Since the pioneer work of O'Regan and Grätzel in 1991, the dye-sensitized solar cell (DSSC) has been a competitive photovoltaic device due to its advantages of low cost, easy of manufacture and remarkable energy conversion efficiency (η) [1, 2]. To boost the total performance of DSSCs, many researchers have optimized components, such as photoanodes [3], electrolytes [4, 5], sensitizers [6, 7], and counter electrodes [8, 9]. As one of the core components, the photoanode undertakes dye loading and charge transport. Among these studied photoanode materials, TiO₂ nanoparticle photoanode has attracted widespread attention due to its eco-friendliness, high chemical stability, and high specific surface area [10]. However, this kind of photoanode film reveals lower light scattering ability due to its smaller particle size [11]. Furthermore, TiO₂ nanoparticle photoanode has disordered electron transport channels, thereby affecting the charge transfer and collection efficiency.

As is well known, TiO₂ hollow structures are more effective to improve light scattering ability and facilitate the electrolyte diffusion than solid ones, which result in a better DSSCs performance [12, 13]. The tailoring of TiO₂ hollow structures is a crucial means to improve the efficiency of DSSCs. Up to present, TiO₂ hollow structures with tailored textural properties are prepared through a variety of methods [14]. The most adopted method for the tailoring synthesis of hollow structure is a template method, including hard templates (mono-dispersed silica [15, 16] or polystyrene beads [17, 18]) or soft templates (surfactant micelles/vesicles [19], emulsion droplets [20] or gas bubbles [21]). For CaTiO₃ templates, Ca²⁺ is easily exchanged from the perovskite Ti-O frameworks in an acidic medium [22]. CaTiO₃ is chosen to act as self-template for the formation of TiO₂ via a topotactic synthetic method, which is in favor of tailoring desirable morphology. Due to the well-matched lattice spacings between CaTiO₃ and TiO₂, TiO₂ can perfectly inherit the CaTiO₃ morphology, guaranteeing a strong mechanical stability of TiO₂ structure.

In addition, the anatase/rutile-mixed phase TiO₂ is an effective photoanode material. The tailoring of the ratio of anatase-rutile is very important to enhance the performance of DSSCs. For the preparation of anatase/rutile-mixed phase TiO₂, two kinds of preparation methods are popular. One of the preparation methods is a mechanical mixing method, which cannot guarantee that the two phases are in close contact and the electron-hole pair separates effectively. The annealing treatment is another preparation method, which easily leads to particle aggregation. Therefore, for the tailoring of TiO₂ hollow structures, it is crucial to adopt a suitable preparation method concentrating various functions.

In this work, the waxberry-like mixed-phase TiO₂ hollow microstructures (WMTHMs) were fabricated via a topotactic synthetic method. WMTHMs are composed of radial TiO₂ nanorods. The ratio of rutile-anatase can be changed by adding FeCl₃. The optimized DSSCs of the assembled double-layer P25 + WMTHMs photoanode obtained an excellent η of 9.12%, which was superior to the DSSC of the pure P25 photoanode (8.12%).

Fig. 1 shows the FESEM, EDS mapping and TEM images of the as-prepared CaTiO₃-Fe_n precursors. As shown in Figs. 1a-c, the as-prepared CaTiO₃-Fe_n precursors are nearly spherical structure in shape and their surfaces are rather smooth. The individual spheres are mutually separated without aggregation. As the amount of FeCl₃ increases, the shape of the CaTiO₃-Fe_n spheres is not changed, but the diameter of CaTiO₃-Fe_n varies from 120 to 160 and 270 nm. The EDS mapping of CaTiO₃-Fe₀ (Fig. 1d) clearly reveals that the three elements (Ca, Ti, and O) are evenly distributed throughout the sample. The EDS mappings of CaTiO₃-Fe₁ (Fig. 1e) and CaTiO₃-Fe₂ (Fig. 1f) indicate that the four elements (Ca, Ti, O and Fe) are distributed over the entire samples. TEM results (Figs. 1g-i) confirm that all CaTiO₃-Fe_n precursors display a solid structure.

