q) (mg g À1 ) was calculated according to eqn (1) (ref. 30) q ¼ (C 0 À C e )V/m (1) where C 0 (mg L À1 ) and C e (mg L À1 ) are the initial and equilibrium N719 concentrations (mol L À1 ), respectively, V is the volume of dye solution (L), and m is the TiO 2 mass (mg). Fits to the Langmuir (eqn (2)) 31 or Freundlich isotherms (eqn (3) and (4)) 31,32 were then calculated by plotting the data according to the relevant equations. For Langmuir, where C e is the equilibrium N719 concentration (mg L À1 ), q is the equilibrium adsorption capacity (mg g À1 ), q m is the maximum adsorption capacity (mg g À1 ) and K L (L mg À1 ) is the Langmuir constant. For Freundlich, eqn (3) was converted to the linear form (eqn (4)).
where q e is the equilibrium adsorbate concentration (mg g À1 ), K f is the Freundlich constant (mg g À1 ), C e is the equilibrium solution concentration (mg l À1 ), and 1/n represents the dimensionless heterogeneity factor.

Results and discussion
To emphasise the importance of the binder within the pastes, two P25/terpineol pastes were prepared; one with ethyl cellulose binder and one without binder. These were doctor bladed onto glass microscope slides and oven sintered at 450 C for 30 min. Fig. 1 initially suggests that the resultant lms appear similar. However, a closer inspection of the binder-free lm reveals ne cracks and a simple adhesion test of applying Scotch tape to the lm and removing it shows that the TiO 2 lm arising from the binder-free paste delaminates easily (as denoted by the dashed red box in Fig. 1) whilst the lm from the binder-containing paste remains strongly adhered to the glass substrate.

Thermal chemistry
The approach taken in this paper is to lower the sintering temperature of binder-containing pastes by including metal peroxide powders which decompose to produce metal oxide particles and release oxygen and heat within the photo-electrode lm during the sintering process. Detailed thermal analyses of selected pastes used in this study are shown in Fig. 2 (all the data are shown in ESI Fig. 2 †). These TGA/DSC data can be used to explain the principle of the metal peroxide combustion effect. In addition, the challenge of lowering the sintering temperature of metal oxide pastes can be understood by looking at the detailed thermal analysis of a P25 paste (Fig. 2a). The data show a three-stage weight loss for P25only pastes with an initial loss of terpineol solvent (ca. 70%) between 60 C and 214 C which is associated with an endothermic peak which reaches its minimum at ca. 190 C in line with that this is an evaporative rather than a combustion process. The second weight loss (ca. 10%, 210-320 C) is associated with a broad, multi-feature exotherm with a shoulder at ca. 240 C (labelled Ia in Fig. 2a) and two maxima at ca. 290 and 320 C. These features are associated with the combustion of ethyl cellulose to CO 2(g) and H 2 O (g) . Although the nal weight loss is small (3%, 320-450 C), it is associated with a strong exothermic peak (maximum at ca. 410 C) which continues up to 450 C (labelled IIIa in Fig. 2a). This is ascribed to the combustion of the residual carbonaceous material from the ethyl cellulose binder which, if not combusted, is believed to reduce dye uptake by blocking surface sorption sites. This residual organic matter gives rise to the brown colouration in under-sintered lms (see Fig. 3). It is key to remove this material to prepare suitable metal oxide surfaces for dyeing. The reason for adding the metal peroxides to the paste is to introduce an oxygen source and heat during binder combustion to accelerate the removal of this residual organic matter.
