The Nature of Electron Transport and visible light absorption in Strontium Niobate -- A Plasmonic Water Splitter

Semiconductor compounds are widely used for water splitting applications, where photo-generated electron-hole pairs are exploited to induce catalysis. Recently, powders of a metallic oxide (Sr$_{1-x}$NbO$_3$, 0.03<x<0.20) have shown competitive photocatalytic efficiency, opening up the material space available for finding optimizing performance in water-splitting applications. The origin of the visible light absorption in these powders was reported to be due to an interband transition and the charge carrier separation was proposed to be due to the high carrier mobility of this material. In the current work we have prepared epitaxial thin films of Sr$_{0.94}$NbO$_{3+{\delta}}$ and found that the bandgap of this material is ~4.1 eV, which is very large. Surprisingly the carrier density of the conducting phase reaches 10$^{22}$ cm$^{-3}$, which is only one order smaller than that of elemental metals and the carrier mobility is only 2.47 cm$^2$/(V$\cdot$s). Contrary to earlier reports, the visible light absorption at 1.8 eV (~688 nm) is due to the bulk plasmon resonance, arising from the large carrier density, instead of an interband transition. Excitation of the plasmonic resonance results in a multifold enhancement of the lifetime of charge carriers. Thus we propose that the hot charge carriers generated from decay of plasmons produced by optical absorption is responsible for the water splitting efficiency of this material under visible light irradiation.


Introduction
Converting solar energy into chemical energy (e.g. splitting water by sun light) with the aid of photocatalysts is a promising way to reduce the increasing demand for fossil fuels 1,2,3,4,5,6,7,8,9,10 . Very few oxide semiconductors have been used as photocatalysts since they need to be chemically robust and their bandgap should be neither too wide nor narrow in order to 3/27 absorb sun light in the visible range efficiently and also satisfy the minimum energy requirement (1.23 eV theoretically but >1.9 eV experimentally) for splitting water into hydrogen and oxygen 11,12,13 . Large bandgap oxides (such as TiO2) are used in photocatalytic water splitting either by reducing their optical bandgap to absorb visible light or incorporating visible light absorbers such as organic dyes, low bandgap quantum absorbers or metal nanostructures 14,15 in the host. In the former case, cationic or anionic doping or combination of both are typically applied to narrow the bandgap 16,17,18,19 . Unfortunately, in most cases the bandgap change achieved is small mainly due to the fact that the discrete defect states are normally very close to the band edges 20,21,22 . In the latter case, hot electrons in the visible light absorber inject into the conduction band of large bandgap oxides, which are subsequently used to reduce water to hydrogen gas. Among the visible light absorbers, metal (e.g. Au, Ag) nanostructures are special as the hot electrons generated by the decay of visible light excitation of surface plasmon resonance (SPR) can be injected in to a large bandgap semiconductor like TiO2 23,24 . However, the materials used for enhancing the photocatalytic activity by using SPR are mainly Au, Ag nanostructures, which are not low cost materials.
Recently a red metallic oxide Sr1-xNbO3 (0.03 < x < 0.20) (in the form of powders) was used in photocatalytic water splitting and the authors proposed a special band structure, in which the electron-hole pairs come from the optical transition from the metallic conduction band (about 1.9 eV above the valence band) to a higher level unoccupied band 25,26 . In this report, the visible light absorption was attributed to the electron's interband transitions and the 4/27 electron-hole pair separation was attributed to the assumption of high carrier mobility although only temperature dependent conductivity was measured. As both charge carrier density and mobility contribute to the conductivity of a sample, simply assuming a large mobility for a highly conductive material may lead to a wrong conclusion. Hence obtaining the mobility as well as other electrical transport properties is crucial for understanding the details of the photo-generated carrier separation process. The optical bandgap obtained from Kubelka-Munk transformation of the reflectance spectrum of the powder form Sr1-xNbO3 is inaccurate since it neglects the plasmonic absorption. The proper way to measure the optical bandgap is to obtain the complex dielectric function, by using Kramers-Kronigtransformed reflectivity or spectroscopic ellipsometry. Epitaxial thin films are required for measuring the intrinsic mobility since the grain boundaries in the film are much less than in powders. Using thin films also allows us to measure the transmittance, reflectance and ellipsometry spectra accurately, which can give proper optical and plasmonic absorptions of this material. Furthermore, both the band structure and the process of the hot electron transition of this material can also be investigated by femtosecond time resolved transient absorption (TA) spectroscopy.
