I. Introduction

Solar energy conversion represents an important area of inorganic research.^1-5 Coordination compounds that absorb sun light can initiate excited state redox reactions that ultimately yield electrical power or useful fuels such as hydrogen gas. A key initial step is photoinduced charge separation, an example of which is given in Scheme 1. Light absorption creates an excited state, D*, that transfers an electron to a proximate acceptor to yield what is hereby referred to as a D⁺,A^- "charge-separated state". The importance of this reaction is that it provides a molecular basis for the conversion of light into free energy in the form of the D⁺ and A^- redox equivalents. The energy is only transiently formed because recombination to ground-state products, D⁺ + A^- D + A, is always thermodynamically downhill.^1-6

Scheme 1

The classical inorganic example of charge separation is electron transfer from the metal-to-ligand charge-transfer (MLCT) excited states of Ru(bpy)₃²⁺ to methylviologen, MV²⁺, where Ru(bpy)₃²⁺ is the chromophoric donor and MV²⁺ is the electron acceptor. Excited-state electron transfer produces a charge-separated state comprised of Ru(bpy)₃³⁺ and MV⁺ that stores about 1.6 eV of free energy (eq 1). Excited-state quenching yields are efficient, but rapid charge recombination lowers the yield of long-lived states significantly in aqueous solution (eq 2).⁷

An intriguing idea has been to convert the energy stored in charge-separated states directly into electrical power. The celebrated "photogalvanic cells" were designed to do this by selective collection of photogenerated D⁺'s and A^-'s at opposite dark electrodes.^10,11 Photogalvanic cells based on Ru(bpy)₃²⁺ as the chromophoric donor and Fe³⁺(aq) as the acceptor have been characterized, although solar conversion efficiencies were and remain quite low.^10,11 A related approach is to have excited-state electron transfer directly to the conduction band of a wide-band-gap semiconductor (Scheme 2).^12,13 Excited-state electron transfer then yields interfacial charge-separated states, D⁺/TiO₂(e^-), that are more easily collected and converted to electrical power.

Scheme 2

When a photocurrent is generated with light energy less than that of the semiconductor band gap, the process is known as sensitization and the light-absorbing dyes are referred to as sensitizers. Polypyridyl compounds of Ru(II) were among the first sensitizers to be studied and continue to be the most promising for real-world applications in regenerative solar cells.^14-16 The mesoporous nanocrystalline TiO₂ thin films developed by Grätzel and co-workers have resulted in an order of magnitude increase in solar energy conversion efficiencies when employed in such cells.¹⁶ The film thickness and high transparency allow interfacial charge-transfer processes to be spectroscopically characterized, with signal-to-noise ratios approaching those obtained in fluid solution.¹⁷

This paper summarizes recent studies of interfacial charge transfer at TiO₂ interfaces sensitized to visible light with (d)⁶ transition-metal compounds.^1-5 We present some background data on the synthesis and characterization of sensitized mesoporous TiO₂ thin films and then focus on surface-mediated photochemical processes relevant to MLCT excited states and interfacial charge transfer. Emphasis is also given to "supramolecular" compounds that were designed to perform more elaborate tasks when anchored to semiconductor surfaces.⁴ Unless otherwise stated, the experimental data described herein were obtained in neat organic solvents or electrolytes. Therefore, the relevance to regenerative solar cells, which generally operate in 0.5 M LiI and 0.05 M I₂ nitrile solutions, is not always clear. The research activity in this area is tremendous, yet many recent discoveries provide new opportunities for solar energy conversion with coordination compounds anchored to semiconductor surfaces.

II. Results and Discussion

A. Sensitized TiO₂ Thin Films. 1. Mesoporous Thin Films. The preparation of nanocrystalline TiO₂ thin films by sol-gel techniques has been well documented in the literature.^18-20 Hydrolysis of titanium alkoxides at controlled pH and particle growth at elevated temperatures yield colloidal solutions of 8-20-nm anatase particles. The solutions are then concentrated, mixed with a stabilizer, and spin-coated or doctor-bladed onto a variety of substrates. A brief heat treatment removes the stabilizers and yields a mesoporous film of interconnected anatase nanoparticles in a spongelike network that is typically 8-10 m thick. Anatase is one of the three common polymorphs of TiO₂.²¹ Interestingly, the (100) plane (Figure 1) has ~1.2-Å-diameter pores that are known to intercalate cations such as H⁺ and Li⁺ when reduced electrochemically.²² These "potential-determining ions" have a profound influence on the energetics of the TiO₂ acceptor states, as is discussed further below.

Figure 1 (100) plane of anatase TiO2.

These mesoporous TiO₂ materials possess many qualities ideal for solar energy conversion and for fundamental studies of interfacial charge separation. Below we list five of the most important.

1. The high surface area is important for sensitizer binding and efficient solar harvesting.^23-25 For ruthenium polypyridyl sensitizers, surface coverages of ~10^-8 mol cm^-2 are typically reported, which is about 100-1000 times that expected for monolayer coverage on a planar semiconductor surface.

2. The 50-70% porosity allows facile diffusion of redox mediators within the film to react with surface-bound sensitizers.

3. The density of unfilled acceptor states can be widely and reversibly tuned in energy for optimization of light to electrical energy conversion efficiencies and for fundamental characterization of excited states and interfacial charge separation.²⁴

4. Injected electrons can be efficiently collected in an external circuit. Carrier transport is an activated process that has been extensively studied.^26,27

5. High transparency in the visible and infrared regions allows spectroscopic characterization of photoinduced electron-transfer reactions. In many cases, the spectroscopy can be correlated with photoelectrochemical properties.^17,18a

We note that these desirable properties are not restricted to TiO₂ and that mesoporous nanocrystalline thin films based on SnO₂, ZnO, and ZrO₂ have been prepared through related sol-gel techniques.^23-25

The nature of the electron-acceptor states in TiO₂ remains poorly understood, and the molecular descriptions an inorganic chemist would like are lacking. Distinctions are often made between delocalized conduction band electrons and electrons trapped to form localized Ti(III) states.^23-25 A commonly held view is that electron injection occurs to the unfilled conduction band states. The injected electron rapidly thermalizes to the conduction band edge, E_cb, where it is trapped to form Ti(III)-like states. Charge recombination is thus from electrons "trapped" in Ti(III) states within the forbidden energy gap. However, we have found no compelling evidence for this distinction between free and trapped carriers and hereby simply refer to reduced TiO₂ as TiO₂(e^-).