As such, an initial screening of P25 pastes containing metal peroxides involved mixing either calcium, magnesium or zinc peroxide into a P25/terpineol paste containing ethyl cellulose binder at a loading of 5, 10 or 15% by wt versus P25. Aer casting onto glass slides, the lms were sintered up to 250-300 C for 30-120 min. Fig. 3 shows that, aer 30 min at 300 C, the P25 control lm changes from yellow to brown. Further tests show that the P25-only lm becomes white only aer heating to 450 C; this colour change was deemed to be a suitable visual indicator of complete binder combustion which would prepare the metal oxide surface area for dyeing. The lms from the metal peroxide pastes all show some residual pale brown colour at combustion temperatures up to 250 C regardless of the metal peroxide loading (see ESI Fig. 3 †). At 250 C, the P25/CaO 2 lm appeared paler than the other lms aer 30 min but increasing the sintering time up to 120 min did not change this. However, sintering these lms at 300 C for 30 min produces white lms for all the P25/peroxide lms whilst the P25-only control lm remains brown (Fig. 3). Hence, 300 C was considered to be the threshold temperature to initiate both metal peroxide decomposition (MO 2 becoming MO and 1/2O 2 ) and binder combustion.
To explore the reasons for the benecial effects of metal peroxides in detail, further TGA/DSC data have been recorded. All the data are presented in ESI Fig. 2 † whilst only the selected data for ZnO 2 are presented here as these data are representative of all the different metal peroxides tested. Thus, TGA/DSC data for powdered ZnO 2 (Fig. 2b) show a small weight loss (ca. 2%) for up to 190 C and then a very rapid weight loss of ca. 16% centred around 200 C which corresponds to the decomposition of ZnO 2 into ZnO (eqn (5)). This is accompanied by a sharp exothermic peak which is labelled Ib in Fig. 2b. This is important because the ZnO 2 decomposition occurs at a slightly lower temperature than the temperature at which binder combustion commences (see IIa in Fig. 2a). This means that the heat or oxygen released from ZnO 2 decomposition will be available to enhance binder combustion at lower temperatures. In addition, because the metal peroxide particles are mixed into pastes, they are subsequently distributed through the printed and sintered lms. Hence, as the metal peroxides decompose, they release oxygen and heat to their local environment where the binder residues exist. This reduces any mass transfer issues which further helps to explain the enhanced binder combustion at lower temperatures.
These effects are illustrated by the data for a P25/ZnO 2 paste (Fig. 2c). As expected, the data show a very similar pattern to the P25 paste because the paste only contains 10% ZnO 2 . However, there is a much more clearly dened and more intense exothermic peak in the DSC signal (labelled Ic in Fig. 2c). This reects the exothermic process associated with ZnO 2  decomposition. As predicted, this is believed to have two main effects; rstly additional and localized heat is released within the lm and secondly the peroxide decomposing releases oxygen into the lm which can help to oxidize the organic binder. This results in a greater loss of residual organic matter from the lm at temperatures lower than 450 C (labelled IIIc in Fig. 2c). By comparison, the TGA/DSC data for CaO 2 and MgO 2 (see ESI Fig. 2 †) show that, while they decompose to produce CaO and MgO, respectively and to release oxygen, the main weight loss takes place at a higher temperature (ca. 400 C). Also, for both CaO 2 and MgO 2 , their decomposition is an endothermic process which will remove energy from the surrounding paste during sintering. Thus, for these peroxides, the localized addition of oxygen should still occur at similar temperatures for binder combustion and therefore should still help this process. However, during any sintering process, the paste must be heated from room temperature to the sintering dwell temperature. Hence, the higher decomposition temperatures for CaO 2 and MgO 2 mean that oxygen will be released slightly later in the process. Overall, both the higher temperature and endothermic nature of the decomposition mean that CaO 2 and MgO 2 should be less effective combustion additives compared to ZnO 2 .
Having established that the addition of metal peroxides can reduce the sintering temperature of TiO 2 -ethyl cellulose pastes, selected lms were sintered at different temperatures and sensitized with N719 before being manufactured into dyesensitized solar cell devices as described in the next section.

