Photoelectrochemical (PEC) cells offer the ability to convert electromagnetic energy from our largest renewable source, the Sun, to stored chemical energy through the splitting of water into molecular oxygen and hydrogen. Hematite (α-Fe2O3) has emerged as a promising photo-electrode material due to its significant light absorption, chemical stability in aqueous environments, and ample abundance. However, its performance as a water-oxidizing photoanode has been crucially limited by poor optoelectronic properties that lead to both low light harvesting efficiencies and a large requisite overpotential for photoassisted water oxidation. Recently, the application of nanostructuring techniques and advanced interfacial engineering has afforded landmark improvements in the performance of hematite photoanodes. In this review, new insights into the basic material properties, the attractive aspects, and the challenges in using hematite for photoelectrochemical (PEC) water splitting are first examined. Next, recent progress enhancing the photocurrent by precise morphology control and reducing the overpotential with surface treatments are critically detailed and compared. The latest efforts using advanced characterization techniques, particularly electrochemical impedance spectroscopy, are finally presented. These methods help to define the obstacles that remain to be surmounted in order to fully exploit the potential of this promising material for solar energy conversion.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
Energy from the Sun can easily provide enough power for all of our energy needs if it can be efficiently harvested. While there already exist a number of devices that can capture and convert electromagnetic energy, the most common—a photovoltaic cell—produces electricity, which must be used immediately or stored in a secondary device such as a battery or a flywheel. A more elegant, practical, and potentially more efficient route to storing solar power is to convert the electromagnetic energy directly into chemical energy in the form of molecular bonds, analogous to the photosynthesis process exploited by nature.1 Biologic photosynthesis effectively rearranges electrons in H2O and CO2 to store solar energy in the form of carbohydrates. However, the extremely low overall efficiency that natural photosynthesis exhibits implies the requirement of vast amounts land and farming resources to meet our energy demands.2–3 Because of this, artificial photosynthetic routes including photoelectrochemical (PEC) and photocatalytic (PC) solar energy conversion have been intensely investigated over the last four decades. Given the abundance of H2O on Earth, the water splitting reaction, H2O→1/2O2 + H2 (E0=1.23 V), is the most appealing pathway for artificial photosynthesis. Indeed solar water splitting would form the basis for a sustainable hydrogen-based energy economy.
A distinction can be made between the different approaches to artificial photosynthesis: PC water splitting systems use a dispersed material in pure water and accordingly produce hydrogen and oxygen homogeneously throughout the solution. This approach is under examination with both inorganic colloid materials4–5 and molecular complexes.6 In contrast, PEC systems employ photoactive materials as electrodes. As in conventional water electrolysis, oxidation (O2 evolution) occurs at the anode, reduction (H2 evolution) occurs at the cathode, and an aqueous electrolyte completes the current loop between the electrodes and an external circuit. One or both of the electrodes can be a photoactive semiconductor, in which a space-charge (depletion) layer is formed at the semiconductor/liquid junction (SCLJ). Upon irradiation, photogenerated carriers are separated by the space-charge field and the minority carriers (holes for an n-type photoanode and electrons for a p-type photocathode) travel to the SCLJ to perform one half of the water splitting reaction. The schematic for an n-type photoanode is shown in Figure 1 a. The advantage of the PEC route is that it allows the spatially separate production, and therefore collection, of H2 and O2.
Despite the difference between PC and PEC water splitting, the requirements of the materials employed are essentially the same, and ever since the seminal demonstration of PEC water splitting with TiO2,7–8 scientists and engineers have sustained a vigorous search for a material combining the essential requirements: a small semiconductor bandgap for ample solar light absorption, conduction and valence band energies that straddle the water oxidization and reduction potentials, high conversion efficiency of photogenerated carriers to the water splitting products, durability in aqueous environments, and low cost. However, to date no single semiconducting material has been found to meet all of these requirements.1 Transition metal oxides such as TiO2 possess adequate stability but only absorb a small fraction of solar illumination due to their large band gap (Eg=3.2 eV for anatase TiO2). Many other semiconductors have smaller bandgaps and appropriate energy levels, such as InP (Eg=1.3 eV); however, they exhibit poor stability in aqueous environments.9 While much work continues to focus on identifying an ideal material using combinatorial10 and ab initio (density-functional theory)11–12 approaches, an alternative approach borrows again from biologic photosynthesis, which uses two photosystems in tandem.
By relaxing the constraint that only one material (whose bandgap energy levels straddle the water redox potentials) can be used, combinations of complementary semiconductors can be employed.13 For example, a photoanode and a photocathode or a photoelectrode and a photovoltaic (PV) device can be used in tandem to afford overall solar water splitting with two or more photon absorption events. Systems delivering a record solar-to-hydrogen (STH) efficiency over 12 % have been demonstrated using III–V semiconductor materials with this approach, but their cost and stability remain major disadvantages.14 Alternatively, inexpensive and stable oxide photoanodes, such as WO3 (Eg=2.6 eV),15–16 with conduction band edges too-low in energy to reduce water, can receive an external bias from an economical PV device such as a dye-sensitized solar cell (DSC), which harvests solar photons not absorbed by the semi-transparent photoanode. This photoanode/DSC concept is shown in Figure 1 b. Using a WO3 photoanode it has demonstrated an overall water splitting efficiency of 4.5 %.1 Unfortunately due to its relatively large bandgap and expected kinetic losses, a maximum of only 8 % STH is possible with WO3.17
However, a tandem water splitting device holds great promise if a photoanode can be developed that combines sufficient light absorption, stability, and performance using raw materials of ample availability and low cost. Iron(III) oxide, Fe2O3, is a promising material in light of these requirements. With a potential to convert 16.8 % of the sun’s energy into hydrogen,17 it has been extensively examined for application to solar water splitting. In this review the advantageous properties of Fe2O3 as well as the challenges it presents to photoelectrochemical water splitting are presented. The most recent efforts at controlling water oxidation at the SCLJ, improving photon harvesting by nanostructuring, and increasing the understanding of this promising material for solar energy conversion are critically examined.
2. Hematite as a Promising Material
Iron is the fourth most common element in the earth’s crust (6.3 % by weight) and because iron is readily oxidized in air to the ferrous (+2) and ferric (+3) states, iron oxide is ubiquitous. For example, both the geological formations in the southwest United States (see frontispiece) and a rusty automobile fender take their distinctive red-brown color from the presence of iron oxides. These chromatic characteristics also exemplify iron oxide’s ability to absorb solar irradiation. This coupled with its abundance and non-toxicity make iron oxide a particularly attractive material for use in solar energy conversion. Since the ferrous and ferric forms of iron are separated by a relatively small energy difference, many well-defined crystalline forms of iron oxide and oxyhydroxide exist in nature. The distinct properties of all of the iron oxides are comprehensively presented in a recent book by Cornell and Schwertmann.18 In this section, we present a brief overview and the latest insights into the important and unique properties of the most important iron oxide for solar energy conversion, α-Fe2O3 (hematite), and look at the seminal efforts to use this material in PEC water splitting.
2.1 Overview of properties
2.1.1 Crystalline and magnetic structure
Hematite is the most thermodynamically stable form of iron oxide under ambient conditions and as such, it is also the most common form of crystalline iron oxide. The iron and oxygen atoms in hematite arrange in the corundum structure, which is trigonal-hexagonal scalenohedral (class 2/m) with space group R-3c, lattice parameters a=5.0356 Å, c=13.7489 Å, and six formula units per unit cell.18 It is easy to understand hematite’s structure based on the packing of the anions, O2−, which are arranged in a hexagonal closed-packed lattice along the  direction. The cations (Fe3+) occupy the two-thirds of the octahedral interstices (regularly, with two filled followed by one vacant) in the (001) basal planes, and the tetrahedral sites remain unoccupied. The arrangement of cations can also be thought of producing pairs of FeO6 octahedra that share edges with three neighboring octahedra in the same plane and one face with an octahedron in an adjacent plane in the  direction (Figure 2). The face-sharing is responsible for a trigonal distortion of the octahedra as the proximal iron atoms are repelled to optimize the crystal’s Madelung energy. Consequently, hematite exhibits a C3v symmetry and there are two different FeO bond lengths, as illustrated in Figure 2. However, the electronic structures of the distorted FeO6 octahedral are thought to be similar to undistorted clusters.18
The arrangement of the oxygen anions and the high-spin (d5) cations naturally affect the orientation of the iron atoms’ spin magnetic moment and thus the observed bulk magnetic properties. The diverse magnetic properties of all the iron(III) oxides have been recently reviewed.20 In brief, hematite is antiferromagnetic at temperatures below 260 K and a weak (parasitic) ferromagnet at room temperature. The latter is due to the ferromagnetic coupling of the spins within the (001) basal planes and antiferromagnetic coupling between iron layers along the  direction.21 Here the trigonal distortion of the FeO6 octahedra produces a slightly canted (about 5°) spin arrangement, causing the destabilization of their perfectly antiparallel arrangement (parasitic ferromagnetism). While the magnetic properties of hematite are not particularly pertinent to its photoelectrochemical performance, the iron spin configuration does influence the optoelectronic and carrier transport properties of hematite. These attributes should be properly understood for its application as a semiconductor for solar water splitting. Decades of research have passed scrutinizing the mechanisms of charge transport and light absorption, and while hematite is still not completely understood, much progress has been made.
