Copper ternary oxides as photocathodes for solar - driven CO 2 reduction

: The sun ’ s energy, though free and virtually limitless, is a largely unexploited resource, as its conver sion into a storable form presents several technological challenges. A promising way of capturing and storing solar energy is in the form of “ solar fuels, ” in a process termed arti ﬁ cial photosynthesis. In a photoelectrochem ical ( PEC ) system, the reduction of CO 2 to carbon - based fuels is driven on the surface of an illuminated semicon ductor electrode. Through the decades, many di ﬀ erent classes of semiconducting materials have been studied for this purpose, to varying successes. Because of their cheap and abundant nature, semiconducting transition metal oxides are good candidates to realize this tech nology in an economic scale and have thus attracted con siderable research attention. In this review article, the progress achieved with a speci ﬁ c class of metal oxides, namely, the copper ternary oxides such as copper iron oxide and copper bismuth oxide, for PEC CO 2 reduction is examined. Although there have been signi ﬁ cant advances in terms of strategies to improve the e ﬃ ciency and stability of these materials, further studies are warranted to address the many challenges to PEC CO 2 reduction and solar fuel production.


Introduction
Solar energy is a free and clean resource abundant enough to supply the world's ever-growing needs [1]. However, due to its diffuse and intermittent nature, it needs to be stored in a transportable form to become a useful energy source [2,3]. In the previous decades, several approaches to harvesting solar power have been developed, to varying levels of technological readiness [3][4][5][6]. Among these technologies is the direct storage of the sun's energy in the chemical bonds of a "solar fuel," in a process resembling an artificial photosynthesis system [7][8][9][10].
Hydrogen is the simplest solar fuel that can be produced via the splitting of water [11]. However, because hydrogen is a gas at standard conditions, it needs to be compressed and/or liquefied to attain a practical volumetric energy density [12]. Liquid organics that can be produced from the reduction of carbon dioxide (CO 2 ), such as methanol and ethanol, are easier to handle, more applicable for direct use, and more compatible with the current energy infrastructure [4,5,8,9,12,13]. From this perspective, the solar-driven reduction of CO 2 to fuels as well as other useful chemicals is considered an attractive route for solar energy harvesting and for closing the anthropogenic carbon cycle [12][13][14][15][16].
Before a fully operational system capable of scalable and long-term production of solar fuels is achieved, the thermodynamic and kinetic challenges to the multi-electron reduction of CO 2 need to be addressed [14,17,18]. CO 2 , with its two C]O bonds and linear symmetry, is an extremely stable molecule [19][20][21]. Its endergonic activation requires a significant energy input due to a structural distortion accompanying electron addition [22][23][24]. Furthermore, the many reaction pathways of the activated CO 2 molecule result in a variety of products, such as formic acid (HCOOH), carbon monoxide (CO), formaldehyde (HCOH), methanol (CH 3 OH), and methane (CH 4 ), making product separation difficult [14][15][16][17][18]. The reduction potentials of CO 2 to these products in aqueous solution, where CO 2 reduction is usually conducted, are given in Table 1. The proximity of the reduction potentials with each other and with the hydrogen evolution reaction (HER) as well as the multiple proton-coupled electron transfers with different kinetic barriers for each step make CO 2 reduction a rather complicated process [25][26][27][28][29].
For PEC CO 2 reduction to be realized in a practical scale, the photoelectrode should be made of low-cost, earth-abundant, and non-toxic materials [3,35,42]. Copper-based oxides, generally p-type semiconductors demonstrating a broad absorption of light, are good candidates as photocathodes in this context [43][44][45][46][47][48]. Because ternary and multinary oxides provide greater flexibility in tuning their properties compared to their binary counterparts, copper ternary oxides have recently gained attention as photocathodes for CO 2 reduction [49][50][51]. Although research is still lacking, there have already been promising results, making it timely to review the current progress. There have been previous reviews on copper ternary oxides, but none so far has focused on their application for PEC CO 2 reduction [45][46][47][48][49][50][51].
This article provides first an overview of the PEC cell and the motivation for the investigation of copper ternary oxides as materials for PEC CO 2 reduction. Then, the latest efforts to improve these materials towards high efficiency and stability photocathodes are examined. Finally, insights on the way forward for these materials, as well as for PEC CO 2 reduction in general, are offered.

.1 Overview of PEC cell
The conversion of solar energy to chemical fuel in a solidstate device requires the linking of light absorption and electrochemical functionalities [3,5,6,15,25]. One configuration is a coupled photovoltaic (PV)-electrolysis system, where the PV module performs the light absorption to produce an electric potential that will drive an electrochemical reaction in the electrolyzer. In another configuration, both functionalities are integrated in a single device based on a semiconductor and the solar energy is directly converted into the energy of the redox products. Many authors claim that the latter setup is more favorable from a fabrication and cost standpoint since there will be less stringent material requirements and fewer physical components, although there is a range of intermediate configurations where the device functionalities are partially decoupled [2,5,6,15,27,29,35].
In brief, a semiconductor is a material that possesses a small or modest region (typically between 1-3 eV) of forbidden energy levels called the band gap. When a semiconductor absorbs photons more energetic than its band gap, one of the possible interactions will be for an electron from its valence band to be excited to the conduction band, leaving a hole in the valence band. Following excitation, the photogenerated charges must effectively separate before recombination occurs. This separation can be facilitated by an electric field, such as that formed at the interface of the semiconductor and an electrolyte solution, transports minority charge carriers to the surface and the majority charge carriers to the bulk (or back contact) [52][53][54]. Accordingly, the role of the semiconductor is primarily to absorb an incident photon, generate an electron-hole pair, and facilitate its separation and transport. Interfacial charge transfer to the adsorbed species to drive the redox reaction is a separate function which may be performed by the semiconductor itself or by a co-catalyst [14][15][16]55]. A more comprehensive treatment of semiconductor physics and the electrochemical behavior of the semiconductor-electrolyte interface under illumination can be found in refs. [52][53][54]56,57].