DSC device data
In this paper, we study sintering and so we have made larger photo-electrodes (1 cm 2 ) than are typically reported for DSC devices to test lm cohesion in devices. 33 To study the reproducibility, we have also made replicate devices and reported the average h AE the error. Whilst this suppresses device efficiencies relative to the highest efficiencies reported in the literature (e.g. for ball-milled P25 pastes 34,35 ), this approach is necessary to accurately test the effectiveness of the different sintering treatments. Table 1 shows IV data for replicate devices made using one layer of metal oxide paste (ca. 7 mm thickness) aer sintering at 300 or 450 C. The approach taken was to keep the photo-electrode thickness constant and vary the type of TiO 2 particles used in the paste. This was because DSC devices tend towards thinner electrodes as the molar extinction coefficient (3) of dyes increases and because solid state devices require thinner electrodes. Thus, two different pastes have been studied in these devices; an in house prepared P25-ethyl cellulose-terpineol paste which is denoted as P25 in Table 1 and a commercial TiO 2 paste (DSL18NR-T from Dyesol Ltd) which is denoted as NRT in Table 1. The data for the TiO 2 -only lms P25 sintered at 450 C can be used as a baseline for the other devices. Thus, Device A for P25-only shows a h of 4.4% with a J sc of 9.01 mA cm À2 and a V oc of 0.71 V. By comparison, Device B gives a better performance (h ¼ 5.0%) mainly due to an improvement in J sc up to 9.80 mA cm À2 . This is ascribed to the DSL18NR-T paste containing smaller, nanoparticulate anatase TiO 2 particles leading to higher dye loading (see ESI Fig. 4 †). In addition the DSL18NR-T paste contains no rutile phase whilst P25 does which is known to produce less efficient DSC devices. 36 The DSL18NR-T paste has also been optimized for DSC devices.
For the P25/peroxide lms sintered at 450 C, the highest efficiencies are observed for the ZnO 2 treated pastes. Thus, for P25/ZnO 2 (Device E) h ¼ 4.4% which is comparable to the P25only device (Device A). By comparison, the NRT/ZnO 2 data (Device F) again show a signicant improvement over the P25 and P25/ZnO 2 data with h ¼ 5.7% which is again mainly due to the improved J sc . However, the NRT/ZnO 2 device also shows a signicant improvement compared to Device B which contains only DSL18NR-T with improved efficiency again related to the improved J sc . This is in line with the P25 data which suggest that ZnO 2 provides an additional benet to the device performance. This may be because the larger ZnO 2 particles may increase light scattering during device operation and/or that the ZnO 2 particles provide an additional benet during sintering leading to an improved surface area and dye loading. Having established the inuence of peroxides on larger devices, some smaller (0.26 cm 2 ) devices were prepared and ultra-fast dyed with N719/SQ1 cocktail solution 27 to study the effect of increased J sc on the peroxide photo-electrodes. This showed that device efficiencies of up to h ¼ 7.5% can be achieved from a binder-containing paste sintered at 300 C (Table 1) which is comparable with previous reports for this dye system. 27 To the best of our knowledge, the performance of these devices is comparable to the highest efficiency devices reported for DSCs fabricated using low temperature methods except that the previous reports used binder-free TiO 2 pastes and pressure rather than thermal sintering. 18,19  Sintering the photo-electrodes at 300 C shows the importance of the metal peroxide additions. For example, sintering P25-only devices at 300 C gives very low h regardless of whether TiCl 4 treatment is used (h < 0.3%, Devices F and G); essentially resulting in non-functional devices. Sintering DSL18NR-T pastes which do not contain ZnO 2 at 300 C produces similar device performances; i.e. no efficiency (data not shown). By comparison, sintering P25/metal peroxide lms at 300 C shows similar device performance data to the respective 450 C sintered lms. The highest efficiency for these devices is again observed for P25/ZnO 2 compared to P25/CaO 2 or P25/MgO 2 (Device J versus Devices H and I).
