Nanowire photochemical diodes for artificial photosynthesis

Abstract

Artificial photosynthesis can provide a solution to our current energy needs by converting small molecules such as water or carbon dioxide into useful fuels. This can be accomplished using photochemical diodes, which interface two complementary light absorbers with suitable electrocatalysts. Nanowire semiconductors provide unique advantages in terms of light absorption and catalytic activity, yet great control is required to integrate them for overall fuel production. In this review, we journey across the progress in nanowire photoelectrochemistry (PEC) over the past two decades, revealing design principles to build these nanowire photochemical diodes. To this end, we discuss the latest progress in terms of nanowire photoelectrodes, focusing on the interplay between performance, photovoltage, electronic band structure, and catalysis. Emphasis is placed on the overall system integration and semiconductor-catalyst interface, which applies to inorganic, organic, or biologic catalysts. Last, we highlight further directions that may improve the scope of nanowire PEC systems.

The integration of nanowire photoelectrodes with suitable inorganic, synthetic or biocatalysts unlocks a unique chemistry.

INTRODUCTION

The sunlight received on Earth over 1 hour is sufficient to cover our total energy demands for 1 year (1, 2). However, essential facilities, industrial processes, infrastructure, and combustion engines rely on fossil fuels, which provide a 20-times higher energy density in comparison with batteries (2–4). As a result, electricity still contributes less than 20% to our energy consumption (1). While a complete electrification of the economy is foreseeable in a distant future, the material resources and capital required to undertake major upgrades of legacy infrastructure pose immediate challenges to current economies. Hence, the sustainable production of carbon-neutral fuels from small building blocks such as water or atmospheric CO₂ is particularly desirable.

Photoelectrochemistry (PEC) provides one alternative toward renewable fuel production. In this case, photoexcited charges can be used to perform electrochemical processes, for example, proton reduction to hydrogen, or water oxidation to yield molecular oxygen. In this configuration, a photocathode is conventionally wired to a separate photoanode for unassisted fuel production, such as overall water splitting. A more elegant approach is presented through the photoelectrochemical diode, a concept that was introduced by Nozik in 1977 (5) and can be seen as a predecessor to the modern artificial leaf (6, 7). In this case, a pair of semiconductors can be directly combined through an ohmic contact, resulting in a design coupling both oxidative and reductive half-reactions for overall fuel production. The resulting device structure mimics the operating principle of natural photosynthesis, in which two complementary light absorbers are necessary to couple carbohydrate synthesis to O₂ evolution. While this approach suits overall water splitting, CO₂ reduction (CO₂R) poses additional challenges including higher catalytic overpotentials or a control of the selectivity.

Semiconductor nanowires present a number of advantages that are key to overcoming those challenges (8). First, nanowires have a high surface area that can accommodate more active sites or allow higher catalyst loadings, translating to an increased turnover of the light-driven reaction. Second, photoelectrochemical nanowires are known to enhance charge separation, as photoexcited carriers only need to travel a short distance to the semiconductor-electrolyte interface. This is crucial for preventing charge recombination in materials with short carrier diffusion lengths, such as transition metal oxides (9). In contrast to a flat semiconductor, nanowire arrays further suppress reflection, which, in turn, improves light absorption. This light-trapping effect occurs because of the orientation of the nanowires and their variable refractive index, which enhance light scattering within the array. Last, nanowire arrays provide a considerable control over the mass transport inside the array, which can be accordingly simulated and adjusted for an increased reactivity (8). In this review, we take a closer look at how these factors govern the performance of nanowire photoelectrodes for solar fuel production. Furthermore, we evaluate the role of the catalyst interface for a wider product scope and extract general design principles toward system integration.

DESIGN PRINCIPLES FOR NANOWIRE PHOTOCHEMICAL DIODES

Because of their unique properties, nanowires provide a multitude of handles to steer the efficiency and selectivity of photoelectrocatalysis. For example, the fabrication of nanowire arrays can affect their light absorption through the choice of diameter, length, and spacing, whereas the presence of heterojunctions, passivating layers, or surface-bound catalysts can alleviate the photovoltage requirements for a given half-reaction. These considerations become particularly relevant in case of unassisted devices combining two light absorbers for overall solar fuel production, where factors including photocurrent or bandgap matching need to be further considered. To provide a complete picture of the opportunities available in nanowire PEC, we take a closer look at these design principles below.

History

To understand the design principles behind nanowire PEC, we first need to look back at its development throughout the years. Semiconductor nanowires have been synthesized since the 1990s through a range of techniques including vapor-liquid-solid (VLS) growth (10, 11), etching (12, 13), or laser ablation (14). While VLS and etching methods remained very popular, an entire range of template-directed methods have been developed in the meantime (15). Originally, a vast amount of literature only reported fundamental studies of their material and physicochemical properties (16, 17). These properties are conventionally probed by means of scanning electron microscopy (SEM), transmission electron microscopy, energy-dispersive x-ray spectroscopy, x-ray diffraction analysis, or surface area techniques, whereas two- and four-point probe stations are necessary to determine the electronic and thermal properties of individual nanowires (15).

Nanowire arrays first found application as photoelectrodes in dye-sensitized solar cells (18–23). In this case, a ruthenium-based dye was coating ZnO nanowires, which replaced the conventional micrometer-thick films of oxide nanoparticles. This proved beneficial in collecting photoexcited carriers by avoiding the trap-limited diffusion process, which accounts for electron transport in nanoparticle films (19). Nanowire-based (photo)electrodes later expanded to a wide range of energy applications beyond photovoltaics (24), including pollutant degradation (25, 26), batteries (27, 28), and supercapacitors (28–30).

Similar nanowires were also used for photoelectrochemical fuel production in a range of configurations. Individual photoelectrodes could either be synthesized in the shape of randomly oriented nanowires on a conductive substrate (31) or as oriented nanowire arrays with a controlled growth (32, 33), depending on the deposition technique and desired properties (Fig. 1, A and B). Some of the earliest reports of overall water splitting involved both nanowire arrangements. For example, bilayer nanowire meshes could be assembled from Ru/Rh-SrTiO₃ and BiVO₄ nanowires (34), whereas TiO₂ nanowires were grown onto larger Si nanowires in a so-called nanotree heterostructure (35). These systems took the original concept of a photochemical diode and expanded it toward a nanowire structure, as illustrated in Fig. 1C (36, 37). As opposed to Nozik’s prototypes (5), where an opaque ohmic contact (metal layers bonded by a silver epoxy cement) meant that both semiconductors need to be irradiated separately, these nanowire designs enabled a single optical path, which corresponds to the natural sunlight irradiation. This concept of nanowire-based photochemical diodes was initially proposed in the Department of Energy Helios Program at Lawrence Berkeley National Laboratory in the early 2000s (38, 39) and was later propagated into the Energy Hub Joint Center of Artificial Photosynthesis.

Fig. 1. Conventional PEC nanowire structures.

(A) Randomly oriented nanowires on a conductive substrate (e.g., FTO|Fe₂O₃) (31). (B) Ordered nanowires, which are grown or etched into a solid wafer (as in the case of Si). (C) Concept of a nanowire photochemical diode, where two nanowire arrays are interfaced through a transparent ohmic contact. (D) Overlap of the photocathode (red) and photoanode (blue) traces for a real device and in an ideal case. (E) Photocurrent traces of the corresponding bias-free PEC tandem devices. In this example, the overall device photocurrent is limited by the lower photocurrent of the photoanode (40). (F) Schematic energy diagram of a photochemical diode coupling a CO₂R reaction (CO₂RR) with water oxidation, as depicted in (C).

PEC performance

While nanowire photoelectrodes can be synthesized in a wide range of shapes and configurations, further factors need to be taken into account when designing a high-performance photoelectrochemical diode. In general, the solar-to-fuel (STF) conversion efficiency of a photoelectrochemical device depends on the photocurrent or product amount obtained under operation. In case of an unassisted tandem PEC device, the bias-free photocurrent and operating voltage correspond to the point where the individual photocurrent traces of the photocathode and photoanode intersect (Fig. 1D) (40). Ideally, both electrodes display high photocurrents, providing a large overlap between the photocathode and photoanode traces, which translates to a high bias-free photocurrent and early onset bias voltage of the corresponding two-electrode PEC device (Fig. 1E). In a real case, photoelectrode onset potentials may be delayed because of high catalyst overpotentials, while photocurrent mismatch between the two electrodes may further decrease the photocurrent overlap. In turn, these effects can shift the operating voltage (V_op), which determines the potential applied onto an electrocatalyst under bias-free operation.

