Solar Water Splitting Using Semiconductor Photocatalyst Powders

Part of the Topics in Current Chemistry book series (TOPCURRCHEM, volume 371)


Solar energy conversion is essential to address the gap between energy production and increasing demand. Large scale energy generation from solar energy can only be achieved through equally large scale collection of the solar spectrum. Overall water splitting using heterogeneous photocatalysts with a single semiconductor enables the direct generation of H2 from photoreactors and is one of the most economical technologies for large-scale production of solar fuels. Efficient photocatalyst materials are essential to make this process feasible for future technologies. To achieve efficient photocatalysis for overall water splitting, all of the parameters involved at different time scales should be improved because the overall efficiency is obtained by the multiplication of all these fundamental efficiencies. Accumulation of knowledge ranging from solid-state physics to electrochemistry and a multidisciplinary approach to conduct various measurements are inevitable to be able to understand photocatalysis fully and to improve its efficiency.


Electrocatalysis Hydrogen Overall water splitting Photocatalysis Semiconductor 

1 Introduction

Solar energy is by far the most abundant renewable energy resource [1]. The total solar energy absorbed by the Earth is 3.85 × 1024 J year−1, which is ~104 greater than the world energy consumption [2]. To compensate for increasing energy demand, solar energy conversion has to be at least partially implemented. To achieve extensive solar energy conversion, the large-scale collection of solar flux is essential. A simple calculation using the air-mass 1.5 global (AM 1.5G) (~1 kW m−2) solar spectrum reported by the National Renewable Energy Laboratory (NREL) predicts that a collection area of ~750,000 km2 is required to meet global energy demands [3]. Therefore, solar energy conversion technologies must have tremendous scalability, irrespective of the conversion method used. For easy storage and transport, producing chemicals and fuels in which the energy is in the form of chemical bonds is preferred. This technology requires water as the sole reactant and directly forms chemical energy (H2) in a single reactor. The reactor contains water and photocatalysts as powders. It does not require any complicated parabolic mirrors or electronic devices. Overall, the simplicity of this type of solar hydrogen production plant makes the system economically advantageous [4].

The photocatalytic system for overall water splitting produces a mixture of H2 and O2 followed by the separation of these products [5]. It is critical to develop highly efficient photocatalysts made from abundant elements using a mass-production process. The overall water splitting reaction is
$$ {\mathrm{H}}_2\mathrm{O}\to {\mathrm{H}}_2+\frac{1}{2}{\mathrm{O}}_2 \Delta G=237\ \mathrm{kJ}\ {\mathrm{mol}}^{-1}. $$
The above reaction (two-electron reaction in this stoichiometry) has a Gibbs free energy of 237 kJ mol−1. The photon energy, E, is expressed by
$$ E=hv=\frac{hc}{\lambda }, $$
where h is Planck’s constant, v is the frequency of the photon, c is the speed of light, and λ is the wavelength of the photon. Thus, a photon energy of 1.23 eV is thermodynamically required to drive overall water splitting, which is equivalent to a wavelength of ~1,000 nm. A photocatalyst for water splitting thus requires a bandgap greater than 1.23 eV. It is also essential to consider the overpotential, i.e., the excess potential beyond the thermodynamic potential required to overcome the activation energy.
As far as solar energy conversion is concerned, analysis of the solar spectrum is essential to determine the important requirements for photocatalyst materials for overall water splitting. Figure 1 shows the solar to hydrogen (STH) energy and the number of photons as a function of wavelength according to the data from the standard AM 1.5G spectrum [3]. The STH efficiency can be calculated using the Gibbs free energy of reaction (1):
$$ \mathrm{S}\mathrm{T}\mathrm{H}=\frac{\mathrm{Output}\ \mathrm{energy}}{\mathrm{Energy}\ \mathrm{of}\ \mathrm{incidence}\ \mathrm{solar}\ \mathrm{light}}=\frac{r_{{\mathrm{H}}_2}\times \Delta G}{P_{\mathrm{Sun}}\times {A}_{\mathrm{Geometric}}}, $$
where \( {r}_{{\mathrm{H}}_2} \) is the hydrogen production rate, PSun is the energy flux of sunlight, and AGeometric is the area of the reactor. The solar energy spectrum (PSun = 1,003 W m−2) has ~93 W m−2 in the ultraviolet (UV) region (λ ≤ 400 nm; 9.3%), ~543 W m−2 in the visible region (400 nm < λ ≤ 800 nm; 54.1%), and 367 W m−2 in the infrared (IR) region (λ > 800 nm; 36.6%). For overall water splitting, hydrogen is the sole product corresponding to an energy of 1.23 eV (or 237 kJ mol−1) equivalent. The theoretically attainable STH efficiency and hydrogen production rate can thus be calculated from the number of photons in the spectrum from shortest wavelength to the respective wavelength at different quantum efficiencies (QE), e.g., 30%, 60%, or 100%:
Fig. 1

Photon number of AM 1.5G as a function of wavelength and theoretical for solar-to-hydrogen efficiency integrated from a low wavelength to the respective wavelength at QEs of 30%, 60%, and 100%

$$ \mathrm{Q}\mathrm{E}(hv)=\frac{2\times {r}_{{\mathrm{H}}_2}}{I_0(hv)}. $$

By analyzing the solar irradiance, the theoretical maximum STH efficiency can be calculated to be ~48% (at 100% QE), which is integrated from the UV to a wavelength of ~1,000 nm (1.23 eV). Based on the definition of STH efficiency, the excess energy from photons greater than 1.23 eV has to be dissipated (mainly as heat). The energy loss becomes more apparent as the wavelength decreases. For UV photons, over half the energy is dissipated (only 1.23 eV can be utilized to generate H2) and consequently the theoretical maximum STH efficiency using only UV light (λ ≤ 400 nm) is only 3.3% (for a QE of 100% at each wavelength). In contrast, the target of STH efficiency is generally set to 10% to become competitive within the hydrogen market. Although the technology is simple, the low STH efficiency requires a large area for the solar reactor. This fact clearly shows why the development of a visible-light-responsive photocatalyst is essential for a high STH process.

As mentioned previously, the benchmark for the STH efficiency is set to 10%. This efficiency corresponds to a hydrogen production rate of ~154 μmol H2 cm−2 h−1 and a photoelectrochemical current of ~8.3 mA cm−2, i.e., a consumption rate of ~260 photons nm−2 s−1 on a flat surface. These values are practically useful for experiments in a laboratory scale. To achieve this efficiency, the development of materials that absorb wavelengths of light up to 600–700 nm (~1.8–2.0 eV) is essential. The challenge is that these materials have to have suitable band positions for water splitting, as discussed below.

The photocatalytic reactions involve various photophysical and electrocatalytic processes on different time scales. Figure 2 shows a general scheme for the photon-induced reaction process for overall water splitting using a solid photocatalyst. Photon absorption initiates non-equilibrium photophysical and photochemical processes. These processes begin with the generation of an exciton, i.e., excitation of an electron in the valence band (VB) or the highest occupied molecular orbital (HOMO) to the conduction band (CB) or the lowest unoccupied molecular orbital (LUMO) [6]. This femtosecond process is followed by relaxation of the electron and the hole to the bottom of the CB and to the top of the VB, respectively, on a similar time scale [6]. Next, the exciton (electron–hole pair) needs to be separated depending on the nature of the semiconductor. The electronic structure should guide the excited electron and hole (polaron) to move independently, fully utilizing the junctions at the semiconductor–catalyst and semiconductor–solution interfaces. Successful charge transfer to the surface and to the cocatalyst is followed by electrocatalytic redox reactions on a time scale longer than microseconds [6]. Because each photon in visible light possesses limited overpotential for water splitting, the presence of such electrocatalysts on the surface of semiconductor is essential.
Fig. 2

Scheme of photon-induced reaction process for overall water splitting by solid photocatalyst

The primary aim of this review is to focus on discussion of the fundamental parameters involved in photocatalytic overall water splitting. The processes involved in photocatalytic reactions, especially water splitting with powder photocatalysts suspended in the liquid phase, are of interest. The steps involved in the photocatalysis for water splitting are divided into the following six processes and the relevant parameters are described:
  1. 1.

    Photon absorption

  2. 2.

    Exciton separation

  3. 3.