It is well-known that TiO₂ can be reduced electrochemically, chemically, or photoelectrochemically and that the resultant blue-black-colored material is stable in the absence of dioxygen or other electron acceptors.²¹ TiO₂(e^-) has characteristic electron paramagnetic repulsion and UV-vis spectra.^21,28-30 A broad absorption that increases with wavelength through the visible region is observed and is accompanied by a blue shift of the fundamental valence-to-conduction band absorption edge. The extinction coefficient of reduced TiO₂ thin films are about 1000 M^-1 cm^-1 at 800 nm.³⁰ We have shown that the electronic spectra of electrochemically reduced TiO₂ was within experimental error the same as those generated by dye sensitization with a sacrificial electron donor.²⁹ Therefore, the TiO₂(e^-)'s can be generated photo- and electrochemically for thermal electron-transfer studies.

Determination of the TiO₂(e^-) "redox potential" is necessary for the estimation of the free energy stored in interfacial charge-separated states. Under all conditions that we have studied, injected electrons in TiO₂ were able to reduce MV²⁺, indicating that the free energy stored is larger than that in the classical Ru(bpy)₃³⁺,MV⁺ charge-separated state, G > 1.6 eV. The standard photoelectrochemical techniques used to estimate E_cb cannot be directly applied to these nanostructured materials.³¹ Rothenberger and co-workers have proposed an accumulation layer model to describe the potential distribution within the TiO₂ particles at negative applied potentials.²⁹ This model assumes that the band-edge positions remain fixed as the Fermi energy is raised into accumulation conditions, which has little literature precedence.²⁸ Nevertheless, the model appears to fit the spectroelectrochemical data and provides the only literature estimates of E_cb available for these materials in organic and aqueous electrolytes.³⁰ Because there is likely to be some band-edge movement as the Fermi energy is raised, we feel that the reported values are best thought of as upper limits; i.e., the true conduction band edge is further from the vacuum level (i.e., more positive) under open-circuit conditions.²⁸

A final point with regard to the energetic positions of the band edges is that they are not singular parameters. They shift when the environment around the semiconductor is changed.²¹ For example, the band-edge positions are known to display a Nernstian (59 mV/pH) shift in aqueous solution due to protonation/deprotonation of surface titanol groups. In nonaqueous solvents such as acetonitrile, the adsorption of cations to the surface has also been shown to have a significant influence on E_cb. For example, E_cb has been reported be -1.0 V vs SCE in 0.1 M LiClO₄/acetonitrile and shifts to ~ -2 V when the Li⁺ cations were replaced by tetrabutylammonium cations.³⁰ The direction of the band-edge shifts has been confirmed by excited-state quenching studies described further below.

2. Sensitizer Binding. Coordination compounds with functional groups for surface binding have been linked to metal oxide surfaces by room-temperature reactions in organic solvents.^23-25 The most common and successful functional groups for excited-state sensitization are based on carboxylic acids. Others based on phosphonates, siloxanes, acetyl acetonates, ethers, phenols, and cyanides have also been reported.³² In some studies, the TiO₂ surfaces were pretreated with aqueous solutions of known pH or with TiCl₄. In many cases, the films were used without pretreatment immediately after the final heating step in the sol-gel process.

Goodenough first proposed that carboxylic acids would dehydratively couple with titanol groups to form surface ester linkages (Scheme 3).³³ We have reported spectroscopic evidence for this and for the more commonly observed "carboxylate" linkage in reactions of Ru(dcbH₂)(bpy)₂(PF₆)₂ and Ru(bpy)₂(ina)₂(PF₆)₂, where dcbH₂ is 4,4'-(CO₂H)₂-bpy and ina is isonicotinic acid, with nanocrystalline TiO₂ and colloidal ZrO₂ films in acetonitrile at room temperature.³⁴ The interfacial proton concentration was intentionally varied by equilibration of the films with aqueous solutions of known pH prior to sensitizer binding. The visible absorption and IR spectral data indicate that a high surface proton concentration yields an ester-type linkage(s) where low proton concentrations favor "carboxylate"-type binding mode(s). The TiO₂ surface saponified the esters in Ru(deeb)(bpy)₂(PF₆)₂, where deeb is 4,4'-(CO₂Et)₂-bpy, and the carboxylate products bound to the surface.³⁴ For a variety of sensitizers with carboxylic acid functional groups reacted with untreated TiO₂, the concentration-dependent binding generally follows the Langmuir adsorption isotherm model with adduct formation constants of 10⁴-10⁵ M^-1 and limiting surface coverages of 10^-8 mol cm^-2. This corresponds roughly to monolayer coverage and about 700 Ru(II) sensitizers anchored to each ~20-nm TiO₂ particle. We emphasize that these values are approximate and that absolute proof of monolayer coverage is exceedingly difficult due to the ill-defined surface area of the mesoporous thin films.

Scheme 3

These data, and those of others in the literature, did not allow direct identification of the surface site(s) involved in the sensitizer-semiconductor bond.^33-38 Deacon and Phillips have tabulated vibrational data for metal carboxylate compounds whose structures were determined crystallographically.³⁹ An empirical relationship between the energy separation of the CO asymmetric and symmetric stretches and the carboxylate-metal coordination mode was found. This same approach has been used to predict the carboxlate binding modes to presumed Ti(IV) sites on the TiO₂ surface.^35-37 The analysis is most consistent with the carboxylate oxygens binding to separate Ti(IV) centers on the anatase surface, in agreement with theoretical studies.⁴⁰