Other important aspects to keep in mind are the bandgap, absorption, and transmittance of the top light absorber, which determine how much light reaches the bottom light harvester in a tandem PEC configuration. Two semiconductors of similar bandgaps will yield a low overall photocurrent, as the top light absorber filters most of the usable solar spectrum, which substantially diminishes the photocurrent of the bottom absorber (41). Accordingly, most PEC and photovoltaic (PV) models predict maximum tandem efficiencies for top and bottom semiconductors with bandgaps of 1.6 to 1.7 and 0.95 to 1.0 eV, respectively, depending on solar spectrum, light concentration, and catalyst activity (42, 43). These theoretical efficiencies can amount to 20 to 30% for solar-to-hydrogen (STH) conversion (43). The optimal configuration becomes more challenging to evaluate for nanowire diodes, as light can be either transmitted directly or filtered through the top light absorber depending on the nanowire density and morphology. A nanostructured electrode may also introduce substantial light scattering.

Stability is also an important factor for these devices to be environmentally and economically practical. The high surface area of the nanowire geometry and the synthetic conditions to produce nanowires generally lead to the presence of surface defects. Surface defects can lead to the pinning of the Fermi level (44), which reduces PEC device efficiency and induces photocorrosion, resulting in the degradation of the photoelectrodes (45). In general, photoelectrodes need to be protected using passivating and charge carrier selective layers (46, 47), whereas high–surface area materials require additional consideration to prevent any site from being exposed to the environment. Demonstrations of stability using diurnal cycling to simulate real conditions are also necessary to show the behavior of the device when returning to open circuit conditions for extended periods of time (48). The potential switching resulting from illumination to dark can induce failure due to corrosion or contamination, as observed in case of planar systems. For example, the lack of SiO_x formed under illumination leads to a rapid corrosion of Si under the dark (49), while carbonaceous impurities (48) and catalyst dissolution (50) can affect the activity of Pt.

Nanowire semiconductor design

Bandgap

Careful design of the light absorber is a first step to improving the overall performance of PEC devices. While one-dimensional nanowires offer several benefits as described earlier, band behavior and nanowire geometry must also be considered. Photoabsorbers require suitable band positions to drive reactions such as water oxidation, proton reduction, or CO₂R. For example, the half-reactions for water splitting are

$$2{\mathrm{H}}^{+}+2e^{-}\to {\mathrm{H}}_2{\mathrm{E}}_{{\mathrm{H}}^{+}/{\mathrm{H}}_2}^0=0\mathrm{V}\text{versus RHE}$$ (1)

$$2{\mathrm{H}}_2\mathrm{O}\to {\mathrm{O}}_2+4{\mathrm{H}}^{+}+4e^{-}{\mathrm{E}}_{{\mathrm{O}}_2,{\mathrm{H}}^{+}/{\mathrm{H}}_2\mathrm{O}}^0=1.23\mathrm{V}\text{versus RHE}$$ (2)

and select half-reactions for CO₂R are

$${\mathrm{CO}}_2+2{\mathrm{H}}^{+}+2e^{-}\to \mathrm{CO}+{\mathrm{H}}_2\mathrm{O}{\mathrm{E}}_{{\mathrm{CO}}_2/\mathrm{CO}}^0=-0.11\mathrm{V}\text{versus RHE}$$ (3)

$$2\mathrm{C}{\mathrm{O}}_2+12{\mathrm{H}}^{+}+12e^{-}\to {\mathrm{C}}_2{\mathrm{H}}_4+4{\mathrm{H}}_2\mathrm{O}{\mathrm{E}}_{{\mathrm{C}\mathrm{O}}_2/{\mathrm{C}}_2{\mathrm{H}}_4}^0=0.08\mathrm{V}\text{versus RHE}$$ (4)

where the potentials are given against the reversible hydrogen electrode (RHE).

The photocurrent can be improved by selecting semiconductors with suitable bandgaps, which can use a wider range of the visible spectrum. A wider bandgap can also increase the photovoltage, resulting in an earlier onset potential for the reaction of interest (51). For bias-free PEC devices, the total photovoltage supplied by the system needs to be larger than the thermodynamic requirement of the two half-reactions. This means that the semiconductors must supply at least 1.23 V for water splitting or 1.34 V for CO₂R to CO coupled with water oxidation. However, reaction kinetics, surface interactions, or local concentration gradients lead to additional overpotentials in real-world systems, which usually contribute a few hundred millivolts to the minimum required voltage.

Band bending and depletion region

Band bending and the width of the depletion region further influence charge extraction and the nanowire photovoltage. Band bending occurs when a semiconductor interfaces with the electrolyte and the Fermi levels equilibrate (Fig. 2A). This bending creates a depletion region in which the formed electric field drives mobile charge carriers away and influences the photovoltage, which arises from the splitting of quasi-Fermi levels that are generated by illumination (52). However, for thin nanowires, the diameter of the nanowire may be smaller than the width of the depletion region (Fig. 2B). This effect can be seen in the recombination rates, which, for a thin n-type nanowire, are given by Eqs. 5 and 6

$$J_{\text{bulk r}}=\frac{q{D}_{\mathrm{p}}{n}_i^2(D-2w)}{8N_{\mathrm{D}}{L}_{\mathrm{P}}^2}(e^{\frac{qV}{k_{\mathrm{B}}T}}-1)$$ (5)

$$J_{\text{depl r}}=J_{\text{depl r},\mathrm{planar}}+\frac{q{n}_{i}Z(D,w)D_p}{2r_{\text{depl r},\max }{L}_{\mathrm{P}}^2}{e}^{\frac{qV}{2k_{\mathrm{B}}T}}$$ (6)

Fig. 2. Band bending in nanowire designs.

(A) Nanowire with a diameter (D) larger than the depletion region width (w). (B) Nanowire with a diameter smaller than depletion region width. (C) Tapered nanowire. (D) Radial homojunction in the nanowire. (E) Radial heterojunction on the nanowire.

J_bulk r and J_depl r refer to the recombination rates in the bulk and in the depletion region for a nanowire, respectively. k_B is the Boltzmann constant. T is the temperature. q is the elementary charge. D_p is the hole diffusion constant. n_i is the intrinsic carrier concentration. N_D is the dopant concentration. L_P is the hole diffusion length. r_depl r,max is the radial position of maximum recombination rate in the depletion region, and Z(D, w) is the effective recombination area as defined by explicit integral over D and w. A more in-depth analysis and derivation of these equations are found in an earlier report (53). For a thin nanowire, the second term in Eq. 6 becomes notable, leading to large recombination rates and a weak interfacial electric field. This weak electric field causes a weak drift current, which means improper charge separation, resulting in decreased photovoltages and photocurrents (54). This also becomes problematic in tapered nanowires (Fig. 2C), where portions of the nanowire are too narrow to properly separate and extract charges, leading to the same effects as described above (53).

One possible strategy to reduce the width of the depletion region while simultaneously increasing the photovoltage is by introducing a buried junction (Fig. 2D). Radial junctions can be easily made on nanowires because of their one-dimensional geometry. Introducing a radial junction by doping p-type silicon nanowires with an n⁺ layer has been shown to increase the photovoltage up to 300 mV by making the band bending dependent on the n⁺p junction rather than the interface with the electrolyte, thus allowing the photovoltage to be maintained across electrolytes of varying pH from 3 to 9 (55) while also showing no degradation after 24 hours (46). This was also shown for gallium phosphide nanowires where the diameter of the nanowire was less than that of the depletion region and resulted in poor performance. When doping the surface with zinc, the width of the depletion region decreased substantially, allowing for higher photovoltages and photocurrents (56). Proper management of nanowire diameters, band bending, and depletion regions allow for more efficient charge separation, which translates to higher photovoltage, photocurrent, and stability. However, excessive concentrations of dopants can lead to increased rates of recombination, so care is needed in forming these buried junctions.

Passivation layers

Photocurrent and stability can also be improved by adding one or more layers onto the nanowire photoelectrodes (Fig. 2E). Inert transparent layers are typically grown on top of the semiconductors to prevent electrolyte exposure, passivate surface states, and promote charge separation by acting as an electron- or hole-selective layer (57, 58). The most commonly used layer, TiO₂, can stabilize Si, GaAs, and GaP for oxygen evolution over 100 hours in 1 M KOH (47). Ideally, the charge carrier should be transported across these layers ballistically to maximize efficiency, but additional layers onto the surface may introduce other effects such as reduced conductivity, band level pinning, and unwanted catalytic behavior (59–61). For example, TiO₂ can get reduced under cathodic conditions, which either leads to oxygen vacancies that trap electrons (61) or the presence of metallic Ti, which behaves as a hydrogen evolution catalyst (62).

