Comparison of electron injection and recombination on TiO2 nanoparticles and ZnO nanorods photosensitized by phthalocyanine

Titanium dioxide (TiO2) and zinc oxide (ZnO) semiconductors have similar band gap positions but TiO2 performs better as an anode material in dye-sensitized solar cell applications. We compared two electrodes made of TiO2 nanoparticles and ZnO nanorods sensitized by an aggregation-protected phthalocyanine derivative using ultrafast transient absorption spectroscopy. In agreement with previous studies, the primary electron injection is two times faster on TiO2, but contrary to the previous results the charge recombination is slower on ZnO. The latter could be due to morphology differences and the ability of the injected electrons to travel much further from the sensitizer cation in ZnO nanorods.


Introduction
Titanium dioxide (TiO 2 ) is the most widely used anode material in dye-sensitized solar cell (DSSC) applications [1], and the highest 2018 The Authors. Published by the Royal Society under the terms of the Creative Commons Attribution License http://creativecommons.org/licenses/by/4.0/, which permits unrestricted use, provided the original author and source are credited.

Samples
Previously developed and tested procedures were used to prepare nanostructures [15,[19][20][21]. In brief, TiO 2 nanoparticle (anatase, average size 20 nm) films were prepared by spin-coating on FTO glasses pre-coated by a thin TiO 2 compact layer [19,20]. The thickness of the TiO 2 nanoparticle layer was determined by scanning electron microscope (SEM) (   photo-anodes were prepared using a hydro-thermal method on indium-tin oxide (ITO)-coated glasses following the procedure described previously [15,18]. The typical length of the nanorods was 800-1000 nm and diameters were around 50 nm (figure 2b). For the studied samples, the specific surface area of the ZnO nanorod samples was estimated to be approximately 10 times smaller than that of the TiO 2 samples. Self-assembled monolayers (SAMs) of Pc were deposited by dissolving Pc in a mixture of BuOH : MeCN 1 : 1 (by volume) at a concentration of 0.03-0.1 mM, immersing the substrates into solution for 30 min and washing away the phys-adsorbed excessive Pc with pure solvent, as described elsewhere [18].

Instruments
Absorption spectra of the samples were measured using a Shimadzu UV-3600 UV-VIS-NIR spectrophotometer. The sample morphology was investigated using a field emission SEM (FE-SEM; Carl Zeiss Ultra 55). Ultrafast transient absorption (TA) responses of the samples were measured using a pump-probe system described previously [20,22] Figure 3. Absorption spectra of TiO 2 |Pc and ZnO|Pc samples after subtracting substrate spectra, and Pc in mixture of BuOH : MeCN. The spectrum of ZnO|Pc sample was multiplied by eight to match the scale.
Topas C, Light Conversion Ltd). White continuum probe pulses were generated by a small fraction of fundamental pulses produced by the generator (Libra F) focused on a sapphire crystal. The timeresolved transient absorption spectra were recorded using an ExciPro TA spectrometer (CDP, Inc.) in two wavelength ranges: 460-770 nm and 850-1050 nm. Home-developed software (decfit.py) was used to process and analyse the pump-probe measurements. The program carries out group velocity dispersion compensation, convolution with an instrument response function and a global data fit to a sum of exponential, stretched exponential and distributed decay functions [23].

Samples
Absorption spectra of the Pc-sensitized samples are presented in figure 3 together with the solution spectrum. The absorption spectra of the substrates, TiO 2 nanoparticle film and ZnO nanorod array were subtracted; therefore, the spectra shown represent absorption of Pc SAMs. Owing to the thickness and morphology differences, the TiO 2 samples had an approximately 10 times higher surface specific area than the ZnO samples [15]. Therefore, to compare Pc spectra in TiO 2 and ZnO, the spectrum of the ZnO|Pc sample was multiplied by 8. The resulting absorption spectra of the Pc SAMs on both ZnO and TiO 2 are reasonably close to each other in shape, and show no evidence of Pc aggregation. One may notice a red shift (by roughly 8 nm) and some differences in the shapes of the spectra in the Q-band area (650-710 nm region) of the SAMs compared with the solution. However, the band width is the same 37 nm (full width half-maximum) in both cases, which indicates no aggregation effect on the spectra of the SAMs.

