Introduction

Converting solar energy to hydrogen by means of photocatalysis is one of the promising ways to solve energy and environmental issues1,2,3,4,5. Although numerous photocatalysts have been searched and designed for overall water splitting in the past decades, the conversion efficiency of solar energy to hydrogen is still low due to the strong recombination of photogenerated carriers. Recently, ferroelectric materials (e.g., PbTiO3, BiFeO3, Na0.5Bi0.5TiO3, and Bi3TiNbO9) with built-in electric field that could facilitate photogenerated carrier separation have attracted wide attention in the field of photocatalysis6,7,8,9. Among them, Bi3TiNbO9 is an Aurivillius-type layered ferroelectric photocatalyst, which possesses a structural distortion induced depolarization field along the in-plane a-axis arising from the dipole interaction between bismuth in the (Bi2O2)2+ layer and oxygen in the (BiTiNbO7)2- perovskite layer (Fig. 1a)10,11. Previous studies of Bi3TiNbO9 mainly focused on the effect of depolarization field on the photogenerated carrier separation and defect engineering for surface active site modification. These synthesized Bi3TiNbO9 based photocatalysts have demonstrated good potential in photocatalytic water reduction or oxidation with the assistance of sacrificial reagents and CO2 reduction12,13,14. In our latest work, Bi3TiNbO9 has revealed efficient and stable photocatalytic overall water splitting by selective exposure of a robust (BiTiNbO7)2- perovskite layer for effective transfer of electron to the cocatalyst, affirming the great promise of Bi3TiNbO9 as a photocatalyst for solar energy to hydrogen conversion15.

Fig. 1: Crystal structure along ac plane.
figure 1

a Bi3TiNbO9. b Bi3TiNbO9-W (Ps-ferroelectric polarization, E-additional electric field).

Full size image

The (Bi2O2)2+ and (BiTiNbO7)2- layers in Bi3TiNbO9 stack along the c-axis direction, resulting in a tendency to expose the {001} facet as the basal surface and the formation of a nanosheet morphology. The photogenerated carriers exhibit an anisotropic migration property in the Bi3TiNbO9 nanosheet photocatalysts, where the photogenerated electrons usually reach the basal {001} facet across the interlayer, but the photogenerated holes prefer to reach the lateral {110} facet within the intralayer13,16. Due to the large interlayer barrier, the photogenerated electrons migrate slowly along the c-axis, which increases the recombination probability of the photogenerated carriers17. Although reducing the thickness of the layered material along the c-axis to obtain an ultrathin structure18 or selective exposure of different layers can shorten the migration distance of photogenerated electrons from the bulk to surface15, to some extent, the nature of poor interlayer charge transport remains unchanged. Therefore, the problem of large interlayer charge transport barrier represents an intrinsic bottleneck in designing efficient layered Bi3TiNbO9 photocatalysts for overall water splitting.

Here, we introduced an additional built-in electric field, perpendicular to the depolarization field in Bi3TiNbO9 nanosheets, by gradient tungsten (W) element doping induced energy band structure adjustment between surface and bulk to break such a bottleneck. Typically, donor dopants can effectively increase the number of free electrons, thus raising the corresponding Fermi level of semiconductors19,20,21,22,23. The concentration of this donor dopant W6+ shows a gradient decrease from surface to bulk along the c-axis (Fig. 1b), giving rise to a doping-depth-dependent variation in the gap between the Fermi level and the conduction band minimum (CBM) from surface to bulk and thus a formation of additional built-in electric field along the c-axis. This introduced electric field could be used for conquering the potential barrier between the layers of (Bi2O2)2+ and (BiTiNbO7)2- for the photogenerated electrons, motivating the photogenerated electrons to the basal {001} facet. Moreover, the W dopant could strengthen the structural distortion, resulting in an enhanced depolarization field along the a-axis for the anisotropic flow of the photogenerated carriers. Combined with selective exposure of the robust (BiTiNbO7)2- perovskite layer, W6+ doped Bi3TiNbO9 exhibits efficient and stable photocatalytic overall water splitting, which is 10.4 times higher than that of the original Bi3TiNbO9. Our work systematically studied the regulation of the crystal structure and electronic structure arising from element doping in Bi3TiNbO9, providing a reference for the application of Aurivillius compound in photocatalytic overall water splitting.