It has also been shown that band pinning could occur when these layers interface with the electrolyte. When using Pt nanoparticles on TiO₂-passivated Si (i.e., the electrolyte was in contact with both Pt and TiO₂), the conduction band was pinned to −0.23 V versus RHE, creating a barrier for the electrons from reducing Fe(III) until the applied potential was 0.75 V versus RHE. If TiO₂ was not used or a Pt film was used instead (so the electrolyte does not contact TiO₂), then the photoelectrode was capable of reducing Fe(III) at 1.28 V versus RHE, which is the expected potential based off the semiconductor photovoltage (46). As a result, the desired reaction may not happen at the expected potential because of the interaction of additional layers on the surface.

Nanowire geometry

Choosing the appropriate nanowire geometry also plays an important role in increasing the efficiency of a PEC device. As mentioned earlier, nanowires with diameters thinner than the width of the depletion layer suffer from poor charge separation. In addition, nanowire geometry has been shown to affect the wavelength of light absorbed. For the same height, silicon nanowires exhibit larger overall optical absorption than their planar counterparts (63), whereas ordered arrays absorb more photons than nonordered samples (64). However, the wavelength of absorbed light is dependent on the diameter, pitch, and length of the nanowires (65). For a nanowire array with a diameter of 50 nm, a pitch of 100 nm, and a length of 1000 to 4000 nm, absorption was near 1 for wavelengths above 440 nm but dropped to 0 for wavelengths below 620 nm (63). This could be shifted to longer wavelengths by increasing the pitch up to 700 nm (66). In general, absorption increases at diameters and pitches of similar length to the incoming light (65). The angle of light incidence also affects light absorption. For example, the largest short-circuit current in silicon nanowires was seen when physically rotating the array by 35° (67), whereas the optimal inclined angle of GaN nanowires was calculated to be 0.6° (68).

The nanowire geometry also controls the surface area, which can influence the photocurrent and product selectivity. In the case of nanowire photoelectrodes, the optimal surface area for a nanowire photoelectrode must balance the photovoltage loss from increased surface area and catalytic enhancement from increased catalyst loading. The activity of a catalyst toward a reaction can be described by the Tafel equation

$$\mathrm{\eta }=a+b\log J$$ (7)

where η is the overpotential, a is the intercept given by the exchange current density, b is the Tafel slope, and J is the current density. The Tafel slope, conventionally reported with units of millivolts per decade, describes how much overpotential is needed to increase the current 10-fold in an ideal system. These values can be compared with the photovoltage loss coming from a textured semiconductor through

$$V_{\mathrm{ph}}(J)=\frac{2.3n{k}_{\mathrm{B}}T}{q}\log (\frac{|J_{\mathrm{sc}}|-|J|}{R|J_0|}+1)$$ (8)

where V_ph is the photovoltage, n is the diode quality factor, J_sc is the short-circuit current, R is the roughness factor, and J₀ is the dark current density or recombination current. To see improvements in performance, the reduction in photovoltage must be smaller than the reduction in overpotential for the specified reaction, or, in other words, reactions with sluggish kinetics are more likely to see improvements due to increased surface despite the lower photovoltage. However, variations in J₀ and n due to fabrication can lead to a disparity between experimentally observed losses of photovoltage and predicted values (69).

Catalysis

Besides increasing the photovoltage, a similar effect on the PEC device performance can be obtained by using selective catalysts with low overpotentials and high turnovers toward a desired reaction. Tailored catalysts can efficiently use photogenerated carriers to perform water oxidation, proton reduction, or CO₂R, alleviating overpotentials occurring at the semiconductor-electrolyte interface. Table 1 shows several examples of electrocatalysts that demonstrate low overpotentials at moderate current densities, making them potential candidates for use in a PEC device.

Table 1. Comparison between the performance of electrocatalysts for oxygen evolution and CO₂R.

These catalysts requiring low overpotentials may be suitable for PEC systems. J_prod, partial current density; FE, faradaic efficiency toward the desired product.

Catalyst	Performance at Jprod = 0.5 mA cm−2		Performance at Jprod = 1.0 mA cm−2		Electrolyte	Stability (hours)	Reference
	E (V versus RHE)	FE (%)	E (V versus RHE)	FE (%)
Oxygen evolution (O2)
Fe-Mn-O nanosheet	1.42		1.44		1 M KOH	2500 cycles	(192)
Single Co on TiO2	1.52		1.54		1 M KOH	42	(193)
MoS2/Co9S8/Ni3S2/Ni	1.27		1.28		1 M KOH	24	(194)
Mn-doped RuO2	1.32		1.34		1 M H2SO4	5.6	(195)
Amorphous IrOx	1.41		1.42		1 M H2SO4	24	(196)
CO2 ➔ CO
Oxide-derived Au	−0.27	77	−0.29	83	0.5 M NaHCO3	8	(197)
Single Fe3+ on N-doped carbon	−0.27	83	−0.29	85	0.5 M KHCO3	12	(198)
Au nanoneedles	−0.23	69	−0.24	75	0.5 M KHCO3	8	(199)
Au-NOLI	−0.37	74	−0.41	93	0.1 M KHCO3	8	(200)
CO2 ➔ HCOO−
B-doped Bi	−0.4	54	−0.44	67	0.1 M KHCO3	13	(201)
Partially oxidized Co atomic layers	−0.18	22	−0.19	30	0.1 M Na2SO4	12	(198)
Single-atom Pb on Cu	−0.32	78	−0.43	89	0.5 M KHCO3	–	(202)
CO2 ➔ CH3OH
5 nm CoNP on N-doped graphene	−0.1	35	−0.15	53	0.1 M NaHCO3	10	(203)
Pd/SnO2 nanosheet	−0.18	45	–		0.1 M NaHCO3	25	(204)
CO2 ➔ C2H5OH
Dense Cu nanowires	−0.58	9	−0.62	13	0.1 M KHCO3	1	(205)
Amorphous CuTi alloy	−0.50	9	−0.60	12	0.1 M KHCO3	3 months	(206)
CoO on N-doped carbon	−0.32	55	−0.37	61	0.5 M KHCO3	5.5	(207)
CO2 ➔ C2H4
CuO nanoplates	−0.45	16	−0.50	25	0.5 M KCl	24	(208)
O2 plasma Cu	−0.63	13	−0.67	12	0.1 M KHCO3	5	(209)
Cu nanoparticle ensemble	−0.66	11	−0.69	16	0.1 M CsHCO3	10	(210)

By taking into account the reaction overpotentials of state-of-the-art catalysts, one can quickly assess which components are needed in a bias-free PEC device. The full cell voltage required to drive a reaction at a partial current density of 0.5 or 1.0 mA cm⁻² can be calculated using Table 1 by subtracting the potentials of the corresponding anodic and cathodic half-reactions. For example, platinum and amorphous iridium oxide catalysts operate at applied potentials of −0.02 V versus RHE for H₂ evolution and 1.42 V versus RHE for O₂ evolution, respectively, amounting to a 1.44 V photovoltage for overall water splitting. Similarly, coupling CO₂R with O₂ evolution demands 1.71 V using oxide-derived Au for CO₂-to-CO conversion or 2.11 V with Cu nanoparticle ensembles for CO₂-to-C₂H₄ reduction.

The activity of a given electrocatalyst can also depend on the local environment, morphology, and its interface with the underlying catalyst support. Semiconductor substrates could lead to increased overpotentials and differing selectivities over catalysts deposited on dark electrodes. For example, in the previous simplified analysis, we would expect TiO₂ to drive overall water splitting given that its photovoltage of ~1.6 V is more than the calculated value of 1.44 V. However, in Fujishima and Honda’s (70) pioneering demonstration of unassisted O₂ evolution, the corresponding cathodic reaction was Fe³⁺ reduction rather than H₂ evolution to ease voltage requirements, suggesting that TiO₂ cannot perform overall water splitting in itself. Indeed, illuminated TiO₂ coupled with O₂ evolution catalysts shows onset potentials above 0.1 V versus RHE (Table 2) instead of an expected onset of around −0.2 V versus RHE, suggesting that ~300 mV of photovoltage is lost. A possible reason for this discrepancy is the level of the conduction band, ~2 eV below the redox potential for oxygen evolution, which causes inefficient charge transfer. The environment may further cause the electrocatalyst to be substantially less active, thus increasing the overpotential by that amount. Hence, the interface between photoelectrode and catalyst is a key aspect, which ultimately enables a successful system integration. This interfacing may take a variety of forms, from ligands allowing a homogeneous dispersion of Au₃Cu nanoparticles along nanowire arrays (71), to anchoring groups that attach molecular catalysts onto passivating oxide coatings, and nanogaps promoting microbial growth (Fig. 3).