XRD patterns of the as-prepared CaTiO₃-Fe_n precursors are shown in Fig. S1a (Supporting information). XRD patterns of CaTiO₃-Fe₁ and CaTiO₃-Fe₂ are similar to that of CaTiO₃-Fe₀. The diffraction peaks (2θ = 23.2°, 33.1°, 38. 9°, 40.7°, 47.5°, 53.3°, 59.3°, 69.5° and 79.1°) correspond to the (101), (121), (211), (220), (202), (301), (042), (242) and (323) planes of perovskite-type CaTiO₃ (JCPDS No. 42-0423). Nevertheless, the Fe-containing species are not detected by XRD, which may be attributed to the low content in CaTiO₃-Fe₁ and CaTiO₃-Fe₂. Fig. S1b (Supporting information) shows the enlarged XRD curves of CaTiO₃-Fe_n in the range of 31-35°. With increasing Fe content, the (121) diffraction peak gradually shifts to a larger angle, verifying the distortion of CaTiO₃ crystal lattice by Fe dopant. It is well known that Ti⁴⁺ (0.75 Å) and Fe³⁺ (0.79 Å) have similar ionic radius, which may lead to the exchange of Ti⁴⁺ and Fe³⁺ in CaTiO₃ for the formation of new species [23]. Fe may also exist in Fe₂O₃ from the reaction of FeCl₃ during a solvothermal process. For CaTiO₃-Fe₁ and CaTiO₃-Fe₂, Fe may exist in Fe₂O₃ or the formation of new species.

The surface chemical states of CaTiO₃-Fe₁ and CaTiO₃-Fe₂ are characterized by XPS (Fig. S2 in Supporting information). Fig. S2a shows the survey spectra of CaTiO₃-Fe₁. For the XPS spectra of Ca 2p (Fig. S2b), the major peaks at 349.8 eV and 346.3 eV correspond to the binding energies of Ca 2p_1/2 and Ca 2p_3/2, respectively. For the XPS spectra of Ti 2p (Fig. S2c), the two major peaks, centered at around 458.1 eV and 463.8 eV, are ascribed to Ti 2p_3/2 and Ti 2p_1/2 of the Ti⁴⁺ states, respectively. For the XPS spectra of O 1s (Fig. S2d), the major peak centered at 529.5 eV can be attributed to the Ti-O band, which corresponds to the O²⁻ state in the lattice of CaTiO₃-Fe₁. For the XPS spectra of Fe 2p (Fig. S2e), the Fe 2p_3/2 peak at 710.2 eV corresponds to Fe³⁺ state [24]. In XPS spectra, there is no significant difference between CaTiO₃-Fe₁ (Figs. S2a-e) and CaTiO₃-Fe₂ (Figs. S2f-j). The results demonstrate that the Fe-containing species exists in CaTiO₃-Fe₁ and CaTiO₃-Fe₂.

Fig. 2 shows the FESEM, EDS mapping and TEM images of the as-prepared WMTHMs-n samples. As shown in Fig. 2a, the WMTHMs-0 is composed of a large quantity of uniform, rough spheres with a diameter of around 0.95-1.10 µm. The high-resolution FESEM (inset of Fig. 2a) displays a waxberry-like TiO₂ hollow sphere. Observingly, WMTHMs-0 consists of tiny short rods with dense state. With increasing the Fe content in the precursor, the diameter of WMTHMs-1 (Fig. 2b) increases to 1-1.30 µm, and the external structure of spheres has changed. The high-resolution FESEM (inset of Fig. 2b) displays the WMTHMs-1 is composed of nanorods with a diameter of about 70-80 nm, and the growth direction of these nanorods is nearly radially on the external surface of sphere. Moreover, the hollow interior is clearly observed from some broken spheres, which is favorable for enhancing the light scattering ability. With further increasing the Fe content in the precursor, as shown in Fig. 2c, WMTHMs-2 shows the same external structure as WMTHMs-1. WMTHMs-2 with a diameter of 1-1.20 µm is also composed of nanorods with a diameter of about 45-55 nm. The EDS mapping images (Figs. 2d-f) indicate that Ti and O elements are distributed homogeneously in three kinds of WMTHMs. The as-prepared WMTHMs-n samples are further characterized by TEM. In Fig. 2g, WMTHMs-0 consists of tiny nanorods and displays a hollow structure. In the case of little amount of FeCl₃ added, WMTHMs-1 (Fig. 2h) and WMTHMs-2 (Fig. 2i) samples are composed of radially arranged nanorods and display a hollow structure. Ca²⁺ ions are exchanged with H⁺ ions in the presence of Na₂EDTA. The out-diffusion rate of Ca²⁺ ions is much faster than the in-diffusion rate of H⁺ ions, which results in the formation of hollow structure [25]. TiO₂ evolves from CaTiO₃ by a dissolution-recrystallization process. Meanwhile, Fe-containing species are removed with the assistance of Na₂EDTA. For WMTHMs-n, Figs. 2j-l demonstrate that the lattice spacings of 0.32 nm and 0.35 nm correspond to the (110) plane of rutile and the (101) plane of anatase TiO₂, respectively.

