Photoelectrochemical cells for solar hydrogen production: Challenges and opportunities on

As the Holy Grail to a carbon-free hydrogen economy, photoelectrochemical (PEC) water splitting offers a promising path for sustainable production of hydrogen fuel from solar energy. Even though much progress has been made over the past decade, the effectiveness and robustness of PEC cells are still far from a mature phase that would allow for widespread deployment. This perspective discusses the key challenges facing the current level of PEC development and proposes experimental approaches and strategies that can be adopted to address the issues. Focuses are mainly placed on the employment of in situ and operando spectroscopic measurements, the introduction of alternative, high value-added oxidation reactions, and the creation of near infrared-responsive photoelectrodes. A brief outlook that may assist the future advancement of PEC technology is also presented.


I. INTRODUCTION
Photoelectrochemical (PEC) water splitting represents one of the most promising energy conversion processes capable of directly producing hydrogen from renewable energy sources. A typical PEC cell consists of a photoactive semiconductor as a working electrode, a counter electrode (usually Pt), and an appropriate supporting electrolyte. When an n-type semiconductor photoelectrode is used, photogenerated holes migrate to the photoelectrode/electrolyte interface and perform water oxidation, while photoexcited electrons travel via the back contact and external circuit to the counter electrode, where they reduce water. In such a practice, the n-type semiconductor serves as the photoanode and its valence band level must be more positive than the H 2 O/O 2 potential (+1.23 V vs RHE) to allow efficient oxygen evolution. The Fermi level of the photoanode, on the other hand, determines the potential of the photoexcited electrons at the counter electrode, which can be otherwise less negative than the H + /H 2 potential (0 V vs RHE) since the potential deficiency can be compensated for by applying an external voltage. Similarly, a p-type semiconductor can function as the photocathode for practical hydrogen evolution, given that the conduction band level is more negative than the H + /H 2 potential. Alternatively, a photoanode and a photocathode can be connected in a two-electrode cell without employing a counter electrode. [1][2][3][4] In this tandem configuration, the overall water splitting reaction, in which oxygen and hydrogen evolution take place at the photoanode and photocathode, respectively, can be driven by light irradiation without applying an external bias. Compared to the typical PEC, which uses a single semiconductor photoelectrode, the tandem PEC cell allows for the pairing up of two semiconductors with complementary materials' properties, offering possible synergistic cooperation for achieving better performance. On the other hand, coupling the typical PEC cell with a photovoltaic cell can also accomplish unbiased solar water splitting, provided that the photovoltaic compartment can contribute sufficient photovoltage to make up for the potential deficiency of the PEC cell. [5][6][7][8] Moreover, multijunction semiconductor architectures can be integrated with an electrochemical catalyst electrode to form a monolithic tandem PEC device, which is also capable of driving overall water splitting without bias. [9][10][11] This type of self-powered, unassisted PEC system is the ultimate goal for accessing solar hydrogen because it spontaneously converts solar energy into solar fuel, a core concept of sustainability.
The ubiquitous availability of water and its low-carbon footprint make PEC water splitting a green and sustainable approach to solve the problem of ever-rising global energy demand. Since Fujishima and Honda demonstrated the first PEC water splitting apparatus using TiO 2 , 12 the number of related publications each year has increased significantly. The high importance and tremendous interest in this field can be highlighted by a simple search on the ISI Web of Knowledge database using "PEC" and "photocatalytic hydrogen" as title keywords. As summarized in Fig. 1, over the past ten years, the number of publications on "PEC hydrogen" and "photocatalytic hydrogen" has both exhibited a significant growth rate, reaching approximately 1000 publications at 2018. For publications with the title of "PEC," the yearly number has exceeded 1300 at 2018. Obviously, the interest in PEC cells continues to maintain a marvelous pace as they remain widely employed in the advancement of solar fuel systems.