    Carrier diffusion

  4. 4.

    Carrier transport

  5. 5.

    Catalytic efficiency

  6. 6.

    Mass transfer of reactants

Figure 3 summarizes the parameters that are highlighted according to the six processes. This review gives a broad and rather conceptual description only, rather than detailed discussion on the specific materials, which can be found in different reviews [7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26]. The author trusts that the understanding of these physicochemical properties leads to “photocatalyst materials by design” where the electronic structure of the semiconductor, interface development, and electrocatalytic properties are fully connected to achieve the complex sequential processes to achieve finally overall water splitting.
Fig. 3

Parameters requiring attention for efficient overall water splitting. Overall water splitting is only successful for high efficiencies at all six gears depicted in the scheme. Different time scales of the reactions are also displayed

2 Photocatalytic Processes

2.1 Photon Absorption and Exciton Generation

Photon absorption by semiconductors generally occurs via the excitation of electrons in the valence band into the conduction band, generating excitons (excited electron and hole pairs). At the macroscopic level, photon absorption by powder suspension systems includes intrinsic and extrinsic absorptions in addition to scattering, reflection, and transmission. A schematic image of the light absorption process by a semiconductor powder is shown in Fig. 4.
Fig. 4

Scheme of photon absorption process using semiconductor powder. See main text for explanation of the notations

A spherical particle with a diameter of 100 nm is irradiated by nearly 107 photons (considering wavelengths <600 nm in the AM 1.5G spectrum) every second [27]. Among those photons, some of them experience scattering by the powder [28]. Scattering can be categorized into Rayleigh and Mie scattering, depending on the size of the particle. The size of a scattering particle is defined by the ratio of its characteristic dimension to the wavelength of the scattered light:
$$ x=\frac{2\pi r}{\lambda }. $$
Scatter diameters much smaller than the wavelength result in Rayleigh scattering, whereas larger diameters result in Mie scattering. Based on the sizes of the photocatalyst particles (a few nanometers to 10 μm), the photons in the UV–vis range with wavelengths of 300–800 nm should be appropriately taken into consideration. The scattering coefficient for Rayleigh scattering varies for small particles inversely with the fourth power of the wavelength (~λ−4), giving higher scattering for shorter wavelengths (i.e., UV > VIS > NIR).

Similarly, some of the photons undergo reflection and transmission through media with distinct refractive indices [29]. The description of light propagation in a complex photocatalyst system may be treated using the Fresnel equations. The reflection of light retains either the geometry or the energy of the incident light. Diffuse reflection can also occur, especially on a rough surface, which retains the energy and is indeed utilized to elucidate the absorption characteristics of the powder semiconductor.

In a photoreactor consisting of a semiconductor powder in suspension configuration, the scattered and reflected light can be further absorbed by other semiconductor particles. This is indeed beneficial for suspension systems compared with film configurations. It is very difficult to measure quantitatively the extent of the scattering and reflection in experiments, so the photocatalytic efficiency is generally associated with the incident photons (which can be separately measured) to the reactor or measured actinometrically [30].

The quantitative measurements of the absorption efficiencies of the photocatalytic powders provide useful information for the development of efficient photocatalyst materials and systems. The absorption spectra indicate the consequences of bandgap excitation, d-d transitions, phonon absorptions, and excitations associated with defect states [31]. The absorption coefficient, α (wavelength dependent), is an important parameter of semiconductors that provides their intrinsic characteristics. Ideally, absorption measurements using photon flux of incident light, Φ0, in transmission and reflectance mode (using an integrating sphere) commonly lead to the following relationships with absorptance (A%), and light that is transmitted (T), specularly reflected (Rs), forward-scattered (S), and back-scattered (Rd) [31]:
$$ {\Phi}_0={A}_{\%}+T+{R}_{\mathrm{S}}+S+{R}_{\mathrm{d}}. $$
The efficiency of the photon absorption process occurring within a sample is thus given by the absorptance A%, which is the fraction of photons absorbed out of all of the photons impinging on the sample. Using the resulting value for absorptance, the absorbance, A, can be calculated using the following relationship:
$$ {A}_{\%}=1-{10}^{-A}. $$
Furthermore, the absorbance can be normalized by the thickness to obtain the absorption coefficient, α, as
$$ \alpha\;\left({\mathrm{cm}}^{-1}\right)=\frac{ \ln (10)\times A}{l\;\left(\mathrm{cm}\right)}, $$
where l is the path length of the light through the sample. The light intensity, I, in the unit of power, as a function of distance, d, is given by
$$ I(d)={\displaystyle \sum^{hv}{\Phi}_0(hv){e}^{-\alpha d}}. $$
The absorption coefficient, α, is preferably determined using absorptance, A%, which gives absorbance, A, though the contribution of scattering is excluded whenever possible. For a practical absorptance measurement, the thin film configuration of semiconductors provides a more precise description, compared to powder configurations, once the film thickness is more rigorously defined [32]. Different semiconductor synthesis techniques for different configurations are reviewed elsewhere [22]. For accurate measurement of the absorption coefficient, the absorption coefficient is linked with the measured transmittance (T) and reflectance (R) [32]:
$$ T=\frac{{\left(1-R\hbox{'}\right)}^2{e}^{-\alpha d}}{1-{R^{\prime}}^2{e}^{-2\alpha d}}, $$
$$ R=R\hbox{'}\left(1+\frac{{\left(1-R\hbox{'}\right)}^2{e}^{-2\alpha d}}{1-{R^{\prime}}^2{e}^{-2\alpha d}}\right), $$
where R′ is the single surface reflectance of the material. These equations lead to determination of the absorption coefficient as a function of measured T and R as follows [33]:
$$ \alpha =\frac{1}{d} \ln \left(\frac{{\left(1-R\right)}^2}{2T}+\sqrt{\left(\frac{{\left(1-R\right)}^4}{4{T}^2}+{R}^2\right)}\right), $$
and, when R is negligible and/or the following inequality is satisfied [34],
$$ R{e}^{-\alpha d}\ll 1, $$
then (12) is simplified to
$$ T={\left(1-R\right)}^2{e}^{-\alpha d}, $$
and, thus,
$$ \alpha =-\frac{1}{d} \ln \frac{T}{{\left(1-R\right)}^2}. $$
From the same spectra, analyzing interference fringes in a weaker absorption range would provide useful information. If the thickness d is uniform, the interference fringes can be used to obtain the position of the interference maxima λ [35]:
$$ 2nd=m\lambda, $$
where n is the refractive index and m is the order of the interference. When m is not known, the adjacent maxima can be used to eliminate m:
$$ {(2nd)}^{-1}={\lambda}_{m+1}^{-1}-{\lambda}_m^{-1}. $$
A complex refractive index nc is described as
$$ {n}_{\mathrm{c}}=n+i\kappa, $$
where κ is the extinction coefficient. Once α(λ) is known, κ(λ) can be calculated from the equation
$$ \kappa \left(\lambda \right)=\frac{\alpha \left(\lambda \right)}{4\pi }. $$
Then the dielectric constant, a measure of the charge retention capacity of a medium, can be obtained. The complex dielectric constant, εr, is described as [36]
$$ {\varepsilon}_{\mathrm{r}}={\varepsilon}_1+i{\varepsilon}_2, $$
where ε1 is the real part of the dielectric constant associated with dispersion:
$$ {\varepsilon}_1={n}^2-{\kappa}^2, $$
and ε2 is dielectric loss, i.e., the imaginary part associated with the dissipative rate of the wave:
$$ {\varepsilon}_2=2n\kappa . $$
These values lead to the dielectric loss tangent, tan δ = ε2/ε1, defined by the angle between the capacitor’s impedance vector and the negative reactive axis, providing an important parameter to quantify the inherent dissipation of electromagnetic energy into heat. The dielectric constant, εr, can be divided into the contributions from the electronic density, \( {\varepsilon}_{\infty } \) and from the motion of ions constituting the material, εvib [37]:
$$ {\varepsilon}_{\mathrm{r}}={\varepsilon}_{\infty }+{\varepsilon}_{\mathrm{vib}}={\varepsilon}_{\infty }+\frac{4\pi }{V}{\displaystyle \sum_p\frac{Z_p^2}{v_p^2}}, $$
where v p is the phonon frequency of mode p, V is the unit cell volume, and Z p is the mass-weighted mode effective born vector, which is proportional to IR absorbance, I p .