The composition of WMTHMs-n is confirmed by XRD (Fig. 3a). The diffraction peaks at 2θ = 37.9° and 48.2° are indexed to the (004), (200) crystal planes of anatase TiO₂ (JCPDS No. 73-1764). Other peaks at 2θ = 36.1°, 41.2°, 54.3°, 56.6°, 62.8°, 69.8° and 72.4°, are indexed to the (101), (111), (211), (220), (002), (112) and (311) crystal planes of rutile TiO₂ (JCPDS No. 78-2485). For all the as-prepared WMTHMs-n, the coexisting anatase and rutile phases are in agreement with TEM. As expected, the anatase/rutile mixed TiO₂ has excellent charge separation efficiency [26, 27]. The ratio of rutile-anatase can be calculated by the following Eqs. 1 and 2 [28]:

(1)

(2)

here, W_r, W_a and A_r represent the rutile content, the anatase content, the integrated intensity of anatase (101) peak, and the integrated intensity of rutile (110) peak, respectively. The ratio of anatase-rutile is listed in Table S1 (Supporting information). The result indicates that the ratio of anatase-rutile is affected by the Fe content of the precursor. With the increase of Fe content in the precursor, the rutile content increases. That is to say, the ratio of anatase-rutile in WMTHMs-n can be successfully controlled by simply adjusting the amount of FeCl₃. XRD patterns of the WMTHMs-1 and WMTHMs-2 show no traces of other phases like Fe₂O₃ or Fe_xTiO_y. It is believed that the ion-exchange reaction may occur between Fe-containing species and H⁺ with the assistance of Na₂EDTA.

As we know, both Ti⁴⁺ ion (0.75 Å) and Fe³⁺ ion (0.79 Å) have similar ion radius, which indicates that Fe³⁺ ions may enter CaTiO₃ lattice to replace Ti⁴⁺ sites [23]. Based on simple charge compensation grounds, the formation of oxygen vacancies would generate due to the substitutional incorporation of Fe³⁺, which can offer space for the atomic arrangement [29]. Besides, the oxygen vacancies increase with increasing Fe content, so the mass fraction of rutile phase increases with increasing Fe content [29]. During transformation CaTiO₃-Fe into TiO₂ through a Na₂EDTA-assisted ion-exchange process, Fe content can affect the ratio of anatase-rutile in WMTHMs.

The co-existence of anatase and rutile phases in the WMTHMs-1 and WMTHMs-2 is further characterized by Raman spectroscopy (Fig. S3 in Supporting information). The peaks appeared at 147 cm⁻¹ (E_g), 397 cm⁻¹ (B_1g), 512 cm⁻¹ (A_1g) and 639 cm⁻¹ (E_g) are well matched with the anatase TiO₂ modes. The three peaks of rutile centered at 239, 445 and 614 cm⁻¹ are attributed to multi-proton process, E_g and A_1g of the rutile modes [30]. It clearly indicates that both WMTHMs-1 and WMTHMs-2 are made of anatase and rutile phases, without other phases like Fe₂O₃ or Fe_xTiO_y, which is also in agreement with XRD results.

The surface characterization of WMTHMs-1 and WMTHMs-2 are performed by XPS. As shown in Fig. S4a (Supporting information), the XPS survey spectra of WMTHMs-1 show the existence of Ti and O elements and the absence of Fe³⁺. In Fig. S4b (Supporting information), the two peaks, presented at around 463.8 eV and 458.1 eV, are attributed to Ti 2p_1/2 and Ti 2p_3/2 of the dominant Ti⁴⁺ state, respectively [31]. In Fig. S4c (Supporting information), the peak at 529.4 eV is ascribed to the lattice oxygen in TiO₂, indicating the presence of an oxygen environment. In XPS spectra, there is no significant difference between WMTHMs-2 (Figs. S4d-f in Supporting information) and WMTHMs-1 (Figs. S4a-c in Supporting information).