The semiconductor may take the form of micrometerto nanometer-sized particles suspended in solution ( Figure 1a) or a photoelectrode, usually a thin film deposited on a transparent conducting oxide substrate in contact with the electrolyte (Figure 1b) [14,32,34]. In the wireless particulate system (commonly referred to as a "photocatalytic" cell), both reduction and oxidation reactions take place on each particle, but at different sites [14,32,58]. In contrast, the photoelectrode configuration allows for the spatial separation of the reduction and oxidation reactions and for the application of an external bias [33,34]. This system is referred to as a PEC cell (or photoelectrocatalytic/photoelectrosynthetic) as it resembles the two-or three-electrode cell construction for electrochemical systems [14][15][16][29][30][31][32][33][34][35]40,41,53]. It is important to note here that catalysis, by definition, involves a thermodynamically favorable reaction (ΔG < 0), where the catalyst only serves to lower the activation energy (E A ) of the reaction. Thus, the use of the terms "photocatalytic" or "photoelectrocatalytic" in the context of CO 2 reduction, a thermodynamically uphill reaction (ΔG > 0), is imprecise, as pointed out by several authors [5,25,27,59]. In this case, an energy input, which may be provided by incident photons, is used to overcome the thermodynamic barrier ΔG rather than to lower the kinetic overpotential E A . However, the semiconductor may be considered catalytic in the sense that it can also lower E A by providing active sites for faster interfacial kinetics. Thus, it is also possible for the reaction to occur at an "underpotential," where the light energy is utilized in such a way that both ΔG and E A are compensated, resulting in a net negative ΔG of the reaction [60,61].
PEC cell constructions vary depending on the number of photoactive materials and reaction compartments [16,35,38,41,42]. In a single photoelectrode configuration, a photocathode (or photoanode), based on a p-type (n-type) semiconductor, drives the reduction (oxidation) half-reaction, with the other half-reaction taking place on a dark anode (cathode), usually a metal electrocatalyst (such as platinum). The basic operation of a three-electrode PEC cell based on a photocathode is illustrated in  Figure 1: (a) Schematic of a photocatalytic cell, where semiconductor particles are suspended in a CO 2 -containing electrolyte. Each particle acts as tiny photoelectrolysis cells where both reduction and oxidation reactions take place. (b) Schematic of a typical two-compartment, three-electrode cell construction of a photocathode-driven PEC cell for CO 2 reduction. Absorption of light (hν) more energetic than the band gap (E g ) of the photocathode results in the generation of electron-hole pairs, separated by the electric field (as shown by the bending of the bands) at the semiconductor-electrolyte interface. Electrons are transported to the surface, where they participate in the reduction of CO 2 , while the holes migrate to the back contact and to the counter electrode (CE) via the external circuit, where they are consumed in the oxidation reaction (in aqueous electrolytes, the oxygen evolution reaction), completing the flow of current in the cell. The addition of a reference electrode (RE) allows the potential of the WE (or more specifically, the Fermi level of the photocathode) to be controlled via a potentiostat, which is useful in the investigation of PEC systems. (Adapted with permission from ref. [14]. Copyright © 2015, American Chemical Society). Figure 1b. A dual photoelectrode configuration is also possible, where the photocathode and photoanode may be in tandem (both semiconductors on a single substrate) or separately illuminated, provided that the semiconductors have matching band gaps and band edge positions [15,35]. Although the ultimate goal for PEC applications is an unbiased cell made of dual photoelectrodes, the photocathode and photoanode can be developed independently [35]. In fact, research on photoanode materials has progressed more and has been the subject of the majority of previous reviews [25,[36][37][38][39]].

Photocathode materials for CO 2 reduction
PEC CO 2 reduction was first reported by Halmann, who studied a p-GaP photocathode irradiated with UV light in a CO 2 -saturated aqueous solution, yielding formic acid (and smaller amounts of formaldehyde and methanol) [62]. Much of the early work on PEC CO 2 reduction involved covalent semiconductors, such as p-GaP [63][64][65][66], p-GaAs [63,[66][67][68][69], p-InP [66,67,70,71], p-CdTe [70,[72][73][74], and p-Si [66,75], being modeled after the several efficient liquidjunction solar cells developed during that period. However, even if their band gaps are suitable for visible light absorption, these semiconductors required extremely negative potentials to effect CO 2 reduction, implying that little to no solar energy was being stored in the products [32]. Modification techniques such as addition of metal co-catalyst [76][77][78][79], polymer coating [80,81], and use of solution mediators [82][83][84][85][86] were needed to improve their catalytic activity as well as their stability against corrosion in aqueous electrolyte. These limitations, as well as their high cost and complicated preparation methods, have motivated researchers to explore new classes of materials for PEC CO 2 reduction. For PEC technology to become economically viable, cheap and earth-abundant materials that can be synthesized using simple and low-cost methods must be used, for which semiconducting transition metal oxides are a good candidate [35,46,49,87,88]. Transition metal oxides are a class of unique materials that exhibit a variety of structures and properties. Depending on the nature of the metal d-orbital, the metal-oxygen bonding can vary between nearly ionic to highly covalent or metallic, resulting in a wide range of electronic behavior, from insulators to semiconductors to superconductors [89,90]. Semiconducting behavior of metal oxides result from intrinsic point defects acting as donor or acceptor states: oxygen vacancies for n-type oxides, while metal vacancies for p-type oxides [90]. Although the first report [91] on the use of semiconducting metal oxides for solar-driven CO 2 reduction came around the same time as Halmann's, research interest had only expanded in the succeeding years, with the advent of nanotechnology and modern computational capabilities [92]. As mentioned earlier, photoanode materials, in particular the most well-studied TiO 2 , have received more attention than photocathode materials [25,[35][36][37][38][39]49]. This is due in part to the fact that there are fewer p-type oxides because of the localized nature of the oxygen 2p orbitals in the valence band of metal oxides that results in large effective hole masses and easier hole compensation [93]. Even so, the small band gap p-type oxides (and thus suitable as photocathodes), namely, Cu 2 O (∼2.0 eV) and CuO (∼1.5 eV), are susceptible to photocorrosion in aqueous electrolyte [43][44][45][46][47][48]. Strategies to enhance the stability and PEC activity of Cu 2 O and CuO have been demonstrated, such as nanostructuring [94][95][96], addition of co-catalyst [97][98][99][100][101][102][103], deposition of protective overlayers [104][105][106][107], and heterojunction formation [61,[108][109][110][111][112][113][114][115][116], to varying successes. The performance of Cu 2 O-and CuO-based photocathodes for CO 2 reduction has been reviewed previously [45,46,48] and will no longer be discussed in this article, but a summary of previous findings is presented in Table 2.
Another promising strategy to address the limitations of binary metal oxides is by expanding the search for photoelectrode materials to multinary metal oxides [35,42,46]. The addition of one or more cations to form more complex oxides provides greater flexibility for tuning the optical and electronic properties [42,45]. For ternary and quaternary oxides, more than 8,000 and 700,000 combinations are possible, respectively [49]. With this immense number, it is likely that a semiconductor with the best combination of properties for PEC CO 2 reduction may be found among this group. Like the case with binary oxides, much of the success with ternary oxides has been achieved with photoanodes, specifically BiVO 4 [35,46,49,51]. There is still a limited understanding of ternary oxide photocathodes and, even more, their application to PEC CO 2 reduction.