The binding of metal cyano compounds to TiO₂ in acidic solutions has also been the subject of considerable research.^41-43 In solution, ferrocyanide has an octahedral geometry with six equivalent C-N bonds, which produces one infrared-active CN stretching band at 2040 cm^-1. Upon attachment to the semiconductor surfaces, the number of infrared-active CN stretching bands was dependent on the adsorption geometry. In principle, surface binding could occur through one, two, or three of the ambidentate cyanide ligands and the corresponding molecular symmetry would be C_4v, C_2v, or C_3v. The C_4v and C_3v symmetries produce two infrared-active modes, and the C_2v symmetry produces three nondegenerate infrared-active modes. Hupp and co-workers observed two bands at 2058 and 2072 cm^-1 and a higher energy band assigned to the bridged CN at 2118 cm^-1 in Raman studies of Fe^II(CN)₆^4-/TiO₂.⁴¹ These data suggested that only one type of coordinated ferrocyanide existed on the TiO₂ surface, linked by a single cyanide ligand with a C_4v symmetry, i.e., Ti^IV-NC-Fe^II(CN)₅. Subsequent FTIR studies were in agreement with this binding geometry.^42,43

3. Sensitizer Energetics. It has been have shown that Ru(II) and other sensitizers anchored to nanocrystalline TiO₂ films can be oxidized and reduced in standard electrochemical cells.³² Cyclic voltammetry and spectroelectrochemistry are thus powerful in situ tools for the determination of formal reduction potentials and absorption spectra of relevant redox states. The mechanism by which the semiconductor-bound sensitizers can be reversibly oxidized and reduced is nonobvious. The Ru(III/II) reduction potentials reside energetically within the forbidden energy gap of TiO₂. Therefore, band-gap illumination would be required for TiO₂ to oxidize these sensitizers. Instead, the dark redox chemistry is initiated by electron transfer with compounds on the underlying substrate (usually fluorine-doped tin oxide) and extends to sensitizers on the nanoparticles by lateral intermolecular electron "hopping".³² Complete oxidation of all of the sensitizers thus requires that a pathway for "hopping" to each sensitizer is present. Bonhote and co-workers have, in fact, shown that a percolation threshold exists for the oxidation of amines anchored to nanocrystalline TiO₂ thin films.⁴⁴ Mobile redox-active compounds in the electrolyte can be used to mediate charge transfer to surface-bound sensitizers in isolated regions of the film.⁴⁵

B. Excited-State Properties. The excited states of (d)⁶ polypyridyl compounds have been extensively studied.^1-5 Radiative and nonradiative decay in the prototypical Ru(bpy)₃²⁺* is mechanistically complex. A number of potential energy surfaces can potentially mediate excited-state decay. In fluid solution, it is well established that upon light absorption quantitative intersystem crossing to a manifold of thermally equilibrated MLCT excited states (or "thexi" states²) occurs. The thexi state consists of three energetically proximate states that have a significant Boltzmann population and behave as a single state near room temperature. It is well formulated as an excited electron localized on a single bipyridine ligand, Ru^III(bpy^-)(bpy)₂²⁺*. Higher in energy is a fourth MLCT excited state and distorted ligand field states.^3,5 It is well-known that the external environment has a profound influence on MLCT excited states.^46,47 Experimental studies of MLCT states on TiO₂ (and other semiconductors) are few mainly because of rapid interfacial charge separation that shortens the excited-state lifetimes considerably. Some aspects of MLCT excited states anchored to TiO₂ are now becoming available through conduction band-edge tuning studies that raise E_cb above (toward the vacuum level) the excited-state reduction potential of the sensitizer, thereby inhibiting charge separation.

1. Energy Transfer. In many important regards, the room-temperature excited-state properties of Ru(dcb)(bpy)₂, where dcb is 4,4'-CO₂^--bpy, anchored to TiO₂ are remarkably similar to those observed in fluid solution.⁴⁸ The most significant difference was that excited-state relaxation on TiO₂ (and ZrO₂) occurred by parallel first- and second-order kinetic processes. The seocnd-order component was attributed to triplet-triplet annihilation reactions that occur in parallel with radiative and nonradiative decay. Excited-state annihilation is facilitated by the close proximity of the surface-bound compounds that affords rapid, lateral, isoenergetic energy transfer across the semiconductor surface (Scheme 4). Monte Carlo simulations indicate that the rate for intermolecular energy transfer is about (30 ns)^-1 at maximum surface coverage. We therefore find that sensitizers with short excited-state lifetimes, < 50 ns, decay exponentially on nanocrystalline TiO₂ surfaces at low irradiances.⁴⁸

Scheme 4

Direct evidence for energy transfer came from studies where both Ru(dcbH₂)(bpy)₂(PF₆)₂ and Os(dcbH₂)(bpy)₂(PF₆)₂ were anchored to the same nanocrystalline TiO₂ film.⁴⁹ The Os compound acts as an energy-transfer trap and reaction 3 occurs with a quantum yield within experimental error of unity and a rate constant > 10⁸ s^-1 when the two sensitizers were anchored to the surface in equal concentration.

2. Ligand-Field Excited States. The presence of low-lying ligand-field states in ruthenium(II) polypyridyl compounds may be inferred from the appearance of ligand loss photochemistry and temperature-dependent lifetimes.^1-5,51 A classical example is cis-Ru(bpy)₂(py)₂²⁺, where py is pyridine.⁵² This compound is nonemissive at room temperature, and photochemical ligand loss is facile. Thermally activated population of ligand-field states is evident. We recently reported that cis-Ru(bpy)₂(ina)₂(PF₆)₂, where ina is isonicotinic acid, was also nonemissive in fluid solution, < 10 ns. When anchored to nanocrystalline TiO₂, the compound was much more photostable and highly photoluminescent with a lifetime of 60 ns.⁵³ The apparent MLCT LF internal conversion activation energy increased to 2500 cm^-1 upon surface binding. Similar increases in this activation energy have been observed for Ru(bpy)₃²⁺ in solid-state media and attributed to destabilization of the ligand-field states by the solid.⁵⁴ Interestingly, a static component was also observed in the temperature-dependent lifetimes of cis-Ru(bpy)₂(ina)₂/TiO₂, which indicated that the intersystem crossing yield was not unity or temperature-independent.⁵³