Transient absorption spectroscopy
The photoinduced reactions in the samples were studied by measuring the transient absorption responses (pump-probe method) in two wavelength ranges, 460-770 nm and 850-1050 nm, with excitation at 695 nm. The time-resolved differential absorption spectra of TiO 2 |Pc and ZnO|Pc samples at selected delay times are presented in figure 4. At a qualitative level, the spectra at a short delay time, e.g. at 0.2 ps, can be attributed to the singlet excited state of Pc formed immediately after the excitation, and at a longer delay time, e.g. at 1 ns, they can be attributed to the Pc cation [24]. The responses of two samples in figure 4 have much in common in terms of spectral shapes, though lifetimes and relationships between different spectral features are clearly different.
The transient absorption response of the ZnO|Pc sample can be fitted reasonably well by a sum of three exponential and one distributed decay functions as discussed in [18]. The resulting decayassociated component spectra are presented in figure 5b. These are spectra of the transient absorption responses associated with corresponding fit functions, exponential and distributed decays denoted as 'exp(. . .)' and 'dist(. . .)', respectively. To obtain an equally good fit of the TiO 2 |Pc data, a combination of two exponents and two distributed decays had to be used (figure 5a). The spectral shapes are highly

Long-lived charge separation
In both cases, the longest-lived components (figure 5) have decay times much longer than the maximum delay time available to the instrument (6 ns); therefore, the calculated decay time values are not accurate. However, the spectra of the components are very much alike and show a relatively narrow absorption band at 865 nm and a broad absorption band around 560 nm in addition to strong bleaching of the Qband around 700 nm. These features are typical for the Pc cation, Pc + [18,24,25]. The relative intensities of the long-lived components are rather different in the two samples, being much stronger in the ZnO|Pc sample. is caused by electron injection to TiO 2 ,

Electron injection
The distributed decay model operates with three parameters [23]: the average (or non-disturbed) time constant (τ inj ) indicated for the corresponding decay components in figure 5, the width of the free and/or reorganization energy distribution ( E), and the sensitivity factor (a). The time constant is approximately two times larger for the ZnO|Pc sample, but the other two parameters are nearly identical for the two samples, E = 0.022 ± 0.004 eV and a = 1.0 ± 0.2 for TiO 2 |Pc and E = 0.028 ± 0.007 eV and a = 1.2 ± 0.3 for ZnO|Pc. This means that the electron injection is two times faster in the TiO 2 |Pc sample, but otherwise the mechanism of electron transfer is the same. There is a fast 0.5 ps component in the response of both samples. It has an effect of a minor absorption decay in the visible part of the spectrum and some reshaping around 850 nm, but it does not lead to the formation of an intermediate state with new distinct spectral properties ( figure 4). This component is tentatively attributed to the thermal relaxation of the singlet excited state and will not be discussed further.

Fast charge recombination at the interface
The last component to discuss is that of the 180 ps time constant in the case of the ZnO|Pc sample and the 23 ps average time constant (distributed decay component) for the TiO 2 |Pc sample. The spectral shapes of this component are rather close for the two samples but the relative intensities are different. It is a minor component for the ZnO|Pc sample and one of the strongest for the TiO 2 |Pc sample. Furthermore, the spectra of these components resemble those of the longest-lived component, with differences mainly seen in the shape of the broad band in the 500-600 nm range and an absorption shoulder (TiO 2 |Pc) or a minor band (ZnO|Pc) in the 950-1050 nm range. The latter can be associated with a Pc anion, which has two characteristic bands: one is close to 1000 nm in the NIR and the other is at 590 nm in the visible part of the spectrum, as was observed for ZnO|Pc covered with a hole-transporting material (electron donor) [18]. The appearance of a Pc anion could be explained by the partial aggregation of the Pc molecules and an intra-aggregate charge separation in the process Pc * − Pc → Pc + − Pc − . This could be a reasonable explanation in the case of the ZnO|Pc sample, though for the ZnO|Pc sample it is a minor process in any case. However, in the case of the TiO 2 |Pc sample, there is no band at 1000 nm and the component shape in the 500-620 nm range does not match that of the Pc anion, having a broad band in the 500-600 nm range rather than a relatively narrow anion band at 590 nm. It is more reasonable to attribute this component to a Pc cation, which leaves a question on the origin of two Pc cations with drastically different lifetimes, 23 ps and 23 ns (TiO 2 |Pc) and minor differences in the spectral shapes.
A typical interpretation of a very diverse decay time of one and the same state is inhomogeneity of the Pc arrangement on the TiO 2 surface. However, this can also be ruled out for two reasons. Firstly, the bulky peripheral groups and rigid core of Pc makes a diverse conformation distribution unlikely. Secondly, the same diversity could be expected for both semiconductor substrates, but in the case of ZnO the fast charge recombination (180 ps) is a minor process, whereas in the case of TiO 2 it is almost the dominating process.