Several copper ternary oxides exhibit p-type behavior, a narrow band gap capable of visible light absorption, and an adequate conduction band potential for CO 2 reduction, as shown in Figure 2. In addition, they appear to be more stable than their binary counterparts, as the additional cation introduces modifications to the electronic structure and to the nature of band transitions that lead to inhibition of Cu self-reduction [49,50]. Along with their inexpensive and nontoxic nature, these copper ternary oxides are good candidates as photocathodes for FE -Faradaic efficiency, STC -solar-to-carbon efficiency.
CO 2 reduction [51]. Still, several challenges limit the actual performance of copper ternary oxide photocathodes. First is the poor charge separation and transport due to low carrier mobility, a common problem among transition metal oxides [87]. Compared to covalent semiconductors, transition metal oxides are more prone to carrier localization or "self-trapping" due to a strong electron-lattice interaction which causes local lattice distortions called the small polaron. Carrier transport then occurs via a small-polaron hopping mechanism, which is characterized by a very small drift mobility due to its thermally activated nature [117,118]. Next, with the increasing complexity of metal oxides, preparation of stoichiometric, highquality films becomes more challenging. Even a small percent of sub-stoichiometry can result in large number of defects that can act as recombination centers and diminish the photoactivity [49,51]. Lastly, although there are encouraging results, the long-term stability needed for PEC applications is yet to be demonstrated [50,51].

Copper ternary oxide photocathodes
In the following sections, the current progress with copper ternary oxides is summarized and critically examined. Indepth focus is given to the relatively more studied oxides, copper iron oxide (CuFeO 2 ) and copper bismuth oxide (CuBi 2 O 4 ), discussing the relation between their crystal and electronic structure, their various preparation methods, the strategies applied to address the aforementioned limitations, and their performance towards PEC CO 2 reduction.

CuFeO 2
CuFeO 2 belongs to a group of Cu(I)-based oxides with the delafossite structure and the general formula CuMO 2 , where M is a metal cation in the trivalent state (M = Al, Co, Cr, Fe, Ga, and Rh) [119]. The delafossite structure comprises of two alternating layers: a planar layer of Cu cations in a triangular pattern and a layer of edge-sharing MO 6 distorted octahedra, where each oxygen is coordinated by one Cu + and three M 3+ cations (Figure 3a and b). Cu and O form linear arrays parallel to the c-axis. Depending on the stacking pattern of the MO 6 layers, delafossites form two polytypes, namely, 3R and 2H, that crystallize in the rhombohedral (R m 3 ̅ ) or hexagonal ( / P mmc 6 3 ) space group, respectively [119,120]. As the band gap of copper delafossites varies widely depending on the M cation, this group of compounds has found a range of electronic applications, including dye-sensitized solar cells, PV and PEC devices, and transparent conducting oxides [45,121].
Among the copper delafossites, CuFeO 2 possesses the narrowest optical band gap   suitable photocathode material for PEC applications [122][123][124]. It exhibits a p-type conductivity, which is believed to arise from native Cu vacancies and O interstitials in its structure [124][125][126]. Electronic structure calculations show that the valence band maximum in CuFeO 2 is dominated by Cu 3d states, with some degree of hybridization from O 2p states, while the conduction band minimum consists of nonbonding Fe 3d states [127]. The lowest optical transition involves an electron excitation from O 2p states to Fe 3d states [128]. This excitation is followed by an ultrafast thermalization of photogenerated holes from O 2p to Cu 3d valence band states, which may be visualized as an electron transfer from Cu 3d states to backfill holes in the deeper O 2p states [129] (Figure 3c). The hole thermalization process facilitates charge separation in the lattice and suppresses electron-hole recombination (because there is no covalent bonding between the Cu and Fe atoms), and hence, is considered responsible for the greater carrier lifetime, improved stability, and superior photoactivity of CuFeO 2 relative to its parent binary oxide, Cu 2 O [126,129]. CuFeO 2 photocathodes have been prepared using a variety of methods, including hydrothermal synthesis [121,130], electrodeposition [122,126], solid-state synthesis [123,131], sol-gel method [132,133], spray-pyrolysis [134], and reactive co-sputtering [127,135].
Based on its band gap, the theoretically achievable photocurrent of CuFeO 2 is 15 mA·cm −2 (based on AM1.5 illumination), but reported photocurrents are much lower (up to 2.5 mA·cm −2 ) [124]. Several studies have been conducted to determine the major limiting factor to the PEC performance of CuFeO 2 . Time-resolved microwave conductivity measurements found that charge carriers in CuFeO 2 are relatively long-lived (as compared to Cu 2 O or α-Fe 2 O 3 ), with a time constant of 200 ns [124]. On the contrary, a very short free carrier lifetime (in the order of picoseconds) was measured via time-resolved optical spectroscopy [127]. The results of these two studies indicate that free carriers generated upon illumination quickly self-trap, forming the observed long-lived excited state that is postulated to be electron thermalizing as a surface-trapped small polaron in the Fe 3d conduction band state [136]. As mentioned in the previous section, a conduction mechanism based on small polaron hopping will result in small carrier mobilities. Another factor is the presence of a high density of surface states on CuFeO 2 , arising from a 10 nm hydroxide or oxyhydroxide surface layer [124]. These surface states act as electron traps, causing severe Fermi level pinning and drastically reducing the photovoltage. Thus, preparation of high-quality films as well as other strategies to address the low carrier mobility and to passivate or remove the surface states is critical to improve the PEC performance of CuFeO 2 .
To increase its conductivity, CuFeO 2 can be doped with a divalent cation (such as Mg 2+ , Ni 2+ , and Sn 2+ ), which substitutes with either one Fe 3+ or three Cu + to contribute additional holes [123,133,137]. Indeed, the hole concentration was found to increase with Mg doping level, but only until a certain threshold where the conductivity of CuFeO 2 shifts to n-type, possibly due to charge compensation by oxygen vacancies, an n-type defect [131,133,138]. A 0.05% Mg-doped CuFeO 2 was studied for PEC CO 2 reduction, yielding photocurrents of up to 1 mA·cm −2 in CO 2saturated 0.1 M NaHCO 3 (pH 6.8) [123]. Formate was observed as the main reduction product (although no quantification was made) in bulk electrolysis experiments performed under blue LED light illumination (470 nm, 2.1 mW·cm −2 ). However, no comparison in the CO 2 reduction activity between the doped sample and an undoped one was made. Alternatively, carrier concentration can also be increased via oxygen intercalation, which can be achieved via conventional thermal annealing [132] or hybrid microwave annealing [139] of as-synthesized CuFeO 2 films. The O interstitials occupy the Cu planes, where there is a large enough space compared to the Fe octahedrons, of the delafossite structure [140]. In one study, the increase in photocurrent of the oxygen-intercalated samples versus as-synthesized CuFeO 2 matched that of the acceptor density, which the authors ascribed to an improved charge separation efficiency due to the higher carrier conductivity [132]. On the contrary, another study found that O interstitials have a negative effect on PEC activity of CuFeO 2 [126]. Because the O interstitials shift the Cu 3d band to lower energy and the O 2p band to higher energy, the valence band maximum changes from Cu 3d to O 2p upon introduction of the O interstitials, precluding the hole thermalization kinetics responsible for the increased carrier lifetimes in CuFeO 2 , as discussed above. These conflicting findings suggest the existence of a more complex underlying defect chemistry in CuFeO 2 that may require further elucidation. To address the presence of surface traps in CuFeO 2 and the slow interfacial kinetics, the addition of a co-catalyst or a protective overlayer has been performed. Noble metals (such as Pt, Ag, and Au) [141,142] and layered double hydroxides [139,143] were employed as co-catalyst, all resulting in an increased photocurrent. Au nanoparticles deposited on diamond-like carbon (DLC) was applied as overlayer for CuFeO 2 [142]. In this strategy, the Au nanoparticles contribute to an enhanced light absorption through its surface plasmon resonance whereas the DLC support prevents agglomeration of the Au nanoparticles and decreases charge trapping surface states on CuFeO 2 . The authors note that further work is needed to optimize the morphology and thickness of the overlayer.