Ruthenium ammine excited states are also known to display efficient ligand-field photochemistry.⁵⁵ In preliminary studies with compounds of the type Ru(ina)(NH₃)₅²⁺ and Ru(dcbH₂)(NH₃)₄²⁺, enhanced photostability after attachment to TiO₂ was also observed.⁵⁶ Excited-state interfacial charge separation was also observed with these sensitizers. The injection yields were dependent on the excitation wavelength and increased substantially when the ammine ligands were deuterated. These initial studies are promising and may ultimately lead to the realization of efficient charge separation with iron polypyridyl compounds whose ligand-field excited states have historically limited their usefulness.⁵⁷

3. Outer-Sphere Reorganization. Sensitizers anchored to semiconductor surfaces are exposed to the external environment and to the semiconductor surface. Therefore, the reorganization energy for charge transfer is expected to include contributions from both. At present, there exist few experimental data on charge-transfer reorganization energies at molecular-semiconductor interfaces. In the classical limit, the optical energy for MLCT absorption, E_ab, and emission, E_em, are related to the Gibbs free energy and the total reorganization energy of the excited and ground states, respectively (eqs 4 and 5).⁴⁷

Cyano compounds of Fe(II) and Ru(II) are known to be highly solvatochromic.^47,58 Outer-sphere interactions with the cyano ligands have a profound influence on the Ru(III/II) reduction potential and hence the color of the compound.⁴⁷ For Fe(bpy')(CN)₄^2- compounds, where bpy' is bpy or 4,4'-(CH₃)₂-bpy, the excited-state reorganization energy was found to be significantly larger on TiO₂ than in fluid solution.⁴³ In fluid solution, _es = 0.10 ± 0.07 eV and increased to 0.30 ± 0.07 eV upon surface binding. The restricted translational mobility of the semiconductor-bound iron compounds and the ambidentate Fe-CN-Ti linkages may underlie the larger reorganization energies.

Ru(dcbH₂)(CN)₄^2- is remarkably solvatochromic.⁵⁹ For example, the lower energy MLCT absorption band of Ru(dcb)(CN)₄^2- was observed at 427 nm in water and at 546 nm in dimethylformamide. The color change was due to a shift of the Ru(III/II) ground-state reduction potential with the solvent. The complex maintains this solvatochromism after attachment to nanocrystalline (anatase) TiO₂ films, although the magnitude of the effect was smaller.⁵⁹ Solvent tuning altered the spectral response of Ru(dcb)(CN)₄/TiO₂ regenerative solar cells in a predictable manner. A recent Raman study has shown that the solvent reorganization energy for cis-Ru(dcbH₂)₂(NCS)₂ decreased by about ¹/₆ when the compound was anchored to TiO₂.⁶⁰ Further studies are needed to provide a more detailed description of the solvation environment of coordination compounds anchored to semiconductor surfaces.

C. Interfacial Charge Separation. The excited-state injection mechanism shown in Scheme 2 is, in fact, only one of three mechanisms identified for interfacial charge separation. The mechanisms differ by the nature of the donor that transfers the electron to the semiconductor: (1) excited state, i.e., Ru^III(dcb^-)(bpy)₂²⁺*; (2) reduced state, i.e., Ru^II(dcb^-)(bpy)₂⁺; or (3) molecule-to-particle charge-transfer complex, i.e., Ru^II-CN-Ti^IV Ru^III-CN-Ti^III. Below we discuss some key observations and unresolved issues for each of these three sensitization mechanisms.

1. Excited-State Electron Injection. Gerischer has described a theory for excited-state electron injection into wide-band gap semiconductors.¹² The rate of interfacial electron transfer at a sensitized semiconductor interface is predicted to be is proportional to the overlap of occupied donor excited states with unoccupied acceptor states

Figure 2 Gerischer's diagrams for excited-state sensitization: (a) favorable overlap for interfacial electron transfer from the sensitizer excited state (low pH); (b) unfavorable overlap (high pH).

There now exists a large body of experimental data supporting ultrafast electron transfer from MLCT excited states to anatase TiO₂.^61,62 Most, but not all, of these studies have focused on the famous "N3" dye first prepared by Nazeeruddin et al., cis-Ru(dcbH₂)₂(NCS)₂.⁶³ In a recent study, an excitation wavelength dependence on the injection process was time resolved.⁶² Femtosecond injection was attributed to the singlet state and a slower picosecond process from the thermally equilibrated triplet state.⁶²

Evidence for ultrafast injection has also come from photoluminescence quenching studies of Ru(dcb)(bpy)₂*/TiO₂ with inorganic cations in organic electrolytes.⁶⁴ The quantum yield for electron injection was reversibly tuned from below detection limits, ~0, to near unity simply by altering the [Li⁺] concentration in an external acetonitrile bath. A Gerischer-type model was proposed to account for this behavior wherein cation adsorption to the TiO₂ surface shifts the semiconductor acceptor states positive on an electrochemical scale (i.e., away from the vacuum level), resulting in better overlap with the chromophoric donor's excited state (Figure 3).

Figure 3 Model to describe the [Li+] dependence of excited-state quenching of Ru(dcb)(bpy)2/TiO2 in acetonitrile. Cation adsorption shifts the acceptor states of TiO2 positive, resulting in better overlap with vibrationally hot excited states of the sensitizer, RuIII/II**. All potentials are versus SCE.

Interestingly, the quenching data were most consistent with injection from vibrationally hot excited states, Ru^III/II**. At low Li⁺ concentrations, excited-state quenching was found to be static on a nanosecond time scale.⁶⁴ In other words, the number of excited states observed immediately after pulsed laser excitation decreased with Li⁺ addition, while the lifetime did not change appreciably. If ultrafast injection were occurring from the thexi state, the lifetime should have dropped to 1/k_inj, contrary to what was observed. An alternative interpretation considered was that the observed static quenching represented sample heterogeneity in the acceptor state energies; i.e., excited states inject only if cations adsorb to physically proximate surface sites. This interpretation was rejected because efficient intermolecular energy transfer across the surface would translate the excited state to such sites, resulting in injection rate constants rate-limited by energy transfer, a behavior that has never been reported.⁴⁹ Cation-induced shifts in E_cb were observed with other Lewis acidic alkali and alkaline-earth metals; the magnitude was correlated with the size-to-charge ratio of the cation.⁶⁴ The application of this behavior for ratiometric photoluminescence sensing of cations was recently explored.⁶⁵

a. Tuning Sensitizer-TiO₂ Electronic Interactions. The appearance of subpicosecond injection rate constants is often attributed to strong electronic coupling between the carboxylic acid groups in the dcb ligand with the semiconductor surface. It was therefore of interest to examine the effects of decreasing the electronic coupling to the surface. Because the radiative and nonradiative rate constants for excited-state decay are relatively slow, nanosecond injection rate constants would still be expected to occur with quantum yields near unity.^1-4 In fact, there may be an as of yet undetermined sensitizer-semiconductor electronic interaction, where charge injection occurs quantitatively and recombination is slowed considerably. Such behavior would be expected to increase the photocurrent efficiency from "black" sensitizers with more negative Ru(III/II) reduction potentials.^23-25