Once the absorption coefficient is obtained, the absorption depth, which describes how far light can penetrate into a material before being absorbed, can be determined by simply taking the inverse of the absorption coefficient α (on a natural logarithm scale). These penetration depths are generally defined as the depths where the light intensity decreases by factors of 1/e (~36%) of the incident intensity. The absorption coefficient and the absorption depth vary with the incident light wavelength: typical density of states of the semiconductor suggests that shorter wavelength light is absorbed closer to the photocatalyst surface than longer wavelength light. Thus, before being absorbed, visible light travels further in a photocatalyst than UV light. Even if a long-wavelength photon is absorbed, it penetrates deep into the bulk of the photocatalyst, so the generated excitons have a long distance to travel to the surface (Fig. 4). Thus, there is a greater probability of excited electrons and holes recombining before they can participate in surface redox reactions. The absorption depth (together with scattering and reflection) also accounts for how many semiconductor particulates are required to absorb effectively the incident light in the photoreactor (the depth of the photoreactor).

The electronic structure of the semiconductor determines various critical characteristics relevant to photocatalysis. Regarding the absorption properties, the electronic structure of the semiconductor not only decides the bandgap and band positions but also the absorption coefficient and direct/indirect nature of the light absorption. The absorption parameters of a given semiconductor primarily depend on its crystal structure, which in turn determines its electronic structure. Essentially, direct bandgaps lead to high absorption coefficients, whereas indirect bandgaps give low absorption coefficients. The symmetry of the crystal structure gives allowed and forbidden energies. Indirect transitions involve both a photon and a phonon because the band edges of the conduction and valence bands are widely separated in k space, as described in Fig. 5. A typical indirect bandgap semiconductor, Si, possesses a typically low absorption coefficient of 1 × 103 to 1 × 105 cm−1, giving absorption depths of up to a few micrometers for visible light (400–800 nm) [38]. Therefore, if Si nanoparticles (less than a micron in diameter) are synthesized for photocatalysis, there is a significant chance that a single photocatalyst particle may not absorb many photons per unit time. Typical absorption coefficients of direct bandgap semiconductors typically fall into the range of 1 × 104 to 1 × 106 cm−1, equivalent to absorption depths of 1,000–10 nm. This range of absorption depths becomes comparable to the typical particle size of the photocatalysts. Therefore, it is important to consider the absorption coefficient to determine the efficiency of the photocatalytic materials and photocatalyst design.
Fig. 5

Electronic structures for (left) direct and (right) indirect bandgap semiconductors and their excitations

For powder semiconductors it is often difficult to address the true absorption coefficient. Alternatively, the measurement of diffuse reflectance is a powerful tool to acquire information about light absorption [29]. Because of the anisotropic nature of the powder system, the Kubelka–Munk function f(R) is frequently used to describe absorption/reflectance:
$$ f(R)=\frac{{\left(1-R\right)}^2}{2R}=\frac{\alpha }{s}, $$
where s is the scattering coefficient. The Kubelka–Munk model is based on the following assumptions [39]:
  • The sample is modeled as a plane layer of finite thickness but with infinite sheet approximation, so there are no boundary effects

  • A perfectly diffuse and homogenous illumination is incident on the surface

  • The only interactions of light with the medium are scattering and absorption; polarization and spontaneous emission (fluorescence) are ignored

  • The sample is considered isotropic and homogeneous and contains optical heterogeneities

  • No external or internal surface reflections occur

  • The scattering coefficient, s, is constant for any layer thickness

Because of the lack of an s value, it is difficult to pin down the absolute value of α. Assuming that the scattering coefficient is independent of wavelength, f(R) is then proportional to α. The Kubelka–Munk function is, therefore, effective for addressing the bandgap measurement.

The absorption coefficients measured in thin films, or through the Kubelka–Munk function for powdered samples, can be used to obtain the bandgap, Eg, of the semiconductor. When α > 104 cm−1, it often obeys the following relationship presented by Tauc and supported by Davis and Mott [40, 41]:
$$ \alpha hv\propto {\left(hv-{E}_{\mathrm{g}}\right)}^{\frac{1}{n}}, $$
where n can take values of 3, 2, 3/2, or 1/2, corresponding to indirect (forbidden), indirect (allowed), direct (forbidden), and direct (allowed) transitions, respectively [31]. Tauc plots, i.e., \( {\left(\alpha hv\right)}^{\frac{1}{n}} \) as a function of hv (n is the same as above), give Eg from the intersection of a tangent to the slope in the linear region of the absorption onset with the baseline. A more detailed description of how to utilize accurately the information of Tauc plots can be found in [31].
The generation rate, G, per photocatalyst particle is the number of electron–hole pairs generated per photon striking the particle with the depth, x. The assumption that each photon directly causes generation of an electron–hole pair can be first described using the Beer–Lambert law approximation [42]:
$$ G=\alpha {\Phi}_{0,p}{e}^{-\alpha x}, $$
where Φ0,p is the photon flux at the particle surface (photons (particle geometric area)−1 s−1). The photon flux density may be directly obtained from the irradiance measurements. As previously mentioned in (9), the light intensity, I, decays as it passes through the semiconductor materials because of light absorption and creates the generation of electron–hole pairs according to [43]
$$ \frac{dI(x)}{dx}=-\alpha I(x)=-{\displaystyle \sum^{hv}G(hv)}. $$

Recent advances in density functional theory (DFT) calculations give quite accurate and reliable estimates for the electronic structures of semiconductors from a given crystal structure, resulting in accurate descriptions of the densities of states (DOS). Essentially, the accurate DOS provides information about the bandgap energy, the band positions (vs vacuum), the connectivity of both the CB and VB within the crystal structure, and the absorption coefficient as a function of wavelength. As discussed later, DFT calculations are now able to describe fundamental parameters for semiconductors with high accuracy, such as the effective mass, exciton binding energy, and dielectric constant, etc. The effective depth (thickness) of the powder suspension can be theoretically estimated from the absorption depth, which is significant information because the concentration (amount) of the photocatalysts per given area and volume can be determined for a given semiconductor.

2.2 Exciton Separation

Upon absorption of light, excitons (electron–hole pairs) are generated [44]. For photocatalytic processes, the subsequent step includes exciton separation to generate excited electrons and holes (free carriers). When the dielectric screening potential and the exciton radius are large, the excitons are Mott–Wannier type, which are typical for bulk semiconductors. If the exciton radius is small, the exciton is Frenkel type, typical for the dielectric characteristics of molecules and organic polymers. Characteristics of these two types of exciton generation are summarized in Fig. 6.
Fig. 6

Frenkel and Mott–Wannier exciton models and their characteristics

The practical measurement of exciton binding energy involves photoemission spectroscopy, optical absorption spectroscopy, photoconductivity screening potential, spectroscopy, and magneto-optical spectroscopy [45]. Using a single dielectric constant εr and Bohr’s quantum theory, the energy En of the exciton series for the Mott–Wannier type can be derived from [46]
$$ {E}_{\mathrm{n}}={E}_{\infty }-\frac{R_{\mathrm{ex}}}{n^2}, $$
where \( {E}_{\infty } \) is the series limit (a constant) and n is a quantum number n = 1, 2, …. Thus, Rex is the exciton binding energy, which represents the energy required to ionize an exciton in its lowest energy state, i.e., the energy separation between the lowest bound state (n = 1) and the series limit, and is much smaller than that for a hydrogen atom [47]. The 1 s state of the exciton is the exciton binding energy, Rex, and can be described as
$$ {R}_{\mathrm{ex}}=\frac{m^{\ast }{e}^4}{2{h}^2{\varepsilon}_{\mathrm{r}}^2}={E}_1\frac{m^{\ast }}{\varepsilon_{\mathrm{r}}^2}, $$
where m* is the reduced effective mass of the electron–hole system (\( \frac{1}{m^{*}}=\frac{1}{m_{\mathrm{n}}^{*}}+\frac{1}{m_{\mathrm{p}}^{*}} \)), e is the elemental charge, and h is Planck’s constant. From the equation, it is observed that the binding energy becomes small if the dielectric constant of the semiconductor is high. For Mott–Wannier excitons, the typical binding energy is less than 10 meV and the radius is ~10 nm. For Frenkel excitons, these values can be as high as 1 eV and ~1 nm. The energy should be compared with the thermal energy (25 meV at room temperature) [37], and efficient separation of excitons requires the binding energy to be lower than this value. For the high binding energies in Frenkel-type excitons, heterojunctions at the molecular levels are essential to separate charges and are prevalent in organic polymer photovoltaic cells. A high delocalization of the charge carriers occurs because of the low effective masses and large collision times associated with defects.