The data also show the importance of TiCl 4 treatment for the metal peroxide-containing lms. For P25/ZnO 2 pastes, the TiCl 4 treatment increases h from 1.1% (Device L, no TiCl 4 ) to 4.1% (Device K, with TiCl 4 ). For the DSL18NR-T/ZnO 2 paste, the effect is even more pronounced with zero h without TiCl 4 treatment (Device N) and h ¼ 5.7% with TiCl 4 (Device M). For DSL18NR-T, it is hard to compare the data as the absence of TiCl 4 produces a non-functional device. However, for the P25/ZnO 2 devices, the main reason is J sc. This effect is ascribed to the TiCl 4 treatment hydrolyzing onto the photo-electrode surface producing a more coherent TiO 2 surface for dyeing. Indeed, it is known that TiCl 4 treatment has little inuence on high temperature sintered devices where all organic matter has been removed and the particles are thermally well sintered together. However, it is noteworthy that the TiCl 4 treatment effectively repairs photoelectrode defects. The differences in performance between the P25 and DSL18NR-T pastes may reect different paste additives which make it more difficult to remove all organic matter from the commercial paste at 300 C. To illustrate the importance of J sc , EQE max for N719 at 530 nm for a P25 device sintered at 450 C is shown in Fig. 4. By comparison of lms sintered at 300 C, the P25/ZnO 2 device shows an EQE of 45% at 530 nm whilst the equivalent P25-only device shows an EQE of <5%.
Finally, the data for NRT/ZnO 2 sintered at 300 C (Device M) shows very similar data to the equivalent device sintered at 450 C with h ¼ 5.7%; again an improvement over the NRT-only device sintered at 450 C (Device B).
Transient photovoltage and photocurrent decay measurements have been used to further study the inuence of sintering temperature, TiCl 4 treatment and ZnO 2 addition to devices ( Fig. 5 and ESI Fig. 5 †). The data show signicantly longer recombination lifetimes for all TiCl 4 treated photo-electrodes compared to their non-TiCl 4 treated analogues. For P25 devices, there is no difference in the transport kinetics which suggests that any increase in J sc observed as a result of TiCl 4 treatment can be attributed to a downward shi in the conduction band as observed in ref. 37. For the 300 C sintered devices however, an apparent decrease in transport kinetics, due to TiCl 4 treatment, may indicate a change in trap density additional to the negative CB shi that is commonly observed as a result of TiCl 4 treatment. The large decrease in the recombination rate observed in devices made with ZnO 2 paste suggests that the ZnO 2 performs a different role from TiCl 4 in the devices and supports our assertion that the peroxide assists in binder combustion and the removal of the residual organic material which otherwise acts as recombination sites. The data also show that the 450 C sintered P25 device without ZnO 2 or TiCl 4 treatment shows comparable recombination lifetimes to TiCl 4 -treated devices suggesting that TiCl 4 treatment does not signicantly affect recombination processes for well-sintered TiO 2 -only photoelectrodes which is in line with many literature reports. However, recombination lifetimes shorten for the 450 C sintered ZnO 2 /P25 device without TiCl 4 treatment but are the shortest for all the 300 C sintered devices which have not been  TiCl 4 treated. These data conrm that both ZnO 2 and TiCl 4 treatment are required for low temperature sintering to achieve adequate recombination lifetimes for effective charge extraction. Overall, these data show that low temperature sintering of binder-containing TiO 2 pastes is possible and with careful control to minimize any over-sintering of the TiO 2 surface can lead to enhanced short-circuit currents. The later materials characterisation work reported in this paper has sought to understand the reasons for these observations.
Studies of varying the ZnO 2 paste loading from 5 to 25% show that a lower ZnO 2 loading (5-10%) gives the best device performance (see ESI Table 1 †). This is mainly due to reduced J sc as the metal peroxide loading increases which is ascribed to lower dye loadings arising from an increased proportion of lower surface area ZnO 2 particles. Studies of extending the sintering time at 300 C from 30 min to 120 min (see ESI Table  2) also show a slight negative impact on the device efficiency due to slight reduction in J sc and V oc . These changes probably reect the increased interactions between the zinc oxide and TiO 2 phases with time which may lead to a loss of surface area and potentially the surface doping of each phase into the other.