Tuning sensitizer-TiO₂ electronic interactions were initially accomplished by introduction of flexible methylene spacers between the carboxylic acid binding groups and the bipyridine ligand or with bimetallic sensitizers.^32,66,67 In both cases, efficient charge injection was realized. More recently, elegant tripodal compounds with phenyl-ethyne spacers between the chromophore and three ester (or carboxylic acid) groups have been designed and synthesized by Galoppini (Figure 4).^68,69

Figure 4 Tripodal sensitizers prepared by Galoppini et al.68

Remarkably, with the tripodal sensitizers shown, subpicosecond injection rates were measured.^68c If the sensitizers were anchored to the surface in an idealized Eiffel Tower like orientation, the Ru(II) center is about 24 Å from the semiconductor surface. Delocalization of the excited state onto the phenyl-ethynes decreases the injection distance. Indeed, this bridge may act as a conduit for electron transfer to the adamantyl group. A concern was that the ultrafast injection process occurred to neighboring TiO₂ particles and not to the one the sensitizer was anchored to. However, the injection dynamics were sensitive to whether bpy or phen was the chromophoric ligand, behavior that is hard to rationalize if injection occurred to a neighboring anatase particle. Furthermore, with related sensitizers based on rigid-rod linkers, we have observed nanosecond injection dynamics by intentionally lowering the density of TiO₂ acceptor states.⁶⁹

We, and others, have shown that the quantum yield for electron injection from a bipyridine ligand that is not directly bound to the semiconductor surface can occur quantitatively (Scheme 5).^32,67,70 For example, light excitation of Re(bpy)(CO)₃(ina)⁺ results in quantitative injection from the Re^II(bpy^-)(CO)₃(ina)⁺* state with k_inj > 10⁸ s^-1.¹⁷ The injection yield after light excitation of cis-Ru(bpy)₂(ina)₂/TiO₂ was lower than that expected based on the excited-state reduction potential.⁵³ It was found that the injection yield increased to near unity when the solution was cooled. Such behavior is exactly the opposite of what one would expect for activated interfacial charge separation based on Gerischer theory.¹² This sensitizer has low-lying ligand-field states, and it was proposed that rapid internal conversion processes underlie the temperature dependence. To our knowledge, this remains the sole example of temperature-dependent electron injection yields at sensitized semiconductor interfaces.⁵³

Scheme 5

b. Is Ultrafast Electron Injection Useful for Solar Energy Conversion? In dye-sensitized solar cells, it remains unclear whether ultrafast injection is necessary, or even desirable, for energy conversion. Nanosecond injection rates might still compete effectively with radiative and nonradiative decay and thus would be expected to be quantitative. However, under conditions where injection was inhibited, efficient lateral energy transfer was observed that could translate the excited state to impurity or defect sites that might ultimately lower injection yields. Undesirable charge-transfer processes with redox mediators or the solvent could also occur if injection were slow. In any case, no one has succeeded in capturing the excess free energy of the injected electron prior to thermalization and phonon release. The same charge-separated state appears to be formed whether vibrational (phonon) relaxation occurs prior to or after injection. A recent example of "trapping" of hot carriers provides some clues as to how hot carriers might be harvested.⁷¹

Electrons injected into TiO₂ from upper vibrational excited states were found to reduce other surface-bound sensitizers and yield long-lived charge-separated intermediates that store ~2 eV of free energy, almost the entire energy of the thexi states.⁷¹ A key to the realization of this behavior was the use of sensitizers whose first reduction potential was below E_cb. In this way, injection must occur exclusively from hot vibrational excited states and, after hot injection, back electron transfer to form the excited state of that sensitizer or the reduced state of a neighboring sensitizer was energetically favored. The first sensitizers shown to do this were M(dcbq)(bpy)₂/TiO₂, where M = Ru or Os and dcbq is 4,4'-(CO₂^-),-2,2'-biquinoline. The trapping yield was found to have an excitation wavelength dependence and increased markedly with photon energy. Charge recombination now occurred by intermolecular charge transfer across the semiconductor surface (Scheme 6). The observation of this behavior opens the door toward fundamental studies of charge trapping at semiconductor interfaces where well-defined molecular compounds trap and store charge. The charge-separated states may be exploited for applications in photocatalysis. It is also likely that other examples will emerge as solar energy researchers utilize ligands with low-lying * orbitals and tune the conduction band-edge position to optimize spectral sensitivity and power conversion efficiencies of dye-sensitized solar cells.⁷¹

Scheme 6

2. Reduced Sensitizer Injection. In this mechanism, a donor present in the electrolyte reductively quenches the excited state and the reduced state, S^-, transfers an electron into the semiconductor. Solar cells based on this mechanism are often referred to as photogalvanic cells.¹¹ A potential advantage of this mechanism is that the reduced sensitizer is a stronger reductant than the MLCT excited state, typically by 0.3-0.5 eV.⁴ Thus, sensitizers that are weak photoreductants may sensitize TiO₂ efficiently after reductive quenching. This mechanism may be exploited to produce large open-circuit photovoltages or enhanced light harvesting in the near-IR regions. The vast majority of early dye sensitization studies were performed on planar electrodes and relied strictly on photoelectrochemical measurements from which it was often impossible to unambiguously distinguish between these two mechanisms.⁷² In special cases, the photoelectrochemistry data quite convincingly demonstrated this mechanism.⁷³ The observation of ultrafast electron injection⁶ coupled with the weak oxidizing power of the excited sensitizers currently in use strongly suggest that an excited-state injection mechanism is operative in regenerative solar cells based on these materials. Nevertheless, absolute proof is lacking because most spectroscopic measurements are performed in a redox-inactive electrolyte. Recent ultrafast injection studies performed in the presence of iodide reveal inhibited charge injection and some evidence for reductive quenching.⁷⁴