Table 2. Performance of representative nanowire photoelectrodes toward solar fuel production.

KP_i, potassium phosphate buffer; NWs, nanowires; μWs, microwires; NPs, nanoparticles.

Structure	J (mA cm−2)	E (V versus RHE)	FY (%)	Product	Bandgap (eV)	Onset (V versus RHE)	Stability (hours)	Conditions	Reference
Photocathodes: H2 evolution
Si NWs|MoS2 nanosheets	16.5	0		H2	1.1	0.26	48	1 sun, 0.5 M H2SO4, pH 0.5	(81)
Au|Cu2O NWs|Ga2O3|TiO2|RuOx	~10	0	–	H2	2	1	120*	1 sun, 0.5 M Na2SO4, 0.1 M NaPi, pH 5	(77)
Si|GaN NWs|Pt	38	0	~100	H2	1.1	0.56	3000	1 sun, 0.5 M H2SO4, pH ~0	(76)
Photocathodes: CO2R
InP|TiO2|Cu NPs	6.5	−0.6†	76.9	H2	1.43	~−0.1†	12	λ = 532 nm, CO2 saturated 0.5 M KCl	(211)
			8.7	CH3OH
n+p-Si|GaN NWs|Sn NPs	−17.5	−0.53	20	H2	1.1	0	10	1 sun, 0.1 M KHCO3	(212)
			5	CO
			76.9	HCOOH
Cu2O|SnOx	−3.7	−0.35	16.1	H2		~−0.1	12	2 sun, λ > 420 nm, 0.5 M NaHCO3	(113)
			74.22	CO
			6.08	HCOOH
n+p-Si μWs|Cu	−20.8	−0.44	55.5	H2	1.1	~0.15	48‡	1 sun, 0.1 M KHCO3, pH 6.8	(69)
			9.5	CH4
			10.5	C2H4
n+p-Si NWs|Cu NPs	~−10	−0.5	55.9	H2	1.1	~0.05	50	1 sun, 0.1 M KHCO3, pH 6.8	(111)
			7.5	HCOOH
			3.1	CO
			4.4	CH4
			24.3	C2H4
			3.1	C2H5OH
			3.9	C3H7OH
Photocathodes: biologic systems
Si NWs/S. ovata	~−0.2	0.14	82	CH3COOH	1.1	0.26	130	1 sun, 20% CO2/80% N2 purged, brackish H2O, pH 6.3 to 6.7	(72)
Photoanodes
TiO2 NWs|ALD TiO2	1.1	1.5	–	O2	3.0-3.2	0.1	–	1 sun, 1 M NaOH	(119)
WO3 nanorods	2.26	1.23	–	O2	2.61	0.5	–	1 sun, 0.5 M Na2SO4	(124)
BiVO4|FeOOH|NiOOH nanoworms	4.5	1.23	>90	O2	2.4	0.2	48	1 sun, 0.5 M KPi, pH 7	(121)
Sn:Fe2O3 NWs	3	1.6	–	O2	2.2	0.8	–	1 sun, 1 M NaOH, pH 13.6	(122)
WO3|W:BiVO4 NWs	3.1	1.23	–	O2	2.6/2.4	0.6	–	1 sun, 0.5 M KPi, pH 8	(140)
Si|Fe2O3 NWs	0.89	1.23	–	O2	1.1/2.2	0.6	–	1 sun, 1 M NaOH	(142)
Branched ZnO NWs|Au NPs	1.7	1.23	–	O2	3.1	0.3	3	1 sun, 0.5 M Na2SO4, pH ∼7.0	(151)

*0.5 V versus RHE.

†V versus the normal hydrogen electrode (NHE).

‡−0.58 V versus RHE.

Fig. 3. Design principles for nanowire photoelectrodes.

Nanowires can be used for proton and CO₂R to small products (blue halo) (A and B), higher value multicarbon synthesis (red) (C and D), or water oxidation (yellow) (E and F). Light harvesting nanowires can form a direct semiconductor-electrolyte interface (F), or they can be coated with suitable electrocatalytic, passivating, or charge-selective nanostructures. These nanostructures include conformal coatings (A), nanoparticles (C), flakes, or nanowires (E). Specific inorganic (C), molecular (B), or biocatalysts (D) are often required to attain high selectivity toward carbon-based products. These catalysts can be interfaced to the nanowire surface through direct deposition (C), anchoring groups (B), or flagella/pili growth (D) (117).

Table 3. Comparison between the performance of nanowire PEC systems for unassisted solar fuel production.

Structure	J (mA cm−2)	FE (%)	Product	STF (%)	Stability (hours)	Conditions	Reference
Photochemical diodes
Si/TiO2 nanotree			H2, O2	0.12	4.5	1.5 sun, 0.5 M H2SO4	(35)
Ru/Rh-SrTiO3–BiVO4 mesh			H2, O2	0.0017	18	450 W Xe lamp, λ ≥ 400 nm, H2SO4, pH 3.5	(34)
Double-band p-GaN/p-In0.2Ga0.8N			H2, O2	~1.8 (H2)	10	~26 sun, H2O, pH ~7.0	(159)
Wired PEC
InP-BiVO4 (side by side)	0.41 ± 0.02		H2, O2	0.5	0.75	1 sun, 1 M KPi, pH 7	(94)
p-Si nanoarray|Pt-Mo:BiVO4|CoPi	0.46	100*	H2, O2	0.57	3.5	1 sun, 0.1 M KPi, pH 5.5	(156)
Au|Cu2O|Ga2O3|TiO2|RuOx-BiVO4|CoPi	~2.4	~100	H2, O2	~3	12	1 sun, 0.2 M KBi, pH 9.0	(77)
Biophotochemical diodes
Si/S. ovata–TiO2	0.35	86 ± 9	CH3COOH, O2	0.38	130	1 sun, 20% CO2/80% N2 purged, brackish H2O, pH 6.3 to 6.7	(72)
		52	PHB	0.2		(+E. coli)
Si/S. ovata–Si multi-jn PV–Pt	~0.65	~80	CH3COOH	~3.6	168	0.25 sun, 20% CO2/80% N2 purged, phosphate-enhanced medium, pH 6.4	(161)

*Assumed.

The overpotential, selectivity, and turnover requirements can be further alleviated by introducing biologic systems such as enzymes or bacteria, which operate near the thermodynamic potential of a given reaction (72). These biocatalysts cover a wide range of stability, cost, and reaction scope (73). While enzymes degrade rapidly outside the cellular matrix and require complex purification steps, they often set a benchmark in terms of catalytic activity and product selectivity because of their low overpotentials and high turnovers surpassing those of inorganic catalysts. An understanding of their active sites can further provide guidelines for the synthesis of high-performance, bioinspired catalysts. On the other hand, bacteria stand out as robust catalysts for hybrid bioinorganic laboratory prototypes and even industrial processes (e.g., fermentation or biofuel production). Those bacteria cultures can be sustained over days and provide a complex range of products through metabolic cascade reactions. In this case, nanowires provide a high–surface area platform to grow bacterial colonies, which can conveniently attach through flagella or pili (72, 73).

PEC PERFORMANCE OF NANOWIRES

Photocathodes

Hydrogen evolution

A great number of studies have looked into H₂ evolution using nanowire photocathodes (74–76). In terms of design, most photocathodes use silicon (75), III-V (76), or metal oxide semiconductors (77, 78), which are directly interfaced to established H₂ evolution catalysts such as Pt (75, 79), MoS₂ (80, 81), and RuO_x (77), as well as Ni-containing (77, 82, 83) or Co-containing (84) compounds (see Table 2 and Fig. 4). While these catalysts can be deposited in the shape of nanoparticles (75, 82), flakes (81), or conformal coatings, their effect on the overall photocurrent is similar, as their high current densities at overpotentials below 100 mV (85) ensure that light harvesting is the limiting step.

Fig. 4. Examples of nanowire photocathodes for H₂ evolution.

(A and D) Si nanowires (NWs) with a MoS₂ catalyst, reproduced with permission from Hu et al. (81). (B and E) Performance of multilayered, heterojunction Cu₂O photocathodes, reproduced with permission from Luo et al. (97). (C and F) Long-term stability of a GaN|Pt photocathode, adapted with permission from Vanka et al. (76).