Thanks to great achievements made over the past decade, the solar-to-hydrogen (STH) energy conversion efficiency of PEC-based cells has approached the target value that can compete with electrolysis systems using photovoltaic cells (approximately 30.0%). 13 Among the different PEC-based technologies, a monolithic tandem PEC setup has thus far exhibited the highest STH efficiency, achieving 19.3% on a RuO 2 -integrated GaAs/GaInAs/GaInP threejunction photocathode. 14 A tandem PEC cell that coupled an ntype GaAs photoanode with a p-type GaAs photocathode, on the other hand, attained an unprecedented STH of 13.1%. 15 From an economic perspective, a tandem PEC system is the most feasible device to implement on a large scale due to its relative simplicity and low cost. A tandem PEC system can even deliver a theoretical maximum STH value of 29.7% at a bandgap combination of 1.60 eV and 0.95 eV for the coupled photoanode and photocathode. 16 Despite the increasing STH performance, the effectiveness and robustness of the current PEC cells is still far from a mature phase needed for widespread deployment. The failure discloses that further optimization of the chemical, electronic, and optical properties of individual photoelectrodes is necessary.

FIG. 1.
Number of yearly publications with title keywords of "PEC" and "photocatalytic hydrogen" since 2008. Here, "others" represent those PEC systems demonstrating applications other than water splitting, e.g., PEC sensors and PEC degradation of organic dyes.
In this perspective, we focus on the key challenges facing the current level of PEC development. We start with a brief introduction of the working principles of PEC water splitting and discuss major issues that impede the current growth of PEC technology. Following that, we propose the experimental approaches and strategies that can be adopted to address the issues. An outlook that may assist the future advancement of PEC technology is presented in the final section. As this perspective highlights only a few studies to discuss the mechanistic understanding of PEC systems, we suggest several excellent review articles with topics concerning the recent advancement of PEC technology for interested readers. [17][18][19][20]

II. CHALLENGES FACING CURRENT PEC DEVELOPMENT
Practical PEC solar hydrogen generation requires reliable photoelectrode materials that meet the following criteria: effective use of charge carriers, reduced overpotential, a broad range of solar spectrum harvesting, high stability for long-term operation, and low cost. As depicted in Scheme 1, effective carrier utilization, reduced overpotential, and broad solar harvesting have been regarded as the core features that need persistent efforts to address. Upon photoexcitation, charge carriers must be extracted from the photoelectrode for participation in surface redox reactions, a process that inevitably competes with charge carrier recombination. Although much progress has been toward understanding charge transfer and recombination in semiconductor materials, the elucidation of charge transfer dynamics in a working photoelectrode under operating conditions is still lacking and remains a topic of intense interest. In this regard, real-time observations of charge transfer dynamics via in situ spectroscopic experiments are highly desirable for realizing the charge transfer scenario of photoelectrodes under PEC operating conditions.
Another important aspect relevant to the charge transfer behavior of photoelectrodes is the introduction of cocatalysts to reduce the overpotential driving the water splitting reaction. The introduced cocatalysts can mediate charge transfer at the photoelectrode surface to improve the surface reaction kinetics. For complete water splitting, the charges of the water oxidation and water reduction reactions need to be balanced. Because of the complex four-proton coupled electron transfer process, the water oxidation reaction is more difficult and slower than water reduction. This highlights the necessity of adding cocatalysts to modify photoanodes to further improve the kinetics of oxygen evolution. Exactly the same concept can be applied to promote the hydrogen evolution kinetics by modifying photocathodes with specific cocatalysts. Nevertheless, the exact role of the cocatalysts in improving the surface reaction kinetics is a source of confusion and debate. This is because the presence of cocatalysts on a photoelectrode surface not only changes the charge transfer kinetics but also alters the local chemical environment and electronic structure. A detailed understanding of the mechanism that underpins the observed performance enhancement is therefore imperative for the widespread implementation