Here we have prepared Sr0.94NbO3+δ films by pulsed laser deposition (PLD) at various oxygen partial pressures on top of insulating LaAlO3 substrates and compared their optical spectra, electronic transport and carrier dynamics properties. We found the optical bandgap to be around 4.1 eV, which was almost independent of the oxygen content although the crystal structure changed from pseudo-tetragonal perovskite to orthorhombic with increasing 5/27 oxygen partial pressure from 5×10 -6 Torr to 1×10 -4 Torr. The bulk plasmon peak for the film grown at 510 -6 Torr was found at about 1.8 eV (688 nm), which is at the appropriate energy for solar-photocatalytic water splitting. The high conductivity (~10 4 S/cm) of the sample prepared at low pressure is mainly contributed by the high charge carrier density (~10 22 cm -3 ) rather than its mobility (2.47 cm 2 / (V•s)) at room temperature. Thus we believe that the photocatalytic activity of Sr0.94NbO3+ under visible to near-infrared irradiation is due to the hot electrons generated from the decay of the plasmon in Sr0.94NbO3 instead of interband absorption transition. Thus Sr0.94NbO3+δ represents an extraordinary material system, which has a large bandgap of 4.1 eV but a degenerate conduction band with a large carrier density exceeding 10 22 electrons/ cm 3 which leads to strong useful plasmonic effects.

Results and discussions
The XRD spectrum of the film deposited at oxygen partial pressure of 5 × 10 -6 Torr is shown in  (Fig. 1(c)). The highlighted open burgess circuit 6/27 indicates an a[100]p type edge dislocation core, where the extra plane on the LaAlO3 side indicates a compressive misfit of the film. As the oxygen partial pressure increases to 1 × 10 -4 Torr, a small shift of the film peaks towards low angles can be observed in the θ-2θ scan and the FWHM of the rocking curve increases (Fig. S1). Structural changes from tetragonal to orthorhombic were observed in the local HRTEM images (Fig. S2). The orthorhombic structure is close to the reported structure of Sr2Nb2O7 (bulk a=3.933 Å, b=26.726 Å, c=5.683 Å), which has an equivalent tetragonal structure with lattice parameters a=b=3.901 Å and c=3.933 Å. The decreasing of the out-of-plane and in-plane lattice parameters with oxygen partial pressure was obtained from the electron diffraction pattern (Fig. S2, Table S1). At the intermediate oxygen partial pressure, mixed structure exists in the film.
It was reported that Sr content strongly determines the crystal structure of the nonstoichiometric SrNbO3 phase 27,28 . Here the elemental content of the films deposited at different oxygen partial pressures are precisely studied (Fig. S3). The cationic contents are identical for the films within the detection limit of PIXE and the Sr/Nb atomic ratio is measured as 0.94/1. The deficiency of Sr content comes from the variation in the target preparation process 29 .
A cut off of the transmission edge is observed near 300 nm ( Fig. 2(a)), which indicates an optical bandgap of ~ 4.1 eV (from Tauc plot-indirect, Fig. S4) and it is almost independent of the preparation oxygen pressure. Both the transmission and reflection of the film prepared at 5  10 -6 Torr were plotted, from which the accurate absorption spectrum could be obtained ( Fig. 2(b)). The minimum reflection is located at around 600 nm, which could 7/27 indicate the rough frequency of its volume plasmon. Hence the reflection spectrum between 500 nm and 1000 nm can be well fit by the Drude Model, and the corresponding plasmon frequency of the fitting curve is 1.6 eV, which has a small difference with the plasmon frequency measured by spectroscopic ellipsometry. The transmission of the films continuously increases with oxygen partial pressure above 600 nm, which indicates absorption along with free carrier absorption (Drude Model) in this wavelength range, where the latter is consistent with the metallic nature of the films and powders 25,30 .