Interfacial charge transfer complex
There are reports on a two-step charge separation at the semiconductor-organic interface [9,11,12,14]. In the first step, a coupled electron-hole pair is formed with the electron in the CB but electrostatically bound to the sensitizer cation, Pc + . This state is also termed the interfacial charge-transfer complex (ICTC) [13]. It may recombine or the electron may leave the interface and become a 'free' carrier in the CB. In the spectral range studied, the measured transient absorption response is dominated by the response of the Pc sensitizer, a cation in both the coupled and non-coupled Pc + states. Therefore, electron migration from the close proximity of the Pc + to the bulk of the semiconductor is expected to have only a minor effect on the transient absorption spectrum shape, which is consistent with the minor spectral differences of the two long-lived components in figure 5.
Taking this two-step charge transfer as a working hypothesis, one could conclude that there is a relatively large probability of charge recombination of the ICTC in the case of TiO 2 |Pc, as the molar absorption coefficients of the 'free' cation Pc + and cation in the ICTC must be almost the same. The gradual decay of the signal at 865 nm within a few tens of picoseconds (figure 6) points to the complex recombination but not to the electron shift to the bulk of TiO 2 . However, this conclusion contradicts previous studies. First of all, terahertz spectroscopy of similar TiO 2 samples suggests that the ICTC dissociates by yielding free electrons in the CB within 1 ps [12,14], much shorter than the 23 ps time constant discussed here. On the contrary, for ZnO samples, this dissociation time is extended to 100 ps and agrees reasonably well with the 180 ps reported here. Secondly, Chegui and co-workers [13] have shown that the fates of the injected electron are very different in TiO 2 and ZnO due to the different electronic band structures, a direct band gap in ZnO and an indirect one in TiO 2 , and ICTC is formed at the ZnO|Pc interface but not at the TiO 2 |Pc interface. Therefore, our results are consistent with an intermediate ICTC complex in the ZnO|Pc sample only. The 23 ps time constant corresponds to an electron diffusion length of a few tens of nanometres in bulk TiO 2 , roughly the size of a nanoparticle. This leads us to the assumption that the fast charge recombination is associated with the probability of the electron remaining in the same nanoparticle into which it was injected. The free electron and Pc + cannot recombine as soon as the electron jumps to another nanoparticle, and 23 ps is an average time for the free electron to move from one TiO 2 nanoparticle to another. The characteristic size, the length, of ZnO nanorods is much larger and the carrier mobility is much higher in ZnO, therefore the same fast recombination mechanism in not efficient in ZnO| samples since free electrons move quickly away from the recombination centre, Pc + . This assumption has no direct proof, however.