of the cocatalyst approach. To this end, operando characterization techniques for studying the cocatalyst-modified photoelectrodes under PEC operating conditions can provide in-depth insight into the surface reaction kinetics for constructing the true working mechanism. Another practical solution to the slow kinetics of oxygen evolution in a PEC cell is to find alternative chemical transformations for the anode reaction while allowing concurrent hydrogen evolution at the cathode. Replacing oxygen evolution with other kinetically faster, technologically valuable oxidation reactions may open new horizons for creating next-generation PEC cells, allowing the realization of artificial photosynthesis. In principle, harvesting a broad range of the solar spectrum can guarantee satisfactory PEC performance as it increases the maximum STH efficiency that can be achieved by a given semiconductor. Figure 2 displays the relationship between the photon harvesting range in terms of bandgap energy and the maximum STH efficiency attainable by some representative semiconductor photoelectrodes. Here, the maximum STH efficiency was calculated by assuming that all photons available in the solar spectrum (AM 1.5 G, 100 mW/cm 2 ) with energy equal to or larger than the bandgap energy are utilized by the photoelectrodes for hydrogen production. This means that 100% of incident photon-to-current efficiency is considered and other factors that can reduce the STH efficiency in practice, such as the light absorption coefficient, charge transfer dynamics, surface reaction kinetics, and band edge positions, are temporarily ignored. On the other hand, the experimentally reported appliedbias photon-to-current efficiency (ABPE) values which stand for a hypothetic STH of individual photoelectrodes 21 are also marked (open symbols) in Fig. 2 for comparison. It is important to mention that the definition of ABPE is essentially different from that of STH. ABPE is an index highly dependent on applied bias, which has been used to evaluate the photoelectrode performance of water splitting in a half-cell configuration without the consideration of Faradaic efficiency. STH, on the other hand, premises spontaneous water splitting on semiconductor photocatalysts without any external bias. Despite the different definitions, the two indexes share the same concept describing the energy efficiency of the produced

PERSPECTIVE
scitation.org/journal/apm hydrogen from the absorbed solar spectrum. It is significant to note from Fig. 2 that the photoelectrodes reported thus far are mostly ultraviolet (UV)-and visible-responsive semiconductors. There are relatively few choices for photoelectrodes that can respond to near infrared (NIR), a spectral region accounting for over 50% of the solar spectrum distribution. Considering the potentially increased STH value at an extended range of photon harvesting, there is an urgent need for seeking promising semiconductor photoelectrodes capable of harvesting photons under NIR illumination.

III. IN SITU ULTRAFAST LASER SPECTROSCOPY
Ultrafast laser spectroscopic techniques such as time-resolved photoluminescence and transient absorption have been widely employed to investigate the charge transfer dynamics of photoelectrodes and their implications in water splitting reactions. [37][38][39][40][41][42][43][44][45][46][47][48][49][50][51] However, these measurements are usually performed in ex situ conditions where operational factors such as constant irradiation, applied bias, and electrolyte presence are not considered. The deduced conclusions may to some extent be incompatible with the phenomena observed experimentally. Therefore, realization of the fate of charge carriers under PEC operating conditions is imperative. Knowledge of the effect of electrochemical bias on charge carrier dynamics is especially vital to understanding the efficiency limitations of photoelectrodes because efficient PEC water splitting usually occurs under bias conditions. The acquired knowledge is also essential for independent optimization of individual photoelectrodes and their further integration in tandem PEC cells.

PERSPECTIVE