2.1.2 Optoelectronic characteristics
The absorption of photons by hematite begins in the near-infrared spectral region where weak absorption bands (with absorption coefficients, α, of the order 103 cm−1) are due to d–d transition states between electron energy levels of the Fe3+ ion, which are split by an intrinsic crystal field.22 While photoexcitation of hematite at these wavelengths has been shown in one case to increase its conductivity,23 sustained photocurrent is not observed in a photoelectrochemical system upon irradiation below the bandgap energy, Eg (which, depending on the method of preparation of hematite, is usually reported to be between 1.9 and 2.2 eV corresponding to λ=650 to 560 nm).24 The absorption coefficient of pure hematite increases abruptly at the bandgap energy, and further up the electromagnetic spectrum α continues to increase with additional absorption features having been observed centered on 2.4, 3.2 and 5.8 eV (with α on the order of 105 cm−1) in samples of polycrystalline hematite.25 Hematite’s strong absorption of yellow to ultraviolet photons in the visible region and transmission of orange to infrared photons gives it a characteristic red color. However, its fairly uniform reflectivity as a function of visible light wavelength gives specular hematite a metallic appearance.
Since the electronic nature of the bandgap in hematite is of great interest to understand its performance as a material for solar energy conversion, much work has focused on this aspect. The Tauc analysis of the bandgap absorption onset, which assumes that the energy bands are parabolic with respect to the crystal momentum, most frequently indicates an indirect (phonon-assisted) bandgap transition.26 However, a few recent reports of a direct bandgap in hematite have been attributed to quantum size-effects.27–28 The initial orbital assignments of the bandgap suggested it was due to an indirect transition of Fe3+ d–d origin,23, 29 and that a stronger direct transition involving a charge transfer from an O 2p orbital to Fe 3d did not occur until 3.2 eV.29 This led to the hypothesis that two different types of p-type charge carriers (holes) could be produced in hematite, depending on the excitation mechanism, and were responsible for the observed difference in photoelectrochemical (PEC) performance as a function of wavelength.30–31 However, the most recent ab initio calculations to determine the electronic structure of hematite by the Hartree–Fock approach32 and density functional theory with a local spin-density approximation and coulomb correlation33–34 both predict that the highest occupied energy states are primarily O p in character and the lowest unoccupied states are from an empty Fe d band. This conclusion is also supported by soft-X-ray (O K-edge) absorption and emission spectroscopy, which, when compared to photoemission spectra from configuration-interaction FeO6 cluster calculations, confirm that the valence band is at least strongly hybridized and indicates further that it is mostly of O 2p character.35 These latest results not only suggest that pure stoichiometric hematite is a charge transfer insulator, and not a Mott–Hubbard type insulator—a detail that affects conduction models—but also contradicts the suggestion that two types of holes, originating from different transitions, could cause different PEC behavior.
2.1.3 Carrier transport
For use as a photoelectrode the efficient transport of majority charge carriers in hematite is essential. The earliest work by Morin on pure sintered polycrystalline samples describes very low electrical conductivities (ca. 10−14 Ω−1 cm−1),36 conduction election concentrations of 1018 cm−3 at 1000 Kelvin, and an activation-energy type electron mobility on the order of 10−2 cm2 V−1 s−1.37 Studies on pure single-crystals showed conductivities below 10−6 Ω−1 cm−1 and were deemed unsuitable for further electronic characterization.38 These unusually small values obliged electrical conduction to be explained by Fe3+/Fe2+ valence alternation on spatially localized 3d orbitals. Modeling studies have had success matching empirical data by describing the conduction mechanism with a small polaron model that includes the effect of the larger size of the Fe2+ ion and the associated lattice distortion (polaron).39–40 Conduction of electrons or holes are then best described by the hopping of polarons with an activation energy. This mechanism causes the mobility of carriers to increase with increased temperature, as the transport is phonon-assisted.41 The localized electron-small polaron model has been supported by further experimental results including pulse radiolysis42 and femtosecond laser spectroscopy on nanoparticles.43
Further studies of hematite single crystals identified a highly anisotropic electron transport with conduction along the iron bilayer (001) basal plane up to four orders of magnitude greater than perpendicular directions (parallel to ).44–45 This discrepancy cannot be explained by the proximity of iron cations alone, as the shortest Fe-Fe distances are actually along the  direction.46 However the anisotropic conductivity can be classically explained considering Hund’s rule and the magnetic structure of hematite. The ferromagnetic coupling of the spins in the (001) basal planes and antiferromagnetic coupling along the  direction create an environment where electrons can move (n-type conductivity) within the iron bilayers (an environment of parallel spins) but are forbidden to hop across the oxygen planes to an iron bilayer with the opposite spins. As such, conduction in the  direction could only involve the movement of holes in the form of Fe3+→Fe4+ electron transfer. This process is significantly slower in hematite.41 While this classical explanation is generally accepted, recent ab initio electron structure calculations combined with electron transfer (Marcus) theory have correctly predicted the large transport anisotropy in hematite, but suggested it arises from the slowness of both hole and electron transport across basal oxygen planes.21 This explanation does not consider electron transport to be forbidden in the  direction, but instead identifies the most important factor that influences the carrier mobility to be the electronic coupling—a quantity found to depend on both a superexchange interaction between the bridging oxygen atoms and the d-shell electron spin coupling.
While the specific details of the conduction mechanism and observed anisotropy are clearly important for orientating hematite crystals in a photoelectrode, its intrinsic conduction properties have been shown to be inadequate for PEC applications. Its conductive properties must be significantly enhanced by adding impurities to act as electronic dopants. Indeed, it is possible to increase conductivities and obtain both p-type or n-type α-Fe2O3 by substitutional doping using atoms such as (but not limited to) Mg2+, Cu2+ (p-type) or with Ti4+, Sn4+, Zr4+, Nb5+ (n-type).41 By substituting at sufficient levels, high carrier conductivities can be attained. For example, Zr4+ was doped into single crystals to give donor densities on the order of 1019 cm−3, conductivities around 0.1 Ω−1 cm−1 , and increased electron mobility (perhaps due to a increase in dielectric constant) of 0.1 cm2 V−1 s−1.38 A review by Shinar and Kennedy outlines the substitutional dopants used for n-type iron oxide and their affect on electronic and photoelectrochemical properties.47 In general, optimum impurity concentrations are ca. 1 at % or less, and while substitutional transition metal impurities have been predicted to introduce inter-bandgap energy states,34 the same dopants have not been reported to significantly change Eg or α.38, 48–50
2.1.4 Photogenerated carrier lifetime
In addition to the conduction of majority charge carriers, a water-splitting photoelectrode must efficiently transport photogenerated minority charges to the SCLJ to attain high conversion efficiency. For this, the lifetime of the photogenerated carriers is an important metric. Luminescence studies on nanoparticles show that the excitation of hematite with photons (3.2 eV) results in fluorescence quantum yields of order 10−5.43 This extremely low value of radiative carrier recombination suggests that fast nonradiative processes such as carrier trapping and phonon coupling are limiting the excited-state lifetime.51 Indeed, the ultrafast dynamics of excited states in hematite have been studied in colloidal nanoparticles, single crystals, and even nanostructured thin films using femtosecond laser spectroscopy.43, 51–52 In nanoparticles it was found that the excited state decay profiles were independent of the pump power and the probe wavelength, and were not affected by lattice doping. However, the lifetime of the excited state was determined to be very short—70 % of the transient absorption had disappeared after just 8 picoseconds and could not be detected after 100 ps.43 Studies on epitaxially-grown thin films (100 nm thick and made using an oxygen plasma assisted molecular beam) reduced the influence of bulk and surface defect states compared to the nanoparticle case, but found similar excited state dynamics, which were described by the following scenario: Initially, hot electrons relax to the conduction band edge within 300 fs. Then their recombination with holes and trap states occurs within 3 ps. The resulting trap states can exist for hundreds of picoseconds before recombination to the ground state.51 The dominant carrier trapping mechanism was ascribed to mid-gap Fe3+ d–d states 0.5–0.7 eV below the conduction band edge, giving them an optical transition of about 1.5 eV.53–54 These are likely the same spin-forbidden transitions intrinsic to hematite which were discussed above. In contrast to the short lifetime observed in hematite at neutral conditions, when a large anodic bias was applied to deplete electrons from an undoped hematite electrode, photogenerated holes were found to have a remarkably long lifetime of 3±1 s.52 Further investigation of the lifetime in doped films under polarization would thus be an interesting next step. The overall consequences of the observed trapping states, low carrier mobility, short photogenerated lifetime will next be discussed in the context of hematite as a photoelectrode for water splitting.