Dynamics of the Pc cation on TiO 2 and ZnO surfaces
The band at 865 nm is a characteristic feature of the Pc cation. Unfortunately, the singlet excited state of Pc also has relatively high absorption at this wavelength, which is the reason why the electron injection (reaction (4.1)) does not lead to an absorption rise at this wavelength in the first few picoseconds of electron injection and cation formation. Nevertheless, at relatively long delay times when the singlet state has already relaxed, e.g. >10 ps, the signal at 865 nm can be used as an indicator of the [Pc + ] population. Comparison of the transient absorption decays at this wavelength is shown in figure 6. Obviously, the charge recombination at the TiO 2 |Pc interface is faster than that at the ZnO|Pc interface, and the most drastic drop of the signal at 865 nm (TiO 2 |Pc sample) takes place within the first 10 ps after the excitation.
A very simplified reaction scheme for both samples is Pc Pc To model the process dynamics, the fit results at 865 nm will be used. Then, we will assume that: (i) the picosecond component is solely due to the relaxation of the singlet excited state, Pc*; thus the average time constants for the singlet state relaxation are τ inj = 2.4 ± 0.2 ps and 5.4 ± 0.7 ps for the TiO 2 |Pc and ZnO|Pc samples, respectively; (ii) only the excited state, Pc*, and cation, Pc + , contribute to the transient absorption at this wavelength, and absorption of two cations in scheme (4.2) is the same; (iii) immediately after excitation only Pc* is formed, which means that the time resolution of the instrument is sufficient to resolve the cation formation and it has zero population at zero time.
Within this model, the population decay of the excited state is where the function dist() is computed as presented in [23], and it depends on the 'non-disturbed' or average time constant, τ inj , the energy distribution width, E, and the sensitivity factor, a. The population decay of the singlet excited state is shown in figure 7 by the red lines. The time profile of the total cation population ('all Pc + ' in figure 7, blue lines) can be computed as the difference between the normalized decays at 865 nm ( A λ=865 nm (t)) and [Pc * ](t), though this does not give an 'absolute' value of the Pc + population since the ratio of molar absorbances of Pc + and Pc* is not known. In other words, the sum 'Pc*' and 'all Pc + ' in figure 7 is the normalized measured decay at 865 nm, or the fits shown in figure 6.
Next, the proportion in the branching reactions of the e − | Pc + state can be evaluated from the relative intensities of the decay components with the time constants 23 ps and 23 ns (in the case of the TiO 2 |Pc sample), which gives τ int = τ c a int a l + a int (4.4) and τ l = τ c a l a l + a int , (4.5) where τ c is the measured time constant of the middle decay component (23 ps for TiO 2 ). The green lines in figure 7 show cation populations minus the population of the e − | Pc + states; figure 7 also presents the dynamics of the 'long-lived' cations. The most reliable part here is the separation of the singlet, Pc*, and cation, Pc + , as it is based on sharp spectral differences between the two states. Division on the 'short'-and 'long'-lived Pc cations (states e − |Pc + and Pc + in scheme (4.2)) is rather phenomenological, but it reflects the experimental observation that the cation Pc + decay has fast, 23 ps, and slow, 23 ns, components with approximately two-thirds accumulated in the fast decay in the case of the TiO 2 |Pc sample. This is not the case for the ZnO|Pc sample, for which the fast decay is the minor component. This model suggests that the relative population of Pc + is higher in the ZnO|Pc sample at delays longer than 20 ps. Although it has to be noted that in complete solar cells the photon-to-current conversion efficiency depends also on how fast the holes from the sensitizer are transferred to the electrolyte in liquid cells or hole-transporting material in solid-state cells. If this process is sufficiently fast, taking not more than a few tens of picoseconds, both systems, TiO 2 |Pc and ZnO|Pc, are expected to be equally efficient. This conclusion is in line with the comparative study of Chandiran et al. [5]. Figure 7 summarizes the discussion of the differences between the TiO 2 nanoparticles (anatase) and ZnO nanorods (wurtzite) as the electron-collecting materials photosensitized by a Pc. The experimental results suggest that the primary electron injection is faster in the TiO 2 sample, in agreement with numerous previous observations. But the charge recombination is also faster in TiO 2 . Furthermore, a relatively large number of the injected electrons recombine with the Pc cation on the semiconductor surface within a few tens of picoseconds in the case of TiO 2 , but not in ZnO. Overall, the yield of Pc cations and thus the yield of the electrons injected into the anode is higher in ZnO at delays longer than 20 ps with the difference increasing with the delay time. This advantage of ZnO nanorods may be due to two factors: higher electron mobility in ZnO than in TiO 2 , and a longer travelling distance in ZnO nanocrystals and nanorods, which helps to separate electrons and surface cations, Pc + , more efficiently in the case of ZnO.

Conclusion
Data accessibility. This article has no additional data.