Results

Photocatalysts characterization

W6+ ion doped Bi3TiNbO9 samples were prepared by a conventional flux method24. Due to the layered structure and the influence of anions in molten salt, all the Bi3TiNbO9-x W (x = 0, 1%, 3%, 5%, 7%, 12% in molar ratio) samples exhibit a typical nanosheet morphology with lateral width of ca. 1 μm and thickness of ca. 50 nm (Supplementary Fig. 1), indicating a trivial effect of W doping on the morphology. The inference of the crystal structure was confirmed by the X-ray diffraction (XRD) characterization. As shown in Supplementary Fig. 2a, the diffraction patterns of all samples are basically the same as the standard card (PDF#39-0233, space group: A21am) of Bi3TiNbO9. Note that there is a gradual variation in the relative intensity of the diffraction peaks at 14 and 24°, which are indexed as the (001) and (110) planes of Bi3TiNbO9, respectively. As analyzed in Supplementary Fig. 2b, the ratio of (001)/(110) shows a volcanic curve with the increase of W6+ ion doping amount. Raman spectra were obtained with 532 nm wavelength excitation to detect the bonding situation and the structure of Bi3TiNbO9-x W (x = 0, 1%, 3%, 5%, 7%, 12%) nanosheets (Supplementary Fig. 3). The Eg modes involving the stretching mode of diagonal oxygen ions on the ab plane of the BO6 octahedron split into B2g at 533 cm-1 and B3g at 573 cm-1, which can be rationalized by the orthogonal distortion of Bi3TiNbO925,26,27. The relative intensity of the two peaks increases with the W doping concentration in a certain range, which is in accord with the trend of ratio of (001)/(110) in XRD, pointing out that W6+ ions are doped into the perovskite layers, thus contributing to structural distortion changes.

The element composition and chemical states of the Bi3TiNbO9-x W (x = 0, 5%) were determined by X-ray photoelectron spectroscopy (XPS) characterization. As shown in the Nb 3d spectrum of the Bi3TiNbO9-5%W (Fig. 2a), there are two additional peaks at 209.79 and 207.02 eV as compared with Bi3TiNbO9, of which the Nb 3d peak positions are in good agreement with those of Nb2O5 (Supplementary Fig. 4f). The chemical states of Bi 4 f, Ti 2p, and O 1 s of Bi3TiNbO9-W are nearly the same as those of Bi3TiNbO9 (Supplementary Fig. 4). These results reveal that W6+ ions occupy the Nb5+ ion sites in the perovskite structure, which is also supported by theoretical results in the following section of Fig. 3. Moreover, there are some impurity peaks that can be assigned to KNbWO6 (PDF #25-0667) shown in the XRD diffraction pattern of Bi3TiNbO9-25% W with heavy doping (Supplementary Fig. 5), also affirming the substitution of Nb5+ by W6+. X-ray absorption spectroscopy (XAS) of the W L3-edge was performed to probe the local coordination environment. The second derivative of X-ray absorption near-edge structure (XANES) spectrum in Supplementary Fig. 6 demonstrates two peaks with an energy difference of about 4.3 eV due to the split 5d states caused by the ligand field. And the splitting gap is in agreement with the value of octahedral wolframates (3.0–5.6 eV)28,29. By further analyzing the Fourier transformed data of the extended fine structure region (EXAFS, Fig. 2b), the fitting curve reveals a W-O bond distance of 1.80 ± 0.01 Å and a coordination number of W atoms of 5.0 ± 0.7, which indicates the unsaturated coordination environment of W atoms in these Bi3TiNbO9-W nanosheets.

Fig. 2: The distribution of the doped element.
figure 2

a Deconvolved XPS spectra of Nb 3d. b Fourier-transform curves of EXAFS data of W L3-edge (rW-O: The bond distance of W-O, CNW: The coordination number of W, R: R factor, σ2: Debye-Waller). c The ratio of W/(Ti+Nb) as a function of etching depth based on XPS of Bi3TiNbO9-W. d Atomic-level cross-section HAADF-STEM image of Bi3TiNbO9-W. e The corresponding EDS mapping images of elements (Bi, Ti, Nb, W).