2.2 Early efforts with hematite for water splitting
Hardee and Bard55 first turned to Fe2O3 as a material for water photolysis in 1976, seeking a photoanode material that was both stable under anodic polarization and capable of absorbing light with wavelengths longer than 400 nm. They prepared thin films of Fe2O3 on Ti and Pt substrates by chemical vapor deposition (CVD) of Fe(AcAc)3 and observed photocurrent from their electrodes under illumination of 500 nm light at a potential of +0.8 V vs. the saturated calomel electrode (VSCE) in 0.5 M KCl as electrolyte. Soon after Quinn et al.56 showed that the (012) face of flux grown single crystals had collection efficiencies of around 20 % at +0.5 VSCE in 2 M NaOH (+1.57 V vs. the reversible hydrogen electrode, VRHE) with 475 nm light. Further work by Kennedy et al. showed that (Pb2+ and Ca2+ doped) p-type doping produced electrodes with high resistivity, but Si4+, Sn4+, and Ge4+ doped n-type hematite performed better than Ti4+ doped hematite.48 Over the next decade numerous reports were published describing continuing studies with pure and doped Fe2O3 made by various routes. Notable extensive photoelectrochemical studies were performed on Ti doped polycrystalline sintered hematite pellets,26, 31 and Nb doped single crystals57—the latter reporting champion performance with an incident photon conversion efficiency (IPCE) of about 37 % at 370 nm and +1.23 VRHE in 1 M NaOH (27 % at+1.06 VRHE).
A good review of these seminal studies on hematite has been given by Lindgren et al.24 and in summary, as an n-type photoanode, hematite was found to have excellent stability and a Faradaic efficiency towards water oxidation of unity. However the first decade of work identified many challenges of employing this material for photoelectrochemical water splitting: (i) a flat band potential, Vfb, too low in energy for water reduction (see Figure 1 a),31, 56 (ii) a large requisite overpotential for water oxidation,31 (iii) a relatively low absorption coefficient, requiring 400–500 nm thick films for complete light absorption,58–59 (iv) poor majority carrier conductivity,54, 57 and (v) a short diffusion length (LD=2–4 nm) of minority carriers.26 Limitation (i) can nominally be overcome with the external bias provided by a PV or a photocathode in tandem. However, the large overpotential required (limitation ii) requires the use of 2 PVs to provide adequate bias.60 This drawback and the remaining limitations deterred much further interest in hematite, as it appeared that highly efficient conversion of solar energy was not possible. Specifically, drawback (iv) requires high doping levels to increase the ionized donor concentration and thus the conductivity. This in turn reduces the width of the space-charge layer.61 For example, donor dopant concentrations in the champion Nb doped single crystals were found to be 5×1019 cm−3 by Mott–Schottky analysis.57 Assuming classical depletion layer theory,61 this would result in a depletion layer width, W, of 7 nm at a band bending of 0.25 V (i.e., Vb=Vfb+0.25 V) and a dielectric constant, ε, of 100. The low absorption coefficient in hematite (challenge iii) then implies that most of the incident photons will not be absorbed in the space-charge region if the geometry of the photoanode is planar because the absorption depth (defined according to the Beer–Lambert law as α−1=depth at which 63 % of the photons are absorbed) for hematite ranges from 120 nm to 46 nm for photon wavelengths from 550 to 450 nm.25 This would not be problematic if LD was large compared to the absorption depth, but, due to the ultrafast recombination of photogenerated holes, LD is only 2–4 nm.26 It is important to note that this value is significantly smaller compared to other oxides used for PEC water splitting. For example, values of LD up to 104 nm and 150 nm have been reported for TiO262 and WO3,63 respectively.
The effect of limitations (iii)–(v) emerges as low quantum efficiencies especially for the longer wavelengths, even when the electrode is placed under large anodic bias. For example, with the highly-efficient Nb doped single crystal electrodes, the peak quantum efficiency was close to 30 % at λ=370 nm, but dropped to below 5 % at 450 nm, and even though all of the incident irradiation was absorbed at 550 nm in the 0.1 mm thick films used, the quantum efficiency at that wavelength was less than 1 %.57 A maximum photocurrent density of 12.6 mA cm−2 is possible with hematite based on its absorption and the AM 1.5G solar spectrum (100 mW cm−2), but the photocurrent delivered by this Nb doped photoanode would be only ca. 0.5 mA cm−2 based on the quantum efficiency reported. This is because the distribution of photons in the solar spectrum puts half of hematite’s maximum possible solar photocurrent density in the wavelength range of 500–600 nm. Indeed, the disaccord between the absorption depth of photons with energy close to Eg (ca. 100 nm) and the carrier harvesting width (W+LD=10 nm) suggests that the standard planar single crystal or sintered disk electrode geometries are not suitable for hematite. However, recent efforts with hematite have shown that optimizing the electrode morphology can significantly increase the water splitting photocurrent density. These recent and exciting advances in hematite structuring are described in detail in section 4. The renewed efforts have also instigated further research into reducing the overpotential required by employing various surface treatments, which are examined in the next section. By combining these two strategies it is possible to approach the performance of an ideal hematite photoanode (Figure 3).
3. Decreasing the Overpotential with Surface Modification
Based on the Vfb usually reported for hematite, an external bias of only 0.3–0.4 VRHE should be necessary to initiate the water splitting reaction.24 Once the applied bias is greater than Vfb, the band bending drives photogenerated holes to the SCLJ. However, the onset of water oxidation photocurrent is usually not observed until 0.8–1.0 VRHE even at a high pH of 13.6 (1 M NaOH) and for single crystal electrodes. The remaining overpotential of ca. 0.5–0.6 V is a major drawback for the implementation of hematite-based tandem cells60 and has been attributed to two distinct surface properties. Firstly there is evidence that mid-bandgap energy states resulting from both oxygen vacancies25, 31, 64 and crystalline disorder65 can trap holes at the surface. This can even result in Fermi level pining in some electrodes.66 Secondly, the oxygen evolution reaction (OER) kinetics are sluggish, not just due to the complicated four-electron mechanism that must occur, but also compared to other oxides semiconductors.67 This may be due to the increased Fe3+ character of the valence band compared to other oxides.31
To overcome the limitation of poor OER kinetics, various catalysts have been attached to the surface of hematite photoanodes. For example, water oxidation by cobalt has been extensively studied and is known to be particularly rapid.68 The treatment of Fe2O3 photoanodes (prepared by a CVD method) with a monolayer of Co2+ resulted in a ca. 0.1 V reduction of the photocurrent onset potential.69 Since this treatment also increased the plateau photocurrent it was good evidence that the reaction rate was increased, and the Co2+ did not just fill surface traps. The application of a recently-reported amorphous cobalt-phosphate (CoPi) based water oxidation catalyst70 on Fe2O3 gave a composite photoanode with a similar photocurrent onset potential to that of the Co2+ treatment.71 However, the increased efficacy of the Co-Pi catalyst at more neutral pH afforded an noticeable enhancement of the overpotential reduction at pH 8.72 A drawback of this approach is the unproductive light absorption by the CoPi catalyst, which allows only a thin layer to be deposited.