The anisotropic nature of exciton binding energy is also critical. The crystal and electronic structures determine the anisotropic nature in the different crystal orientations. These differences are currently possible to predict using DFT calculations, and it is therefore an effective method for this investigation because of its high accuracy. Inoue also discussed the correlation between distorted metal-oxygen crystal structures and photocatalytic activity; high distortion creates an anisotropic electronic field upon exciton generation which assists in separation [17].

2.3 Carrier Diffusion and Recombination

Once the exciton is separated, free charge carriers have to transfer to the surfaces for successful photocatalysis [48]. For the electrons and holes, the ability to move around in a material and transport charge is called mobility (electron mobility and hole mobility, respectively). The diffusion coefficient, D, and the mobility of the charge carrier, μ, are connected through the Einstein relations [37]:
$$ D=\frac{k_{\mathrm{B}}T}{e}\mu, $$
where kB is the Boltzmann constant and e is the elemental charge. Thus, the mobility (in a specific direction) can be obtained as
$$ \mu =e\frac{\tau_{\mathrm{c}}}{m*}, $$
where τc is the collision time of the charge carrier and m* is the effective mass. Now the flow of electrons and holes are considered current in a given direction, e.g., the x-direction. There are two driving forces for the carrier movement: the concentration gradient of the carriers (change in concentration) and external electric fields [44]. When a concentration gradient of the carriers is present, they distribute by themselves from regions of high concentration to regions of low concentration only through thermal motion. The movement of charge results in a so-called diffusion current. In a gradient of electrons or holes, \( \frac{d{p}_x}{dx} \) and \( \frac{d{n}_x}{dx} \) are non-zero, generating diffusion currents described with diffusion coefficients, Dp and Dn, and charges, +q and −q, according to Fick’s law (p for holes, n for electrons, respectively). When an electric field is present, the potential gradient also causes a drift current. The drift and diffusion currents make up the total current in a semiconductor. The total current density (for one vector) is driven by the carrier gradients and the potential gradients:
$$ \begin{array}{l}{\overrightarrow{J}}_{\mathrm{total}}={\overrightarrow{J}}_{\mathrm{p},\mathrm{diffusion}}+{\overrightarrow{J}}_{\mathrm{n},\mathrm{diffusion}}+{\overrightarrow{J}}_{\mathrm{p},\mathrm{drift}}+{\overrightarrow{J}}_{\mathrm{n},\mathrm{drift}}\\ {} =-q{D}_{\mathrm{p}}\frac{d{p}_x}{dx}+q{D}_{\mathrm{n}}\frac{d{n}_x}{dx}+pq{\mu}_{\mathrm{p}}{\overrightarrow{E}}_x+nq{\mu}_{\mathrm{n}}{\overrightarrow{E}}_x.\end{array} $$
Under equilibrium conditions, the total current density should be zero. In powder semiconductor systems there are no external electric fields by choice. Hence the carrier mobility of the photocatalyst must be high because it strongly relies on the diffusion process of the excited carriers. However, if the doping or electronic structure is not uniform in terms of potential, there is a concentration gradient which can create an electric field within the semiconductor and result in non-zero current densities, which may cause electrons and holes to transfer. The anisotropic nature of the electronic structure within a semiconductor is also of great importance. The anisotropy guides electrons and holes to move to different crystal orientations, which allows them to avoid recombination. The gradient is even facilitated by junctions intentionally added between the metal, semiconductor, and electrolyte. This result is discussed in the following section.

Carrier mobility is one of the most important parameters that determines overall photocatalytic efficiency. The mobility is affected by temperature, doping concentration, and the magnitude of the applied field. It also depends on the effective masses of the electrons and holes, which are determined primarily by the electronic structure of the semiconductor. The carriers with small effective masses have large mobilities. As a result, the holes are significantly less mobile than the electrons. The doping concentration also has a significant influence on the mobility. When the doping concentration is low, the mobility can be considered independent of the doping concentration. When the concentration of dopants becomes high, the mobility of the carriers decreases monotonically. These factors are closely associated with the recombination process. Practically, the resistivity, charge carrier concentration, and resultant mobility of the semiconductors can be measured by the van der Pauw technique with the Hall measurement [49, 50], although this method requires a high quality semiconductor slab.

The minority carrier lifetime, a consequence of the charge carrier concentrations and charge mobilities, is the intrinsic indicator of whether a semiconductor material is an effective photocatalyst. The minority carrier lifetime is the average time a typical minority carrier exists before recombination. In other words, the lifetime from indirect transitions is inversely proportional to the trap density. It is denoted τ n for electrons in a p-type material and τ p for holes in an n-type material. The minority carrier lifetime is measured optically or electronically using spectroscopic techniques [51]. One of the methods is using the simultaneous measurement of the light-induced photoconductance of the sample and the corresponding light intensity. The transient mode using a short light pulse used to be the common method, but the recent alternative method allows for measurement under quasi-steady state or quasi-transient illumination. For quasi-steady-state and quasi-transient carrier lifetime measurements, effective carrier lifetimes τeff can be given in the following generalized form [51]:
$$ {\tau}_{\mathrm{eff}}=\frac{\Delta n(t)}{G(t)-\frac{d\Delta n(t)}{dt}}, $$
where G is the generation rate and Δn is the time-dependent values of the excess carrier density.
Once the carrier lifetime is measured, the minority carrier diffusion length, denoted Lp for holes in an n-type material and Ln for electrons in a p-type material, represents the average distance the excess minority carrier travels from where it was generated to where it is annihilated:
$$ {L}_{\mathrm{n}}\equiv \sqrt{D_{\mathrm{n}}{\tau}_{\mathrm{n}}}, {L}_{\mathrm{p}}\equiv \sqrt{D_{\mathrm{p}}{\tau}_{\mathrm{p}}}. $$
Examples of the lifetime as a function of carrier concentrations for Si (indirect bandgap) are shown in Fig. 7. Law et al. show that the minority carrier lifetime can be fitted numerically with the following equation [52]:
Fig. 7

Hole (left) and electron (right) lifetimes in heavily doped n-type and p-type silicon, respectively. Copyright 1991, IEEE. Reprinted, with permission, from [52]

$$ \tau =\frac{\tau_{\mathrm{o}}}{1+{N}_{\mathrm{D}}/{N}_{\mathrm{ref}}+{\tau}_{\mathrm{o}}{C}_{\mathrm{A}}{N}_{\mathrm{D}}^2}. $$
The parameters used in the fit for n-type and p-type Si are (τo = 10 μs, Nref = 1 × 1017 cm−2, and CA = 1.8 × 10−31 cm6 s−1) and (τo = 30 μs, Nref = 1 × 1017 cm−2, and CA = 8.3 × 10−32 cm6 s−1), respectively. If the photocatalysis depends on the diffusion of excited carriers, the carrier lifetime and relevant diffusion length must be long for successful photocatalysis. This consideration is often associated with the semiconductor designed with “high crystallinity” and least defects [22].
The photocatalytic efficiency is decreased when the excitons or free carriers recombine: free carriers are no longer able to move because they are participating in covalent bonds in the crystal. The electron in the conduction band recombines by returning to the valence band whereas a hole in the valence band recombines when an electron annihilates it by falling from the conduction band. An electron can also be captured by a trap or recombination center. Recombination can be categorized into the following three types, as shown in Fig. 8:
Fig. 8

Types of recombination processes

  • Band-to-band recombination

  • Shockley-Read-Hall recombination (defects) [53, 54]

  • Auger recombination [55]

The types of recombination strongly depends on the electronic structure of the semiconductor. For direct bandgap semiconductors, band-to-band recombination is dominant, generating radiative processes such as luminescence. This recombination depends proportionally on the density of available electrons and holes. For indirect bandgap semiconductors, the band-to-band recombination is negligibly low, but recombination through defect levels, so-called Shockley–Read–Hall recombination, is dominant because the recombination is facilitated by the exchange of thermal energy with a phonon. Surface recombination occurs in a very similar manner. The surfaces and interfaces often act as trap sites because they contain impurities and abrupt terminations (the presence of dangling bonds which are electronically active). Auger recombination is a process in which an electron and a hole recombine in a band-to-band transition, but with the resulting energy transferred to another electron or hole. The involvement of a third particle affects the recombination rate so the Auger recombination has to be treated differently from band-to-band recombination.