Overall, Si remains the most popular light absorber for nanowire PEC (86, 87). The best Si nanowire electrodes display photocurrents plateauing above 30 mA cm⁻² (67, 81, 88, 89) and an onset potential of around 0.4 to 0.5 V versus RHE (Fig. 4A) (67, 82, 90). To attain these values, photocathodes often use n⁺p-Si wire arrays (67, 90), whereas p-Si arrays are limited to photocurrents of only 10 to 20 mA cm⁻² and onset potentials for proton reduction of 0.1 to 0.3 V versus RHE (67). These buried junctions decoupling the photovoltage from the band bending at the semiconductor-liquid interface can also be obtained using TiO₂ coatings, which have an additional role as passivating layers (80).

A higher photovoltage can be obtained using III-V semiconductors such as GaN (76, 91, 92), GaP (56), GaAs (93), InP (94, 95), and InAs (96), which translates to an onset potential of 0.5 to 0.6 V versus RHE for H₂ evolution (Fig. 4C). In those cases, TiO₂ coatings can be again used as protective interfaces between the semiconductor and catalyst, which facilitate charge extraction and increase the electrode stability under operation (93, 94, 96). Accordingly, GaAs|TiO₂|Pt photocathodes sustained a photocurrent density of −11 mA cm⁻² over 67 hours at −0.6 V versus RHE, whereas unprotected electrodes were completely degraded after 36 hours of operation (93). A similar way of improving the onset potential and photocathode stability is by surface modification of the nanowires (92, 95). Sulfur-dissolved oleylamine was used to produce an inorganic, protective sulfide coating onto InP nanowires (95), whereas an organic 1,2-ethanedithiol enhanced the carrier lifetime by suppressing chemisorbed ─OH and ─O bonds at the GaN nanowire surface (92). However, the highest stability was reported using a n⁺-p Si|n⁺-GaN nanowires|Pt photocathode, which could operate in 0.5 M H₂SO₄ under 1-sun irradiation for 3000 hours (see Fig. 4F and Table 2) (76).

The onset potential can be further increased beyond 0.6 V versus RHE for H₂ evolution using Cu₂O (97) or CuO (98–100) nanowire photocathodes. However, Cu₂O undergoes photodegradation in water; therefore, a careful surface passivation is required (Fig. 4E). This is typically achieved using wide-bandgap metal oxide coatings including TiO₂, ZnO (100), or NiO_x (101), which act as electron-selective layers facilitating charge separation. A careful choice of materials can improve the band alignment between the conduction band level of the light absorber and that of the electron-selective layer, reducing photovoltage losses. Accordingly, the onset potential of Cu|Cu₂O|AZO|TiO₂|RuO_x nanowire heterojunction photocathodes could be shifted from 0.6 to 1 V versus RHE, by replacing the aluminum-doped zinc oxide (AZO) coating by a Ga₂O₃ layer (77, 97). The fill factor of the electrode could be further improved by adding a hole-selective layer between a Au metal substrate and Cu₂O, which improves the overall fuel production in a bias-free tandem configuration (78), as we discuss later in more detail.

CO₂ reduction

Another reaction of interest to realizing artificial photosynthesis is the photoelectrochemical conversion of CO₂ into value-added chemicals (Fig. 5). While the redox potentials for CO₂R reactions (e.g., $E_{{\mathrm{CO}}_2/\mathrm{CO}}$ = −0.11 V versus RHE and $E_{{\mathrm{C}\mathrm{O}}_2/{\mathrm{C}}_2{\mathrm{H}}_4}$ = 0.08 V versus RHE) are similar to that of the hydrogen evolution reaction (HER; $E_{{\mathrm{H}}_2\mathrm{O}/{\mathrm{H}}_2}$ = 0 V versus RHE), the complexity of the reaction kinetics for CO₂ activation and conversion often results in needing large overpotentials to drive the reaction. In addition, reaction selectivity is essential, as there is competing HER, as well as the broad scope of products that are produced from CO₂R (102). Unlike HER where a monolayer of Pt is sufficient to drive appreciable currents (75), CO₂R catalysts often require large catalytic surface areas to achieve similar currents at modest overpotentials. To further add a layer of complexity, accessing and selecting multicarbon products are an interplay between the current density and driving potential (103).

Fig. 5. Nanowire photocathodes for CO₂R.

(A to F) Au₃Cu nanoparticles assembled on Si nanowire arrays, reproduced with permission from Kong et al. (71). (A) Schematic of the nanoparticle deposition; (B) linear sweep voltammetry (LSVs) and faradaic efficiencies toward CO production; and (C to F) SEM and corresponding elemental mapping of Cu (D), Au (E), and Si (F). Scale bars 200 nm. (G to I) Si|Cu photocathodes for multicarbon production, reproduced with permission from Kempler et al. (69). (G) False-color SEM image and (H and I) PEC performance. (J to L) Cu₂O|SnO_x photocathode, reproduced with permission from Zhang et al. (113). (J) Schematic depiction and (K and L) PEC activity. (M to O) Cu nanoparticles on Si nanowires for selective ethylene synthesis, reproduced with permission from Roh et al. (111). (M) Schematic of the Si nanowire/Cu nanoparticles assembly, (N) faradaic efficiencies, and (O) stability test.

The necessity of using high–surface area catalysts for CO₂R leads to the problem of parasitic light absorption when using planar systems. An earlier work showed that the optical transmittance of a copper film was 0% at a thickness of 100 nm, which only increased to 15% when decreasing the thickness to 35 nm. However, decreasing the thickness changed the selectivity toward methane from 60 to 0% due to the grain boundary oxidation of a thinner film (104). To overcome these limitations, various strategies have been used such as catalyst patterning to allow light transmission or direct irradiation, back-side illumination through conductive glass substrates, or choosing a photocathode material that also has catalytic activity. Patterning the substrate to selectively deposit the catalyst on portions of the surface allows light to pass through exposed areas, but this only slightly mitigates the problem as the photocurrent density is still usually reduced by ~50% (104, 105). Another method is the use of a metal prism array or dielectric nanocone structure to direct the incoming light to maximize transmittance, but this limits the area of catalyst coverage and catalyst options, which leads to generally the need for large overpotentials (106, 107). Choosing a photocathode that acts as the catalyst removes these considerations altogether but greatly limits materials selection, and such a photocathode remains to be discoveredfor selective aqueous CO₂R.

On the other hand, the unique structure of nanowire photoelectrodes bypasses these limitations and shows greater light absorption instead due to its enhanced light trapping properties as described earlier. When comparing the same mass loading of Au₃Cu nanoparticles on planar Si and Si nanowires for CO₂ conversion to CO, it was found that the total photocurrent was larger for the nanowire system over the planar system (Fig. 5, A and B) (71). This was also seen for a silicon microwire system where a large mass of Cu (330 μg cm⁻²) was photoelectrodeposited onto the wires but was still able to generate a photocurrent of over 30 mA cm⁻² under 1-sun illumination (Fig. 5, G and H) (69). On a planar substrate, this would correspond to a film thickness of approximately 370 nm where the expected optical transmittance is 0%.

Catalysts deposited onto nanowire photoelectrodes also see a lower onset potential and a higher selectivity toward the desired CO₂R product despite generally having a smaller photovoltage than their planar counterparts. For example, for CO₂R to CO, a planar InP photocathode covered with TiO₂ and Au showed a current density of 2 mA cm⁻² at −0.11 V versus RHE while etching the InP into a nanopillar array increased it to 4.53 mA cm⁻² at the same potential. Along with the increase in current density, the faradaic efficiency toward CO also increased from 53.2 to 84.2%, which corresponds to an almost fourfold increase in partial current density toward CO production (61). This was also seen when using nitrogen-doped graphene quantum sheets onto Si photoelectrodes, where the onset potential shifted positively by 250 mV when using nanowires instead of a planar substrate and the faradaic efficiency toward CO increased from 90 to 95% (108). Similar effects were observed for CO₂R to HCOO⁻ using photoelectrodeposited tin nanoparticles on Si. The faradaic efficiency toward formate was improved from 33% on planar Si to 40% on Si nanowires, while the total production of formate was increased fivefold (109). The origin behind this shift in onset potential and increased selectivity has not been fully explored and has been often attributed to superior light absorption, but this would only explain short-circuit currents and not currents within catalytically limited regimes. We can draw from studies done in the electrochemical CO₂R community and best attribute these benefits to increased surface area of the catalyst and an increase in pH within the wire network to suppress HER (110).