Strong spectroscopic evidence for the reduced sensitizer injection mechanism was recently reported.⁷⁵ Interfacial electron transfer was shown to be rate-limited by reductive quenching of the sensitizer excited state. The sensitizer used in this work was Ru(dcbH₂)(bpy)₂(PF₆)₂, and the electron donor was phenothiazine, PTZ. A drawback of the PTZ donors is that they produce negligible photocurrents in regenerative dye-sensitized nanocrystalline solar cells.⁷⁵ Iodide is the sole electron donor identified that yields solar energy conversion efficiencies > 10% under one sun of AM 1.5 solar irradiation.^23-25 Ru(dcb)(bpy)₂²⁺* and N3* are weak photooxidants that do not efficiently oxidize iodide by dynamic mechanisms.

In more recent work, Ru(II) sensitizers based on bipyrazine that were potent excited-state oxidants capable of efficient iodide oxidation were characterized.⁷⁶ Efficient reductive quenching was observed by a variety of electron donors. Unlike previous studies, neither the reduced or excited state of the sensitizer transfers electrons to TiO₂ or SnO₂ nanoparticles efficiently. Cage escape yields were quite high. Long-lived charge separation was realized with the reduced sensitizer bound to the surface and a mobile oxidized donor. These studies demonstrated a general approach for photogeneration of charge-separated states that store >2 eV of free energy.⁷⁶

3. Metal-to-Particle Charge Transfer. There exists a relatively small, but important, class of inorganic and organic compounds that form charge-transfer complexes with the TiO₂ surface.⁷⁷ Interfacial chemistry between the compounds and the TiO₂ surface produces color changes that cannot be explained by trivial acid-base chemistry, decomposition, or aggregation. Such absorption bands were first observed by Grätzel and assigned as molecule-to-particle charge-transfer transitions.⁷⁸ There now exist a large number of theoretical and experimental results that support this assignment.⁷⁷ Molecule-to-particle charge-transfer interactions raise the interesting issue of where the molecule stops and the extended solid begins. Experimentally, they are less complicated because one is assured that each absorbed photon is converted to an interfacial charge-separated state. In contrast, the efficiency of charge separation by excited-state sensitization has been shown to be a function of the excitation wavelength, pH, temperature, and ionic strength.⁷⁷

Metal cyanides, [M(CN)_x]^4- (M = Fe^II, Ru^II, Os^II, Re^III, Mo^IV, or W^IV, x = 6, 7, or 8), such as ferrocyanide, Fe^II(CN)₆^4-, bind to TiO₂ through ambidentate cyano ligands. Fe^II(CN)₆^4- does not absorb light beyond 380 nm, but a deep orange color with an absorption maximum centered at 420 nm was observed for Fe^II(CN)₆^4-/TiO₂. The visible absorption was attributed to a MPCT complex formed between Fe^II(CN)₆^4- and surface Ti⁴⁺ ions, Fe(II) Ti(IV). Sensitized photocurrents and transient absorption studies support this assignment.⁷⁸

Lian et al. studied interfacial electron transfer with a Fe^II(CN)₆^4-/TiO₂ nanoparticle in D₂O with subpicosecond infrared spectroscopy.⁷⁹ A mid-infrared absorption was assigned to TiO₂ electrons in the semiconductor. Consistent with a direct MPCT sensitization mechanism, the injection rate constant could not be time-resolved with a 50-fs instrument response function. TiO₂(e^-) Fe^III(CN)₆^3- charge recombination was found to be nonexponential, with rates that showed little dependence on the pH, particle size, or pump power. This suggested that charge recombination was mostly geminate. The complex kinetics were attributed to a distribution of trap-state energetics at variable physical locations relative to the ferric center. Electrons in deep trap states and at longer distances recombined more slowly because of a smaller electronic coupling matrix element.⁷⁹

We reported the first example of a coordination compound, Fe(bpy)(CN)₄^2-, designed to sensitize TiO₂ to visible light by two distinct charge-transfer pathways (Scheme 7).^42,43 The absorption spectra of the Fe(bpy)(CN)₄^2- compound anchored to TiO₂ were well modeled by a sum of metal-to-ligand (Fe bpy) and metal-to-particle (Fe(II) Ti(IV); MPCT) charge-transfer bands. Charge separation could not be time-resolved, while recombination was well described by a second-order kinetic model. An ionic strength dependent quantum yield for charge separation measured after Fe bpy excitation was taken as evidence for the MLCT excited-state sensitization pathway. The metal-to-particle pathway gave the expected ionic strength independent yield of unity. The MLCT bands were solvatochromic, while the MPCT bands were not. The total reorganization energy for the MPCT of Fe(CN)₆^4-/TiO₂ was estimated to be ~0.6 eV, which compared well with that for MLCT sensitization, ~0.3 eV, of single-crystal electrodes in aqueous solution.¹⁵

Scheme 7

D. Interfacial Charge Recombination. Recombination of the injected electron with the oxidized metal center generates ground-state products and wastes the energy stored in the interfacial charge-separated state. For efficient photocurrent generation in regenerative solar cells, iodide oxidation must be faster than charge recombination. It has been known for some time that recombination occurs on a micro- to millisecond time scale while injection is orders of magnitude faster. The origin of this fortuitous difference in interfacial electron-transfer rates has been the subject of much discussion.^23-25 Our studies first indicated that charge recombination was not slow because of inherently sluggish rate constants but because the process was second-order in nature.⁶⁴ Systematic measurements where the concentration of charge-separated states was independently varied with irradiance or cation adsorption were reported. The abstracted second-order rate constants were found to be independent of the number of interfacial charge-separated pairs photocreated. These data represent the strongest evidence to date that interfacial charge recombination is a second-order process.⁶⁴