2.4 Carrier Transport

For efficient photocatalysis, the free carriers should transfer to active sites according to the type of semiconductor. For intact semiconductors, the carriers have to transfer via diffusion. Thus, the efficiency is associated with the carrier lifetime. For improved charge separation, the interface (surface modification) is essential to creating new effective electronic structures [48]. The space charge region formed at the interface describes this unique charge distribution according to the junction created [56]. The key parameter to determine the space charge layer is the carrier density. For extensive charge separation, one of the driving forces of charge transfer is creating and utilizing band bending, which is strongly driven by the newly-created electronic structure at the interface. The interfaces involve metal–semiconductor, semiconductor–semiconductor, and semiconductor–electrolyte pairs [57]. The electronic structure should have a slope or bending for the free carriers to be transferred.

Where band bending is achieved, the potential description relative to the reference level is drawn in three-dimensional space. Figure 9 depicts the band bending that occurs when an n-type semiconductor is in contact with a solution [58]. In this case, the majority charges (electrons) near the interface of the semiconductor transfer into the solution until the potentials are equilibrated, i.e., the Fermi level of the semiconductor (EF,s) equals the redox potential (E0) in the solution while preserving the band edge positions (pinning). This transfer of the majority carrier creates a depletion layer. Band bending can be described by the Poisson equation as follows [44]:
$$ \frac{d^2\Psi (x)}{d{x}^2}=-\frac{e{N}_{\mathrm{D}}}{\varepsilon_0{\varepsilon}_{\mathrm{r}}}\left(0\le x\le W\right), $$
where ND is the majority carrier density, W is the depletion layer width, ε0 is the static permittivity in vacuum, and εr is the static dielectric constants of the semiconductor. Solving this for Ψ gives [44]
Fig. 9

Scheme of semiconductor–electrolyte junction and influence of particle size on depletion layer

$$ \Psi (x)=-\frac{e{N}_{\mathrm{D}}}{\varepsilon_0{\varepsilon}_{\mathrm{r}}}\left(Wx-\frac{1}{2}{x}^2\right)-{V}_{\mathrm{B}}. $$
When x = W,
$$ W={\left\{\frac{2{\varepsilon}_0{\varepsilon}_{\mathrm{r}}{V}_{\mathrm{B}}}{e{N}_D}\right\}}^{1/2}, $$
where e is the elementary charge and VB is the potential drop in the space-charge layer or the band bending that originates from the potential barrier height for the majority carriers to pass over on transferring from the semiconductor to the solution.
The flat-band potential, EFB, is the potential where band bending diminishes and the band becomes flat. The potential, EFB, is used to estimate the intrinsic band edge of the majority carriers, assuming that the surface states do not affect the surface potential. The absolute position of the majority band edge of the semiconductor can be estimated by determining the majority carrier concentration from Mott–Schottky measurements [58]. This potential is expressed as the flatband potential, EFB, where the band bending becomes zero. The space-charge capacity of the semiconductor, C, is given by the Mott–Schottky equation as follows [59]:
$$ \frac{1}{C^2}=\frac{2}{e{\varepsilon}_0{\varepsilon}_{\mathrm{r}}{A}^2{N}_{\mathrm{D}}}\left({E}_{\mathrm{app}}-{E}_{\mathrm{FB}}-\frac{kT}{e}\right), $$
where A is the surface area and Eapp is the applied potential. The capacity of the electrode with a target semiconductor is measured as a function of the applied potential. The Mott–Schottky plot, \( \frac{1}{C^2} \) vs the applied potential, can be extrapolated to \( \frac{1}{C^2}=0 \) to derive the flat-band potential EFB. The slope not only gives n-type or p-type information, but also an estimate of the majority carrier concentrations, ND, from knowing εr and A. Fabrication of high-quality electrodes using powder semiconductors is critical because the Mott–Schottky relation is only applicable to “ideal” semiconductors, where the bulk and surface are uniform with preferably known surface areas. The surface states also significantly affect the results of the plot because the pH effects follow the Nernstian relationship of −59 mV pH−1 for many semiconductors [60].
When metallic particles are deposited on the semiconductor surfaces, there is a new electronic structure at the metal–semiconductor interface [61]. The theory of metal–semiconductor interfaces has also been developed; the details of this theory can be also found in the work of Tung [62]. Simplified schematics for metal–semiconductor interfaces are shown in Fig. 10. The initial local electronic structure at the interface is strongly influenced by the relative positions between the work function of the metal and the Fermi level of the semiconductor. In short, there is the formation of a barrier, a so-called Schottky barrier, or of an energetically-smooth interface, a so-called Ohmic contact. The generalized criteria to generate such Ohmic contacts are ϕSC,n-type > ϕm and ϕm > ϕSC,p-type, and the opposite is applicable to Schottky junctions. Additionally, a high concentration of dopants may lead to Ohmic contact because of tunneling effects. The Schottky barrier height is a complex problem because it depends on the atomic structure of the metal–semiconductor interface. The reduction of the metal and its hydride formation also causes a change, namely a reduction in the barrier height [62]. For interface driven potentials, the difference between the work function, ϕm, and the flatband potential of the semiconductor, ϕSC, gives the diffusion voltage, VD:
Fig. 10

Schematic images showing different metal–semiconductor interfaces

$$ e{V}_{\mathrm{D}}=\left|{\phi}_{\mathrm{m}}-{\phi}_{\mathrm{SC}}\right|. $$
The maximum Schottky barrier height is described as the potential difference between the work function and the electron affinity of the semiconductor, χSC:
$$ {\phi}_{\mathrm{B}}=\left|{\phi}_{\mathrm{m}}-{\chi}_{\mathrm{SC}}\right| \left(\mathrm{Schottky}\ \mathrm{limit}\right) $$
Each surface atom has a dangling bond, forming new surface state. When such surface states or interface states pin the Fermi level of the system, independent of the work function of the metal, the barrier height becomes smaller than the Schottky limit (another limit, called the Bardeen limit) [63]. In reality, the barrier comes between the Schottky and Bardeen limits. Using empirical equations, the barrier can be described as
$$ {\phi}_{\mathrm{B}}=\mathrm{S}\left|{\phi}_{\mathrm{m}}-{\chi}_{\mathrm{SC}}\right|+\mathrm{A}{E}_{\mathrm{g}} $$
where S varies between 0 and 1, A is a constant, and Eg is the bandgap of the semiconductor. The previously discussed Schottky barrier model only describes semiconductor–metallic ideal contacts. The classic model fails to describe realistic porous ion-permeable electrocatalysts (i.e., oxyhydroxide cocatalysts for water oxidation). Recently, a model was proposed to describe so-called adaptive junctions and non-permeable metallic junctions (e.g., Ni(OH)2/NiOOH vs IrO x ) where the potential drop develops only on the semiconductor for the former case whereas the potential drop occurs both in the electrolyte and in the semiconductor for the latter case [64]. The ion-permeable adaptive junction has the main benefit of redox-active species that charge-up, creating adaptive barriers and increasing the apparent photovoltage generated at the interface. In contrast, an impermeable metallic buried junction creates a constant barrier–height interface [65, 66].
Figure 11 summarizes the transient potential shift caused by a single electron injected into a hemispherical or cubic metal particle. It assumes a constant capacitance of 20 μF cm−2 for metals, although the capacitance actually varies slightly with the potential and the facet [27]. As the particle becomes larger, a single electron generates a smaller potential shift. This small potential suggests that multiple electrons have to reach the particle within a reasonable time scale to cause a significant negative shift of the potential. The smaller particles require fewer electrons to shift the potential, but the contact area with the semiconductor also decreases, suggesting that there is an optimum metal particle size for the charge-up effect derived from excited electrons. This charge-up occurs up to the chemical potential of the electrons injected from the semiconductor (conduction band), and it is equilibrated at the catalyst on the surface.
Fig. 11