Full size image
Fig. 3: Additional built-in electric field introduced by gradient doping.
figure 3

a The calculated band structures of Bi3TiNbO9 with different W-doping concentrations. The calculated bandgap values are also given. The VBM (set to be 0 eV) and the Fermi level are denoted by black dotted and red dashed lines, respectively. The insets show an enlarged distribution of the Nb 4 s energy levels highlighted by the cyan dashed rectangles. The black dashed lines in the insets represent the averaged energy of the Nb 4 s levels. b The calculated density of state (DOS) of Bi3TiNbO9 with different W-doping concentrations. The DOS within an energy window from 2.0 to 3.5 eV nearby the CBM is shown in band structure plots. For the cases of B-doped Bi3TiNbO9 with doping concentration of 3.2, 6.3, and 12.5%, the local DOS value of W dopant is multiplied by a factor of 16, 8, and 4, respectively. c The band alignment of pristine Bi3TiNbO9 and W-doped Bi3TiNbO9 with different doping concentrations, by taking the Nb 4 s energy level as a reference. d Schematic diagram for the built-in electric field of Bi3TiNbO9 induced by gradient W-doping.

Full size image

The distribution of W elements in Bi3TiNbO9-W nanosheets was confirmed by XPS depth profiles and elemental mapping analysis. As shown in Fig. 2c, the atomic ratio of W/(Ti+Nb) decreases gradually from 22.6% to 13.3% with increasing the etching depth by argon ions up to around 20 nm, proving the gradient doping of W. In addition, the combined XPS and ICP-OES investigations reveal that the W/(Ti+Nb) ratio on the surface is much higher than that in the bulk for all W-doped samples (Supplementary Fig. 7). Notably, the difference between the measured surface and bulk W/(Ti+Nb) ratio increases rapidly with W-doped concentrations of higher than 5%. These results suggest that the diffusion of doping atoms in the material is limited and the saturation doping concentration is about 3.5%. The spatial distribution of W was further examined by energy-dispersive X-ray spectroscopy (EDS). Viewing from Supplementary Fig. 8, the elements Bi, Ti, Nb, O, and W show a homogeneous distribution, demonstrating the homogenous doping of the W element in the basal {001} facet of the Bi3TiNbO9-W nanosheets. To further study the W distribution along the c-axis, the two stacked Bi3TiNbO9-W nanosheets were cut by focused ion beam, and their cross-section region was inspected by high-angle annular dark-field scanning transmission electron microscopy (HAADF-STEM, Supplementary Fig. 9). The layered structure of Bi3TiNbO9-W nanosheets can be clearly identified in Fig. 2d, in which the double-layer Bi atom is the (Bi2O2)2+ layer and the single-layer Bi atom is the (BiTiNbO7)2- layer, respectively. The microstructure at the interface between two nanosheets suggests that the outermost layer of Bi3TiNbO9-W is the (Bi2O2)2+ layer. The corresponding EDS mapping images show a gradient W distribution feature in a single Bi3TiNbO9-W nanosheet, with a maximal W content on the surface (Fig. 2e), further indicating the gradient doping of W along c-axis.

Density functional theory (DFT) calculations explored the effect of W doping on the electronic structure of Bi3TiNbO9. Based on the analysis of the ion valence and radius (Ti4+: 0.605 Å, Nb5+: 0.64 Å, W6+: 0.60 Å), W atom may replace both the Ti and the Nb atoms in Bi3TiNbO9-W. Therefore, we have firstly calculated the formation energies of W-doped Bi3TiNbO9 (2 × 2 × 1) with one Ti or one Nb atom replaced by one W atom, aiming to identify the specific doping site for the W atom in Bi3TiNbO9. As a result, the formation energies are calculated to be 4.00 and 2.59 eV for W replacing Ti and Nb, respectively. This indicates an energetical preference for W dopants to occupy the Nb sites and is in good agreement with the XPS spectra (Fig. 2a) and XRD patterns (Supplementary Fig. 5). Therefore, we only consider cases of W replacing Nb for electronic structure calculations of Bi3TiNbO9-W. As shown in Fig. 3a, Bi3TiNbO9-W exhibits smaller bandgap values (2.35–2.50 eV) compared with that (2.79 eV) of pristine Bi3TiNbO9, due to a contribution from the W 5d orbitals with lower energy to the CBM (see Fig. 3b). Moreover, the bandgap values of Bi3TiNbO9-W decrease from 2.50 to 2.38, then to 2.35 eV with increasing the W-doping concentration from 3.2 to 6.3, then to 12.5%, respectively. UV-vis absorption spectra of Bi3TiNbO9-x W (x = 0, 1%, 3%, 5%, 7%, 12%) samples in Supplementary Fig. 10 also demonstrated a slight red shift of the absorption edge with the increase of W doping concentration.