Other approaches to improve the performance of CuFeO 2 include formation of novel nanostructures that enhance light absorption and charge separation. CuFeO 2coated amorphous SiO 2 microspheres were assembled on an fluorine-doped tin oxide (FTO) substrate to form a monolayer opal photocathode with glass-like transparency ( Figure 4a) [144]. Due to the two-dimensional photonic architecture, the photocathode demonstrated self-light harvesting. Despite the enhanced light absorption, the incident photon-to-current efficiency (IPCE) was low due to the very small thickness (∼2 nm) of the CuFeO 2 shell that cannot support a depletion layer. To improve the charge separation, the CuFeO 2 -coated microspheres were coated with an outer shell of CuAlO 2 , forming a heterojunction between the two oxides [145]. It was initially shown that a host-guest architecture for CuFeO 2 and CuAlO 2 (a thin film of CuFeO 2 on a highly transparent CuAlO 2 scaffold) facilitates hole transfer, while blocking electron transfer from the absorber to the scaffold, reducing charge recombination and attaining higher photocurrents [146]. An inverse opal structure was also   Figure 4b) [143]. In this structure, light absorption was enhanced owing to multiple internal light scattering in the macropores. The inverse opal structure also shortened the electron path to the electrolyte, resulting in a two-fold increase in photocurrent versus that of a planar CuFeO 2 . In another study, Nb-doped TiO 2 nanotubes were employed as substrate for CuFeO 2 , creating a p-n composite photocathode for the PEC reduction of CO 2 [147]. Nb doping improved the heat stability of the TiO 2 nanotubes, so that the structure can retain its integrity after the calcination step in the formation of CuFeO 2 . The authors claimed that a Z-scheme heterojunction was formed, where the weaker photogenerated electrons from TiO 2 recombined with the holes from CuFeO 2 , allowing the more reductive CuFeO 2 electrons to reduce CO 2 . Formaldehyde and ethanol were the CO 2 reduction products detected after photoelectrolysis carried out in CO 2 -saturated 0.1 M NaHCO 3 electrolyte under a 250 W Xe lamp illumination, with ethanol dominating at higher bias potentials.
Several reports have found that composites of CuFeO 2 with binary copper oxides (Cu 2 O or CuO), typically resulting from nonstoichiometry of precursors during preparation, perform better than phase-pure CuFeO 2 [127,[148][149][150][151]. In particular, CuFeO 2 /CuO photoelectrodes produced much higher photocurrents than CuFeO 2 or CuO alone [127,148]. These studies hypothesize that a type II heterojunction forms between the CuFeO 2 and CuO phases. This heterojunction is then responsible for an increased carrier lifetime and presumably, to the observed improved photoactivity of the mixed-phase photoelectrodes. In a proposed scheme ( Figure 5), photogenerated electrons from the higher E cb of CuFeO 2 can migrate to CuO, whereas those from the lower E cb (an optically forbidden gap) recombine with photogen-erated holes from CuO, which cannot effectively perform water oxidation as compared to holes from CuFeO 2 [152]. This internal recombination allows for an improved charge separation.
Based upon these findings, an unbiased CuFeO 2 / CuO-Pt cell under circumneutral pH was devised [148]. The cell produced formate for over a week, with a solarto-formate efficiency of 0.7-1.2% and Faradaic efficiency of 90%. The CuFeO 2 /CuO electrode, synthesized via electrodeposition, had a bulk Cu/Fe atomic ratio of 1.4 (measured via inductively coupled plasma-mass spectrometry) and a double-layered structure where CuFeO 2 was located in the bottom region, while CuO was uniformly distributed throughout, as determined via transmission electron microscopy-energy-dispersive X-ray spectroscopy. Thus, CuO was segregated as a secondary phase during calcination of the precursor in the presence of air. Notably, the authors did not detect other products (or possibly, were below detection limits), especially H 2 which is known to compete with CO 2 reduction. For comparison, single phase CuFeO 2 (Cu/Fe = 1.06) and CuO electrodes were tested under identical experimental conditions. Formate production rates were much lower, and more importantly, O 2 evolution in the Pt counter electrode was observed in neither. In a follow-up study, they investigated the stability of the CuFeO 2 /CuO photoelectrode for 7 days of PEC testing [150]. X-ray diffraction (XRD) analysis of the used samples showed a decrease in the intensity of CuFeO 2 and CuO peaks, while new peaks ascribed to Cu 2 O were found, indicating occurrence of cathodic photocorrosion. Reannealing of the used electrodes restores the performance similar to that of the as-synthesized samples. In another study, the effect of varying the Cu/Fe ratio on the selectivity of the CO 2 photoreduction process was investigated [149]. The authors found that decreasing the Cu/Fe ratio (more Fe-rich) favors the formation of acetate, a reduction product with C-C coupling. However, a consequence is that the Faradaic efficiency lowers (Figure 6a and b). Because minimal H 2 evolution was observed in all Cu/Fe ratios, the authors attributed the Faradaic efficiency loss to self-reduction of the photocathode. X-ray photoelectron spectroscopy (XPS) of the tested samples revealed almost no surface Fe, even for the most Fe-rich sample. Correspondingly, acetate formation is observed to subside after 10 min of reaction, revealing that surface Fe atoms act as active sites for acetate formation [149].
