New materials for the light-induced hydrogen evolution reaction from the Cu–Si–Ti–O system

Cu-containing photocathodes are generally limited by fast photocorrosion under working conditions. Hence stabilization of these materials is a key factor in their potential application for the light-induced hydrogen evolution reaction (HER). In order to identify new materials, oxidized Cu–Si–Ti metallic thin film precursor materials libraries were evaluated using a combinatorial approach. High-throughput photoelectrochemical characterization using an automated optical scanning droplet cell was performed on a material library to analyze doping and alloying effects on the light-induced HER. The results revealed that compositions near Ti-doped CuSiO3 (dioptase and copper-polysilicate) and Si-doped Cu3TiOx act as comparatively stable and highly active materials for HER.


Introduction
Photoelectrochemical (PEC) water splitting 1 is the scientic endeavor of absorbing sunlight using a semiconductor to drive an electrochemical reaction on its surface to obtain hydrogen as an energy carrier. This work contributes to the ongoing search for new Cu-O-based p-type materials exhibiting promising properties (photovoltage, photocurrent and stability) for solar water splitting. The Cu-Si-Ti-O system was explored for stable absorber materials. 2 In the materials driven eld of photoelectrochemistry, research is directed towards understanding of the fundamental processes in metal oxide photoanodes 3 but only a few photocathode materials 1 suitable for PEC such as Cu 2 O (cuprite) 4 or CuFeO 2 5 are known. Most of the known metal-oxide photocathode materials are however unstable in aqueous solution 2 unless protected by layers of stable materials like TiO 2 . 6 The reason for the low stability of p-type oxides such as cuprite is due to photoreduction of a metal-oxygen bond when hydrogen evolves on the surface. In cuprite this happens as the reduction potential of Cu(II) to Cu(0) lies within the bandgap. 7 Most of the well-known p-type metal oxides own their conductivity type to Cu-vacancies as for example in cuprite 8 and CuFeO 2 . 5 In this work, new materials are searched for in the Cu-Si-Ti-O system to broaden the choice of p-type absorbers that could be employed in an all oxide tandem solar water splitting device. 9,10 The necessary stability enhancement of Cu-containing metal oxides such as cuprite 7,11 is investigated by combinatorial alloying with Si and Ti as third elements in an exploratory search for a stable photoabsorber material.
Combinatorial materials science in this work involves combinatorial sputter deposition of gradient thin lm materials libraries 12,13 and their high-quality and high-throughput indepth characterization. 14,15 Different possibilities have been proposed that enable the deposition of multinary composition spreads like metal nitrate printing, 16 evaporation 17 or reactive magnetron sputtering. 15 Combinatorial optimization of surface protection of cuprite by TiO 2 for photovoltaic applications has been suggested, 18 but a remaining problem for PEC is the formation of pinholes that drastically reduces the lifetime of a related device. 19 The Cu-Si-Ti-O system was chosen, as on the Ti-and Si-rich corners of the quasi-ternary composition space (i.e. the oxygen content is neglected) two well-studied stable oxides exist. However, these materials are either insulators (SiO 2 ) or have a large bandgap 20 (TiO 2 ). It has been shown that CuO-based semiconductors can be partially stabilized 21 by alloying with Ti. However, the addition of Ti into the lattice is associated with deep donor states that act as non-radiative recombination centers, leading to lower photoactivity as compared to CuO or Cu 2 O. 18,22 These deep donor states give rise to slightly higher optical absorptivity at higher wavelengths assuming a reduced bandgap by e.g. doping or alloying TiO 2 with various elements. 23 Investigating such effects requires photocurrent spectroscopy at a variety of applied potentials. 24 In this work we used combinatorial wedge-type multilayer deposition 13 to fabricate simultaneously binary and ternary composition spread type metallic thin lm materials libraries in the Cu-Si-Ti system that are subsequently oxidized in air to obtain ternary and quaternary materials in the Cu-Si-Ti-O system. High-throughput PEC characterization of the material libraries was performed to identify stable and highly active materials for HER.