1. Driving Force Dependence. There exists compelling evidence that charge recombination falls in the Marcus kinetic inverted region. The temperature and driving force dependence of the reaction rate constants are consistent with inverted behavior.^79-83 Furthermore, the reorganization energies for back electron transfer to the Ru(III) center are ~0.5-1.0 eV, which is much less than the estimated driving force. Therefore, inverted kinetic behavior is expected and is observed. For example, Hupp and co-workers sensitized TiO₂ with a series of Fe(CN)₅L^3- compounds in which the Fe(III/II) potentials were modified through variations in L.⁸⁰ Back electron transfer was quantified by nanosecond absorption spectroscopy, and log k_bet decreased as the relative driving force increased. A similar behavior has been seen after MLCT sensitization of colloidal solutions and thin films.^81,82 The inverted interfacial reactions are known to be thermally activated, with H* increasing with the driving force.^79-83

We and others have found conditions where charge recombination was insensitive to the formal reduction potential of the sensitzer.^84,85 This behavior was attributed to slow transport of the injected electron to the oxidized sensitizer as the rate-determining step for charge recombination. Charge transport and recombination in sensitized TiO₂ have both been shown to be second-order in electron concentration. Nelson has modeled recombination data with the Kohlrausch-Williams-Watts model, which is a paradigm for charge transport in disordered materials.⁸⁵ The rates increased significantly when additional electrons were electrochemically introduced into TiO₂.⁸⁶ It therefore appears likely that a high density of trap states and a relatively small number of injected electrons will give rise to this behavior. Because the number of trap states at colloidal semiconductor interfaces is generally expected to be quite high, this observation implies that long-lived charge separation will occur at most sensitized nanocrystalline semiconductor interfaces provided that charge separation occurs.

2. Electronic Coupling. Researchers have recently correlated back-electron-transfer rate constants with the physical location of the oxidized sensitizer's lowest unoccupied molecular orbitals. This work compliments the studies of supramolecular sensitizers described below. The rates were found to decrease exponentially with distance over the first few angstroms, and a value of 0.95 ± 0.2 Å^-1 was estimated.⁸⁷ Attempts to slow recombination by fixing the distance between the sensitizer and semiconductor surface with conjugated spacers have been less successful.^68,88 Rate constants that showed very little distance dependence or did not vary in a systematic way with the expected distance have been reported. This research deserves further study.

An alternative approach for tuning electronic coupling and interfacial charge-separated state lifetimes has been to place a thin layer of a second metal oxide between the sensitizer and TiO₂.^89,90 Two classes of materials have been investigated in this regard, insulators and semiconductors. The conduction band edge of the semiconductor layers was designed to lie energetically above that of TiO₂, thereby promoting vectorial charge transfer away from the oxidized sensitizer. With insulators, the strategy was to have the excited electron tunnel through the insulating barrier and lower the electronic coupling for back electron transfer. These materials have indeed enhanced charge-separated state lifetimes, although they have not increased the photoelectrochemical energy conversion efficiencies over what has been realized with sensitized TiO₂ films that did not contain additional oxide layers.¹⁶

In a recent study, the effects of insulating layers of Al₂O₃ or ZrO₂ between cis-Ru(dcb)₂(NCS)₂ and TiO₂ were reported.⁹⁰ A good correlation was found between the recombination dynamics and the point-of-zero charge, pzc, of the metal oxide. In this context, the pzc refers to the pH where the net surface charge is zero. The material with the most basic pzc, Al₂O₃ (pzc = 9.2), was found to yield optimal behavior for inhibiting charge recombination. In a novel extension of this same approach, a cis-Ru(dcb)₂(CN)₂/TiO₂ thin film was coated with a layer of Al₂O₃, to which a ruthenium phthalocyanine complex was attached to produce RuPc/Al₂O₃-Ru(dcb)₂(CN)₂/TiO₂ thin films. Light excitation was found to initiate electron-transfer reactions that ultimately produced an electron in TiO₂ and an oxidized phthalocyanine. About half of these states lived for 5 ms prior to recombination. It is noteworthy that the sol-gel process was sufficiently mild that it allowed these complex molecular heterostructures to be fabricated.⁹⁰

E. Supramolecular Sensitizers. "Supramolecular" sensitizers have allowed intramolecular "hole" hopping and "stepwise" electron injection processes to be quantified. This work provides an example of how the principles of stepwise charge separation, originally developed in the field of supramolecular photochemistry, can be applied to solid-state materials.⁴ The systems studied were designed as proof-of-principle examples, without any pretension to compete with the sensitizers commonly used in regenerative solar cells.

1. Stepwise Electron Injection. With the binuclear Rh-Ru compounds shown above, the rhodium unit was bound directly to TiO₂ and the chromophoric ruthenium donor was fixed away from the semiconductor (Scheme 8).⁹¹ In fluid solution, the Ru MLCT excited state is quenched by intramolecular electron transfer to the Rh unit. Back electron transfer to Ru(III) was rapid, and only a short-lived MLCT excited state was observed spectroscopically. On TiO₂, it was found that the Rh acceptor levels lie between the TiO₂ conduction band edge and the MLCT excited state. Light excitation thus resulted in an unprecedented electron "hopping" from the Ru to Rh to the semiconductor nanocrystallite. Under the experimental conditions studied, about 40% of the electrons that arrive at Rh were found to transfer electrons to TiO₂, the rest recombined to Ru(III). This branching ratio presumably reflects different orientations of the compound on the TiO₂ surface. When employed in regenerative solar cells, the photocurrent efficiency was rather low, mainly because of low charge injection yields. Nevertheless, the results suggest a general strategy to slow recombination between the injected electron and oxidized sensitizer.