scitation.org/journal/apm scales, which promotes hole transfer dynamics to increase the photocurrent density for water oxidation. 52 As Fig. 3(b) shows, the fs-ns transient kinetics probed at two spectral regions (750 and 575 nm) revealed essentially different dependences on the applied electrochemical bias. Absorption at 750 nm was assigned to hole absorption within α-Fe 2 O 3 , whereas the bleach signal at 575 nm was attributed to electron trapping by localized states close to the conduction band edge. For the transient kinetics at 750 nm, the application of an anodic bias resulted in a substantial retardation of decay kinetics, from 6 ps decay half-time at flat-band potential (+0.5 V RHE ) to 200 ps at +1.4 V RHE . These data provided evidence that anodic bias can greatly increase the charge carrier lifetime in a metal oxide photoanode, which is attributed to the generation of electric fields in the space charge layer that aid the spatial separation of charges. On the other hand, the anodic bias accelerates the decay kinetics at 575 nm and causes its inversion to a negative transient signal at long times. As shown in Fig. 3(c), the localized states within the space charge layer can become unoccupied under anodic bias to enable possible electron trapping. Following photoexcitation, the relaxation of excited state electrons into these unoccupied states (i.e., electron trapping) can result in bleaching of the ground state absorption, further leading to a negative transient absorption, as observed at 575 nm under anodic bias. Note that this electron trapping process is initiated on sub-picosecond time scales and may compete with electron/hole recombination during the hole transfer process, thus enabling more efficient extraction of electrons to the external circuit, and much importantly, more effective hole transfer to the photoelectrode surface. The requirement to retard ultrafast electron/hole recombination by space charge layer formation and the consequent electron localization is possibly one of the reasons for the large overpotentials required for photocurrent generation on α-Fe 2 O 3 photoanodes. The observations from this study further infer that intrinsic and extrinsic defects may create localized and trap states close to the conduction band edge to enable the electron trapping event. Therefore, doping α-Fe 2 O 3 with suitable heteroatoms (for example, Si for the current case) may increase the density of such localized states within the space charge layer, providing an additional, beneficial mechanism for enhancing the water oxidation efficiency. Clearly, in situ ultrafast spectroscopic studies can provide distinct spectral assignment of photoexcited electrons and holes, probe their fate upon photoexcitation, and disclose which charge transfer process is the main efficiency limiting factor under PEC operating conditions. By inspecting the in situ charge dynamics data in working photoelectrodes, one can gain further insight into the electronic and structural factors that dictate the charge transfer dynamics in photoelectrodes under operating conditions. The underlying origins can supply scientists with guidelines for optimizing photoelectrodes to ensure better charge carrier utilization. Especially for distinctive semiconductor heterostructures such as type-II and Z-scheme photoelectrodes, the identifications of charge transfer pathways under operating conditions may provide fundamental design criteria to engineer photoelectrodes for advanced PEC applications.

IV. OPERANDO X-RAY SPECTROSCOPY
In addition to charge transfer dynamics, understanding the reaction kinetics at photoelectrode surfaces is also critical for determining the overall PEC performance. In particular, for photoanodes, the oxygen evolution reaction has been recognized as the main bottleneck for achieving efficient overall water splitting due to its complex, multiple-electron transfer processes. To improve the surface reaction kinetics, semiconductor photoelectrodes are usually modified with specific cocatalysts. For water oxidation, practical cocatalyst candidates include Co-Pi, 54 2  69 ) have demonstrated remarkable performance as efficient cocatalysts. However, the exact role of such surface modification is ambiguous, causing inconsistent results under different experimental conditions. Even if the same cocatalyst is used to modify the same photoelectrode, discrepancy in the resultant effectiveness is constantly encountered. By probing the local chemical, electronic, and structural states of a selected atomic species using operando X-ray spectroscopy, one can acquire empirical knowledge about the key steps of surface reactions to elucidate the mechanistic role of cocatalysts and establish a plausible kinetics model. In fact, operando X-ray spectroscopy has been widely employed to study the reaction kinetics of electrochemical catalysts. 70,71 By identifying the nature of catalytic sites and monitoring the dynamics of the active sites at atomic scales during a catalytic process, the rational design of electrochemical catalysts with desired activity, stability, and selectivity becomes achievable. Despite such advances, the use of operando X-ray spectroscopy to investigate the reaction kinetics of photoelectrodes in a PEC cell has been relatively scarce.