The transient potential shift caused by a single electron injected into a hemispherical or cubic metal particle assuming a constant capacitance of 20 μF cm−2

There are a few examples of direct measurements of the potential shifts on metal particles driven by light absorption by semiconductor particles [67, 68, 69, 70, 71]. The semiconductors exhibit shifts in the Fermi level under irradiation, which provides the driving force for redox reactions. The equilibrated potential of the metal particles under irradiation was more negative than the H2 evolution potential at a given pH, which is consistent with the steady-state H2 evolution observed. This shift depends on the light intensity, and this charge-up phenomenon plays an essential role in the photocatalysis. It is also possible to measure the potential of semiconductors under various conditions (electrolyte, solution) and applied potentials with and without irradiation when a semiconductor photocatalyst is employed as an electrode [72]. The rest potential measurement of a semiconductor should provide a good estimate of the potentials at equilibrium under dark conditions and steady-state light illumination. A lack of equilibration is often observed between the metal/semiconductor and the redox potential in the solution; this can be attributed to corrosion of the semiconductor, formation of a surface film (e.g., an oxide), or the inherently slow electron transfer across the interface [73], making the situation more complex. It is thus preferable to measure the potentials directly during the photocatalytic process.

The size of the photocatalyst affects a number of parameters. The smaller particles lead to a high specific surface area, a shorter travel distance for the charge carriers to the surface, a lower degree of band bending, and possibly a wider bandgap by quantum size effects [57]. Although the travel distances of the generated electrons and the holes to the surfaces are minimized by decreasing the size of the particles [16], a high surface area does not directly increase, and often even decreases, the photocatalytic activity [74]. The surface is a “defect” possessing dangling bonds and potential determining ions, which have different potential states and may serve as recombination sites for the excited carriers [57]. If the particle radius is smaller than the width of the space charge layer, the degree of depletion does not penetrate into the bulk, as shown in Fig. 9. Because of the small volumes of the particles, the number of carriers in one particle is very limited, leading to limited influence by the semiconductor–electrolyte interface [22]. Under such conditions, the barrier height ΔV at a distance l from the center of the spherical particles is given by the following equation [75]:
$$ \Delta V=1/6{\left(l/{L}_{\mathrm{p}}\right)}^2, $$
where Lp is the Debye length of the semiconductor. The lower degree of band bending leads to a higher probability that the photogenerated charge carriers transfer simply via diffusion.

The electrolyte is strongly influenced by the surface state and potential-determining ions on the surface [76]. In water, the isoelectric point of the semiconductor provides a useful indication of whether the surface is negatively or positively charged [77]. Therefore, it is an interesting approach to isolate the bare surface from the water electrolyte by using some oxide (e.g., SiO2, Al2O3, or TiO2) [78, 79, 80]. The oxide protective layer isolates the semiconductor surface from the electrolyte, thus avoiding the photocorrosion prevalent for some semiconductor compounds [78]. Knowing the majority carrier density and the size of the cocatalysts, their location (separation distance) needs to be selected appropriately.

2.5 Catalytic Efficiency

On the semiconductor surface, the photogenerated charge carriers need to be successfully consumed by an effective electrocatalytic process. To achieve efficient water splitting under visible light irradiation where there is no significant overpotential for electrocatalysis, the electrocatalysts need to transfer the received electrons and holes to the relevant reactants in the water splitting redox reactions. The efficiencies of these electrocatalytic processes can be measured separately. The water splitting reactions can be described by the following two half reactions:
$$ \begin{array}{l}2{\mathrm{H}}^{+}+2{e}^{-}\to {\mathrm{H}}_2\;\left(\mathrm{acid}\right) \mathrm{or} 2{\mathrm{H}}_2\mathrm{O}+2{e}^{-}\to {\mathrm{H}}_2+2{\mathrm{OH}}^{-}\;\left(\mathrm{base}\right)\\ {}E = 0\ \mathrm{V}\ \mathrm{v}\mathrm{s}.\ \mathrm{R}\mathrm{H}\mathrm{E}\end{array} $$
$$ \begin{array}{l}2{\mathrm{H}}_2\mathrm{O}\to {\mathrm{O}}_2+4{\mathrm{H}}^{+}+4{e}^{-}\;\left(\mathrm{acid}\right) \mathrm{or} 4{\mathrm{O}\mathrm{H}}^{-}\to {\mathrm{O}}_2+2{\mathrm{H}}_2\mathrm{O}+4{e}^{-}\;\left(\mathrm{base}\right)\\ {}E = 1.23\ \mathrm{V}\ \mathrm{v}\mathrm{s}.\ \mathrm{R}\mathrm{H}\mathrm{E}\end{array} $$
In general, hydronium ions (protons) are more easily reduced than water molecules [81], and hydroxyl ions are more easily oxidized than water molecules [82]. Therefore, extreme pH conditions are generally chosen for water electrolysis. The pH effects are further discussed in the following section. An alkaline electrolyzer is a commercialized technology for water splitting. Alkaline electrolytes seem to be chosen because many materials (e.g., nickel- or iron-based) are stable during electrolysis, maintaining the high activity of the hydroxyl ions as the reactants. Typical conditions can be found in the literature: cathode NiZn, anode Pt/IrO2, at 1.55 V, 2.5 kA m−2, 80°C [82].
Electrochemical reactions occur at the steady-state potential of the electrocatalysts on the surface of photocatalysts. The potential is determined as a consequence of all the photophysical and photochemical events discussed up to now. If there is a way to shift the potentials of the electrocatalysts immobilized on the surface of semiconductor photocatalysts, as discussed previously, then it is possible to obtain the rates from electrocatalytic knowledge, which can be measured separately. Electrocatalytic activity can be described in the form of the Tafel equation by neglecting the reverse reaction. The reaction rate for the photoelectrochemical half reaction, r (mol s−1), can be described by [73]
$$ r=\frac{i_0}{nF} \exp \frac{\alpha nF\left({E}_{\mathrm{D}}-{E}^0\right)}{RT}, $$
where i0 is the exchange current of the given metal (A), α is the transfer coefficient, n is the number of electrons involved in the reaction, F is the Faraday constant, ED and E0 are the Fermi level of the metal and the redox potential in solution, respectively, R is the universal gas constant, and T is the absolute temperature. The main problem to be solved is to determine precisely the potential applied to the photocatalyst and cocatalyst. Photoexcited electrons and holes are generally known to undergo relaxation immediately after excitation to the levels at the edges of the CB and VB, respectively. The reduction in the bandgap implies a reduction in the driving force for the reaction, which is associated with the potential difference between the semiconductor band edges and the redox potential in the solution (ED − E0). Therefore, it is more difficult to achieve photocatalysis at high rates with smaller bandgap materials. Photocatalysts would simply provide an external bias (i.e., apply a potential) to metals when under photoexcitation.

Identifying the electrocatalytic activity indicators is very important. For hydrogen evolution, the conventional volcano plot reported by Trasatti [83] or more recently by Nørskov and co-workers [84] for H2 production in acid solution with different metals shows that there is an optimal value for the free energy of hydrogen adsorption on metals. Similarly, for the oxygen evolution reaction, the volcano trend for metal oxides as a function of several thermodynamic descriptors has been claimed by several authors [85, 86, 87]. Markovic and coworkers reported that islands of nickel or cobalt species on noble metal surfaces (such as Pt) further enhance the water redox reactions for both hydrogen and oxygen evolution [88]. Mixed oxyhydroxides, such as iron-nickel, and perovskites have also been reported as low overpotential electrocatalysts that do not use noble metals [89, 90, 91, 92]. The metal particle sizes alone significantly affect the overall efficiency of both the electrochemical and photocatalytic reactions [69, 93]. Therefore, the metal particle size should also be rigorously accounted for, but this type of data analysis provides good hints for forming strategies to develop highly efficient photoabsorbers using efficient cocatalysts.