As is well known, the core electrons of transition metals normally do not hybridize with the O 2p electrons in metal oxides, giving rise to the core electron level with very low energy. Such a core electron level does not obviously change by changing the chemical environment and has been widely adopted as a reference for band alignment30,31. Here, taking the Nb 4 s level as the reference, the band alignment of Bi3TiNbO9 and Bi3TiNbO9-W with different doping concentrations is given in Fig. 3b. As we can see, the valance band maximum (VBM) of Bi3TiNbO9-W exhibits a slight decrease with increasing the doping concentration. Combined with the gradually reduced bandgap values, a more pronounced decrease of the conduction band minimum (CBM) can be observed with increasing the doping concentration for Bi3TiNbO9-W, compared with pristine Bi3TiNbO9 (Fig. 3c). Typically, the introduction of W6+ ions in Bi3TiNbO9 could give rise to a shallow donor doping effect, caused by the unsaturated coordination environment of W atoms in these Bi3TiNbO9-W nanosheets as analyzed in Fig. 2b and can be confirmed by the Mott–Schottky curves shown in Supplementary Fig. 11. The gradient donor W doping could contribute to a gradual decrease in the gap between the Fermi level and the CBM from the bulk phase to the surface, forming an additional built-in electric field along the c-axis, which could provide a driving force for the migration of photoexcited electrons from bulk to the surface of Bi3TiNbO9-W (Fig. 3d)32.

To understand the structural distortion caused by W doping, Rietveld refinement was carried out via the General Structure Analysis System (GSAS), as shown in Figs. 4a, b. There is a slight shrinkage at the a-axis after W doping, while the b-axis maintains unchanged properties basically. The corresponding b/a value increases from 1.0055 to 1.0067, implying an enhancement of the polarization along the a-axis33,34. The piezoelectric coefficient (d33) and dielectric constant (εr) were further used to investigate the ferroelectric polarization characteristics of the samples. Figures 4c, d shows the piezoelectric responses near the resonant frequency for Bi3TiNbO9 and Bi3TiNbO9-W samples at different driving voltages, respectively. It can be clearly observed that the piezoelectric resonant peak amplitude (Amax) increases with the driving voltages. The linear relationship between \({A}_{\max }/Q\) and Vac can be obtained by fitting the harmonic oscillator model, where Q is the quality factor and Vac is the driving ac voltage35,36, as shown in Fig. 4e. The piezoelectric coefficient d33 values are derived to be 1.24 and 2.29 pm/V for Bi3TiNbO9 and Bi3TiNbO9-W, respectively. The remarkable increase in d33 indicates that the ferroelectric polarization intensity (Ps) of Bi3TiNbO9 is significantly enhanced after W doping37,38. Figure 4f shows the dielectric constant as a function of frequency at room temperature. Generally, the polarization of dielectric materials is contributed by electronic, ionic, orientation, and space charge polarization39,40. The contribution of space charge polarization decreases with the increase of frequency, giving rise to an inverse correlation with frequency for the dielectric constant (Fig. 4f). Comparing the dielectric constants of Bi3TiNbO9 and Bi3TiNbO9-W, it can be inferred that W doping could elicit an increase in the polarizability characteristics of Bi3TiNbO9, in accordance with the shrinkage of the a-axis by the Rietveld refinement (Fig. 4a, b) and the increased piezoelectric coefficient (d33) values in Fig. 4e.

Fig. 4: Structural polarization characteristics.
figure 4

X-ray diffraction Rietveld refinement: (a) Bi3TiNbO9, (b) Bi3TiNbO9-W. The piezoelectric response near the contact resonant frequency under different driving voltages: (c) Bi3TiNbO9, (d) Bi3TiNbO9-W. e The out of-plane piezoelectric coefficient (d33) values. f Dielectric constant versus frequency at room temperature of Bi3TiNbO9 (cyan point) and Bi3TiNbO9-W (purple point).