Minguzzi et al. performed operando X-ray absorption spectroscopic (XAS) measurements on IrOx-deposited α-Fe 2 O 3 (α-Fe 2 O 3 /IrOx), a prototype for cocatalyst modified photoanodes with enhanced photocurrents at reduced bias. 72 As sketched in Fig. 4(a), the PEC cell contains a frontal window that allows illumination by both visible and X-ray photons. The full XAS spectra in Fig. 4(b) were collected under +1.4 V RHE bias at the Ir-L III edge, which displayed a near-edge structure dominated by an intense white line (WL) peak caused by the photoexcitation of 2p electrons to the empty Ir 5d band. Noticeably, light illumination resulted in an increased WL peak area, meaning that the 2p to 5d transition of the Ir atoms became increasingly probable. Because no significant spectral difference was observed at the extended fine structure region (EXAFS, for E > 11 235 eV), the local structure of the Ir centers was considered unchanged under illumination. Therefore, the enhanced amplitude of the WL under illumination was solely attributed to a less filled 5d band, suggesting the prevalence of hole transfer from the photoexcited α-Fe 2 O 3 to the IrOx cocatalyst. To further realize the charge transfer scenario under PEC operating conditions, the differential (light-dark) spectra at the edge region (XANES) were monitored and compared at four different applied bias values: +0.1 V RHE , at which electron accumulation establishes; +0.4 V RHE , which is close to the flat-band potential; +0.8 V RHE , where the energy band is slightly bent; and +1.4 V RHE , at which the energy band is strongly bent. As Fig. 4(c) shows, the differential spectra at bias ≤+0.8 V RHE were all characteristic of an increased spectral weight at low energy and an otherwise decreased spectral weight at high energy. Such a spectral redshift trend indicated a partial decrease in the oxidation state of Ir, reflecting photoelectron injection from the photoexcited α-Fe 2 O 3 into the IrOx at low bias conditions. This observation points out the presence of other possible charge transfer pathways between α-Fe 2 O 3 and IrOx that are not limited to hole transfer to IrOx. On the other hand, the differential spectrum at +1.4 V RHE displayed an intense and positive peak at the WL region of the Ir-L III edge, meaning that the density of 5d empty states is higher under illumination as a result of hole transfer from α-Fe 2 O 3 to IrOx. In this situation, IrOx behaves as a hole trap, attracting photogenerated holes from α-Fe 2 O 3 to decrease the occupancy of the Ir 5d band. Combined with the results obtained from low bias conditions, it can be considered in Fig. 4(d) that the overall charge transfer scenario on α-Fe 2 O 3 /IrOx is the consequence of a combination of both electron and hole transfer processes. In this picture, IrOx acts as a recombination center and may pose an unfavorable effect on oxygen evolution, especially when the loading of IrOx is high. This explains the substantially low photocurrent recorded on α-Fe 2 O 3 /IrOx at +1.4 V RHE (approximately 2 μA/cm 2 ), in which the IrOx loading was considerably high in order to acquire high-quality XAS spectra. In fact, the fraction of IrOx involved in the charge transfer processes PERSPECTIVE scitation.org/journal/apm associated with oxygen evolution was merely 5%, as can be determined by integrating the differential WL peak in Fig. 4(c) and comparing this peak area with the corresponding peak area recorded in the dark. This phenomenon suggested that the oxygen evolution reaction mostly occurs at the α-Fe 2 O 3 /electrolyte interface. This outcome further reveals a surface modifier function for cocatalysts that is not directly related to improved reaction kinetics but rather associated with the modification of the electronic structure of photoelectrodes in terms of surface state passivation [73][74][75][76] and suppression of back electron-hole recombination. 77 This observation also outlines the necessity of engineering the microstructural features of cocatalysts in order to optimize the charge transfer processes associated with the surface reaction kinetics. For example, for Mnoxide-modified Nb:SrTiO 3 photoelectrodes, 78 the Mn-oxide particle cocatalyst showed inferior PEC activity enhancement over the Mnoxide thin film cocatalyst due to undesired charge recombination as a result of simultaneous electron and hole transfer into the Mn-oxide particles.
For most of the semiconductor photoelectrodes, their longterm operation in a PEC cell is hindered by significant photocorrosion associated with the accumulation of photoexcited charge carriers and subsequent redox reactions with the semiconductor itself. 79 A substantial effort has been made to resolve the photocorrosion issue by protecting the photoelectrode surface with specific corrosion-resistant materials. Such a surface protection strategy may either prevent direct contact of photoelectrodes with electrolyte 80 or mediate the charge transfer kinetics at the photoelectrode surface, 32 therefore leading to increased durability for extended PEC operation. However, similar to the ambiguous mechanism of cocatalyst modification, the protective layer approach lacks mechanistic understanding, and developing a generalized methodology that can address the instability issue for all kinds of photoelectrodes has been challenging. By employing operando XAS to identify the electronic and structural state evolution on the protective layers of photoelectrodes, one may acquire the in-depth understanding of the working principles of protective layers and convey a viable guideline to the reproducible creation of stable photoelectrodes for sustainable solar hydrogen generation. Operando XAS techniques allow the development of a detailed picture of the electronic and structural localization of charge carriers at the photoelectrode surface under PEC operating conditions. Understanding the electronic and structural changes under operando conditions can be an important step toward gaining mechanistic insights into the reaction kinetics at the photoelectrode surface, which lays a solid foundation for rationally designing reliable photoelectrodes by means of cocatalyst modification and protective layer introduction.