The material often reported as the most effective catalyst for the OER is IrO2.73–74 For example, nanoparticles (ca. 2 nm diameter) deposited onto a glassy carbon electrode achieve quantitative Faradaic efficiency of water oxidation and an unequalled current density of 0.5 mA cm−2 at an overpotential of 0.25 V.75–76 The application of these nanoparticles to the surface of hematite by electrophoretic deposition resulted in an impressive shift of the photocurrent onset by about 200 mV giving J=0.3 mA cm−2 at 0.9 V and 1.16 mA cm−2 at 1.0 VRHE.77 This specific result is detailed further in the following section, but the fact that even the best OER catalyst leaves a significant overpotential remaining (0.3–0.4 V) suggests that the other surface limitation—that is, surface trapping states—still plays an important role for hematite photoanodes.
The presence of trapping states on the surface has been verified by using sacrificial electron donors in the electrolyte,53 and been shown to be controllable by altering the oxidizing environment of the hematite preparation conditions.64 However, since this is not always possible when using a preparation method designed for nanostructuring, a method to address surface trapping states on already-prepared hematite is necessary. It has recently been shown that surface trapping states can be passivated with extremely thin Al2O3 overlayers made by atomic layer deposition (ALD).78 Even just one ALD cycle of Al2O3 on a Fe2O3 photoanode shifted the photocurrent onset potential by about 100 mV. The subsequent application the Co2+ catalyst further reduced the overpotential and resulted in a record photocurrent at 0.9 VRHE of over 0.4 mA cm−2. This result shows that separately addressing both of the surface limitations can afford a significant reduction of the overpotential, however, to enable the realistic use of tandem water splitting device with a hematite photoanode, a further reduction of the overpotential is required. Essentially, the maximum (plateau) photocurrent density (up to 12.6 mA cm−2) must be obtained at only ca. 0.6 VRHE. With typical onset behavior, this implies an overpotential of only 0.1–0.2 V.
A different approach, recently reported by Hu et al. holds promise to reach the high photocurrents at 0.6 VRHE. Here the authors observed a shift of Vfb in the cathodic direction by 150 mV (as measured by the Mott–Schottky technique) using a surface modification with fluoride.79 Attaining unassisted water splitting with hematite could even be possible with this technique if Vfb could be moved cathodic of 0 VRHE. In a practical device a combination of all three surface treatment strategies would be employed to afford the highest photocurrent density at the lowest applied potential.
Overall surface treatments have shown great promise in reducing the applied potential necessary to onset photocurrent with hematite photoanodes. However, increasing the solar photocurrent from the ca. 0.5 mA cm−2 in the single crystal case towards the maximum possible 12.6 mA cm−2 is more critical for establishing the viability of hematite as a material for solar energy conversion. Methods to address this challenge are discussed in the next section.
4. Increasing Photocurrent by Morphology Control
The small carrier harvesting depth (W+LD) discussed in section 2.2 suggests that the ideal morphology of a hematite photoanode would be one where all of the material is within 10–20 nm of the SCLJ. Of course, one could simply deposit a 10 nm layer of hematite on a transparent substrate using a traditional deposition method. The relatively poor absorptivity of such a film could then be overcome by stacking multiple layers in tandem, assuming the substrate used was a transparent conductor such as fluorine doped tin oxide (FTO). This approach was suggested by Itoh and Bockris in 1984.59 While this could fundamentally resolve the issue, it is cumbersome and expensive to implement, and in practice the thin iron oxide films were found to exhibit poor performance due to the increased recombination of the photogenerated holes.58 Fortunately the recent development of tools to control the dimensions and morphology of materials at the nanometer length scale has offered more practical solutions for hematite. The application of nanostructuring techniques to PEC solar hydrogen production has indeed led to advances with many materials. A general overview of this topic is presented by recent review articles.80–81 This section will analyze how different deposition techniques have allowed the morphology control of hematite and affected its performance as a photoanode.
4.1 Porous thin films from solution-based colloidal methods
A straightforward way to create nanostructured photoelectrodes is to coat a dispersion of nanoparticles and a porogen onto a conductive substrate. Drying and then heating these films in air burns away the porogen and sinters the remaining oxide, leaving a porous structure of interconnected particles. This concept has been employed effectively with TiO2 for the DSC, and was first attempted with hematite in 1994.82 In this seminal work, Fe2O3 sols created by the hydrolysis of FeCl3 were combined with Triton-X100 before they were doctor-bladed onto FTO and sintered at 560 °C. Micrometer-thick, porous films of necked hematite were observed to have good adhesion to the substrate and a primary particle size in the 25–75 nm range. However the incident photon-to-current efficiency (IPCE) of these photoanodes towards water splitting was quite low—on the order of 1 % at 0.4 VSCE in 0.1 M NaOH (1.4 VRHE) with 400 nm incident irradiation. The IPCE was 100 times lower when the anode was illuminated from the hematite/electrolyte interface as compared to the substrate/hematite interface and the quantum efficiency did not improve when LiI was added as a hole scavenger. This led the authors to conclude that charge carrier recombination was the critical factor controlling the photocurrent. The higher quantum efficiency of similar particles when dispersed in electrolyte83 pointed to grain boundaries in the porous films to be the cause of the excessive recombination and poor performance. This limitation was later addressed by altering the film thickness to optimize the light absorption/carrier transport issue.84 However, no significant improvement was obtained.
Recently the hypothesis that the limitation of these porous films was the transport and collection of majority carriers was confirmed, and relatively high water splitting photocurrents were obtained after successfully incorporating dopants at sufficient concentration.19 In this work hematite nanoparticles 5–10 nm in diameter, which had been prepared by thermal decomposition of Fe(CO)5 in the gas phase, were dispersed in 2-propanol with acetylacetonate and hydroxypropyl cellulose and coated on FTO or Pt substrates. While no water splitting photocurrent was observed with these electrodes upon sintering at 400 or 700 °C, 20 min at 800 °C was sufficient to diffuse and activate dopants (Sn4+ or Pt4+ from the substrate) and afford relatively high photocurrents. Under standard illumination conditions (AM 1.5G 100 mW cm−2) 0.56 mA cm−2 at 1.23 VRHE and over 1.0 mA cm−2 before the onset of dark current (1.55 VRHE) were obtained.19 The same high sintering temperature of 800 °C was found to be necessary to afford photoactivity even when (Ti4+) dopant atoms were included in the colloidal dispersion.85 This observation suggests that a minimum activation energy is required to overcome a barrier for the diffusion and activation of doping cation impurities in the hematite lattice, and is supported by the report of a drastic change in the diffusion coefficient of 55Fe3+ isotopes in hematite86 and the further increase in the C3v crystal distortion (see section 2.1),87 both of which have been reported to occur at approximately the same temperature. The latter of these effects was shown to be directly correlated to an increased absorption coefficient in films sintered at these high temperatures suggesting that this lattice relaxation is an important aspect influencing the optoelectronic properties of hematite.19
Another effect of the high-temperature treatment is the increase of the feature size from the original 5–10 nm to more than 50 nm via sintering (see Figure 4 c). This large particle size prevents the photocurrent densities from surpassing the single crystal or sintered disk approaches. To overcome this limitation a method to independently control the feature size and functionality with nanostructured oxide electrodes using an encapsulation approach was recently developed.88 First, the as-deposited films were subjected to a “morphological” annealing to remove the porogen, give good interparticle connections, and set the feature size. Next the films were encapsulated with mesoporous SiO2 using a solution-based precipitation approach. The SiO2 acted as a confinement scaffold during the high temperature, “functional” annealing (800 °C) which followed to diffuse and activate the dopant (Ti4+ included in the deposition solution). The subsequent removal of the SiO2 by dissolution revealed the functional electrode with a feature size controlled by the first annealing step. The outcome of applying this approach is shown in Figure 4.