Experimental
Synthesis and pre-characterization of thin lm materials libraries Thin lm ternary Cu-Si-Ti composition-spread type materials libraries were deposited in a combinatorial magnetron sputtering system (DCA Finland, CMS 600/400LIN) using a wedgetype deposition method. 13,25 The substrate was a polycrystalline Al 2 O 3 on which a 20 nm Au back contact was sputtered prior to the deposition of the materials library. The individual layers of Cu-Si-Ti were sputtered from elemental 4 inch high purity (>99.99%) sputter targets (Lesker). To avoid surface hydroxides, the substrate coated with the Au back contact was annealed for 10 h at 150 C prior to the Cu-Si-Ti deposition. The total thickness of the as-deposited thin metal precursor lms was 300 nm, calculated from measuring individual sputtering rates before the deposition. The compositional spread of the Cu-Si-Ti materials library was analyzed by automated energy dispersive X-ray spectroscopy (Inca-EDX, Oxford Instruments). Subsequently the metallic thin lm precursor materials library was oxidized at 600 C for 8 h in air using a conventional furnace.
The measurement areas for the different high-throughput characterization methods are shown in Fig. 1. Around each electrochemical measurement spot four synchrotron X-ray diffraction (XRD) measurements were performed.

Synchrotron diffraction
Wide-angle synchrotron XRD was carried out at BL9 at the DELTA Synchrotron (Technical University Dortmund). The experimental details and setup of the diffraction experiment can be found elsewhere. 26 The X-ray spot is about 1-2 mm 2 in diameter. The X-ray detector was a 2D MAR345 (marXperts). The X-ray energy was 20 keV. XRD patterns on the 15 Â 15 grid were recorded by mounting the sample on a computer-controlled ve-axis stage. For phase analysis the 2D patterns were integrated. Background subtraction was performed according to the algorithm by Sonneveld & Visser. 27 Fig. 1 shows the real space coordinates of each XRD measurement area. Phase region analysis was aided by machine learning algorithms on XRD patterns by using the approach of clustering similar XRD pattern by their geometric distance as described elsewhere. 28 Photoelectrochemistry Photoelectrochemistry (PEC) measurements were performed by means of an automated scanning droplet cell 14 with circular measurement areas of 1 mm diameter on a 16 Â 16 grid as shown in Fig. 1. Prior to each measurement the open circuit potential (OCP) was determined rst in the dark and then under illumination. A positive shi in open circuit potential (DOCP) is expected for a p-type material. 29 It was shown previously that the OCP and its shi correlate with the occurrence of different phases. 24 PEC measurements were performed in a three electrode setup with a Ag/AgCl/3 M KCl (210 mV vs. NHE) as reference electrode and a Pt wire as counter electrode in 0.1 M KH 2 PO 4 at pH 7. The applied potential (E appl ) was calculated against the reversible hydrogen electrode (RHE) for better comparability: E RHE ¼ 210 mV + E appl + (59 mV Â pH). The linear sweep was started at 923 mV (vs. RHE) with a negative slope to À50 mV at a scan rate of 1 mV s À1 . A 150 W Xe-Lamp (Hamamatsu) was used as light source and the power of the incident light was around 100 mW cm À2 . The lamp shutter was opened and closed for 5 s each.

Results and discussion
High-throughput XRD for combinatorial phase mapping By utilizing machine learning algorithms on XRD patterns obtained from gracing-incidence synchrotron measurements, 28 four major phase regions were identied in the Cu-Si-Ti-O materials library, marked by Roman numerals in Fig. 2. These regions exhibit a good correlation between changes in color and phase boundaries as shown by the RGB background image in    33 who report an enthalpy of formation of À42.2 kJ mol À1 for dioptase from CuO (tenorite) and SiO 2 (quartz).
The bright white color in the Si-O rich side of Fig. 2 may indicate a detached lm as no metal color (Au back contact) is visible.