V. ALTERNATIVE OXIDATION REACTION
Compared with hydrogen, which is targeted as a viable solar fuel, the oxygen evolved from the other half reaction of a PEC cell is less valuable because of its abundance in the atmosphere. In particular, the reaction kinetics of oxygen evolution is considered critical due to the complexity of the four-proton coupled electron transfer processes for generating one oxygen molecule. To overcome the large kinetics barrier, a considerable portion of energy is wasted as overpotential when driving the oxygen evolution reaction on the photoanodes. In fact, it is not necessary to conduct water oxidation at the photoanode side in order to allow concurrent hydrogen evolution to occur at the photocathode. Recent studies have explored possible alternatives to the oxygen evolution reaction in PEC systems. For example, a unique PEC-based direct methanol fuel cell has been proposed by employing semiconductor photoelectrodes for conducting methanol oxidation. [81][82][83] By controlling the window of applied bias to solely drive methanol oxidation (+0.6 V RHE ∼ +0.8 V RHE 84 ), the oxygen evolution reaction can be replaced with methanol oxidation under PEC conditions. It should be noted, however, that methanol oxidation might not be an appropriate alternative to oxygen evolution in a PEC cell, considering its likewise multiple electron transfer nature. From a kinetics viewpoint, the alternative oxidation reactions should occur more easily with lower anodic overpotential and be able to provide electrons and protons to feed the hydrogen evolution reaction. In this regard, organic oxidation reactions that are compatible with the hydrogen evolution reaction and involve only one-or two-electron/proton transfer are particularly appealing. 85 Relative to traditional homogeneous organic oxidation reactions, using illuminated photoelectrodes to drive oxidation reactions in a heterogeneous manner allows chemists to reap many practical advantages. For instance, adding an oxidizing reagent is no longer necessary because a homogeneous oxidizing reagent can be replaced by the oxidizing power rendered by photoelectrodes. In addition, the need to remove the excess oxidizing reagent from the reaction solution is discarded, which mitigates the need for product purification and lowers the overall cost.
For most n-type oxide semiconductors, the top of the valence band is primarily composed of O 2p orbitals, endowing them with considerably positive valence band levels. Particularly for BiVO 4 and WO 3 , the highly oxidative valence band and relatively narrow bandgap make them intriguing candidates to drive a wide spectrum of alternative oxidation reactions. Among the various organic oxidations, biomass valorization, which transforms biomass waste into industrially important molecules, provides a viable alternative to oxygen evolution in a PEC cell. As an intermediate from the polydehydration of sugar, 5-hydroxymethylfurfural (HMF) has received a great deal of attention for its promising use as a carbon-neutral feedstock for the production of other polymeric materials. 86,87 As shown in Fig. 5(a), oxidation of HMF produces many derivatives that can further serve as monomers for the synthesis of functional polymers. For example, the fully oxidized derivative 2,5-furandicarboxylic acid (FDCA) can replace terephthalic acid as a biorenewable monomer to synthesize polyethylene terephthalate resins. [88][89][90] As a result, many different catalytic approaches have been devised to realize the oxidation of HMF into FDCA. [91][92][93][94][95] Using semiconductor photoelectrodes to drive HMF oxidation in a PEC cell offers possibilities of realizing photosynthetic biomass valorization. Choi and coworkers demonstrated the first use of solar energy to carry out HFM oxidation on a BiVO 4 photoanode and hydrogen evolution on a Pt cathode. 96 Figure 5(b) illustrates the concept of combining biomass valorization and hydrogen production in a PEC cell. Using 2,2,6,6-tetramethylpiperidine-1-oxyl (TEMPO) as a redox mediator, the PEC oxidation of HFM into FDCA could be achieved with nearly 100% yield and 100% Faradaic efficiency. Figure 5 and Au electrode (under electrochemical conditions). A significant reduction in onset potential of approximately 700 mV was observed for the PEC case, suggesting that the TEMPO-mediated HFM oxidation could be kinetically facilitated on the BiVO 4 photoanode. This outcome can be understood from the fact that the valence band level of BiVO 4 has sufficiently strong oxidizing power to initiate the mediated HFM oxidation. The photooxidation products were further analyzed in the PEC cell operated at +1.04 V RHE , where electrochemical oxidation did not occur. The result in Fig. 5(d) shows the complete conversion of HMF to FDCA as 40 C of charge passed. Significantly, the Faradaic efficiency of FDCA production was calculated to be 93% at 