These effects become more pronounced when reducing CO₂ further into multicarbon products such as ethylene, ethanol, and propanol. At low mass loadings of 5 μg cm⁻² of 7 nm Cu nanoparticles on Si nanowires, the CO:C₂H₄ ratio is 3:1 at −0.5 V versus RHE but changes to 1:8 when increasing the mass loading to 40 μg cm⁻² and reaches a faradaic efficiency toward C₂H₄ of 24.3% (111). This was attributed to the necessity of building up intermediates to access multicarbons, which occurs with more surface area. On the other hand, when copper was interfaced to planar silicon photocathodes at low mass loadings to prevent parasitic light absorption, the faradaic efficiency toward C₂H₄ could only reach 6% and mostly favored single-carbon products across all potential ranges (112).

Interestingly, nanowire photoelectrodes also see enhanced stability for CO₂R over their planar counterparts despite the nanowires generally introducing surface states that often lead to failure modes. During cycling, planar p-Si underwent a gradual deactivation within 5 cycles, while the nanowires did not. This was attributed to nanowires having a hydrogen- or fluorine-functionalized surface from etching that prevented oxidation (109). However, in a different work, x-ray photoelectron spectroscopy data showed a notable increase in the SiO₂/Si ratio on silicon nanowires after 1 hour, although it sustained stable CO₂R for over 50 hours, suggesting that prevention of oxide growth may not be the reason behind its stability (Fig. 5O) (111). Furthermore, a TiO₂-covered InP nanopillar array saw improved stability toward methanol production over its planar counterpart despite both having passivation layers to protect the surface (61), while SnO_x-modified Cu₂O nanowires sustained almost 80% faradaic efficiency toward CO for 11 hours (Fig. 5, J to L) (113). Another possible reason for this stability could be the nanowire morphology, which is better able to prevent delamination of the catalyst from the surface than in flat substrates (69).

Investigations into the charge transfer between photocathode and catalyst may help elucidate the reasons behind the benefits of using nanowire photoelectrodes for CO₂R, such as lower onset potentials, higher selectivity, and increased stability. This effect becomes prominent when using molecular catalysts where the orientation or the catalyst/surface interaction is a key consideration. When using [Ni(bpy)₂] as a catalyst, the nanowires resulted in a 300 mV lower onset potential for CO₂R over a planar Si substrate. The team concluded that improved light harvesting would fail to account for such difference in potential and suggested that this discrepancy was caused by a difference in charge transfer. Because nanowires are multifaceted, while planar silicon only has one facet exposed, the electron transfer pathways are likely formed on Si nanowire/[Ni(bpy)₂] interfaces that favor the process (114).

Biological systems such as bacteria have also been used as catalysts because of their ability to accept charges from semiconductor surfaces and their excellent selectivity at low overpotentials (115). The same considerations above for CO₂R mean that bacteria have often been used for dark cathodes because of their low optical transmittance (116) but exemplify that nanowires are an excellent substrate for forming these biohybrid interfaces. In addition, single-cell level imaging showed that Shewanella oneidensis MR-1 can recognize and show preferential attachment to nanowires (117). As a result, Sporomusa ovata has been interfaced to silicon nanowire photocathodes to reduce CO₂ into acetate with a faradaic efficiency up to 90% for 200 hours. Although S. ovata can only catalyze CO₂ under anaerobic conditions, stable CO₂R was demonstrated under aerobic conditions when Pt was added because of mass transport maintaining anaerobic conditions within the wire assembly (72). This demonstrates a unique feature of the nanowires, which can create and maintain a different environment from the bulk conditions playing a role in many catalytic reactions.

Photoanodes

In terms of oxygen evolution, transition metal oxides remain the most encountered photoanode materials. The wide bandgaps of 2.5 to 3.5 eV are accountable for their robustness under operation, which can span up to several weeks for oxide nanostructures in aqueous electrolytes (see Table 2) (41, 118). The performance is enhanced in case of the nanowires because of their light trapping and scattering effect. For example, while TiO₂ films are often transparent, the photocurrent density of TiO₂ nanowire array electrodes could be improved from 0.125 to 0.73 mA cm⁻² at 1.5 V versus RHE by increasing the nanowire length from 0.28 to 1.8 μm (119). However, a wider bandgap also restricts the amount of light that can be harvested over the visible spectrum. To address this challenge, several strategies have been explored including doping, building core-shell heterojunctions, and nanowire sensitization (Fig. 6).

Fig. 6. Strategies to improve the performance of nanowire photoanodes.

(A and E) Doping of TiO₂ photoanodes and corresponding IPCE spectra (E), reproduced with permission from Hoang et al. (129). (B, C, F, and G) Core-shell nanowire heterojunctions. (B, F, and G) WO₃|BiVO₄ electrode with corresponding IPCE spectra (F) and LSV plots (G), reproduced with permission from Rao et al. (140). (C) ZnO|SnO₂|BiVO₄ nanowire array, reproduced with permission from Bielinski et al. (139). (D) Sensitization of ZnO nanowires with CdTe quantum dots, reproduced with permission from Chen et al. (146). (H) LSV traces of TiO₂ sensitized with CdS, reproduced with permission from Ai et al. (145).

While nanowire photoanodes including TiO₂ and ZnO absorb mostly in the ultraviolet (UV) region, other oxides including WO₃ (120), BiVO₄ (121), and Fe₂O₃ (122) extend their absorption in the visible spectrum (Table 2). However, even those suffer from low photovoltages, which may be due to a high recombination, low charge carrier mobility, or noncontributing excitations (9, 123). Therefore, WO₃ only displays an onset potential of 0.5 to 0.7 V versus RHE for oxygen evolution (51, 124), whereas the 0.8 to 1.0 V versus RHE onset of hematite (125–127) is even more positive. Hence, one strategy is to dope wide-bandgap oxides with atoms such as nitrogen, carbon, tin, or titanium.

N doping of TiO₂ not only increases the incident photon to current efficiency (IPCE) in the UV region but also shifts the light adsorption to higher wavelengths (Fig. 6, A and E) (128, 129). For example, the IPCE of TiO₂ nanowire photoanodes with a cobalt oxygen evolution catalyst (OEC) was increased from around 60% to above 80% by N doping, while the IPCE spectrum of doped samples already showed activity at 500 nm irradiation (129). While doping in itself decreased the photocurrent density of TiO₂ (from 0.38 to 0.23 mA cm⁻² at 1.23 V versus RHE) and shifted the onset potential positively (from 0.2 to 0.5 V versus RHE), the Co OEC addition achieved the highest photocurrent of 0.61 mA cm⁻² at 1.23 V versus RHE for doped samples (129). Similarly, the IPCE of ZnO nanowire arrays could be improved from ~40% to over 80%, whereas the IPCE onset shifted from roughly 400 to 450 nm, resulting in a photocurrent density of 0.4 mA cm⁻² at 1.0 V versus Ag/AgCl (130). On the other hand, tetravalent atoms such as C or Sn have been shown to increase the IPCE of TiO₂ in the UV range without shifting the absorption onset (128, 131). Hydrogen treatment has been further used to reduce Ti⁴⁺ to Ti³⁺ species, yielding dark TiO₂ nanowire photoanodes with IPCE values approaching 100% at wavelengths below 400 nm (131–133).

In case of hematite, Ti (31) and Sn (122, 134) have been used as dopants. Although no notable shifts were observed in the onset potentials, these treatments did produce photocurrent densities as high as 1 mA cm⁻² for Ti:Fe₂O₃ (31) and 3 mA cm⁻² for Sn:Fe₂O₃ (122) at 1.6 V versus RHE, as opposed to the photocurrents around 0.2 mA cm⁻² of undoped samples at the same potential. The formation of oxygen vacancies and Fe²⁺ through β-FeOOH decomposition in an oxygen-deficient environment provided a similar effect, resulting in a photocurrent of 3.37 mA cm⁻² at 1.5 V versus RHE (135).

Another route to increase the performance of nanowire photoelectrodes is by making core-shell structures (Fig. 6, B, C, F, and G). In this design, the shell may act as a catalyst (136), passivating layer (137), or heterojunction improving charge separation (138, 139). For example, a W-doped BiVO₄ coating increased the photocurrent of a WO₃ nanowire photoanode from 1.1 to 3.1 mA cm⁻² at 1.23 V versus RHE, even in the absence of a dedicated OEC (140). Core-shell heterojunctions can also consist of two (140) or more (141) complementary light absorbers, which are able to harvest a larger portion of the solar spectrum. For instance, the onset potential of hematite for O₂ evolution could be shifted ~0.4 V earlier by depositing Fe₂O₃ onto Si (142). The shells themselves can be further nanostructured, thereby maximizing light absorption and the photoactive surface area. These branched nanowires have been previously demonstrated by growing TiO₂ nanorods onto Si nanowire array backbones (143) or by growing ZnS nanowires onto ZnO rods (144).