The effect the encapsulation technique had on the nanostructure was immediately evident by eye, when comparing the resulting electrode to a control heated to 800 °C without a confinement scaffold. The larger feature size in the control electrode resulting from sintering without the SiO2 scatters visible light and the electrode appears translucent with a high opacity. In stark contrast, feature sizes in the confined electrode (500 °C morphological annealing) remain too-small to scatter ambient light and the electrode remains transparent (Figure 4 a). Scanning electron micoscopy images clearly show the difference of the final feature size, which remains at ca. 25 nm with the encapsulation approach and grows to 100–200 nm in the control case (Figure 4 b and c). As the encapsulated electrode’s smaller feature size allows the photogeneration of holes closer to the SCLJ, a marked increase in the water oxidizing photocurrent was also observed. The photocurrent of the encapsulated electrode reached 1.86 mA cm−2 at 1.43 VRHE with a maximum photocurrent of 2.34 mA cm−2 before the onset of the dark current while the control electrode gave only a maximum of 1.5 mA cm−2 (Figure 4 d). This remarkable increase of photocurrent was shown to be due an increase of the photon conversion efficiency (IPCE) especially in the longer wavelength range.
Despite the demonstrated efficacy of the encapsulation technique certain drawbacks prevent it from reaching further improvement of the photocurrent. First, a considerable anodic shift of the photocurrent onset potential was observed (Figure 4 d). This might result from the diffusion a large concentration of Si from the confinement scaffold into the surface of hematite creating defect states, and could potentially be addressed with a subsequent etching step. Second, the conditions found to give a the maximum photocurrent were with a 500 °C morphological annealing giving a feature size of ca. 25 nm, however attempts to use a lower morphological annealing temperature resulted in films with less photocurrent despite a smaller feature size (which should allow more photogenerated holes to reach the SCLJ). Altering the film thickness suggested that a tortuous path for electron conduction with many grain boundaries limited electron transport in films prepared with a morphological annealing temperature less than 500 °C. Ultimately, while the solution-based nature of this approach generates much interest in this technique higher photocurrents would be possible if the electron transport limitation could be avoided.
4.2 Fe2O3 nanowire arrays
An obvious solution to the problem of majority carrier transport in hematite films created from the colloidal approach is to use nanometer-sized rod or wire arrays. An array of single-crystal nanorods with diameters in the 10 nm range, attached and oriented orthogonally to a conducting substrate would eliminate grain boundaries, and provide a direct path for electron collection while still allowing photogenerated holes to efficiently reach the SCLJ. A simple method to create hematite arrays on a variety of substrates from the controlled precipitation of Fe3+ in aqueous solution was first reported by Hagfeldt and co-workers,89 and investigated for water photoelectrolysis soon after.27 Bunches of individual 5 nm nanorods with an average diameter of 50 nm and a length of 0.1 to 1.5 μm (see Figure 5, right) were investigated in perpendicular and parallel orientations to the substrate. While the authors were able to show a small improvement when controlling the orientation (5 % IPCE at 360 nm for the perpendicular nanorods versus 3 % for the parallel) the overall photocurrents remained low under white light illumination, even with the hole-scavenging iodide present. The large difference between the electrode performance when illuminating from different sides of the semitransparent photoanode observed in this work,67 and a recent report90 examining the surface photovoltage on electrodes prepared in the same way, suggests that bulk or surface defects are the major factors limiting the performance hematite prepared by this route. Indeed, aqueous methods of preparing hematite typically pass through a phase containing iron hydroxide (e.g., akaganeite, lepidocrocite or goethite) and despite the fact that primarily hematite is detected after annealing at 500 °C, it has been shown that at temperatures up to 800 °C, a nonstoichiometric composition remains in hematite prepared by aqueous methods.91 As such, residual hydroxyl groups or vacancies are likely to blame.
Another facile method to produce hematite nanowires is by the simple thermal oxidation of iron foils and has been reported by many groups.92–96 Due to the increased volume of the oxide over the metal, when foils of iron are thermally oxidized under the right conditions, arrays of Fe2O3 nanowires spontaneously grow from the surface of the foil. With diameters of 20–40 nm and lengths of up to 5 μm (see Figure 5, left) these nanowire arrays would have large surface area, sufficient light absorption and a direct path for the conduction of electrons to the substrate (since basal planes are oriented perpendicular to the substrate95), making them a very attractive morphology for hematite. This simple thermal oxidation method has been demonstrated on multiple (even transparent) substrates,96 and methods for doping the wires have been employed.92 However, no convincing reports of activity towards water oxidation have appeared. This may be due to the defects present in nanowires prepared this way or the presence of suboxides near the interface of the substrate which greatly enhance recombination.97
Recently, a method to prepare hematite nanorod arrays using an anodic aluminum oxide (AAO) template was reported.98 Gold was sputtered onto one side of a 60 μm thick AAO membrane with 200 nm diameter pores. Electrodeposition of iron oxide from aqueous solution, annealing at 500 °C, and the subsequent removal of the AAO by dissolution gave the final structure. The length of the rods could be varied by changing the deposition duration. Despite the absence of any intentional doping, a 10 μm thick rod array showed water oxidation photocurrent densities of about 1 mA cm−2 at a potential of 1.5 VRHE under AM 1.5 illumination. However, no defined photocurrent plateau was observed, suggesting a majority carrier transport limitation in spite of the nanostructure. The incorporation of dopants and annealing at 800 °C may afford higher photocurrents with this approach. However, based on the known minority carrier transport properties of hematite the pore size should also be reduced to less than 25 nm to afford a significant increase in photocurrent.
4.3 Electrochemical Fe2O3 nanostructuring
The recent development of nanostructuring techniques using potentiostatic anodization has provided another possible route to create structured hematite photoelectrodes. Prakasam et al. first showed that iron foils could be nanostructured using anodization in a glycerol-based electrolyte containing 1 % NH4F+1 % HF+0.2 % HNO3.99 Ordered nanopores were observed with pore size ranging from 20 to 250 nm and depths up to 600 nm depending on the anodization voltage and time. Under simulated solar illumination these photoanodes produced a photocurrent of 0.05 mA cm−2 at 0.4 VSCE in 1 M NaOH (1.45 VRHE). Work by the same group on anodized TiFeO electrodes is notable here due to the high photocurrent densities reported (1.1 mA cm−2 at 1.4 VRHE in 1 M NaOH) despite the presence of both hematite and rutile in the photoanode prepared at the optimized conditions.100 In contrast, very well defined nanotube arrays of pure iron oxide created from iron foils have been subsequently reported by a different research group.101 In this work, a single anodization step with 0.1 M NH4F+3 vol % water in ethylene glycol created nanotubes with walls less than 50 nm and lengths of about 1.5 μm. After an optimized annealing treatment these electrodes were found to be a mixture of both hematite and maghemite by XRD and had only small photocurrents (160 μA cm−2, 1.23 VRHE under AM 1.5 illumination compared to 120 μA cm−2 dark current). However, a double step anodization procedure with the addition of sodium tripolyphosphate in the electrolyte, and the subsequent annealing created a dendrite-like morphology situated over the tube array. Electrodes prepared this way exhibited a photocurrent of about 1 mA cm−2 under AM 1.5 illumination and at 1.23 VRHE. A large carrier concentration (1021 cm−3) was found via Mott–Schottky indicating a large amount of dopant or impurity present. The authors attribute the high photocurrent to the double layered structure which includes both a large surface area for water oxidation and a vertically orientated nanotube array for electron transport.
In subsequent work by the same group, a sono-electrochemical anodization method was employed to obtain either nanoporous or nanotubular Fe2O3, depending on the anodization conditions.102 The optimization of conditions gave nanotubes with walls an impressive 5–6 nm thick and micrometers in length (see Figure 6, left). Annealing these films resulted in pure hematite which gave similar photocurrents (ca. 1 mA cm−2 at AM 1.5 and 1.23 VRHE) and similar carrier densities to the previous work. While the initial tube size presented in this work is remarkable and an important step for obtaining hematite with the ideal morphology, the tubes were initially amorphous Fe2O3. The annealing step necessary for crystallinity and photoactivity almost certainly demolishes the meticulous nanostructuring through the sintering of the material.