40 C, suggesting that the TEMPOmediated HMF oxidation is kinetically faster than water oxidation and is the dominant oxidation reaction occurring on the BiVO 4 photoanode. The nearly 100% yield and 100% Faradaic efficiency for FDCA production from HFM photooxidation highlight that PEC operation could be one of the most efficient, practical, and environmentally benign routes for biomass valorization. This demonstration also suggests that solar-driven biomass conversion can be a viable anode reaction, which has the potential to increase both the integrity and utility of PEC systems for solar fuel production. A similar idea has been implemented to realize biodiesel manufacturing on a BiVO 4 photoanode, which can selectively oxidize glycerol in a PEC cell to produce high value-added dihydroxyacetone with a conversion selectivity of 63.6%. 97

VI. NIR-RESPONSIVE PHOTOELECTRODES
The first requirement for effective solar hydrogen production using semiconductor photoelectrodes lies in the proper bandgap energy. Photons with energy smaller than the bandgap pass through the semiconductor without being absorbed, while photons with energy exceeding the bandgap dissipate their energy as heat via phonon emission. To minimize these two fundamental energy losses, a multijunction approach has been proposed and utilized in the photoelectrode design for creating a sophisticated PEC system. The idea is based on the use of a large bandgap semiconductor to first absorb high-energy photons, accompanied with a small bandgap semiconductor for harvesting the low-energy photons. This concept has been realized by introducing heterojunctions in individual photoelectrodes, 45,47,98,99 constructing two bandgap-optimized photoelectrodes in a tandem PEC cell 1,3,7,100 and devising heterotype dual photoelectrodes. 31 Despite intense research efforts, major progress has only been made on the combination of UV-and visiblesensitive components as the multiple junctions in the photoelectrode design. The development of NIR-responsive photoelectrodes has been relatively slow. Note that the energy distribution of solar light is approximately 6.8% in UV (λ < 400 nm), 38.9% in visible (λ = 400-700 nm), and 54.3% in the NIR range (λ = 700-3000 nm). Significantly, NIR light with a wavelength longer than 1000 nm represents a vast source of untapped energy. The creation of NIRresponsive photoelectrodes has therefore been seen as a prerequisite for realizing full-spectrum-driven (UV-vis-NIR) solar hydrogen production.
Conventional NIR-responsive materials are limited to certain narrow bandgap semiconductors such as lead and mercury chalcogenides. The high toxicity, poor stability, and reduced redox power due to bandgap narrowing restrict their utilization as practical photoelectrodes. Localized surface plasmon resonance (LSPR) is a unique optical property that has been extensively studied in noble metal nanostructures such as Au, Ag, and Cu. Au nanostructures APL Materials have attracted especially significant attention because of the possibility of tuning the plasmonic absorption across the visible to NIR region. 98 Such absorption tunability has significant implications in utilizing the NIR spectrum where conventional semiconductor materials offer only limited choices. However, the low efficiency of plasmonic energy conversion, resulting from the ultrafast relaxation and recombination of hot carriers, has hindered the use of plasmonic metal nanostructures for solar hydrogen production. 101 Recently, heavily doped, nonstoichiometric semiconductors such as Cu 2−x S, 102 Cu 2−x Se, 103 Cu 2−x Te, 104 WO 3−x 105 , and MoO 3−x 106 have demonstrated significant LSPR absorption primarily in the NIR region. The introduction of dopants (i.e., vacancies) generates a high density of free charge carriers for these semiconductors, inducing LSPR in the relatively low energy NIR region. In contrast to the LSPR in metals, which is attributed to oscillation of the intrinsic free electrons, the LSPR in these self-doped semiconductors arises from free carriers induced by the vacancies of the nonstoichiometric crystals. For example, the copper vacancies in Cu 2−x S induce the formation of abundant free holes, while the oxygen vacancies in WO 3−x produce plentiful free electrons. Importantly, the LSPR frequency of the self-doped semiconductors can vary with changes in the degree of doping, which is not possible to achieve in metals. As a representative example, the LSPR bands of Cu 2−x S and Cu 2−x Se nanocrystals progressively increased in intensity and blueshift with increasing density of Cu vacancies. 107,108 Control over the Cu vacancy density, i.e., the x value, further allowed for dynamic tuning of the LSPR frequency and possibly the capability of harnessing the entire solar energy spectrum, rendering plasmonic semiconductors as potential candidates to fill the gap of harvesting the NIR spectrum.