The BET specific surface area and pore size distribution of WMTHMs-n samples are investigated by N₂ adsorption-desorption measurement. In Fig. 3b, all N₂ adsorption-desorption isotherms show typical Ⅳ-type isotherms with an H2 hysteresis loop according to IUPAC classification, which are characteristic of mesoporous materials [32, 33]. The WMTHMs-0 and WMTHMs-1 samples exhibit ideal specific surface area of 64.76 m²/g and 68.05 m²/g, respectively, which are higher than WMTHMs-2 (43.41 m²/g) and the commercial P25 (~50 m²/g). Therefore, the WMTHMs-0 and WMTHMs-1 samples can facilitate the dye adsorption of DSSCs. Fig. 3c reveals the corresponding pore size distribution of WMTHMs-n samples. With the increase of Fe content in the precursor, the pore diameter has a decrease. The pore size of the as-prepared WMTHMs-0, WMTHMs-1 and WMTHMs-2 are 19.2, 15.2, and 12.6 nm, respectively. The relatively wide pore size distributions are beneficial to electrolyte diffusion. This result indicates the Fe content of the precursor exerts a great effect on the textural structure.

The J-V curves are displayed in Fig. 4a and their corresponding photovoltaic characteristics (J_sc: short-circuit current density, V_oc: open-circuit voltage, FF: fill factor, and η) are summarized in Table 1. From Table 1, the four cells have a similar FF, while both J_sc and η exhibit an increase and then decrease with the increasing Fe content in the precursor. The cells based on WMTHMs-0 and WMTHMs-1 scattering layer photoanodes have achieved higher η (8.38% and 9.12%) than that of the Cell-P25 without scattering layer (8.12%). However, introducing an additional scattering layer of WMTHMs-2, the η of Cell-P25 + WMTHMs-2 decrease to 7.96%. The η initially increases from 8.38% to 9.12% with the increasing Fe content in the precursor, which is attributed to the synergistic effects of the appropriate ratio of anatase-rutile, the convenient one-dimensional electron transport channel, the superior light scattering ability and the higher dye loading amounts. The mixed-phase TiO₂ can form a synergistic effect between anatase and rutile crystals. The photo-generated electrons coming from N719 inject into the rutile phase TiO₂ and then migrate to the anatase phase TiO₂. Such staggered bandgap can suppress charge recombination, thereby improving the efficiency of DSSCs. However, when the Fe content of the precursor is further increased, the η decreases from 9.12% to 7.96%, which may be due to the unsuited ratio of anatase-rutile, the inferior light scattering capacity and the lower dye loading amounts. Compared with CaTiO₃-Fe_n light scattering layer, the DSSCs based on the WMTHMs-n show higher efficiency, as shown in Fig. S5 and Table S2 (Supporting information).

It is well known that the J_sc is closely related to the light scattering capacity and the amounts of dye loading. To study the effect of light scattering capacity on the J_sc, the diffuse reflection property of the as-prepared photoanode films without N719 dye loading is investigated. In Fig. 4b, the diffuse reflection property of P25 + WMTHMs-0, P25 + WMTHMs-1 and P25 + WMTHMs-2 films is obviously higher than that of the pure P25 film, which is expected to exhibit a better light scattering capability of the WMTHMs-n layer. For the WMTHMs-n, besides that the comparable sphere size to the wavelength of visible light plays an important role in enhancing light scattering, their hollow structure can confine the incident light within the photoanode by light refraction between the wall and the air-filled pore, as shown in Fig. S6 (Supporting information).

The amounts of N719 dye anchored on the four films are investigated using 0.1 mol/L NaOH solution. UV-vis absorption spectra of dye molecules are shown in Fig. 4c. The order of dye loading capability is: Cell-P25 + WMTHMs-1 (16.57 × 10⁻⁸ mol/cm²) > Cell-P25 + WMTHMs-0 (15.25 × 10⁻⁸ mol/cm²) > Cell-P25 (13.53 × 10⁻⁸ mol/cm²) > Cell-P25 + WMTHMs-2 (12.27 × 10⁻⁸ mol/cm²), which are closely related to their BET specific surface areas and ratio of anatase-rutile. Due to the higher light scattering capacity and bigger dye loading capability, Cell-P25 + WMTHMs-1 demonstrates an optimum J_sc.

The differences in J_sc of the four cells are further investigated by measuring EQE spectra. Fig. 4d shows the EQE spectra as a function of wavelength for the four cells. The high EQE values in the short wavelength range are mainly attributed to the higher dye loading capability, and the somewhat higher EQE values in the long wavelength range of 600-750 nm are ascribed to the more efficient light scattering of the WMTHMs-n layer. Considering that dye loading capability and the diffuse reflection, Cell-P25 + WMTHMs-1 possesses higher EQE values over a wide range than Cell-P25. However, Cell-P25 + WMTHMs-1 shows lower EQE values than Cell-P25 in the short wavelength range, whereas it is just the opposite in the long wavelength range.