Sensitization is another popular alternative to increase the light absorption of oxide photoanodes (Fig. 6, D and H). In this case, quantum dots of II-VI semiconductors such as CdS (145), CdTe (146), or CdSe (147, 148) are deposited onto TiO₂ or ZnO nanowires. Water oxidation catalysts such as Co-Pi can again improve the photoanode onset potential and fill factor (145), whereas additional heterojunctions with light absorbers such as Cu₂O have recently proved beneficial, yielding very early onset potentials below −1.0 V versus RHE (149, 150). Alternatively, TiO₂ and ZnO nanowires can be decorated with Au nanoparticles, which achieve a broadband UV-visible absorption through plasmonic effects (151, 152).

Besides the material choice and fabrication methods, the photocurrent and onset potential can also be adjusted by depositing suitable oxygen evolution reaction (OER) catalysts. IrO₂ and RuO₂ have been established for water electrolysis, providing overpotentials of 0.3 to 0.4 V for O₂ evolution in strongly acidic or alkaline media (153). These noble metal catalysts can also be used for photoanodes (154), resulting, for example, in improved fill factors and early onset potentials of 0.2 V versus RHE for TiO₂ nanowire photoanodes (35). However, as described earlier, earth-abundant catalysts including Co, Fe, or Ni compounds are more popular for PEC applications, where reactions are often conducted under (near) neutral pH, at current densities below 10, or even 1 mA cm⁻² (85).

Overall systems

As established earlier, two complementary light absorbers are often required for overall fuel production. These systems can span a wide range of designs, from planar, buried junction tandem devices (41), to two-electrode tandem PEC devices (2), and Z-scheme photocatalyst sheets with solid mediators (155). Hence, a number or nanowire-based designs can be envisioned, including the nanowire photochemical diode (Fig. 7). While the concept of nanowire photochemical diodes, as depicted in Fig. 1C and proposed by Nozik, is challenging to implement in practice, other related systems can operate similarly (Table 3). Namely, these systems provide a single optical light path, take the macroscopic shape of a panel, and consist of two semiconductors of different bandgaps, which are interfaced either directly or through an ohmic contact.

Fig. 7. Nanowire PEC systems for overall solar fuel production.

(A to D) Nanowire photochemical diodes for water splitting. (A and B) Bilayer Ru/Rh-SrTiO₃–BiVO₄ nanowire mesh, reproduced with permission from Liu et al. (34). (C and D) Si/TiO₂ nanotree, adapted with permission from Liu et al. (35). (A and C) False color: SEM images. (B and D) Energy schemes. (E and F) Bacteria-nanowire interfaces for light-driven acetate production. (E) S. ovata bacteria are interfaced to a Si nanowire electrode. Unassisted acetate production is performed by wiring the Si nanowire photocathode to a TiO₂ photoanode, reproduced with permission from Liu et al. (72). (F) Wood-Ljungdahl pathway for acetate production within an S. ovata bacterium, adapted with permission from Su et al. (161). (G) Efficient nanowire photochemical diodes require further development on the light harvesting and catalysis sides. A positive shift in the cathodic signal can be obtained by using catalysts with lower overpotentials for CO₂RR and HER or photocathodes providing higher photovoltages and photocurrent outputs. Similar effects can increase the activity of a photoanode on the OER side. A suitable integration of the oxidative and reductive sides in a single–optical path system is required to maximize light absorption.

A first category involves wired PEC systems for overall water splitting. In this case, a photoanode can be wired to a photocathode in an overlapping, single light path (tandem) or side-by-side arrangement. Electrodes are often facing the light source to make use of the light trapping advantages of the nanowire geometry. For example, appropriate catalysts result in earlier onset potentials for Si nanowire photocathodes (0.45 V versus RHE) and BiVO₄ photoanodes (0.3 V versus RHE), resulting in an overlap at 0.38 V versus RHE, with a steady-state STH efficiency of 0.57% for the corresponding tandem PEC device (156). The overlap can be improved by introducing photocathodes with early onset potentials, such as the Cu₂O nanowire heterojunction electrodes. Their onset potential of 1 V versus RHE shifts the operating potential of Cu₂O-BiVO₄ tandem devices to around 0.6 V versus RHE, resulting in a 3% STH efficiency (77). A side-by-side arrangement is beneficial if the electrodes are deposited on opaque substrates or if the light absorption and scattering through the first semiconductor is too strong. However, care must be taken when calculating the STF efficiency, as the geometric areas of both light absorbers add up in a two–optical path configuration (94, 157).

Photocatalyst systems, where no charges are flowing through an external circuit, are closer to Nozik’s photochemical diode design from Fig. 1C. In this case, semiconductors can be interfaced either directly, forming heterojunctions (158) and ohmic contacts (35), or via solid-state conductive mediators such as ITO nanoparticles (155). This contact between nanowires becomes particularly important because a limited contact area, insufficient charge separation, or charge recombination at the interface can substantially decrease the overall fuel production. For instance, an eightfold decrease in the rates of H₂ and O₂ evolution was observed when replacing a mixed Ru/Rh-SrTiO₃–BiVO₄ nanowire mesh with the corresponding bilayer mesh (Fig. 7A). For the latter, only nanowires in the vicinity of the Ru/Rh-SrTiO₃–BiVO₄ layer interface contributed to the charge separation and subsequent fuel production (34). One way of avoiding parasitic contact losses is by growing nanowire structures directly on top of each other. Accordingly, a 0.12% STH efficiency was obtained by partly covering Si rods with TiO₂ nanowires in a so-called nanotree heterostructure. In this configuration, UV light is used by TiO₂ for O₂ evolution, whereas the visible light is transmitted to the exposed Si wire underneath for proton reduction (Fig. 7C), fulfilling the requirements of a single light path (35). A similar strategy could be applied to other types of heterostructures, e.g., the growth of CdS nanowires on BiVO₄ platelets for overall water splitting (158). This nanowire photochemical diode could be further integrated by introducing heterojunctions along a single nanowire. Accordingly, p-GaN/p-In_0.2Ga_0.8N nanowires, where the two light absorbers have bandgaps of 3.4 and 2.6 eV, were estimated to reach a ~1.8% STH efficiency under ~26-sun irradiation (159).

As noticeable from the previous examples, most systems are limited to overall water splitting because additional overpotentials of several hundred millivolts occur for CO₂R (160). However, these limitations can be overcome using biologic systems operating at very low overpotentials, close to the standard electrode potentials. Taking advantage of an overpotential <200 mV of the bacterium S. ovata, a Si-TiO₂ nanowire PEC tandem was able to perform unassisted acetate production at a 0.38% STF efficiency and faradaic efficiency up to 90%. The versatility of this approach toward making higher-value multicarbon products was demonstrated by feeding the acetate to a genetically engineered Escherichia coli. This bacterium could convert the acetate to n-butanol, polyhydroxybutyrate (PHB) polymer, and different isoprenoid natural products via acetyl–coenzyme A (CoA), attaining a 0.20% overall STF efficiency for the PHB synthesis (72). More recently, a close packing of S. ovata bacteria onto conductive Si nanowire arrays resulted in an STF of 3.6% for acetate production, when driving the process by an externally wired, multijunction Si PV cell (161).

The performance of nanowire systems also depends on their integration into suitable photoreactors. On a macroscopic level, these nanowire electrodes and the corresponding tandem devices are solid plates, which makes their setup and operation similar to those of other PEC systems (2). Normally, photoelectrodes as those depicted in Fig. 1 (A and B) can be operated in two-compartment cells, where ion-selective membranes separate the two compartments avoiding product crossover. In case of the proposed nanowire diode (Fig. 1C), the ohmic contact would provide a physical separation between the two sides of the diode, which results in overpotentials due to pH gradient buildup. This can be avoided by using strongly basic or acidic electrolytes. Under neutral conditions, these nanowire diodes would require flow systems with separators (162, 163) or integrated membranes (157) to prevent product crossover while maintaining an even pH distribution. Alternatively, the diodes could simply operate in a one-compartment reactor. The resulting explosive product mixtures can be filtered using gas-separating membrane units, as demonstrated recently for large-scale photocatalyst sheets (164).