With the unique LSPR features, the nonmetal, self-doped plasmonic semiconductors can be integrated with other common semiconductor photocatalysts to perform hydrogen production under extended light irradiation. In Fig. 6, we highlight two unprecedented examples of exploiting the peculiar LSPR property of nonmetal plasmonic nanostructures for solar hydrogen production. Zhang et al. reported the design of W 18 O 49 /g-C 3 N 4 heterostructures as Z-scheme photocatalysts that can harvest photon energies spanning from UV to NIR regions. 109 At 800 nm of irradiation, the W 18 O 49 /g-C 3 N 4 photocatalysts achieved an apparent quantum yield of 0.016% for hydrogen production. Recently, Lian et al. also proposed an NIR-responsive plasmonic photocatalyst comprising CdS/Cu7S 4 heterostructured nanocrystals. The apparent quantum yield reached 3.8% at 1100 nm, which is a record-breaking performance under NIR irradiation. These two pioneering works have paved the way for creating a new paradigm of photoelectrodes that presently use untapped NIR solar energy, which can boost the development of full-spectrum-driven PEC systems for much securing solar hydrogen production.

VII. SUMMARY AND OUTLOOK
How to resolve the problem of the ever rising global energy demand is one of the greatest challenges in the 21st century. Solar hydrogen production using PEC water splitting has emerged as an attractive and long-term answer to this big question. Before the widespread deployment of PEC technology becomes possible, there are several issues to be addressed. First, using in situ and operando spectroscopic techniques to investigate the charge dynamics scenario and surface reaction kinetics of a working photoelectrode is imperative. The acquired knowledge is essential for independent optimization of individual photoelectrodes and their further integration in tandem PEC cells that can achieve the ultimate goal of accessing solar hydrogen without an external bias. Second, the slow kinetics of oxygen evolution at the photoanode has been the main bottleneck for achieving efficient overall water splitting in a PEC cell. Finding other kinetically faster, technologically valuable oxidation reactions as alternatives to the oxygen evolution reaction may open new horizons for creating next-generation PEC cells that allow for the realization of artificial photosynthesis. Last but not the least, designing a new paradigm of photoelectrodes based on the untapped NIR solar energy can boost the development of fullspectrum-driven PEC systems for much securing solar hydrogen production. In this regard, nonmetal, self-doped plasmonic semiconductors have emerged as potential candidates to fill the gap of harvesting the NIR spectrum.
To date, the highest STH efficiency reported for a PEC-based system is 19.3%, 14 which is approaching the target value that can potentially compete with electrolysis systems using photovoltaic cells (30.0%). 13 Considering the simple setup and reasonable cost, PEC cells still hold great promise for realizing solar hydrogen production on an industrial scale. Developing highly efficient yet essentially robust photoelectrodes capable of harnessing the full solar spectrum is the core task in this field. We envision that further PERSPECTIVE scitation.org/journal/apm advancement of PEC technology will continue to demand and benefit from the detailed understanding of the working principles of photoelectrodes under operating conditions. Moreover, in the interest of fully utilizing intermittent solar energy, integrating PEC cells with other electrochemical devices such as batteries 111 and capacitors 112 to simultaneously harvest, store, and release solar energy is appealing as well. Overall, efforts inspired from a broad range of scientific areas will be highly desirable to allow future breakthroughs in the promotion of PEC technology.