To provide a theoretical understanding of CO 2 reduction on CuFeO 2 , the mechanism of CO 2 adsorption and activation on CuFeO 2 surface was investigated [153]. In this study, the authors found that the defect-free (011) surface was inert towards CO 2 adsorption but that oxygen vacancies result in a negative charge accumulation on nearby Fe atoms, consequently serving as potential active sites for adsorption. Oxygen vacancies on many metal oxides are well known to play a role in CO 2 adsorption [154]. In another study, the effect of the heterogeneity of the photoelectrode on the selectivity of the PEC CO 2 reduction was investigated [152]. In their DFT calculations, the authors employed two models, a homogeneous structure (HMS) and heterogeneous structure (HTS). The former was based on a homogeneous distribution of Cu, Fe, and O in the entire film (40% CuFeO 2 and 60% CuO), while the latter was based on a CuO layer (3 layers) on top of CuFeO 2 (2 layers). Bader charge analysis showed that in both the HMS and HTS models, Cu atoms were more enriched with electrons than the surrounding Fe or O atoms, indicating that Cu atoms serve as active sites for the electrophilic CO 2 molecule. Further analysis showed that monodentate coordination of an O atom of CO 2 to a Cu site on the HTS surface was the most kinetically preferred pathway for formate production. Hence, these findings can be correlated to the higher rate of formate production with the heterogeneous CuFeO 2 /CuO photoelectrodes. Nevertheless, the surface Fe atoms can also act as electron rich sites; however, adsorbed CO 2 on Fe sites are preferentially reduced to surface-bound *CO or *C species that blocks C-H bond formation (a prerequisite to formate production) [155]. This can account for the acetate formation observed in Fe-rich photoelectrodes [149].

CuBi 2 O 4
CuBi 2 O 4 ("kusachiite") is a p-type oxide with a tetragonal crystal structure (space group P4/ncc). It comprises square planar [CuO 4 ] 6units stacked along the c-axis in a staggered manner, with Bi 3+ ions positioned between the stacks and connected to six O 2ions with three different bond distances (Figure 7a) [156]. Accordingly, a c-axis projection of the crystal structure shows straight channels formed around Bi 3+ ions (Figure 7b) [157]. The isolated [CuO 4 ] 6stacks, which are unlike the edge-sharing oxygen octahedra or tetrahedra commonly found in metal oxides, define the unique crystal structure of CuBi 2 O 4 and play a significant role in its electronic properties [158]. Initially identified as a promising photoactive material in a highthroughput screening study, CuBi 2 O 4 possesses an optical band gap of 1.5-1.8 eV, making it a good photocathode for PEC devices [157,159]. Unlike in CuFeO 2 , the Cu ions in CuBi 2 O 4 adopt a +2 oxidation state and thus, an open shell 3d 9 configuration that allows intra-atomic electronic transitions [160]. Initially, the valence and conduction band edges of CuBi 2 O 4 were considered to be primarily of O 2p and Cu 3d character, respectively. Cu vacancies, the most probable defect in CuBi 2 O 4 , introduce states near the valence band edge that gives rise to its p-type behavior [161]. However, a recent first principle and comprehensive spectroscopic characterization study revealed that the valence band maximum and conduction band minimum arise from the occupied and unoccupied spin states, respectively, of the Cu 3d x²-y² (hybridized with O sp 3 orbitals) that formed from the splitting of Cu 3d levels in the square-planar crystal field of the [CuO 4 ] 6units [157]. The orbital composition of the valence and conduction bands of CuBi 2 O 4 is depicted in Figure 8a. Owing to the strong hybridization between Cu 3d and O sp 3 orbitals, Cu 3d character is present across the whole valence band up to the maximum. Contrary to initial reports that the valence band maximum has a significant Bi 6s character, Bi 6s only occurs deep into the valence band. On the other hand, beyond its minimum which consists of the unoccupied Cu 3d x²-y² spin state, the conduction band is predominantly of Bi 6p character [157,160].   [161]. While the Cu d-d transition accounts for the lowenergy visible light absorption of CuBi 2 O 4 , it does not significantly contribute to the photoresponse, as localized d-d transitions are known to be less efficient in converting photons to photocurrent than O 2ligand to metal cation charge transfer (LMCT) transitions in openshell metal oxides [157]. This can very well explain why even though CuBi 2 O 4 exhibits an absorption onset at low energies (1.5-1.8 eV), the IPCE and absorbed photon-tocurrent efficiency become practically significant only at higher energies (1.8-2.25 eV), as reported in many experimental studies [156,[162][163][164][165][166][167][168]. Meaningful photocurrent generation in CuBi 2 O 4 requires the absorption of higher energy photons that brings about the O 2p to Cu 3d LMCT transition (or O 2p to Bi 6p at even higher energies) [160].

Consequently, the lowest-energy optical absorption in CuBi 2 O 4 corresponds to a Cu d-d excitation, not the O 2p to Cu 3d charge transfer transition initially proposed
Due to the nature of the optical transitions, the photogenerated electrons and holes are confined in the [CuO 4 ] 6units that are isolated from each other, as mentioned above (Figure 8b). Consequently, there is a lack of DOS overlap among [CuO 4 ] 6units that would serve as conduction paths for the minority and majority carriers. This results in a localization of electrons at discrete Cu sites, and hence, to polaron formation that is stabilized in part because of the attainment of a transient stable electronic configuration (from Cu 2+ 3d 9 to Cu + 3d 10 ). Consequently, electronic conduction in CuBi 2 O 4 proceeds via small-polaron hopping between [CuO 4 ] 6units along the c-axis, as the nearest cation to Cu is another Cu along the c-axis (Figure 8c) [156,158]. Similar to previous discussions, this small-polaron hopping conduction mechanism is responsible for the poor charge carrier transport in CuBi 2 O 4 . Based on its band gap of 1.5-1.8 eV, the maximum theoretical photocurrent for CuBi 2 O 4 under AM1.5 illumination is 19.7-29.0 mA·cm −2 [164]. Actual photocurrents are therefore expected to be lowered by the inferior charge transport, and even more by the sluggish CO 2 reduction interfacial kinetics [157]. Therefore, strategies to address the inherent limitations of CuBi 2 O 4 are requisite for CuBi 2 O 4 to be a suitable photocathode for PEC applications.
To achieve superior PEC performance, preparation of high quality CuBi 2 O 4 photocathodes is of foremost importance. CuBi 2 O 4 films have been prepared using simple techniques such as electrodeposition [162,166,169], dropcasting [156,170,171], and spin-coating [167,168]. However, these techniques usually resulted in incomplete surface coverage, which can reduce the photoactivity because exposed FTO can act as sites for back reactions [172,173].