Nanostructured hematite photoanodes also have been prepared by the electrodeposition of precursors from solution. Recently McFarland and co-workers reported a method to deposit iron hydroxides from FeCl3 solutions under cathodic polarization. The subsequent annealing at high temperature (700 °C) then resulted in porous hematite films.28, 104 This method readily allows for the incorporation of dopants which were found to have an effect on the morphology of the sintered electrodes. Feature sizes down to 40 nm were observed and an IPCE of 8 % with 400 nm illumination at 1.2 VRHE in 1 M NaOH with 15 % Mo doping were reported. In a subsequent report, an isovalent substitutional dopant Al3+ was added to the hematite to modulate the lattice strain—a factor predicted to benefit polaron migration and offer a novel way to increase conductivity.105 However, the generally observed large particle size obtained with this technique remains a limitation to higher photocurrent densities.
Another example of electrodeposition has been recently reported by Spray and Choi103 using an anodic electrodeposition. They were able to demonstrate impressive morphology control ranging from wires arrays (see Figure 6, right) to porous films by varying the solution pH. The films were photoactive in an electrolyte containing iodide, but water oxidation photocurrents were not reported suggesting surface recombination issues. In general, while impressive morphologies can be obtained with electrochemical deposition techniques, water splitting photocurrents have been limited by the quality of the material produced.
4.4 Spray pyrolysis techniques
Despite the deliberate attempts to nanostructure hematite using the methods described above, a simple spray pyrolysis technique consistently produces electrodes with superior performance. First reported in 1984,106 an aqueous or ethanolic solution of Fe3+ is simply sprayed onto a hot (400–500 °C) substrate in air and photocurrent densities in the 0.5–1.0 mA cm−2 range are typically obtained under simulated solar illumination (1 M NaOH, 1.6 VRHE)66, 107–109 with the exception of one report110 of higher photocurrent attributable to excess UV in the light source.111 Surprisingly, reasonable photoactivity can be obtained with this method, even without intentionally adding dopants to increase the conductivity of the Fe2O3. Commonly used precursors such as FeCl3 or Fe(AcAc)3 likely leave chlorine or carbon residues behind that act as electric dopants. Films produced are generally compact (not porous) and the surface area limits the photocurrent produced under standard illumination conditions. However, the ease and reproducibility of this technique has led to valuable fundamental studies on dopants107, 112 and substrate effects.108 Rigorous optimization has also produced films with a very high quantum efficiency (IPCE=25 % at 400 nm and 1.42 VRHE).112 In a recent report from Duret and Grätzel, the spray pyrolysis technique was modified to create nanostructured films by using an ultrasonic spray nozzle to produce very fine droplets of the precursor solution which were entrained in a rapid flow of air and transported horizontally over a heated substrate.66 Under optimized conditions, a slow growth rate (ca. 100 nm h−1) produced hematite with a leaflet type structure consisting of 100 nm-sized platelets of 5–10 nm thickness bundled into 50 nm sheets oriented perpendicular to the FTO support. These ultrasonic spray pyrolysis (USP) films performed remarkably as water-splitting photoanodes, producing 1.3 mA cm−2 under AM 1.5G (100 mW cm−2) simulated solar irradiation in 1 M NaOH at 1.23 VRHE. The IPCE was found to be 16 % with 375 nm illumination at the same applied potential, and over 30 % at 1.6 VRHE. Subsequent work on these films revealed the importance of Si doping, which was found to be leeching from the silicon tubing used in the previous work.113–114 In addition, the platelets were found to be preferentially orientated with the (001) basal planes normal to the substrate and the flat surface of the platelets presenting the basal plane. While this morphology should facilitate electron transport to the FTO substrate it also requires the holes to be conducted across oxygen layers. Both this detail and the poor kinetics of water oxidation were thought to be restricting the photocurrent produced.
4.5 Atmospheric pressure chemical vapor deposition
Continuing efforts to improve the performance beyond the USP films has led to the development of a simple, yet effective process to make nanostructured films of hematite from the thermal decomposition and oxidation of iron pentacarbonyl in an atmospheric pressure chemical vapor deposition (APCVD) reactor.113 Briefly, Fe(CO)5 vapors were transported, along with tetraethyl orthosilicate (TEOS) as a source of silicon dopant, by a carrier gas (dry air), and directed vertically onto a heated substrate. The poor thermal stability of the Fe(CO)5 (the half-life for decomposition at 300 °C is 5.3 ms115) caused the homogenous nucleation of nanoparticles in the gas phase.116 These particles were then subjected to a thermophoretic force which limited their approach to the hot substrate and resulted in fractal-like cauliflower structures characteristic of a diffusion-limited aggregation mechanism (in contrast to surface reaction-limited).65 In 2006 Kay et al. reported APCVD films optimized with a substrate temperature of 420 °C and using a carrier gas flow rate of 2 L min−1. Cauliflower-type structures with a minimum feature size of 5–10 nm grew on FTO substrates at rate of about 100 nm min−1. The best film produced a photocurrent density of 1.8 mA cm−2 at 1.23 VRHE under standard illumination in 1 M NaOH and was further increased to 2.2 mA cm−2 by treating the surface with Co2+ ions. Similar to the films prepared by the USP method, the silicon dopant was found to affect the nanostructure, producing a smaller feature size in both cases and an alignment of the basal crystal plane normal to the substrate.69, 113
In continuing work with this deposition technique ferrocene and tetramethyl orthosilicate (TMOS) were found to be viable sources of iron and silicon, respectively.117–118 The use of ferrocene is notable for replacing the toxic iron pentacarbonyl, however the photocurrent density obtained remained less than 1 mA cm−2. With the standard Fe(CO)5/TEOS system Cesar et al. showed a large temperature effect on the nanostructure which corresponded to the performance of the electrodes.65 Higher temperatures were found to decrease the deposition rate, lead to a larger particle size, and reduced the photocurrent density. A Mott–Schottky analysis gave the calculated donor density, Nd, of the Si doped Fe2O3 films as 1.7×1020 cm−3—close to the predicted 6×1020 cm−3 if the known incorporation of silicon (1.6 at % Si)69 was substitutionally incorporated into the lattice as Si4+. Based on classic depletion layer theory,61 this led the authors to suggested that a space-charge layer of ca. 5 nm could still be present and aiding hole transport the SCLJ even in the small features on the top of the APCVD film. In addition, changing the deposition time defined an optimum thickness (500 nm) for photocurrent production from front side illumination (i.e., upon the hematite/electrolyte interface). A slow decrease in photocurrent was observed for thicker films resulting in less than 1 mA cm−2 for 1 μm thick films. This implied an electron transport limitation, which was subsequently addressed by tuning the particle to precursor ratio.
An important parameter governing the structure and quality of nanostructured prepared by particle-assisted CVD systems, like the APCVD of Fe(CO)5, is the particle to precursor ratio.119 By changing the carrier gas flow rate, keeping the same concentration of iron precursor, and re-optimizing the other experimental parameters, the amount of time the precursors resided in the critical region of particle nucleation and growth above the deposition substrate was effectively tuned. It was found that using a carrier gas flow rate three times higher than the previous result (6 L min−1) enhanced the preferential orientation of the basal planes normal to the substrate (enhancing majority carrier transport), but did not change the primary particle size.120 This resulted in an optimized thickness of 800 nm, and photocurrents above 3 mA cm−2 (see Figure 7). To improve upon the overpotential reduction found with the Co2+ treatment, the IrO2 nanoparticle OER catalyst was then deposited onto the Fe2O3 cauliflowers by an eletrophoretic process. This produced a 200 mV shift in the photocurrent onset potential and photocurrent of 3.3 mA cm−2 at 1.23 VRHE under standard testing conditions.77
The photocurrent produced was further verified by measuring the photoanode IPCE at 1.23 VRHE. Over 50 % of the photons with λ=300 nm, 39 % at 400 nm, and a remarkable 20 % at 500 nm were effectively harvested using this nanostructure. Moreover, the integration of the product of the IPCE and the standard AM 1.5G solar photon flux over the entire wavelength range gave a photocurrent of 3.01 mA cm−2, verifying the light source used for the J–V measurements. The photocurrent was found to be stable for many scans, however under extended measurement the IrO2 nanoparticles detached from the surface and the photocurrent onset potential shifted back to the original value. Despite this limitation, which could possibly overcome using an appropriate linking strategy, this result clearly demonstrated that water oxidizing photocurrents over 3 mA cm−2 at relatively low bias potentials are possible with hematite using the combined strategy of morphology control and surface modification. This demonstration also thrusts the performance of hematite past the other oxides studied for PEC water oxidation to be the most efficient at converting solar light into energy stored in a tandem cell. If the bias required was applied by two PV cells in tandem each delivering 0.6 V and 6.6 mA cm−2, a total solar to hydrogen efficiency of 4.8 % would be obtained.