Full size image

The difference of photogenerated charge transport behavior between Bi3TiNbO9 and Bi3TiNbO9-W was comprehensively analyzed. As shown in Fig. 5a, the electrons in Bi3TiNbO9 tend to migrate along the c-axis, but might be confined by interlayer barrier between the of (Bi2O2)2+ and (BiTiNbO7)2- layers, while the holes mostly migrate along the direction of the polarization electric field within the ab-plane13,16. The additional built-in electric field introduced by the gradient doping in Bi3TiNbO9-W could weaken the potential barrier between the layers of (Bi2O2)2+ and (BiTiNbO7)2-, thus promoting the migration of the photogenerated electrons to the basal {001} surface. Meanwhile, the enhanced depolarization field of Bi3TiNbO9-W along the a-axis could improve the transport of the photogenerated holes to the lateral surface (Fig. 5b). The surface photovoltage (SPV) spectra of Bi3TiNbO9 and Bi3TiNbO9-W were executed to further explore the charge transfer characteristics (Fig. 5c). Owing to the restriction on the migration of the photogenerated electrons along c-axis, the Bi3TiNbO9 exhibits a positive SPV feature in a consistent optical response range (300–400 nm)41,42,43. In contrast, the negative signals of Bi3TiNbO9-W imply that the electron diffusion to the surface is largely promoted by the modified energy band structures, where the additional build-in electric field induced by the W gradient doping could facilitate the transport of the photogenerated electrons from bulk to the {001} surface. The difference in the charge transfer processes is responsible for the higher hydrogen evolution activity and lower oxygen evolution activity of Bi3TiNbO9-W compared with that of Bi3TiNbO9 (Supplementary Fig. 12). The SPV signal on timescales of microseconds to milliseconds is attributed to the drift (Peak I) and diffusion (Peak II) processes of photogenerated carriers (Fig. 5d)44. Bi3TiNbO9 presents a positive peak II lasting for 12.3 ms before decaying to zero, while Bi3TiNbO9-W exhibits a negative peak II lasting for 16.7 ms. The extension of the photoresponse time might be ascribed to the enhancement of the polarization intensity along the a-axis direction and the additional electric field along the c-axis direction in Bi3TiNbO9-W, which can promote the anisotropic migration and effective separation of photogenerated charges.

Fig. 5: Photogenerated charge transfer behavior.
figure 5

Transfer of photogenerated charged carriers of Bi3TiNbO9 (a) and Bi3TiNbO9-W (b) along ac plane (Ps-ferroelectric polarization, E-additional electric field). Surface photovoltage (SPV) spectra (c) and transient photovoltage (TPV) response (d) of multiple Bi3TiNbO9 (cyan point) and Bi3TiNbO9-W (purple point) crystals.

Full size image

The surface catalytic activity of Bi3TiNbO9 and Bi3TiNbO9-W, especially oxygen evolution reaction (OER), was also studied by DFT calculations. Considering its intrinsic ferroelectricity, there exists a depolarization field along the \([\bar{1}00]\) direction of bulk Bi3TiNbO9 (see Supplementary Fig. 13a). In this case, the migration of photogenerated holes towards \((\bar{1}\bar{1}0)\) and \(\left(\bar{1}10\right)\) facets can be accelerated by the depolarization field, making the \((\bar{1}\bar{1}0)\) and \((\bar{1}10)\) facets more active for water oxidation reaction due to the accumulation of holes7,45. Since the \((\bar{1}\bar{1}0)\) facet and the \(\left(\bar{1}10\right)\) facet are equivalent in crystallography, only the reaction coordinate for OER on the \((\bar{1}\bar{1}0)\) facet of Bi3TiNbO9 and Bi3TiNbO9-W was considered in this work (see Supplementary Fig. 13b). Supplementary Fig. 13c, d present the calculated free energy profiles of water oxidation on the \((\bar{1}\bar{1}0)\) facets of Bi3TiNbO9 and Bi3TiNbO9-W. As we can see, Bi3TiNbO9 exhibits an overpotential of 0.78 V for OER at the Nb site, while the overpotential for OER at the W site on the \((\bar{1}\bar{1}0)\) facet of Bi3TiNbO9-W is calculated to be 0.54 V. Basically, a lower overpotential indicates higher OER activity. However, the potential of photoexcited holes in Bi3TiNbO9 is roughly estimated to be ~2.40 V15, suggesting that the photoexcited holes can easily overcome the overpotentials for OER on the \((\bar{1}\bar{1}0)\) facet of both Bi3TiNbO9 and Bi3TiNbO9-W. Moreover, as shown in Fig. 3d, the additional built-in electric field along c-axis induced by gradient W doping may impede the migration of photoexcited holes from bulk towards the surface as validated by the SPV signal reversion from positive to negative after W doping (Fig. 5c, d). In fact, it has been reported that a low concentration of surface-reaching photogenerated holes can extremely limit the photocatalytic oxygen evolution performance46. Therefore, Bi3TiNbO9-W exhibits an even lower oxygen evolution yield (see Supplementary Fig. 12b) due to the compromise of the minor advantage of reduced OER overpotential and major disadvantage of less surface-reaching holes by W doping.