In addition, nanoscale phase impurities that are potentially deleterious to the integrity of the photoelectrode and that may not be detected by XRD can also be formed [174]. More advanced techniques such as spray pyrolysis [173], reactive co-sputtering [157,175], and pulsed laser deposition [176,177] yielded highly phase-pure, homogeneous CuBi 2 O 4 films. Furthermore, phase-purity was better achieved with rapid thermal processing (RTP) (10 min at 650°C) than conventional furnace heating (CFH) (72 h at 500°C) of pulsed-laser-deposition-prepared CuO/Bi 2 O 3 precursor [178]. This finding was attributed to nucleation and grain growth processes during heating (Figure 9a). The high heating rate in RTP rapidly brings the CuO/Bi 2 O 3 precursor to a high-temperature steady state that enables fast diffusion of Cu ions into the Bi 2 O 3 layer. This yields a homogeneous distribution of CuBi 2 O 4 nucleation sites after 5 min; rapid grain growth then results in the formation of a singlephase CuBi 2 O 4 . In contrast, the slower heating rate in CFH causes grain growth to occur even while Cu diffusion is still taking place. The resulting CuBi 2 O 4 grains impede further diffusion so that phase transformation of the precursor is incomplete.
Carrier diffusion length in CuBi 2 O 4 was found to be in the order of 10-60 nm, as determined from time-resolved microwave conductivity measurements [156]. Compared to the light penetration depth of 244 nm (for a wavelength of 550 nm), the short carrier diffusion length leads to excessive electron-hole recombination, severely limiting the PEC performance [156,164]. As with other photoelectrodes, the diffusion pathway of charge carriers can be shortened through nanostructuring, which may also increase internal light scattering and active sites for interfacial reactions [34][35][36][37][38][39][40]. Recently, a novel template-assisted synthesis strategy was devised to prepare a CuBi 2 O 4 film with a nanodendritic structure, with the trunks and subbranches having a radius around 90-150 nm (Figure 9b) [179]. The nanodendritic CuBi 2 O 4 demonstrated a photocurrent (measured in N 2 -saturated 0.1 M Na 2 SO 4 electrolyte under AM1.5 G illumination) twice that of a planar film, which the authors ascribed to a shorter diffusion length of the CuBi 2 O 4 /electrolyte interface as well as a higher specific surface area. To improve its photostability, the nanodendritic CuBi 2 O 4 was conformally coated with a TiO 2 protective overlayer.
Enhancing hole transport to minimize electron-hole recombination can significantly improve the performance of CuBi 2 O 4 photocathodes. To this end, doping and use of hole transport layers have been investigated. Doping of CuBi 2 O 4 with Ag resulted in an increase in hole concentration, as substitution of Ag + with Bi 3+ is charge-compensated by free holes (no Ag impurities were formed) [172].
Owing to the enhanced hole transport, Ag-doped CuBi 2 O 4 exhibited higher photocurrent and better stability (against anodic photocorrosion) than undoped CuBi 2 O 4 [172]. Hole transport layers (HTL) address the mismatch between the work function of FTO (a degenerately doped n-type semiconductor), the most commonly employed conducting substrate, and the Fermi level of the semiconductor photoelectrode. Due to such mismatch, a Schottky barrier may form at the FTO/CuBi 2 O 4 interface, impeding hole collection in the back contact and enhancing carrier recombination [180]. Gold [162] and Cu-doped NiO [180] were found to be effective HTLs for CuBi 2 O 4 . The energy band alignment of the FTO/Cu:NiO/CuBi 2 O 4 photocathode is shown in Figure 10a.
Another strategy to improve the PEC cell performance of CuBi 2 O 4 is the addition of conformal protective overlayers to address its photocorrosion. The function of the overlayer should not be limited to passivation of surface states and blocking of CuBi 2 O 4 /electrolyte contact; ideally, the overlayer should also act as an electron-selective contact that enhances charge separation [181,182]. Like in HTLs, addition of an overlayer will require a precise engineering of the interface such that energy levels are properly aligned and that no recombination traps are formed [175]. To this end, the multilayer scheme initially developed for Cu 2 O [104] was adopted for CuBi 2 O 4 with CdS as the buffer layer between CuBi 2 O 4 and the TiO 2 protection layer [181]. In another study, high throughput methodology was employed to identify a suitable multinary metal oxide overlayer for CuBi 2 O 4 [175]. An optimized Cu 1.5 TiO z overlayer was well-matched with CuBi 2 O 4 , forming a heterojunction that favored electron transport to the surface (Figure 10b). A unique, facile approach to introduce a protective overlayer on CuBi 2 O 4 involves its pre-irradiation to form reduced Cu states on its surface [182]. The surface states induce an additional downward band bending upon illumination that contributes to improved charge transport and hence PEC cell performance (Figure 10c).
Tailoring the Cu/Bi ratio to deliberately introduce phase impurities has also been applied as a strategy to achieve better PEC cell performance of CuBi 2 O 4 . Similar to CuFeO 2 , nonstoichiometric CuBi 2 O 4 films likely contain binary oxide phases, that is, CuO if Cu/Bi >1/2 or α,β-Bi 2 O 3 otherwise, since CuBi 2 O 4 is a line compound (there is no Cu x Bi 2−x O 4 solid solution) [174]. Because CuO and Bi 2 O 3 are both photoelectrochemically active, the electronic and optical properties of these nonstoichiometric films can be tuned with the Cu/Bi ratio [163]. Since CuO has a narrower band gap (∼1.5 eV), the optical band gap redshifts and the films become progressively darker with higher Cu/Bi ratio [183]. Also, with higher Cu/Bi ratio, the flat band potential shifts to less positive values (away from valence band) because of a lower Cu vacancy concentration (recall that Cu vacancies introduce p-type behavior of CuBi 2 O 4 ) [164,167,175]. Taking advantage of this concept, a photocathode with a Cu/Bi ratio gradient across its thickness was devised by placing into contact three films of varying stoichiometries (Cu/Bi = 1/3, 1/2, and 1) and thus, varying Fermi levels (Figure 11a) [164]. When placed in contact, the Fermi levels of the three films equilibrate by distributing the free holes, so that the conduction and valence bands bend at each interface (Figure 11b). The forward-gradient film, that is, with Cu/Bi ratio increasing from substrate side to electrolyte side (or Cu vacancy concentration decreasing), exhibited the correct band bending where photogenerated holes are transported to the FTO back contact, while the electrons are transported to the surface (Figure 11c), and resulted in the best photoresponse versus a homogeneous and a reverse-gradient film.