Dark reactions involving the back reaction (H2 and O2 going to H2O), a typical problem for cogeneration of H2 and O2 from overall water splitting, must be prevented. Noble metals typically cause the back reaction either catalytically or electrochemically (oxygen reduction reaction). Successful suppression of the oxygen reduction reaction without affecting the hydrogen evolution activity has been achieved using chromium and other materials [94, 95]. For Cr, the experiments suggest that the Cr layer works as a selective membrane for H2 but is not permeable for O2, as depicted in Fig. 12 [94]. Metallic nickel does not seem to cause oxygen reduction even under the conditions of hydrogen evolution, and nickel (hydr)oxide works for oxygen evolution and is distributed on the semiconductor surfaces [95].
Fig. 12

Metal-Cr core-shell nano-structure for selective hydrogen evolution that does not cause back reaction from H2 and O2 to H2O [94]

It is obvious that when either reduction (H2 formation) or oxidation (O2 formation) is accelerated, the potential shifts in the direction reflecting the remaining electrons or holes. Enhancement of the rate of reduction or oxidation improves the overall efficiency of water splitting based on this charge-up theory, which is determined by the photon flux and the electron efficiency from the photocatalyst to the metal particles because the accelerated electron or hole process affects the potential, which in turn perturbs the rates of the counter-side process. In this sense, there is no rate-determining step in overall water splitting, where the reactions occur in parallel for reduction (electron path) and oxidation (hole path). In other words, each component in the respective steps during photocatalysis should be improved to achieve overall water splitting. To demonstrate this concept, co-loading of H2 and O2 evolution cocatalysts was found to be effective to some extent [96]. Integrated studies are essential to establish highly efficient photocatalysts for overall water splitting.

2.6 Mass Transfer

In the study of photocatalysis, the main research focus has been developing efficient materials, including cocatalysts. The mass transfer of the reactants has often been overlooked during the photocatalytic efficiency determination. Photocatalytic water splitting is nothing but electrocatalysis, which is driven by the photocatalyst-assisted excited carriers. Therefore, rigorous and quantitative arguments from electrocatalysis regarding the thermodynamic and kinetic information should be appropriately put into practice. For example, a two-compartment photoelectrochemical cell is able to separate the oxidation and reduction products (e.g., O2 and H2). However, the H2 evolution reaction causes an increase in pH (according to the reaction at (44)), and the O2 evolution reaction causes a decreases in the pH (according to the reaction at (45)). The complete isolation of ions leads to a high concentration overpotential (shifting the thermodynamic potential of 59 mV pH−1), which stops the reaction and determines the overall efficiency. Therefore, the use of an ion-exchangeable membrane is mandatory [97]. Nafion or an alkaline membrane typically works in extreme pH media. When acidic conditions are chosen, the development of a non-noble metal electrodes with acid tolerance is required. The recent development of metal phosphide materials is of significant interest because they contain only abundant transition metals, such as Ni, Fe, and Co [98, 99, 100]. One of the most significant benefits of co-producing an H2/O2 mixture is avoiding this pH gradient, which minimizes the concentration overpotential. The coproduction of H2/O2 also avoids the use of (potentially expensive) membranes in the photoreactor and makes it possible to operate at neutral pH, which is otherwise impossible even for highly buffered solutions [97].

An interesting consideration regarding the activity of the reactants is the use of water vapor as a reactant (water liquid vs water vapor). Using vapor phase water has advantages such as the easily controlled supply and simple reactor designs, e.g., a fixed bed, for powder systems [101]. However, it encounters considerable difficulties because of the additional adsorption term for water vapor, which may strongly decrease the overall efficiency. In contrast, liquid phase water or associated ions (H+ or OH) as reactants can achieve high activities (close to unity for water, or increasing or decreasing pH).

One of the benefits and simultaneously demerits of overall water splitting is the cogeneration of H2 and O2, which allows for the direct use of pure water or eventually seawater because there is no significant concentration (activity) loss in the reactor [97]. The H2-O2 coproduction system allows for nanometer separation of the reduction and oxidation sites, thus minimizing the concentration gradient of the ions. The study of water electrolysis under neutral pH is therefore very important. In water at pH 0 or 14, one H+ or OH ion is present among 55 H2O molecules. Kinetically speaking, the reactions with hydronium ions (protons) or hydroxyl ions are more facile than those with water molecules for reduction and oxidation, respectively [82]. In neutral conditions, buffering actions are effective based on the reactant switching over varying pH. For electrochemical measurements, the supporting electrolyte is an essential component to avoid solution resistance (iR drop). To date, it seems that there is a lack of information in the literature regarding the supporting electrolyte effects in overall water splitting. The effects of the supporting electrolyte (identity and concentrations) have to be rigorously and quantitatively taken into account because they may cause reactant switching and additional adsorption onto active sites, which may enhance or decrease the overall catalytic efficiency.

2.7 Other Considerations and Standardization of Measurements of Photocatalysis

There cannot be a sole kinetically relevant step in photocatalysis, as discussed previously. The concept of the rate-determining step is only applicable for a series of sequential reaction steps. In photocatalysis, there are always parallel pathways for efficiency loss (either the forward direction for photocatalysis or recombination). Thus, the total efficiency relies on the first-order rate constants of all processes associated with the lifetime (reciprocal of the sum of the first order: \( \tau =\frac{1}{{\displaystyle \sum^i{k}_i}} \)). For better visualization of the photocatalytic efficiencies of the involved processes, multiplication of the efficiencies for the designated processes wherever isolated should be used. After all, the photocatalytic activity for solar energy conversion can be compared based only on the effects of the absorbed (or incident) photons, i.e., quantum yield or quantum efficiency, at a specified wavelength and its integration through varied wavelengths. (The use of terminology such as quantum yield and quantum efficiency in photocatalysis can cause some confusion because they are different in homogeneous and heterogeneous catalysis [30]. In addition, the solar cell or the photoelectrochemical community use quantum efficiency.) It is important to note that the quantum efficiency is not a function that is proportional to the surface area or the mass of the photocatalyst. The photocatalytic activities consider the fate of photons whether reacted or recombined. Therefore, the rates per surface area or per mass of catalyst cannot be used as indicators to compare the intrinsic photocatalytic activities of the materials unless those parameters are of particular interest for comparison [102].

It is important to compare the results obtained using lasers and conventional lamps (including solar radiation and lights with band pass filters) when discussing the behavior of photocatalyst powders because the photon flux (the number of photons per unit area) drastically affects the nature of the chemical process. The photocatalytic process induced by laser radiation readily causes multielectron reactions. It is important to consider carefully the light intensity whenever discussing the photocatalytic activity. It is even recommended to report the photon distribution for all of the illumination used as a function of wavelength.

Many studies use dye degradation to compare the (visible light) photocatalytic activity of different materials. Using photon-responsive dyes as a substrate for a photocatalytic test has long been known to bring ambiguity to the activity results [103, 104, 105]. More importantly, promoting this test as a measure of photocatalyst activity must be avoided because there is a tremendous contribution from the dye-sensitized pathway; the excited dye is used in photocatalysis rather than bandgap excitation of the semiconductor [104]. Furthermore, the International Organization for Standardization (ISO) has developed standardized experimental procedures for these tests, which should be followed rigorously [105].

3 Concluding Remarks

Efficient overall water splitting to date is limited to the UV range of light where high STH is not expected according to the solar irradiance [106, 107]. It is essential to develop visible-light-responsive photocatalyst materials. Although cocatalyst design has enabled water splitting using visible light [108, 109], the efficiency has to be greatly improved. This review focuses on the fundamental parameters involved in the photocatalytic processes for overall water splitting. As demonstrated in this chapter, photocatalytic water splitting is a complex process involving photon absorption, exciton separation, carrier diffusion, carrier transport, catalytic efficiency, and mass transfer of the reactants. Isolation of such parameters and their quantitative measurements and descriptions are becoming more and more important for developing novel materials.

Powder semiconductors have tremendous potential for solar fuel generation, partly because of their synthesis scalability. They can also be synthesized by wet chemistry [22], which allows chemists to contribute greatly to the field of solar energy conversion. It is, however, noted that the quality of the semiconductor powders needs to be very high, preferably at the level used in solar cells. Based on the above discussion, the four most critical key parameters on which to focus research investigations to improve photocatalytic performance are:
  1. 1.

    Electronic structure: determining most of the semiconductor properties that are strongly supported by advanced DFT calculations

  2. 2.

    Charge carrier concentrations: correlating with carrier lifetime and mobility

  3. 3.