While this is a considerable step towards high performance hematite, the photocurrent density remains far from the 12.6 mA cm−2 possible. This is because, despite absorbing almost all of the light (with hν>Eg) the quantum conversion efficiencies are still relatively low for wavelengths near the band edge (500<λ<600) where, half of hematite’s possible solar photocurrent is available. This can be primarily attributed to the cauliflower-type morphology, which has an appropriate feature size of 10 nm at the SCLJ, but the individual cauliflower structures consist of thick stems which increase the number of photons absorbed far from the SCLJ.
4.6 Extremely thin absorber approach
The techniques discussed until now attempt to increase the photocurrent by nanostructuring the hematite itself. However, another strategy that can place all of the hematite in proximity to the SCLJ and include sufficient material for light absorption is to coat an extremely thin layer on a high-surface area, nanostructured scaffold in analogy to the DSC121 and extremely thin absorber (ETA) PVs.122
This concept was recently demonstrated using a nanostructured WO3 as the host scaffold.123 In this work the scaffold was coated with a thin layer (ca. 60 nm) of Fe2O3 by the APCVD method and a 20 % increase in the photocurrent was observed over control samples prepared on flat FTO substrates. This enhancement was attributed to an increase in the absorbed photon to current efficiency (APCE), especially in the wavelength range from 500 to 600 nm. However, the thickness-optimized 60 nm layer of hematite was too thick based on the known hole transport limitations. Attempts to use thinner films of iron oxide resulted in poor performance, similar to other reports of thin films Fe2O3,58 and pointed to a high recombination of photogenerated holes at the hematite/scaffold interface.
This limitation has recently been addressed by including a SnO2108 or a SiOx124 buffer layer the problematic interface. For the latter report, a simple spay deposition was used to make highly uniform compact films of Fe2O3 on FTO substrates to first examine the effect of the thickness on the interfacial recombination. A minimum thickness of 22 nm was required to observe appreciable photoanode behavior. However when a SiOx buffer layer was deposited on the FTO, reasonable photocurrents were obtained with only a 12.5 nm thickness of Fe2O3. In addition, comparing films of similar thickness with and without the buffer layer showed less recombination upon examining photocurrent transient behavior. Moreover, it was found that the buffer layer significantly altered the film nucleation and growth characteristics, and consequently the crystallinity. Unfortunately, the best ultrathin hematite films with employed buffer layer showed slightly lower values for the APCE compared to the state-of-the-art APCVD films. This is possibly due to the high reflectivity of the compact and flat films. In general, the ETA/buffer layer/scaffold approach provides a clear path to overcoming all of hematite’s limitations by decoupling the light absorption and the charge transport. Research using hematite with this approach is relatively new and further investigations into buffer layers and appropriately structured scaffolds should ultimately produce photocurrents higher than Fe2O3 prepared by the APCVD technique.
4.7 Nanostructuring method comparison
Several different approaches have been taken to nanostructure Fe2O3 for use in photoelectrochemical water splitting, and at least as many methods of characterizing the electrodes have also appeared. As such it can be difficult to compare the results in the literature, especially when variations in light sources give large overestimations of photocurrents due to superfluous UV light. A recent review addresses these issues and outlines standardized techniques for assessing photoelectrodes.125 While not all of the reports discussed in this section have followed these suggestions it can still be helpful to compare the major accomplishments from the different structuring techniques and the resulting performance. This is done in Table 1.
Table 1. Comparison of hematite photoanode structuring methods
1.3 mA cm−2 at 1.23 VRHE and AM 1.5 (100 mW cm−2) and IPCE 26 % at 400 nm and 1.43 VRHE
Overall, it is clear there is a strong dependence of the performance of synthetic hematite on the deposition technique. Interestingly, while methods such as spray pyrolysis and CVD consistently produce electrodes photoactive for water oxidation at low temperatures (400–500 °C), solution-based methods such as the colloidal and electrodeposition approaches require much higher temperatures (700–800 °C) to produce photoactive hematite. This is certainly related to the quality of the prepared material in terms of crystallinity, impurity concentrations, and dopant incorporation, and a further understanding of this aspect is important to pursue.
Also worth noting is that the IPCE values at ca. 400 nm rarely exceed that of the single crystal or sintered polycrystalline electrodes prepared during the first decade of work. The champion electrodes made by APCVD show only a modest increase in the conversion efficiency with 400 nm photons. The explanation for the significant increase in photocurrent observed with the APCVD and the colloidal solution-deposited electrodes has, of course, been the nanostructuring, that has allowed increases conversion of the longer wavelengths of light. Further work on the nanostructuring will, no doubt, see solar photocurrents rise, but the increase in IPCE beyond 40 % at 400 nm and 1.23 VRHE will require further research directed at increasing the understanding of hematite photoanodes.
5. Advanced Understanding by using Electrochemical Impedance Spectroscopy
Many electrochemical techniques have been used to characterize hematite photoanodes with the goal of understanding its performance and limitations. The most often employed— electrochemical impedance spectroscopy (EIS)—has proven to be valuable for ideal systems like single crystals. However, the many different preparation techniques and the advent of nanostructured electrodes have complicated efforts to understand and compare results obtained from EIS. Fortunately, recent efforts have provided significant advances in applying this classic technique to increase the understanding of hematite photoelectrodes. In this section the latest developments characterizing hematite photoanodes with EIS are discussed.
As a modulation technique, EIS consists of applying a time-dependant signal to the electrode, usually a sinusoidal perturbation of bias potential, and measuring the complex response of the electrode. By varying the excitation frequency, processes with different time constants can be resolved. The further analysis of the frequency-dependant response using an electronic model can separate the different aspects of charge transport, charge trapping, and charge transfer at the interface with electrolyte. During an EIS experiment the electrode dark- or photocurrent density is typically recorded while applying an alternating current (AC) signal with frequency, f (0.1 Hz<f<106 Hz), superimposed on the bias potential. A major advantage of EIS over other spectroscopic techniques is that it allows sample characterization under working conditions (in electrolyte under bias potential). The application of the well-known Mott–Schottky (MS) relation allows the extraction of Vfb and Nd by plotting the inverse square of the capacitance versus the applied bias. Typically, the capacitance is determined by fitting the frequency responses with a simple resistor–capacitor (RC) circuit model (see Figure 8 a). A single frequency,30, 50, 56 multiple frequencies,105 extrapolating to an infinite frequency,54, 57, 126 or an entire range of high frequencies65, 127–128 have been used to fit the circuit model. Consistent values of Vfb from 0.4–0.6 VRHE and Nd on the order of 1017 cm−3 (for undoped) to 1021 cm−3 (for heavily doped samples) have been reported for both planar and structured electrodes.24, 127–128
However since the classic MS relation relies on a simple parallel-plate capacitor model the application of EIS data from real systems has often been problematic. For example, frequency dispersion in MS plots is often observed low frequencies. This has been attributed to non-idealities such as roughness of the samples, relaxation of the dielectric constant in the space charge region, and surface trapping states,129 and explains the preferential use of high frequency to determine the sample capacitance. In addition, MS plots of hematite photoanodes have exhibited large deviations from linearity when Vb≫Vfb. This phenomenon has been attributed to deep energetic levels partially ionized in the space charge region (Goodman model).130 To account for this, the observed EIS response was fitted with a more sophisticated circuit model taking in to account these intrabandgap states (Figure 8 b). Using this model, these states were found to lie 0.5–0.6 eV below the conduction band30, 54, 131 and associated with a trapping time constant about 0.4 ms. Some studies also established the presence of shallow donors, or trap states located at 0.1–0.2 eV under the conduction band using this model at low bias potential. Nevertheless McAlpine and Fredlein132 observed severe limitations to this model and more recent studies using nanostructured hematite have not reported similar behavior, possibly due to small features exhibiting low bulk-to-surface ratio. For nanostructures, a superlinear trend of the MS plot has also been observed. This has recently been explained in a terms of decrease in the surface area contributing to the capacitance.65 As the hematite photoanode is biased in the anodic direction, the large overpotential causes the full depletion of the smallest surface features or curvature of the nanostructured surface resulting in a decrease of the effective electrode area and thus a decrease in capacitance.