Photoelectrochemical characterization
Visualization of the multidimensional dataset of linear sweeps voltammograms under chopped light illumination over the ternary composition spread is shown by color coding a ternary diagram with the photocurrent densities at two potentials as shown in Fig. 4. The photocurrent is calculated as the difference between the current density under illumination and in the dark aer equilibration.
The potential of 0.716 V vs. RHE at which the photocurrents are plotted in Fig. 4a was chosen as reference as it roughly corresponds to the photocurrent onset potential of the materials. The potential of 0.316 V vs. RHE in Fig. 4b is the potential at which the highest photocurrent density was observed. Apparently only regions at quasi-binary compositions of Cu-Ti-O and Cu-Si-O with Cu >50 at% and the fourth element (Si or Ti) <20 at% are photoactive.
In the following sections, these two regions are discussed as quasi-binary. For Ti-O and Si-O rich compositions, no photocurrents were observed which is attributed to a too large band gap for signicant visible light absorption. The Cu-rich corner of the quasi-binary Cu-Ti-O composition space at the le side of the ternary diagram in Fig. 4a and b shows highest photocurrent densities of up to À420 mA cm À2 at 0.316 V vs. RHE at high Cu concentrations of around 70 at%, Ti contents <30 at%, and Si contents <20 at%. Deviations from these apparently optimal compositions result in signicantly lower photocurrent densities. Especially the incorporation of Si into the Cu-Ti-O compositions results in a decay of photoactivity.
The dark current for Cu-rich samples (Cu >60 at%) started to increase at 500 mV vs. RHE signicantly, indicating lower stability as compared to Cu 57 Si 40 Ti 3 O x most likely due to the reduction of Cu 2+ ions (peak at 350 mV vs. RHE). For Cu 57 Si 40 -Ti 3 O x no reduction peak was observed at this potential suggesting that in Cu 57 Si 40 Ti 3 O x copper is stabilized, although the photocurrent is signicantly smaller as compared to Cu-rich phases (Cu >60%). However, optimization of the layer thickness, nanostructuring or decoration by co-catalyst materials for Fig. 3 Synchrotron-XRD for phase region IV near a nominal composition of Cu 50 Si 50 O x that is identified to be a mixture of dioptase and copper-polysilicate. This phase region exhibits promising PEC properties. Black stars denote substrate peaks from Au/Al 2 O 3 . The black indexes are corresponding to peaks from dioptase whereas the red indexes followed by the Roman numeral II denote those of copperpolysilicate. Fig. 4 Cu-Si-Ti-O quasi-ternary diagram with color-coded photocurrent density at a bias potential of (a) 716 mV (the potential close to the photocurrent onset potential) and of (b) 316 mV (the potential of highest photocurrent density of À420 mA cm À2 ).  were recorded (Fig. S1 and S2 ESI †). For both quasi-binary systems especially at high Cu concentration and low concentrations of Ti or Si, respectively, the dark current increased rapidly with increasing bias potential. The stability could be improved by decreasing the Cu-content or increasing the Ti or Si content, respectively. However, Cu reduction peaks are observed at Ti <20%. Comparing both quasi-binary material compositions, the addition of Si leads to more stable lms than the addition of Ti.
The photocurrent densities at applied potentials >0.650 V vs. RHE as well as the photocurrent onset potential is essentially the same for all selected measurement areas. All samples depicted in Fig. 5 show small overshoots within the 5 s measurement time under illumination (inset in Fig. 5), associated to low surface recombination of photogenerated charge carriers. The current increase at potentials <0.2 V vs. RHE was attributed to gas evolution.

Conclusions
Two new photocathode materials with compositions close to Si:Cu 75 Ti 25 O x and Ti:Cu 50 Si 50 O x were identied as light absorber materials with promising photocurrent densities and onset potentials on a combinatorial materials library analyzed by high-throughput methods. Both identied materials are ptype semiconductors that show slightly enhanced stability over Cu 2 O 11 and CuO due to comparably lower dark currents at potentials that match the reduction of Cu-cations. The two identied materials show promising photovoltages. Dioptase may contain crystal water in the narrow hollow Si 6 O 18 silicate rings, affecting its electronic properties. 30 Dioptase is assumed to be a promising material for further studies, as the comparably large hollow silicate rings might show catalytic 31 properties as here solely Si-O bonds are exposed to the electrolyte. Due to its complex crystal and electronic structure dioptase could be used as base material for the optimization of photocathodes by introducing additional dopants (e.g. Zr, Hf, Zn, Al, Nb). As dioptase as well as Cu-polysilicate consist of transition metal oxides with partially lled d-states classical DFT methods like PBE fail to conclusively address accurately the electronic structure of the semiconductor and suggest metallic behavior. 30 Therefore more accurate (i.e. DFT+U) calculations are needed to gain a deeper understanding of the material especially for optimization through doping.