Photocatalytic performance

Photocatalytic overall water splitting performance tests were carried out by loading Rh/Cr2O3 cocatalyst on the surface of Bi3TiNbO9 nanosheets47,48,49. As shown in Fig. 6a, the photocatalytic performance displays a volcanic curve with the W/(Ti+Nb) ratio increasing, which is consistent with the trend of structural distortion (inferred in Supplementary Fig. 2b, 3). When the ratio of W/(Ti+Nb) reaches 5%, that is, the saturation concentration of doping (Supplementary Fig. 7), Bi3TiNbO9-W exhibits the optimal photocatalytic overall water splitting activity with the average hydrogen and oxygen evolution rates of 43.99 and 20.66 μmol h-1, respectively, which are 5.55 times higher than those of Bi3TiNbO9. Considering the very close specific surface area (1.2655 versus 1.3352 m2 g-1) of the samples without and with W dopant, one can conclude that the additional built-in electric field has an essential effect on the photocatalytic overall water splitting activity enhancement. Once the doping amount exceeds the limit value, the formed impurity phase may be enriched on the surface of the sample, affecting light absorption characteristics and reducing photocatalytic activity. Furthermore, this gradient doping strategy in Bi3TiNbO9 system shows universality, demonstrating an efficient photocatalytic overall water splitting properties by employing Mo as the dopant as well. However, the activity of Bi3TiNbO9-Mo is only half of Bi3TiNbO9-W, as shown in Supplementary Fig. 14, which might be mainly attributed to the formation of defect transition levels below the conduction band edge induced by the Mo dopants. In contrast, W doping does not introduce any defect level in the band gap for the recombination of photogenerated carriers50.

Fig. 6: Activity and stability of photocatalytic overall water splitting.
figure 6

a Comparison of the activity of Bi3TiNbO9-W with different dopant concentrations. Stability test of photocatalytic overall water splitting (λ ≥ 300 nm): (b) Bi3TiNbO9, (c) Bi3TiNbO9-W, (d) Bi3TiNbO9-W-PL.

Full size image

Time-dependent photocatalytic overall water splitting performance tests were used to evaluate the stability of Bi3TiNbO9-W. As shown in Figs. 6b, c, the activities of both Bi3TiNbO9 and Bi3TiNbO9-W decay rapidly during the long-term test, which can be ascribed to the self-reduction of Bi3+ in the surface (Bi2O2)2+ layers, as revealed in our previous work15. The XRD pattern shows that the diffraction peaks of Bi3TiNbO9 and Bi3TiNbO9-W remain unchanged after the photocatalytic overall water splitting reaction (Supplementary Fig. 15), indicating that the samples have a highly stable crystal structure. Adjusting the surface structure of Bi3TiNbO9 by acid etching and selectively exposing the robust perovskite layer (denoted as PL) have been successfully demonstrated to be an effective way to inhibit photocorrosion. We adopted such a strategy to remove the (Bi2O2)2+ layer on the surface of Bi3TiNbO9-W (the resulting sample is named as Bi3TiNbO9-W-PL). The robust surface states of the obtained Bi3TiNbO9-W-PL catalysts can well balance the transfer of the photogenerated electrons to the perovskite layer-terminated surface and the photogenerated holes to lateral surface, corresponding to a weak and negative SPV signal demonstrated in Supplementary Fig. 16 Therefore, the Bi3TiNbO9-W-PL exhibits an efficient and stable activity (106.71 μmol h-1 for H2 and 47.94 μmol h-1 for O2) for overall water splitting in a 10 h long-term test (Fig. 6d), which is 10.4 times higher than that of pristine Bi3TiNbO9. The apparent quantum yield (AQY) of Bi3TiNbO9-W-PL was calculated to be 0.57% at 365 nm, being higher than the reported value for ferroelectric photocatalytic materials claimed (Supplementary Fig. 17, 18 and Supplementary Tables 1, 2).