Materials characterisation
The aim of this part of the work is to try to understand how the addition of metal peroxides to TiO 2 -ethyl cellulose pastes affects the phases and morphology of the resulting metal oxides. Their surface chemistry has also been studied using BET and dye loading isotherms because of the strong correlation between device performance and J sc . Thus, a key aim of this work is to understand if the completion of sintering and any presence of residual organic matter is linked to the surface area and dye loading. Unless otherwise stated, the data for P25-ethyl cellulose pastes containing 10% ZnO 2 (by wt) are presented as these pastes gave rise to the most efficient DSC devices.
First, looking at the structural phases present, XRD data show that all the lms contain anatase 38 and rutile 39 TiO 2 phases in an approximately 4 : 1 ratio as expected for the P25 lm. The data for a sintered P25/ZnO 2 lm (Fig. 6) show additional peaks at 31.7 and 34.4 2q, which can be attributed to ZnO 40 resulting from the decomposition of ZnO 2 . The data for P25/ CaO 2 or P25/MgO 2 lms were essentially identical to the P25-only lm with no evidence for the formation of a second metal oxide phase which suggests that the CaO 2 or MgO 2 forms amorphous products on decomposition (see ESI Fig. 6 †). The P25-only XRD data also show a slight narrowing of the diffraction lines (see ESI Fig. 6 and 7 †) for lms sintered at 450 C compared to the lms sintered at 300 C which is in line with the increased crystallinity and inter-particle necking. Interestingly, a similar but subtle effect is observed for the P25/peroxide lms sintered at 300 C. It is difficult to identify the reasons for this with absolute condence because XRD measures an "average" across the whole sample. However, this might reect the generation of localized heat during peroxide decomposition as the TGA shows that this is an exothermic process. In turn, this might increase the TiO 2 crystallinity and/or might reect increased inter-particle interactions leading to an increased crystallite size.
SEM data (Fig. 7) clearly show two different types of particles interspersed within the lm. The majority of particles are ca. 25 nm in size and the EDX analysis (see ESI Fig. 8 †) shows the presence of Ti and O conrming these as P25 TiO 2 nanoparticles ( Fig. 7top). The data also show larger irregularly shaped agglomerations of particles which are 100-300 nm in size and the EDX data of these particles show the presence of Zn and O, conrming these to be the ZnO particles arising from the thermal decomposition of ZnO 2 . Interestingly, the surface morphology of particles appears different depending on the sintering temperature. Thus, aer sintering at 450 C, the surface appears smooth and the particles appear to be singular whilst the 300 C sintered particles are larger (120-600 nm), have a much more irregular surface and appear to be made up of many smaller particles. This suggests that, whilst the TGA data show that the loss of O 2 from the ZnO 2 occurs rapidly at ca. 200 C, the atomic rearrangement into a more ordered ZnO structure is not complete aer 30 min at 300 C.