The complex refractive index,̃( ) = ( ) + ( ), and the loss function, -Im[ -1 (ω)], spectra of the 5x10 -6 Torr sample extracted from spectroscopic ellipsometry data are shown in Fig. 2(c) and 2(d). The extinction coefficient spectrum, (ω), of the sample (Fig. 2(d)) shows that it has a Drude peak below 2 eV (typical of a metal) and a first interband transition peak (indicating the bandgap of the film) above 4.1 eV, consistent with its transmission spectrum ( Fig. 2(b)). Between these two peaks, the (ω) is featureless, indicating the lack of major optical transitions within the 2 -4.1 eV energy range. Meanwhile, the loss function spectrum of the sample (Fig. 2(d)) shows a large peak at ~1.5 -2.1 eV with a peak position of ~1.8 eV (688 nm), indicating the existence of a bulk plasmon at that energy 31 . From n(ω) and (ω) spectra, the normal-incident reflectivity of the film can be obtained using Fresnel equations, as shown in Fig. 2(e). This reflectivity is consistent with the spectrum measured by UV-Visible spectroscopy. From the reflectivity, the Kubelka-Munk function of the film can be obtained (Fig. 2(e)), and it can be seen that the shape of the function resembles the previous reported results, with an apparent absorption edge at ~2 eV. Since there is no peak 8/27 in the (ω) spectrum at around that energy, this absorption edge does not come from intraor interband transition as previously reported. Instead, this absorption edge is plasmonic in origin because it coincides with the plasmon peak at ~1.8 eV in the loss function spectrum.
The origin of hot electrons in SrNbO3 under irradiation with visible light, which can be used for water splitting, can be interpreted by the plasmon model. When SrNbO3 is under irradiation, there will be a resonant collective oscillation of the electrons in the conduction band, which is the surface plasmon. Hot electrons are generated by the decay of the plasmon and transferred to co-catalysts (e.g. Pt), where the H + reduction reaction can take place. The holes left in the SrNbO3 can drive the oxidation reaction at the surface of this material. However, it was reported that the visible light absorption was attributed to the electrons' interband transitions and the electron-hole pair separation was attributed to a possible high mobility of the electrons, which is inconsistent with our model. To resolve this problem, the carrier transport properties were studied for this material.
The conducting property of the film is strongly dependent on the oxygen partial pressure, where a transition of metallic to semiconductor transport behavior is clearly seen when increasing the deposition oxygen partial pressure (Fig. 3(a)). The electron density of the most conductive sample (prepared at 5 × 10 -6 Torr) reaches 10 22 cm -3 and it is almost independent of the measurement temperature, which agrees with the reported data 32 (Fig. 3(b)) indicative of a degenerate Fermi level. In contrast, the electron mobility is only 2.47 cm 2 /(V•s) at room temperature, which is not outstanding compared with other oxides (e.g. TiO2, BaSnO3) and semiconductors 33,34,35,36 (Fig. 3(c)). So the high conductivity of this material is due to the 9/27 high carrier density and not the carrier mobility. The absence of significant internal electric field to avoid electron-hole recombination also implies that an interband transition model is not suitable. As the oxygen content increases in the film, the sample becomes more insulating. Both the charge carrier density and the mobility decreases with oxygen partial pressure, which is consistent with the observed two crystal structures of the materials 37 .