Discussion

In summary, we have successfully synthesized a series of W-doped Bi3TiNbO9 (Bi3TiNbO9-W) nanosheets. The concentration of this donor W dopant shows a gradient decrease from the surface to the bulk along the c-axis, which exhibits favorable characteristics for photocatalytic water splitting. One is that the gradient distribution of the W dopant can introduce an additional built-in electric field from the surface to the bulk to conquer the potential barrier between the layers of (Bi2O2)2+ and (BiTiNbO7)2- along the c-axis for the migration of the photogenerated electrons. The other is that ion substitution in the perovskite layer could strengthen the structural distortion of Bi3TiNbO9, which can enhance the depolarization field along the a-axis. As a result, the effective anisotropic migration and separation of the photogenerated carriers is fulfilled. Combined with the selective exposure of the robust (BiTiNbO7)2- perovskite layer, our synthesized Bi3TiNbO9-W-PL catalyst shows an efficient and stable photocatalytic overall water splitting, which is 10.4 times higher than that of the pristine Bi3TiNbO9. This gradient doping strategy introduced in this work might provide guidance for the design of highly efficient and stable Aurivillius-type layered photocatalysts for overall water splitting.

Methods

Preparation of W-doped Bi3TiNbO9 nanosheets

Bi2O3 (0.699 g, Sinopharm (China)), anatase TiO2 (0.08 g, Aldrich, 99%), Nb2O5 (0.133 g, Aladdin, 99.9%), NaCl-KCl (8 g, Sinopharm (China), a molar ratio of 1:1 and an eutectic temperature of 657 °C), and an appropriate amount of WO3 (Sinopharm (China)) were mixed by an agate mortar for 30 min, and then transferred into an alumina crucible and calcined at 800 oC for 2 h. The obtained product was washed with hot deionized water to completely remove the flux reagents and dried in air, and denoted as Bi3TiNbO9-x W (x = 0, 1%, 3%, 5%, 7%, 12% in molar ratio). The synthesis process of Bi3TiNbO9-Mo is the same as that of Bi3TiNbO9-W, except that MoO3 instead of WO3 is used.

Preparation of Bi3TiNbO9-W-PL nanosheets

250 mg of the Bi3TiNbO9-W sample was dispersed into 125 mL 0.32 mol L-1 hydrochloric acid solution and continuously stirred for 5 h. The sample is then separated by filtration and washed with a large amount of deionized water until neutral.

Cocatalyst deposition

150 mg of the sample was dispersed in 100 mL of aqueous methanol solution (10 vol%). Photoreduction of rhodium chloride (0.9 mg, RhCl3, Alfa, 99.9%) and potassium chromate (0.9 mg, K2CrO4, Sigma-Aldrich, 99%) was carried out in steps under vacuum. The product was washed thoroughly with distilled water and dried.

Sample characterization

The Scanning Electron Microscope (SEM, Nova NanoSEM 430) was used to examine the morphology of the samples. X-ray diffraction patterns of the samples were collected on a Rigaku D/max 2400 spectrometer using Cu Ka X-rays of wavelength 1.54056 Å. The Raman spectra were collected on a Lab-Ram HR 800 spectrometer with an excitation wavelength of 532 nm. The UV-vis diffusion reflection spectra were obtained using a spectrophotometer (Jasco V-770) equipped with an integrating sphere in diffusion reflection mode. X-ray photoelectron spectroscopy (XPS) measurement was conducted on Escab-250 (Thermofisher Scientific, America) with a monochromatic Al Kα X-ray source. All binding energies were referenced to the C 1 s peak (284.6 eV) that arises from adventitious carbon. XPS depth profile analyses were performed using an Ar+ ion gun to etch stepwise layers at a rate of ~0.1 nm s-1. Elemental analysis was carried out by a means of inductively coupled plasma atomic emission spectroscopy (ICP-OES). The piezoresponse force microscopy (PFM) of the powder samples were performed by a Scanning Probe Microscopy (SPM, Bruker Dimension Icon) system operated in Vertical PFM mode with Pt/Ir-coated Si cantilever tips. The dielectric measurements were performed using the two-probe impedance method with an Impedance Analyzer (TH2828A) over the frequency range from 100 Hz to 2 MHz. Cross-sectional HAADF-STEM images were acquired using an aberration-corrected STEM (Titan Themis 60–300 X-FEG microscope (FEI)) with double aberration (Cs) correctors from CEOS and a monochromator operating at 300 kV.

Photocatalytic activity tests

Photocatalytic water splitting tests were conducted in an automatic testing system (Beijing Perfectlight Technology Co., Ltd., Labsolar-6A) with a 250 mL reaction container and a set temperature of 10 °C. The specific operation process is as follows: The sample loaded with cocatalyst (50 mg) was dispersed into 100 mL of pure water. After the air in the system was completely removed, a 300 W xenon lamp (Beijing Perfectlight Technology Co., Ltd., PLS-SXE300D) at a wavelength of λ  ≥  300 nm with an intensity of 349 mW cm−2 was used as the light source. The gases generated from the system were analyzed at given time intervals by gas chromatography (Agilent Technologies, 6890 N).