Indeed, composites of CuBi 2 O 4 with its secondary oxide CuO, both as randomly intermixed [163,168] or as separate layers [184,185], have demonstrated better PEC cell performance than CuBi 2 O 4 or CuO alone. The simultaneous formation of both CuBi 2 O 4 and CuO phases from a common precursor ensures a sufficient interfacial contact between the two phases, a requirement to establish an effective heterojunction [163,185]. However, the exact alignment of the heterojunction is still unclear. In one study, a layered CuO/CuBi 2 O 4 was synthesized where CuBi 2 O 4 acts as the electrocatalytic phase and as a protective layer for the relatively less stable CuO [184]. The authors claimed that a type II heterojunction forms at the interface, where photogenerated electrons are injected to CuBi 2 O 4 , while the holes to CuO (Figure 12a). In a different study, a CuBi 2 O 4 /CuO heterojunction with CuO deposited on CuBi 2 O 4 was prepared [185]. CuO and CuBi 2 O 4 still exhibited a type II alignment but with electron flow towards CuO (Figure 12b). Aside from CuO, composites of CuBi 2 O 4 with Bi 2 O 3 [163,186,187], as well as other oxides [188,189], have been reported in literature with varying successes in the improvement of PEC activity. Despite this significant progress with CuBi 2 O 4 , majority of the efforts were directed towards H 2 O reduction; studies on the improvement of CuBi 2 O 4 as a photocathode for CO 2 reduction are few. Electrodeposited CuBi 2 O 4 films were studied for the PEC reduction of CO 2 [166]. The film thickness was first optimized based on the tradeoff between an enhanced light absorption and an increase in number of recombination centers. The optimized sample was used as photocathode for CO 2 reduction (in a CO 2 -saturated 0.1 M KHCO 3 electrolyte, under a 300 W Xe lamp illumination and a bias of 0 V vs Ag/AgCl). CO was the only detected product in the headspace after 4 h of reaction, but with a very low yield (0.0155 μmol·cm −2 ). In another study, no product was even detected despite the conduction band being in a thermodynamically feasible position for CO 2 reduction [190].
These results indicate that aside from the inherently poor charge transport kinetics, the bare CuBi 2 O 4 surface shows little activity for CO 2 reduction and that similar strategies to those discussed in the preceding paragraphs are necessary to drive CO 2 reduction on CuBi 2 O 4 . TiO 2 was employed as a protective overlayer on nanoporous CuBi 2 O 4 , using two different overlayer thicknesses (50 nm and 1 μm) [191]. However, TiO 2 was not conformally deposited on the CuBi 2 O 4 surface; there were regions where there was contact between CuBi 2 O 4 and the electrolyte. CO and H 2 were the detected products in PEC measurements in a CO 2 -saturated bicarbonate electrolyte under an applied bias of −0.50 V (vs reversible hydrogen electrode (RHE)). Measured photocurrents and the overall yield were both larger for the sample with thicker TiO 2 due to a more efficient charge separation versus the sample with thinner TiO 2 , where the space-charge region in the TiO 2 /electrolyte interface cannot be fully supported by the thin overlayer. However, selectivity for CO was lower for the sample with thicker TiO 2 layer; the authors attributed this to the effect of the nanostructure on mass transport ( Figure 13). In the   [184]. Copyright © 2021, Elsevier) and (b) electron moves from CuBi 2 O 4 to CuO to electrolyte (Reprinted from ref. [185]. Copyright © 2020, Royal Society of Chemistry). sample with thicker TiO 2 layer, the channels or pores were narrower so that local pH changes near the interface were more pronounced. Because CO 2 reduction is more favored at higher pH, CO 2 is more quickly exhausted near the interface, and the process becomes mass transfer limited. Consequently, the electrons are instead consumed for the HER.

Other copper ternary oxides
Other copper delafossites, namely, CuCrO 2 and CuGaO 2 , have been investigated as photocathodes for CO 2 reduction. CuCrO 2 is, however, a wide band gap p-type semiconductor (2.95-3.15 eV) implying that UV illumination is necessary to obtain a photoresponse [192]. CuCrO 2 prepared via solution combustion synthesis was tested for its photoactivity towards CO 2 reduction [193]. The as-synthesized sample contained phase impurities, namely, CuO (4 wt%) and α-Cr 2 O 3 (6 wt%); annealing in Ar at 700°C allowed the formation of a more phase-pure sample. As expected, onset of IPCE was near the UV region (438 nm). Long-term photoelectrolysis of the Ar-annealed sample in CO 2 -saturated bicarbonate, under UV illumination, resulted in the formation of CO, CH 4 , and H 2 in the gas phase, while HCOOH and CH 3 OH in the liquid phase, with a total Faradaic efficiency of 85-90%. Although photocurrents decreased significantly after one hour, the authors claimed that their CuCrO 2 photocathode was relatively more stable than Cu 2 O. Because unmodified CuCrO 2 was proven capable of CO 2 reduction, there is much room for improvement especially by making it responsive to visible light. CuGaO 2 is another wide band gap p-type semiconductor (3.3-3.75 eV) which has been employed as hole transport layer in dye-sensitized solar cells [194,195]. The photochemical activity (in powder form) of CuGaO 2 and its solid solution with CuFeO 2 and CuGa 1-x Fe x O 2 towards CO 2 reduction was investigated [194]. The indirect band gap of 2.55 eV in pure CuGaO 2 (which was not previously observed in thin film CuGaO 2 ) was observed to shift to 1.5 eV upon Fe alloying, due to a crystal strain that modifies the electronic structure of CuGaO 2 . Observed reduction products from an aqueous CO 2 solution upon illumination with broadband light were CO and CH 4 (trace). However, the amount of CO evolved did not correlate with the degree of Fe substitution, so it is likely that the indirect band gap absorption does not result in an efficient charge separation and accordingly contribute to the photocatalytic process. In a different study, CuGaO 2 was employed as a scaffold for a Ru(II)-Re(I) supramolecular photocatalyst for CO 2 reduction [196]. PEC testing using the hybrid photocathode in a CO 2 -saturated aqueous solution under visible light irradiation (>460 nm) produced both CO and H 2 , with a total Faradaic efficiency of 81%. A PEC cell employing the Ru(II)-Re(I)/CuGaO 2 hybrid photocathode coupled to a CoO x / TaON photoanode with a Z-scheme configuration was then devised. Under visible light illumination and biasfree condition, CO and H 2 were produced at the cathode, while O 2 evolution was observed at the anode.
Other Cu 2+ -based ternary oxides that have been investigated as photocathodes for CO 2 reduction are CuFe 2 O 4 , Cu 2 V 2 O 7 , and the copper niobates CuNb 2 O 6 and Cu 3 Nb 2 O 8 . CuFe 2 O 4 is a ternary oxide with an inverse spinel structure and a band gap of 1.39-1.42 eV. Methanol was the only product detected in the liquid phase from PEC reduction in a CO 2 -saturated bicarbonate solution under visible light irradiation (>470 nm), with a corresponding Faradaic efficiency of 62% [197]. Addition of CdS [198], graphene oxide [199], and polyaniline (PANI) [200] as co-catalysts resulted in a better PEC performance, as reflected in the higher IPCEs of 12.09, 8.02, and 7.1% (at 470 nm), respectively, Figure 13: Effect of mesostructure on activity and selectivity of CuBi 2 O 4 /TiO 2 nanocomposite for PEC reduction of CO 2 . (Reprinted with permission from ref. [191]. Copyright © 2021, Elsevier).