To further understand the role of the plasmon in the catalytic process, the transient absorption spectroscopy and time-resolved pump-probe spectroscopy were used to characterize the carrier dynamic process in strontium niobate. The Fig. 3 (d) and (e) show the various excitation wavelength dependent differential reflectance (ΔR/R) spectra with different delay time. Two peaks located near 600 nm (positive ΔR/R, 2.07 eV) and 670 nm (negative ΔR/R, 1.85 eV) are observed in the transient reflection spectra. The sign of differential reflection signal would be usually opposite to the sign of differential transmission signal 38,39 . Therefore, the positive 600 nm peak can be attributed to the optical absorption of the excited electrons (hot electrons), which might be the transition from the valence band to the deep trapped states. The negative 670nm peak can be attributed to both the transition from deep trapped states to conduction band and the optical absorption of plasmonic resonance which is consistent with the plasmonic resonance peak measured by the ellipsometry spectroscopy. It should be noted that though the transient absorption 10/27 spectrum can show the signal from deep trap states, the transition process related with these states cannot be used for the photocatalytic water splitting process because the deep trapped states can only act as the recombination centers for photo generated electron-hole pairs. The intensity of differential transient reflection spectra excited by 685 nm pump light at 0.5 ps delay is very strong compared with that of other delays which can be attributed to the strong absorption of the plasmon resonance. Fig. 4 (a) shows the Landau decay process of plasmon resonance 40 . The Landau decay process is very short which is in the time scale of a few hundred femtoseconds. The lifetime of Landau process as a function of excitation wavelength is presented in Fig. 4 (c). The lifetime increases to its maximum when the excitation wavelength is 685 nm which is near the plasmon peak of SrNbO3. Fig. 4 (b) shows decay curve of the transition from deep trapped states to the conduction band. However, because the density of unfilled states in the conduction band depends on the thermal dissipation process of hot carriers, thus this decay lifetime corresponds to the thermal dissipation process of hot carriers in SrNbO3 film. We can see the lifetime of this process can be as long as 400 to 550 ps when it is excited by the pump pulse with energy higher than that of the plasmon, while it will decrease to about 250 ps when the pump energy is lower than that of the plasmon. This shows that plasmon resonance can increase the lifetime of hot carrier thermal dissipation process which may also enhance the photocatalytic activity of SrNbO3 as the lifetime of the hot electrons is long enough for the carriers to convert water into gases before recombination.
The energy band structures of SrNbO3, SrNbO3.4 and SrNbO3.5 are calculated using density 11/27 functional theory (DFT) and shown in Fig. 5. In the calculations, the perovskite structure was assumed for the stoichiometric SrNbO3 compound, and the extra oxygen atoms for the hyperstoichiometric compositions were assumed to order into planar defects, as illustrated by the structural figures in the left panels of Fig. 5. This structural model is consistent with electron microscopy analyses that will be reported elsewhere 41,42 . The Fermi level of SrNbO3 is located in the conduction band, which implies that this material is metallic even though the bandgap is as large as 4.1 eV (we note the calculations predict a smaller bandgap relative to this experimental value, as is typically found from DFT). The Fermi level of SrNbO3.4 is located near the bottom of the conduction band, so the conductivity is poorer than that of SrNbO3 as there are fewer states for the free carriers leading to a lower carrier density. These Conversion TOPAS-C optical parametric amplifier to generate 350 nm as the pump beam. The intensity of the pump beam was attenuated using a neutral density filter and modulated using an optical chopper at a frequency of 500 Hz. The smaller portion of the beam was used to generate white light by passing through a 1 mm sapphire plate, which acted as the probe beam. The white light beam was further split into two portions: one was used as the probe and the other was used as the reference to correct for the pulse-to-pulse intensity 14/27 fluctuation. The pump beam was focused onto the sample surface with a beam size of 300 μm, and it fully covered the smaller probe beam (diameter: 100 μm). The reflection of the probe beam from the sample surface was collected with a pair of lens and focused into a spectrometer. Very thick film samples were used to minimize the signal contribution from the substrate. The delay between the pump and the probe pulses was controlled by a computer-controlled translation stage (Newport, ESP 300). Pump probe experiments were carried out at room temperature. During the measurements, the pump and the probe energies were kept low enough to minimize damage to the samples.

Theoretical calculations:
The atomic and electronic structure of SrNbO3+x compounds were performed employing spin-polarized density-functional-theory (DFT) calculations, using the Perdew-Burke-Ernzerhof (PBE96) 43 exchange-correlation potential, and the projectoraugment wave (PAW) method 44,45 , as implemented in the Vienna ab-initio simulation program (VASP) 46 . In these calculations Sr 4s4p5s, Nb 4p5s4d, and O 2s2p orbitals were treated as valence states, employing the PAW potentials labeled "Sr_sv", "Nb_pv" and "O" in the VASP PBE library. The cutoff energy for the plane-wave basis set was set to 450 eV, and the DFT+U approach due to Dudarev et al. 47