Sorption data have been measured either using N 2(g) sorption data at À196 C tted to the BET model isotherm or using N719 dye solutions sorbed at 22, 40 or 50 C which have been tted to the model Langmuir or Freundlich isotherm. The dye uptake data have been measured using relatively low initial dye concentrations passively dyed and at equilibrium to study the effects of photo-electrode composition (AEZnO 2 ) and sintering conditions (300 vs. 450 C) rather than to optimise sensitization. Thus, fast dyeing 27 higher initial dye concentrations would be expected to show different responses but that is beyond the scope of this paper. P25/ZnO 2 lms were chosen as these devices show the best device responses compared to the P25 or other P25/peroxide lms ( Table 1). The N 2 sorption BET data show a surface area of 54 m 2 g À1 for P25 sintered a 300 C which drops to 45 m 2 g À1 aer sintering at 450 C. This is expected because one aim of raising the temperature to 450 C is to sinter particles together and create inter-particle necking. This can only be achieved if the TiO 2 surface atoms become mobile. If this occurs, then the surface tension will provide a driving force towards smoothing the surface to the lowest surface area. In this case, 450 C is not sufficient to complete this process but is sufficient to sinter the particles together. For the P25/ZnO 2 data, a similar trend is observed but with a surface area of 51 m 2 g À1 at 300 C and 42 m 2 g À1 at 450 C. The situation is complicated for these samples partly because they contain a mixture of TiO 2 particles along with a smaller number of ZnO particles which have resulted from the decomposition of ZnO 2 but also because the surface area measurement is an average across the whole. Looking at neat ZnO 2 powder rst, this has a surface area of 14 m 2 g À1 at RT which increases slightly to 18 m 2 g À1 aer sintering at 300 C but drops to 7.8 m 2 g À1 at 450 C. These trends are typical for a material such as ZnO 2 which releases gas during decomposition as this effectively bursts out of the material creating an increased surface area during the process. 41 However, the resultant ZnO particles sinter rapidly as the temperature is increased further to 450 C resulting in a subsequent loss of surface area. Thus, whilst the "average" surface area of P25/ZnO 2 might be expected to be slightly lower to reect the addition of lower surface area ZnO 2 particles, it is not possible to separate out what is happening to the surface of the TiO 2 particles alone. In addition, N 2 will physisorb to residual organic matter and also include this surface area in the data whilst N719 dye will only chemisorb to "free" metal oxide surfaces. Finally, N 2 is much smaller than N719 and so their sorption characteristics will be different. Hence, although the BET data are useful in highlighting trends between organic matter containing under-sintered samples and over-sintered samples (where the surface area begins to be lost), dye loading data have been measured and analysed in some detail to provide the most accurate picture of the metal oxide surface for dye binding.
The dye sorption data show relatively higher dye sorption at a low initial dye concentration. However, as the initial dye concentration increases, dye sorption increases but to a relatively lesser extent until the data reach a plateau (Fig. 8top). This is expected as there are a xed number of sorption sites in the metal oxide lms and, as these become increasingly lled fewer free sites remain, and greater dye concentrations are required to partition more dye on the surface. However, in these measurements, the dye concentration decreases throughout the experiment as the dye adsorbs until equilibrium is reached between the remaining free sites and the remaining dye concentration. Thus, if the dye concentration is proportional to the driving force for dye sorption and this drops throughout the sorption process, the initial dye concentration should be predictive of the nal dye loading.
Comparing the two isotherm models, the R 2 values are consistently higher when the data are tted to the Langmuir isotherm (R 2 > 0.997) compared to the Freundlich isotherm (R 2 0.962-0.983); see ESI Fig. 9-11 and ESI Tables 3-6. † The Langmuir model assumes monolayer sorption of adsorbates onto identical sites in separate sorption events whilst the Freundlich model assumes that adsorption takes place on heterogeneous surfaces and is not restricted to monolayer sorption. 33 On the basis of the higher correlations, the Langmuir data have been used to analyse dye uptake using P25 lms sintered at 450 C as a control to compare to P25/ZnO 2 lms sintered at either 450 or 300 C.