An additional issue using EIS with nanostructures is illustrated in Figure 8 d. Here a typical impedance response of a nanostructured hematite photoanode (prepared by APCVD) is shown in as a Nyquist plot (real vs. imaginary impedance) at several bias potentials. The classically used high frequency response can be observed at low impedance (close to the origin of the plot), and is characterized by a semicircle easily-fit by the simple R-(RC) circuit as shown in Figure 8 a. At low frequency signals (large impedance, far from the origin), a second semicircle appears, and at certain potentials (e.g., 0.8 VRHE in Figure 8 d) this second feature is not easily distinguishable from the high frequency response. Gomes and Van Maekelbergh129 interpreted the two semicircles in the Nyquist plot in terms of a two-step electron- or charge-transfer process mediated via surface states or reaction intermediates. This hypothesis in combination with the recent development of high surface area nanostructures has led some authors to believe that the Helmholtz layer capacitance could vary significantly over the applied potential during EIS. Indeed recently, a varying Helmholtz layer capacitance has been taken in account with a model comprising a resistance in series with two RC elements. One element represents the space charge region and the other corresponds to charge transfer and Helmholtz layer (see Figure 8 c).133–135 Surface states are not specifically addressed with this model but are assumed to be closely coupled with the measured Helmholtz layer. Aroutounian et al.135 introduced a more complex system including the double RC model and a Warburg element characterizing a diffusion-limited process (in the space charge or in the Helmholtz layer). While this model fit well to their observations with Fe2O3 sintered pellets doped with 1 % Ti, it has yet to be verified by application to another system.
In general, EIS has been shown to be useful to assert the limitations of hematite in a PEC system, especially through the determination of the flat band potential and consequently the overpotential required for water splitting. The deeper understanding of the complex impedance spectra obtained by EIS leads to detailed characterization of electronic features, such as intra-bandgap states or surface states that presumably cause the two major drawbacks of hematite for water splitting. Nevertheless, the impedance response—especially at low frequencies—of hematite is complex and not yet fully understood. Further work will eventually lead to increase the knowledge on the electronic behavior of iron oxide.
6. Summary and Outlook
Decades of research examining the properties of hematite have identified specific limitations that prevent high solar light harvesting efficiencies. However, hematite’s positive characteristics (i.e., bandgap of 2.0 eV, exceptional stability, and abundance) make it a unique and obvious choice for solar energy storage via photoelectrochemical water splitting. Here we have seen how the optoelectronic properties create a conundrum that can be addressed by precisely controlling substitutional doping and the morphology. In combination with specific techniques to address electrochemical losses at the surface, these nanostructuring approaches have shown a remarkably rapid increase in the photocurrent produced by hematite at standard conditions. An overview of the progress in recent years is shown graphically in Figure 9.
From 2000–2005 the spray pyrolysis and ultrasonic spray pyrolysis methods produced higher photocurrents than the single crystal and polycrystalline sintered disks studied in the first decades of work with hematite photoanodes. In the past five years cauliflower-type films prepared by a particle-assisted APCVD method have produced record water splitting photocurrents. After coating these films with a layer of IrO2 nanoparticles to reduce the overpotential for water oxidation, photocurrent densities over 3 mA cm−2 at 1.23 VRHE and standard illumination conditions have been demonstrated. These sudden advances suggest that a photocurrent density approaching the maximum possible 12.6 mA cm−2 can be produced with hematite using the appropriate morphology and surface treatments. Indeed, many recent reports have demonstrated rational nanostructuring techniques that offer a clear path to further increasing the photocurrent. Noteworthy advancements in solar photocurrents using solution-based colloidal methods, and electrodeposition techniques indicate that many routes to highly photoactive hematite exist. Based on the progress and limitations of all of the different approaches to morphology control, the nanowire array and the ETA approaches offer the greatest possibility for improvement. For the former approach, attaining single-crystalline nanowires of the appropriate diameter (10–20 nm), the proper orientation, and doping will realize high photocurrents. A templating technique98 will likely be successful in this regard. On the other hand, the ETA approach has an attractive advantage as doping, and crystal orientation may not be necessary, but the suitable scaffold and buffer layer must be developed to eliminate interfacial recombination. Parallel with developing these approaches to morphology control, more work needs to be done to further understand hematite’s intrinsic limitations. For example, the sub-50 % quantum efficiencies (even at short wavelengths in optimized single crystal photoanodes), the ultrafast carrier recombination, and the interband-gap states found in hematite suggest photocurrents approaching 12.6 mA cm−2 will be difficult, if not impossible, to achieve given our lack of understanding. However, recent combinatorial efforts to optimize both the alloy composition10, 136 and doping in hematite137 may indicate new avenues for investigation. In addition, a significant reduction of the overpotential for water oxidation is urgently required. In comparison to the extensive work done on TiO2,138 relatively little is known about the surface chemistry of hematite. Only recently have some important and interesting reactivity differences on hematite surfaces have been observed,139–140 and further work on this topic is necessary.
Finally, in many ways the development of Fe2O3 reported here can be considered analogous to the development of silicon photovoltaics. At the time of the beginning of the terrestrial photovoltaic industry in the 1970s, the conventional wisdom was that crystalline silicon solar cells were hopelessly expensive and too inefficient to be a viable energy source.141 However the fact that silicon possessed ideal properties for a photovoltaic device could not be ignored, and today silicon solar cells are the dominate form of photovoltaic energy conversion thanks to the extensive technological effort put into its development. Hematite similarly possesses many ideal characteristics for solar water splitting. The recent developments presented here give strong prospect that continued efforts will bring photoanodes with performances sufficient for commercial application.
7. Symbols and Abbreviations
α Absorption coefficient
AC Alternating current
ALD Atomic layer deposition
APCVD Atmospheric pressure CVD
CVD Chemical vapor deposition
DSC Dye-sensitized solar cell
Eg Bandgap energy
EIS Electrochemical impedance spectroscopy
FTO Fluorine-doped tin oxide
J Current density
LD Diffusion length
Nd Donor density
OER Oxygen evolution reaction
RHE Reversible hydrogen electrode
SCE Saturated calomel electrode
SCLJ Semiconductor/liquid junction
STH Solar-to-hydrogen conversion efficiency
Vb Applied bias potential
Vfb Flat-band potential
VRHE Volts vs. RHE
VSCE Volts vs. SCE
W Depletion layer width
The frontispiece photograph of the Valley of Fire State Park (Nevada, USA) was graciously provided by Frank Kovalchek.
Dr. Kevin Sivula is a native of Minneapolis, MN (USA). He completed his B.Ch.E degree at the University of Minnesota in 2002, and then went on to investigate materials for bulk-heterojunction photovoltaic devices at UC Berkeley under the supervision of Prof. Jean Fréchet. In 2007 he joined the Laboratory of Photonics and Interfaces at EPFL to further study solar energy conversion. He is now leading the research activities of PECHouse, a Swiss centre of excellence devoted to the development of materials for solar water splitting.
Florian Le Formal was born in France in 1984. He obtained his M.Sc. in materials science at the European School of Chemistry, Polymers and Material Science in Strasbourg (France, ECPM). He is a currently a PhD student in the Laboratory of Photonics and Interfaces at EPFL. As a main topic, he is studying photo-electrochemical water splitting using hematite photoanodes. His major research interests are the photochemistry and the electrochemistry of metal oxide semiconductors.
Prof. Dr. Michael Grätzel directs the Laboratory of Photonics and Interfaces at EPFL. Here he has pioneered research on energy and electron transfer reactions in mesoscopic materials and their application in solar energy conversion systems, optoelectronic devices and lithium ion batteries. He received his doctorate in Natural Science from the Technical University of Berlin, and has since been awarded honorary doctorates from the Universities of Delft, Uppsala, and Turin. He is a member of the Swiss Chemical Society, a fellow of the Royal Society of Chemistry, and an honorary member of the Société Vaudoise de Sciences Naturelles.