Copper ternary oxide photocathodes for CO 2 reduction  445 FE, Faradaic efficiency. (compared to 5.1% for bare CuFe 2 O 4 ) as well as a higher selectivity for methanol formation, 72, 87, and 73%, respectively (compared to 62% for bare CuFe 2 O 4 ). This improvement was attributed to an enhanced electron-hole separation and to an increased CO 2 adsorption for the case of PANI. A direct Z-scheme system of Cu 2 V 2 O 7 (with a band gap of 2.01 eV) and g-C 3 N 4 was devised [201]. Under illumination of light with wavelength of 400-700 nm, the Cu 2 V 2 O 7 /g-C 3 N 4 system produced a much higher photocurrent response than Cu 2 V 2 O 7 or g-C 3 N 4 alone. CO 2 photoreduction (under a photocatalytic scheme) resulted in the formation of CH 4 , CO, and O 2 with rates of 305, 166, and 706 μmol·g -1 catalyst·h -1 , respectively. CuNb 2 O 6 , belonging to a class of metal niobates with an orthorhombic columbite structure, was found to be photoactive for PEC CO 2 reduction even without any co-catalyst [202]. It showed photoresponse under visible light illumination (150 W tungsten-halogen lamp), consistent with its optical band gap of 1.77 eV. Photocurrents measured in CO 2 -saturated 0.1 M NaHCO 3 solution (pH 7) were up to six times greater than those measured in an N 2 -saturated 0.1 M Na 2 SO 4 solution (pH 7), indicating that the photocurrent was being used in the reduction of CO 2 , although no efforts were done to identify and quantify the reduction products. Another copper niobate, Cu 3 Nb 2 O 8 (with band gap of 2.5 eV), was also found to photoelectrochemically reduce CO 2 under AM 1.5 G illumination, with CO as the primary product [203]. However, the Faradaic efficiency for CO evolution was found to be around 9% (at -0.20 V vs Ag/AgCl), indicating that majority of the cathodic photocurrent was being used in the photocorrosion of the Cu 3 Nb 2 O 8 film. Indeed, chronoamperometry tests (in CO 2 -saturated 0.5 M NaHCO 3 solution at -0.20 V vs Ag/AgCl) showed a 98% decrease in the initial photocurrent only after 20 min. Reannealing of the tested films restored the initial photocurrent to that of the as-synthesized film, corroborating the cathodic photocorrosion of Cu 3 Nb 2 O 8 , and the use of protective layers or addition of co-catalysts will be needed to improve its PEC performance. The results of PEC CO 2 reduction studies on the copper ternary oxides, where the reduction products were identified and quantified, are summarized in Table 3. Although beyond the scope of this review, there are many research efforts in recent years to explore other classes of materials as photocathodes. Non-copper ternary oxides, such as AgRhO 2 [204], BiFeO 3 [205,206], CaFe 2 O 4 [207,208], Ca 2 Fe 2 O 5 [209,210], LaFeO 3 [211,212], NiFeO 4 [213], and PbMoO 4 [214], have been prepared as photocathodes but their application towards PEC CO 2 reduction is yet to be investigated. On the other hand, copper-based chalcogenides, such as CuInS 2 [215][216][217][218], CuGaS 2 [219], CuGa 3 Se 5 [220], Cu(In,Ga)Se 2 [221,222], Cu 2 ZnSnS 4 [223,224], and Cu 2 ZnGeS 4 [225], which have been widely used in solar cells due to their favorable transport properties, have recently been investigated as CO 2 reduction photocathodes ( Table 4). Although their efficiency and stability are comparable with the copper ternary oxides, the cost and toxicity (primarily of In and Ga), as well as the complicated preparation methods that limit possibilities for scale-up, are major issues that need to be addressed [226].

Summary and outlook
In the search for new p-type materials suitable as photocathode for PEC applications, copper ternary oxides have recently gained attention due to their favorable band structure and earth-abundant nature. This article examined the current progress achieved with CuFeO 2 and CuBi 2 O 4 by summarizing the various strategies to improve their efficiency and stability towards CO 2 reduction. Other copper ternary oxides, CuCrO 2 , CuGaO 2 , CuFe 2 O 4 , Cu 2 V 2 O 7 , CuNb 2 O 6 , and Cu 3 Nb 2 O 8 , were also reviewed. For these photocathodes, CO, HCOOH, HCOH, CH 3 OH, and CH 4 as well as C 2 H 5 OH and (CH 3 ) 2 CO were the observed CO 2 reduction products (Table 3). However, photocorrosion is still significant in these materials. Several studies are in search of selective inhibitors that can allow only specific reactions to proceed, aiding the overall kinetics. Polymers and membranes have been used recently as the protective overlayer that simultaneously fosters CO 2 binding sites [227] and modifies the surface hydrophobicity to improve selectivity [228].
Most of the papers reviewed here approach the improvement in the copper oxide photocathodes from the viewpoint of charge transport. In Cu 2 O, it has been observed that surface recombination and bulk transport were the contributors to the lifetime of charges [229]. A close connection between semiconductor and conducting support leads to faster charge separation; the challenge is to determine the type and structure of the support and the copper oxides. However, a system of efficient and stable CO 2 reduction relies not only on charge transport but also on chemical reaction kinetics and mass transfer of reactants and products. Theoretical studies on CO 2 adsorption and reaction mechanisms are being done by several groups [230,231]. Similarly, there have also been theoretical and experimental works to understand CO 2 reduction active sites in copper and copper-based semiconductors [228,232]. Studies such as these can guide synthesis of structures with complementary adsorption and active site.
It remains a prospect whether a semiconductor material with the correct compositional mix that satisfies all requirements as photoelectrode can be identified. More likely, this semiconductor will come from complex metal oxides where there are greater possibilities for tuning and optimization of material properties. With this large material space, theory on interfacial phenomena, such as CO 2 activation and intermediates, should be refined much further to establish advanced descriptors for computational screening of new complex oxides. High throughput methodologies for the synthesis of a large library of candidate materials and testing of their relevant properties for PEC applications will further accelerate materials discovery. It must be noted, however, that with increasing complexity of multinary oxides, the cost and ease of processability, along with sustainability considerations critical towards large scale production, become more difficult to attain.
In the past few decades, significant achievements have been made in the field of PEC solar fuel generation. At its current state, further advances will rely not only on materials development, but also on device modeling and simulation and techno-economic analysis. For PEC CO 2 reduction to become a viable, carbon-neutral technology that can harness the enormous, undeveloped potential of solar energy, many key challenges remain to be solved.