Scheme 8

2. Hole Hopping. Intramolecular "hole" transfer has been used to regenerate the ground state of the Ru(II) chromophore at sensitized TiO₂ interfaces.^92-95 The first compound reported to perform this function was Ru(dcb)₂(4-CH₃-4'-CH₂-PTZ-2,2'-bipyridine)²⁺, where PTZ is the electron donor phenothiazine, and is shown in Scheme 9.⁹¹ In fluid methanol solution, electron transfer from the PTZ group was moderately exergonic (<0.25 eV) and had an approximate rate constant of ~2.5 × 10⁸ s^-1. The corresponding charge recombination step was faster than the forward one, like the Ru-Rh compounds, so no appreciable transient accumulation of the electron-transfer product was observed.

Scheme 9

When attached to TiO₂, MLCT excitation resulted in a new charge-separated state with an electron in TiO₂ and an oxidized PTZ group, abbreviated PTZ⁺-Ru/TiO₂(e^-). Mechanistically, there were two possible electron-transfer pathways available to reach this charge-separated state. In the first, charge injection was followed by oxidation of the PTZ donor by the Ru(III) center, PTZ-Ru^II*/TiO₂ PTZ-Ru^III/TiO₂(e^-) PTZ⁺-Ru/TiO₂(e^-). In an alternative pathway, reductive quenching by the PTZ group is followed by charge injection into the semiconductor, TiO₂-Ru^II*-PTZ TiO₂-Ru^II(dcb^-)⁺-PTZ⁺ PTZ⁺-Ru/TiO₂(e^-).

Nanosecond transient absorption revealed the appearance of the PTZ⁺-Ru/TiO₂(e^-) interfacial charge-separated state.⁹² It was not clear whether this state was formed by interfacial electron transfer from the excited or reduced state. However, as previously discussed, excited-state electron injection into TiO₂ often occurs on a femtosecond time scale under similar conditions, so the excited-state pathway seemed most probable under the experimental conditions employed. After injection, electron transfer from PTZ to the Ru(III) center (-G ~ 0.36 eV) produces the charge-separated state PTZ⁺-Ru/TiO₂(e^-). Recombination of the electron in TiO₂ with the oxidized PTZ to yield the ground state occurred with a rate constant of 3.6 × 10³ s^-1. Excitation of a model compound that did not contain the PTZ donor, Ru(dcb)₂(dmb)₂²⁺, where dmb is 4,4'-(CH₃)₂-bpy, under otherwise identical conditions gave rise to the immediate formation of a charge-separated state, TiO₂(e^-)-Ru^III, that recombined with an average rate constant of 3.9 × 10⁶ s^-1. Therefore, translating the "hole" from the Ru center to the pendant PTZ moiety inhibits recombination rate constants by about 3 orders of magnitude. Note that, at the time of these studies, we had not yet discovered the second-order nature of charge recombination.⁶⁴ The time-resolved data were fit to a sum of exponentials, and an average rate constant was reported. Because these measurements were made under identical conditions of surface coverage and laser irradiance, we feel that these internal comparisons are still meaningful.

The Ru-PTZ and model compounds were tested in regenerative solar cells, with iodide as an electron donor. The photocurrents were within experimental error the same for the two sensitizers. However, the open-circuit photovoltage, V_oc, was observed to be about 100 mV larger for Ru-PTZ/TiO₂. The enhancement was even more pronounced in the absence of iodide, 180 mV larger over 5 decades of irradiance. The diode equation (eq 8) predicts a 59 mV

The bimetallic sensitizer [Ru(dcb)₂(Cl)-bpa-Os(bpy)₂(Cl)](PF₆)₂, abbreviated Ru-bpa-Os, where bpa is 1,2-bis(4-pyridyl)ethane, was anchored to TiO₂ for interfacial electron-transfer studies (Scheme 10).²¹ Pulsed light excitation of a Os-bpa-Ru/TiO₂ material immersed in a 1.0 M LiClO₄/acetonitrile bath at 25 C results in rapid interfacial electron transfer and intramolecular electron transfer (Os(II) Ru(III)) to ultimately form an interfacial charge-separated state with an electron in TiO₂ and an oxidized Os(III) center, abbreviated Os^III-bpa-Ru/TiO₂(e^-).⁹⁴ This same state could also be generated after selective excitation of the Os(II) group with red light. The rates of intramolecular and interfacial electron transfer are fast, k > 10⁸ s^-1, while interfacial charge recombination, Os^III-bpa-Ru/TiO₂(e^-) Os^II-bpa-Ru/TiO₂, required milliseconds for completion. The results here show a general strategy for promoting rapid intramolecular electron transfer (Os(II) Ru(III)) after interfacial electron injection and a "remote" electron injection process that occurs after direct excitation of the Os(II) chromophore. Unfortunately, there was no evidence for an enhanced lifetime of the charge-separated state, presumably because the Os(III) center was proximate to the semiconductor surface. Grätzel and co-workers have reported related studies with supramolecular sensitizers and have emphasized their potential application in photochromic devices.⁹⁵ Interestingly, these workers found long-lived charge separation, like that described for the PTZ-Ru^II/TiO₂ system above, in some cases while not in others. Presumably, driving force and semiconductor-molecular orientations are two key parameters. Additional experiments have been reported in this area and more are required before this interesting interfacial behavior can be fully understood.

Scheme 10

III. Conclusion

It has been over 25 years since inorganic chemists first directly observed photoinduced charge separation.⁷ These studies have progressed and naturally evolved toward solid-state materials that are more applicable to solar energy conversion in the real world. The mechanisms for interfacial charge separation and recombination are now understood in considerable molecular detail, and it is now relatively easy to generate interfacial charge-separated states that store >1.5 eV of free energy for time periods of milliseconds. Aside from electrical power generation in regenerative solar cells, a significant future challenge is to intercept these states to drive reactions that produce useful fuels. The mesoporous nanocrystalline thin films that are under active investigation provide exciting new opportunities for fundamental research in this area. The ability to trap and store a high density of redox equivalents in a semiconductor material offer the real possibility for multielectron-transfer photocatalysis relevant to water splitting and environmental remediation. Future prospects for solar energy conversion with coordination compounds anchored to semiconductor surfaces appear to be very bright.

Acknowledgment

The Division of Chemical Sciences, Office of Basic Energy Sciences, Office of Energy Research, U.S. Department of Energy, is gratefully acknowledged for research support. We thank the National Science Foundation for support of the environmental chemistry aspects of this work.