The dark J-V characteristics reflects the recombination of injected electrons with I₃⁻. In Fig. 4e, the dark current densities of the Cell-P25 + WMTHMs-n are 4.09, 5.51 and 2.35 mA/cm² at 0.80 V, respectively. On the contrary, the P25 + WMTHMs-2 electrode displays a low current density, indicating the low recombination loss. The relatively high dark current density of P25 + WMTHMs-1 electrode means a high interface recombination loss, which could be explained by the enlarged recombination. The high recombination loss leads to a decrease in V_oc, which is agreement with the V_oc value.

To further make out the electron transport and interface recombination process in the four cells, EIS of Cell-P25 + WMTHMs-n are measured (Fig. 4f). In Fig. 4f, the smaller semicircle is assigned to the charge transfer resistance (R_ct1) at the electrolyte/Pt counter electrode, while the larger semicircle is assigned to the charge recombination resistance (R_ct2) at the TiO₂/dye/electrolyte interfaces. The C_µ represents the chemical capacitance. The data are analyzed using Z-view software with an equivalent circuit (inset of Fig. 4f). The EIS parameters are listed in Table 1. The R_s and R_ct1 values are almost identical because of the similar substrate, counter electrode, and I⁻/I₃⁻ electrolyte. From Table 1, the R_ct2 of the four cells are distinguishing, and the order of R_ct2 value is: Cell-P25 + WMTHMs-2 (95.38 Ω) > Cell-P25 (69.19 Ω) > Cell-P25 + WMTHMs-0 (63.12 Ω) > Cell-P25 + WMTHMs-1 (61.36 Ω). Based on the fitted R_ct2 and C_µ2, the electron lifetime (τ_e = R_ct2 × C_µ2) values are calculated to be 58.23, 55.16, 89.45 and 65.13 ms for Cell-P25 + WMTHMs-0, Cell-P25 + WMTHMs-1, Cell-P25 + WMTHMs-2 and Cell-P25, respectively. Obviously, the Cell-P25 + WMTHMs-1 shows the shortest τ_e, manifesting that Cell-P25 + WMTHMs-1 has the quickest electron recombination process. This may be attributed to the large surface area of WMTHMs-1 introduced many defects. This leads to rapid recombination of electrons at the photoanode/dye/electrolyte interface, which in turn reflects a shortening τ_e for Cell-P25 + WMTHMs-1 [34]. The shorter τ_e for the Cell-P25 + WMTHMs-1 also further supports its lower V_oc. The undesired V_oc for P25 + WMTHMs-1 electrode can be compensated by the high J_sc due to the efficient light scattering and superior dye loading capability, making Cell-P25 + WMTHMs-1 endow a marked efficiency.

In this study, the anatase/rutile mixed TiO₂ hollow spheres were fabricated via a topotactic synthetic method using monodispersed CaTiO₃ precursor templates. The ratio of anatase-rutile was controlled through adding FeCl₃ content. WMTHMs-1 sample exhibited the best light scattering property and highest specific surface area. A double-layer photoanode consisting of WMTHMs-n used as light scattering layer and P25 as underlayer was designed. The DSSCs of the assembled double-layer P25 + WMTHMs-n photoanodes obtained a η of 8.38%, 9.12% and 7.96%, respectively. A maximum η of 9.12% was achieved by using the P25 + WMTHMs-1 bilayer photoanode, showing a marked improvement compared with the pure P25 photoanode (8.12%). The improvement of the efficiency was mainly attributed to the structure of WMTHMs-1, which can provide multiple scattering centers to enhance the light harvesting ability, the direct pathways for fast electron transfer, the appropriate ratio of anatase-rutile for quick charge separation, the staggered bandgap structure, and the high specific surface area for adsorbing dye.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Acknowledgements

This work is financially supported by the National Natural Science Foundation of China (No. 21965013), the Natural Science Foundation of Hainan Province (No. 220RC590), and the Graduate Student Research and Innovation Program of Hainan Province (No. hsyx2019-17).

Supplementary materials

Supplementary material associated with this article can be found, in the online version, at doi:10.1016/j.cclet.2021.07.032.