Surface Photovoltage (SPV) measurement

The system included a lock-in amplifier (SR830, Stanford Research Systems, Inc.) with a light chopper (SR540), a 500 W xenon lamp with a monochromator (SBP500, Zolix) as a light source, and a photovoltaic cell. The monochromatic light chopped with a frequency of 23 Hz is focused on the photovoltaic cell. The photovoltaic signal is amplified by a lock-in amplifier and transmitted to a computer for recording.

Transient Photovoltage (TPV) measurement

The system included a Nd: YAG laser source (Q-smart 450 Quantel) providing laser radiation pulse (wavelength: 355 nm, pulse width: 5 ns and Intensity: 100 µJ), a 500 MHz digital phosphor oscilloscope (TDS 5054, Tektronix) with a preamplifier (5003 Brookdeal Electronics), and a parallel-plate capacitor-like sample chamber.

Calculation detail

The density functional theory (DFT) calculations were performed by using Vienna Ab-initio Simulation Package (VASP)51. The interaction between ions and valence electrons was described using projector augmented wave (PAW) potentials52, and the exchange-correlation interaction was described by generalized gradient approximation (GGA) with the Perdew-Burke-Ernzerhof (PBE) functional53. To overcome the well-known self-interaction error in DFT, the GGA + U method was used to treat the d electrons (\(U=4{{{{{\rm{eV}}}}}}\) for Ti 3d and Nb 4d, \(U=3{{{{{\rm{eV}}}}}}\) for W 5d)54. The plane wave energy cutoff was set to be 500 eV. The valence electron configurations of Bi, Ti, Nb, O, and H atoms are 5d106s26p3, 3s23p63d24s2, 4s24p64d45s1, 2s22p4, and 1s1, respectively.

To describe the W-doped Bi3TiNbO9, a \(2\times 2\times 1\) supercell containing 16 Ti and Nb atoms was constructed, subsequently, one, two, and four Nb atoms were replaced by W atoms, corresponding to a W-doping concentration of 3.2, 6.3, and 12.5%, respectively. In order to calculate the Gibbs free energy (ΔG) of different intermediates (OH*, O*, OOH*) adsorption on the \((\bar{1}\bar{1}0)\) facet of Bi3TiNbO9, a simplified model only with the perovskite layer was adopted for the four-electron oxygen evolution reaction (OER), since the perovskite layer was demonstrated to be active for OER55,56. As for the W-doped case, one surface Nb atom was replaced by a W atom. For this model, except for one five-fold-coordinated Ti, Nb, or W atom on the \((\bar{1}\bar{1}0)\) surface was left to be exposed as the reaction site for OER, all other unsaturated O and transition metal (Ti, Nb, W) atoms were saturated by H atoms and hydroxyl groups, respectively. The vacuum thickness was set to be 16 Å for both the b and c directions. The atomic structures were fully optimized with the convergence criteria for residual force and total energy of 0.02 eV Å−1 and 10−6 eV, respectively.

The Gibbs free energy (\(\Delta G\)) was calculated from the DFT simulation by adding the free energy correction according to following equation,

$$\Delta G=\Delta {E}_{{{{{{\rm{DFT}}}}}}}+\Delta {G}_{{{{{{\rm{cor}}}}}}}$$
(1)

where the free energy correction (\(\Delta {G}_{{{{{{\rm{cor}}}}}}}\)) for free molecules and adsorbed species was obtained by using the VASP toolkit56, based on the following definition,

$${G}_{{{{{{\rm{cor}}}}}}}={ZPE}+{E}_{{{{{{\rm{vib}}}}}}}+{E}_{{{{{{\rm{tran}}}}}}}+{E}_{{{{{{\rm{rot}}}}}}}+{PV}-{TS}$$
(2)
$${G}_{{{{{{\rm{cor}}}}}}}={ZPE}+{E}_{{{{{{\rm{vib}}}}}}}-{TS}$$
(3)

where ZPE is the zero-point energy, \({E}_{{{{{{\rm{vib}}}}}}}\), \({E}_{{{{{{\rm{tran}}}}}}}\), and \({E}_{{{{{{\rm{rot}}}}}}}\) are thermal corrections to the internal energy from vibrational, translational, and rotational motions of molecules, respective, the PV term converts internal energy to enthalpy, and the entropy term TS converts enthalpy to Gibbs free energy.