First comparing the data for P25 and P25/ZnO 2 lms sintered at 450 C, the values for adsorption capacity (q m ) are consistently higher for the P25/ZnO 2 lms which is in line with the higher J sc of these devices ( Table 2). This is despite the slightly lower surface area of the P25/ZnO 2 lms measured by N 2 sorption. As discussed previously, the N 2 BET data measure an "average" surface area of the entire sample surface (TiO 2 , ZnO and any residual organic matter). However, we expect the dye sorption data to only measure a monolayer of N719 dye chemisorbed to any available metal oxide surfaces. Thus, we believe the increased dye loading in P25/ZnO 2 samples reects the improved removal of residual organic matter from the metal oxide surfaces that improves the metal oxide surfaces for dye uptake. However, the device efficiency data (ESI Table 1 †) show that increasing the ZnO 2 loading in the lms from 5 to 15 to 25% reduces the device performance. This suggests that the increases in dye uptake are associated with improvements in the surface of TiO 2 particles. This is in line with earlier assertions that the main role of ZnO 2 is to release oxygen and heat to aid binder combustion which removes organic matter and increases the TiO 2 surface area for dye binding.
The data in Table 2 also show that adsorption capacity is higher for P25/ZnO 2 lms sintered at 300 C than P25 or P25/ ZnO 2 lms sintered at 450 C. This is to be expected based on the higher BET surface area of P25/ZnO 2 lms sintered at 300 C. However, as discussed previously, the situation is complicated because the BET data are an average of all the material data in the sample. The increased dye uptake for P25/ZnO 2 at 300 C suggests that the presence of ZnO 2 results in greater removal of residual organic matter from metal oxide surfaces (which in this sample must be dominated by the much more abundant TiO 2 particles). In addition, the higher BET surface area and increased dye loading aer 300 C sintering suggest that, as long as the residual binder can be removed at lower temperature, the metal oxide particles lose less surface area than if sintered at 450 C. However, this benet can only be realised with increased dye loading if a combustion agent such as ZnO 2 is added to help combust the residual binder. This benecial inuence on dye loading has not been realised before because it has not been previously been possible to sinter binder-containing pastes at low temperature. Table 2 also shows that the N719 adsorption capacity increases on all lms with dyeing temperature which suggests an endothermic adsorption process. In practice, solvent volatility limits the dyeing temperature and so these data have only been measured up to 50 C. Low R L values indicate favourable dye uptake. 30 The data show that R L values decrease with increasing dyeing temperature and also across the series P25 450 C > P25/ZnO 2 (450 C) > P25/ZnO 2 (300 C). Thus, P25/ZnO 2 lms sintered at 300 C and dyed at 50 C show the most favourable N719 uptake in line with the highest N719 adsorption capacity. This is in line with the J sc and h data which suggests that, as expected, J sc is linked to dye loading (Table 1).

Conclusions
The data show that, for the rst time, binder-containing pastes can be sintered at low temperature by using metal peroxide combustion promoters. This has the advantage that metal peroxides are low cost powders which are safe and easy to use with commonly used paste formulations. The most effective metal peroxide when using ethyl cellulose binder-containing pastes is ZnO 2 which is believed to be because ZnO 2 decomposes at similar temperatures to ethyl cellulose combustion so that the oxygen released during this process enhances binder combustion and removal as CO 2(g) at lower temperature. In addition, the by-products of ZnO 2 decomposition are relatively large particles of ZnO which may enhance light scattering within the device whilst not limiting the device performance because ZnO is used as a photo-anode material in DSC devices in its own right.
The sorption data show the importance of considering BET surface area data arising from multi-layer N 2 physi-sorption as the average of the whole sample surface area whilst dye loading data relate to chemisorbed dye monolayers. Furthermore, dye loading data aer sintering at lower temperatures show, for the rst time, that this can actually give rise to higher dye loadings which is linked to the removal of the residual binder at lower temperatures along with a reduction in the loss of metal oxide surface area associated with 450 C sintering. This suggests that sintering at 450 C actually slightly over-sinters TiO 2 resulting in a loss of surface area and lowered dye loading. However, it remains important to remove the vast majority (and ideally all) of the organic binder in order to maximize the number of dye binding sites and resulting J sc . Thus, a general model of sintering would be to sinter at the lowest possible temperature to remove the organic material, optimize dye loading and enable the formation of inter-particle connections to ensure the lms are mechanically robust.