SEMICONDUCTOR CHARGE TRANSFER

While considerable efforts have been made in terms of nanowire geometry and synthesis, doping, heterogeneous layers for passivation or band alignment, catalyst discovery and integration, or semiconductor material discovery, a fundamental goal is to articulate correlations between each of these components and the overall PEC performance: mainly, onset potential, photocurrent, selectivity, and stability. Understanding charge transfer through these multicomponent systems can further inform the design of efficient PEC devices. For instance, there was a positive shift of 300 mV using [Ni(bpy)₂] (114) and 230 mV using [Co(TPA)Cl]Cl (165) on silicon nanowires compared to planar Si for PEC CO₂R, which was attributed to the multifaceted nature of the silicon nanowires, allowing for more efficient charge transfer. Those charge carrier dynamics in photoelectrochemical diodes span a wide extent of time scales. Charge migration or transfer after photoexcitation occurs on femtosecond to picosecond time scales (166). Depending on the degree of charge separation and extraction, recombination rates occur on nanosecond to millisecond time scales (167), while surface reactions span millisecond to second time scales depending on the kinetics of the reaction. The most prevalent techniques that have been used to study charge transfer are electrochemical impedance spectroscopy (EIS) and transient absorption, as described below.

EIS using an alternating current measures the current response to a sinusoidal potential perturbation as a function of frequency and can be used to differentiate sources of overpotentials that may be difficult to detangle otherwise. It is a powerful technique that examines frequency-dependent processes such as charge transfer and diffusion but requires careful analysis and knowledge of the system to use representative equivalent circuits (168, 169). Most often for (photo)electrochemical systems, EIS is used as a supporting technique to confirm differences in activity. For example, in a CdS nanowire system decorated with Ni₃S₂ for water reduction, the optimal molar ratio was found to be 1:10 for Ni₃S₂ to CdS. The Nyquist plots from EIS revealed that this molar ratio gave the smallest charge transfer resistance under illumination when Ni₃S₂ was interfaced with CdS. However, the resistance was larger than pure Ni₃S₂ by three orders of magnitude, showing that the enhanced activity comes from a better charge extraction from the CdS (170). In addition, EIS also helped explain how dual-etching TiO₂ and doping with W enhanced the photocurrent by 225%. By creating a Mott-Schottky plot, flat-band potentials and charge carrier densities were extracted from the EIS data. The dual-etched–doped TiO₂ exhibited a higher flat-band potential of −0.60 V versus Ag/AgCl and a larger charge carrier density of 5.04 × 10¹⁸ cm⁻³, while the pristine TiO₂ had a lower flat-band potential of −0.89 V versus Ag/AgCl and a lower charge carrier density of 3.86 × 10¹⁸ cm⁻³. The higher flat-band potential facilitates more efficient charge transfer, while the charge carrier density leads to a larger photocurrent (171).

Another powerful alternative to elucidate charge transfer is by looking at excited state lifetimes through transient absorption. Charge transfer processes are on the order of femtosecond to picoseconds, which makes ultrafast spectroscopy a candidate for measuring their kinetics. Transient absorption probes charge carrier dynamics by applying a pump to photoexcite the electrode and measuring the change in absorption over time. In the CdS/Ni₃S₂ example above, the impedance spectroscopy was complemented by ultrafast transient absorption spectroscopy to get a clearer understanding of why the 1:10 molar ratio of Ni₃S₂ to CdS ratio had the best performance. The average lifetime of the excitons for pure CdS was calculated to be 686 ps but decreased to 185 ps with the addition of 10% Ni₃S₂. However, upon further addition of Ni₃S₂, the average lifetime increased to 461 ps. This was attributed to the intimate contact of the Ni₃S₂ and the fast transfer of charges to Ni₃S₂ from the CdS. On the other hand, further increasing the molar ratio to 3:10 led to the formation of flocculent structures that led to decreased contact with the CdS, resulting in increased lifetimes (170).

With the advent of core-level extreme UV (XUV) ultrafast transient spectroscopy, it is possible to get element-specific information while also probing electron and hole dynamics independently. Unlike UV, visible, or infrared spectroscopy, which has overlapping features from all the elements within the light path, XUV or x-rays are capable of core-to-valence transitions with a unique signature for each element (172). This technique was used to measure charge transfer kinetics in a Ni-TiO₂-Si junction by probing the Ti M_2,3 edge, Ni M_2,3 edge, and Si L_2,3 edge and found that the photogenerated hole travels ballistically through the 19 nmthick TiO₂ layer into the Ni on a femtosecond time scale. Afterward, the holes back-diffuse through the TiO₂ and recombine at the Si/TiO₂ interface on picosecond time scales (173). This confirms previous studies that saw thick layers of amorphous TiO₂ being used as passivation and transport layers without losses in conductivity (47, 57).

In addition to these two more representative techniques, other unique approaches have been used to track charge transfer in photoelectrodes such as scanning electrochemical spectroscopy (45), single-particle photoluminescence spectroscopy (174), and Kelvin probe force microscopy (175). While offering important insights, these approaches are not readily generalizable to studying a wide range of photoelectrodes due to their setup. However, the development and advancement of all mentioned techniques will be crucial in gaining deep insights into the photoelectrochemical process and design.

OUTLOOK

As discussed above, a careful selection of the light absorbers and catalysts can substantially influence the performance of a nanowire PEC device toward solar fuel synthesis. These same criteria apply to the fabrication of efficient photochemical diodes. To achieve a sufficient overlap between the photocurrent traces of the photocathode and photoanode (Fig. 1D), stable photoelectrodes with high photovoltages and photocurrent outputs must be developed. These high photovoltages can be attained by either introducing new semiconductor materials or carefully designing the device structure. Earlier photoelectrochemical onset potentials can be further obtained by interfacing these light harvesters with newly discovered CO₂RR, HER, and OER catalysts, which display high selectivities at low overpotentials (Fig. 7G). Below, we investigate further strategies that may help maximize light absorption, photocurrent, and photovoltage.

In terms of photovoltage, ferroelectric materials can achieve open circuit voltages higher than the bandgap through the bulk photovoltaic effect. In these cases, charge separation occurs at domains oriented along the polarization direction. Accordingly, these materials do not require p-n junctions, as charges separate spontaneously. While this effect has been observed for BiFeO₃ crystalline films (176), similar effects are hypothesized to occur along ferroelectric perovskite nanowires such as CsGeBr₃ (177). Ferroelectric BaTiO₃ and SrTiO₃ shells have been demonstrated to alter the onset potential, photocurrent, and fill factor of TiO₂ nanowire photoanodes, depending on the poling direction (138, 178).

An opportunity to enhance the photocurrent is given by plasmonics (152, 179, 180). In this case, Au or Ag nanoparticles are deposited on nanowire photoanodes such as TiO₂ or ZnO (179). This improves the kinetics of oxygen evolution and photoactivity due to the localized surface plasmon resonances and hot electron injection (181). While Au nanoparticles often enhance light absorption within the same wavelength range (181, 182), a choice of different nanoparticle shapes (152) and electrode morphologies (151) can extend light absorption far within the visible region (151, 152). This strategy stands out in the case of nanowire photoelectrodes, as nanowires constitute an excellent scaffold to immobilize plasmonic nanoparticles at a greater loading without blocking light, resulting in notable enhancements in terms of light absorption.

A different approach to overall solar fuel production is offered by organic transformations (183–186). In this case, the thermodynamically demanding O₂ evolution ($E_{{\mathrm{H}}_2\mathrm{O}/{\mathrm{O}}_2}$ = 1.23 V versus RHE) can be replaced by organic oxidations, which already occur at lower applied potentials of 0.3 to 0.6 V versus RHE (187). This induces a negative shift in the oxidation curve, which improves the overlap, resulting in higher photocurrents under no applied bias voltage.

At last, to bring this technology closer to real-world applications, more focus needs to be placed on the high-throughput, scalable fabrication of stable nanowire PEC devices. While few reports have looked into the scalability of nanowire photoelectrodes (15, 188), strategies have been recently proposed to improve stability (76, 189). A self-passivation of the Si substrate to SiO₂ helps sustain the performance of a Si|nanowire GaAs|TiO₂|NiO_x photoanode, even as individual nanowires begin degrading (189). Efforts have also been made to deposit nanowire photoelectrodes onto flexible substrates (190), which are compatible with modern fabrication techniques (191). By combining these scalable deposition techniques with the unique control of nanowire synthesis, photochemical diodes may come closer to practical applications under real-world operating conditions.