    Electrocatalysts: enabling the redox reactions at low overpotentials

  4. 4.

    Interface: minimizing electronic barriers and protecting unstable components


Integrating the knowledge gained from studying these parameters can enable the concept of “photocatalysts by design,” which can lead to improvements in photocatalytic efficiency to allow us to meet our energy demands through solar H2 production.



This study presented in this chapter was supported by King Abdullah University of Science and Technology (KAUST). The author thanks Dr. A. Ziani, Mr. A.T. Garcia-Esparza, Mrs. E. Nurlaela, and Mr. T. Shinagawa at KAUST for proofreading the manuscript.


  1. 1.
    Lewis NS, Nocera DG (2006) Proc Natl Acad Sci 103:15729CrossRefGoogle Scholar
  2. 2.
    International Energy Agency (2010) World Energy Outlook 2010. International Energy Agency, ParisGoogle Scholar
  3. 3.
    National Renewable Energy Laboratory (NREL) (1999)
  4. 4.
    Pinaud BA, Benck JD, Seitz LC, Forman AJ, Chen Z, Deutsch TG, James BD, Baum KN, Baum GN, Ardo S, Wang H, Miller E, Jaramillo TF (2013) Energy Environ Sci 6:1983CrossRefGoogle Scholar
  5. 5.
    Takanabe K, Domen K (2011) Green 1:313CrossRefGoogle Scholar
  6. 6.
    Turro NJ, Ramamurthy V, Scaiano JC (eds) (2010) Modern molecular photochemistry of organic molecules. University Science, SausalitoGoogle Scholar
  7. 7.
    Nozik AJ (1978) Annu Rev Phys Chem 29:189CrossRefGoogle Scholar
  8. 8.
    Nosaka Y, Ishizuka Y, Miyama H (1986) Ber Bunsenges Phys Chem 90:1199CrossRefGoogle Scholar
  9. 9.
    Memming R (1988) Top Curr Chem 143:79CrossRefGoogle Scholar
  10. 10.
    Hagfeldt A, Grätzel M (1995) Chem Rev 95:49CrossRefGoogle Scholar
  11. 11.
    Kaneko M, Okura I (eds) (2002) Photocatalysis science and technology. Kodansha/Springer, Tokyo/BerlinGoogle Scholar
  12. 12.
    Domen K (2003) In: Horvath IT (ed) Encyclopedia of catalysis. Wiley, HobokenGoogle Scholar
  13. 13.
    Maeda K, Domen K (2007) J Phys Chem C 111:7851CrossRefGoogle Scholar
  14. 14.
    Kamat PV (2007) J Phys Chem C 111:2834CrossRefGoogle Scholar
  15. 15.
    Osterloh FE (2008) Chem Mater 20:35CrossRefGoogle Scholar
  16. 16.
    Kudo A, Miseki Y (2009) Chem Soc Rev 38:253CrossRefGoogle Scholar
  17. 17.
    Inoue Y (2009) Energy Environ 2:364CrossRefGoogle Scholar
  18. 18.
    Walter MG, Warren EL, McKone JR, Boettcher SW, Mi Q, Santori EA, Lewis NS (2010) Chem Rev 110:6446CrossRefGoogle Scholar
  19. 19.
    Abe R (2010) J Photochem Photobiol C 11:179CrossRefGoogle Scholar
  20. 20.
    Maeda K, Domen K (2010) J Phys Chem Lett 1:2655CrossRefGoogle Scholar
  21. 21.
    Hisatomi T, Minegishi T, Domen K (2012) Bull Chem Soc Jpn 85:647CrossRefGoogle Scholar
  22. 22.
    Takanabe K, Domen K (2012) ChemCatChem 4:1485CrossRefGoogle Scholar
  23. 23.
    Tong H, Ouyang S, Bi Y, Umezawa N, Oshikiri M, Ye J (2012) Adv Mater 24:229CrossRefGoogle Scholar
  24. 24.
    Tachibana Y, Vayssieres L, Durrant JR (2012) Nat Photonics 6:511CrossRefGoogle Scholar
  25. 25.
    Osterloh FE (2013) Chem Soc Rev 42:2294CrossRefGoogle Scholar
  26. 26.
    Hisatomi T, Takanabe K, Domen K (2015) Catal Lett 145:95CrossRefGoogle Scholar
  27. 27.
    Takanabe K, Domen K (2014) Photocatalysis in generation of hydrogen from water. In: Tao F, Schneider WF, Kamat PV (eds) Heterogeneous catalysis at nanoscale for energy applications. Wiley, Hoboken, pp 239–270Google Scholar
  28. 28.
    Bohren CF, Huffman DR (eds) (2004) Absorption and scattering of light by small particles. Wiley, WeinheimGoogle Scholar
  29. 29.
    Dahm DJ, Dahm KD (eds) (2007) Interpreting diffuse reflectance and transmittance. NIR, ChichesterGoogle Scholar
  30. 30.
    Braslavsky SE, Braun AM, Cassano AE, Emeline AV, Litter MI, Palmisano L, Parmon VN, Serpone N (2011) Pure Appl Chem 83:931CrossRefGoogle Scholar
  31. 31.
    Chen Z, Dinh HN, Miller E (eds) (2013) Photoelectrochemical water splitting, standards, experimental methods, and protocols. Springer, New YorkGoogle Scholar
  32. 32.
    Wemple SH, Seman JA (1973) Appl Opt 12:2947CrossRefGoogle Scholar
  33. 33.
    Di Giulio M, Micocci G, Rella R, Siciliano P, Tepore A (1993) Phys Status Solidi A 136:K101CrossRefGoogle Scholar
  34. 34.
    Lodenquai JF (1994) Sol Energy 53:209CrossRefGoogle Scholar
  35. 35.
    Swanepoel R (1983) J Phys E Sci Instrum 16:1214CrossRefGoogle Scholar
  36. 36.
    Chen LF, Ong CK, Neo CP, Varadan VV, Varadan VK (2005) Microwave electronics: measurement and materials characterization. Wiley, ChichesterGoogle Scholar
  37. 37.
    Le Bahers T, Rérat M, Sautet P (2014) J Phys Chem C 118:5997CrossRefGoogle Scholar
  38. 38.
    Green MA (2008) Sol Energy Mater Sol Cells 92:1305CrossRefGoogle Scholar
  39. 39.
    Džimbeg-Malčić V, Barbarić-Mikočević Ž, Itrić K (2011) Technical Gazette 18:117Google Scholar
  40. 40.
    Wood DL, Tauc J (1972) Phys Rev B 5:3144CrossRefGoogle Scholar
  41. 41.
    Schubert EF (ed) (2006) Light-emitting diodes, 2nd edn. Cambridge University Press, CambridgeGoogle Scholar
  42. 42.
    Bae D, Pedersen T, Seger B, Malizia M, Kuznetsov A, Hansen O, Chorkendorff I, Vesborg PCK (2015) Energy Environ Sci 8:650CrossRefGoogle Scholar
  43. 43.
    Sze SM, Ng KK (eds) (2006) Physics of semiconductor devices. Wiley, New YorkGoogle Scholar
  44. 44.
    Kittel C (2005) Introduction to solid state physics, 8th edn. Wiley, HobokenGoogle Scholar
  45. 45.
    Kim DW, Leem YA, Yoo SD, Woo DH, Lee DH, Woo JC (1993) Phys Rev B 47:2042CrossRefGoogle Scholar
  46. 46.
    Liang WY (1970) Phys Educ 5:226CrossRefGoogle Scholar
  47. 47.
    Bastard G, Mendez EE, Chang LL, Esaki L (1982) Phys Rev B 26:1974CrossRefGoogle Scholar
  48. 48.
    Gerischer H (1984) J Phys Chem 88:6096CrossRefGoogle Scholar
  49. 49.
    van der Pauw LJ (1958) Philips Res Rep 13:1Google Scholar
  50. 50.
    Heaney MB (2000) Electrical conductivity and resistivity. In: The measurement, instrumentation and sensors handbook. CRC, Boca RatonGoogle Scholar
  51. 51.
    Nagel H, Berge C, Aberle AG (1999) J Appl Phys 86:6218CrossRefGoogle Scholar
  52. 52.
    Law ME, Solley E, Liang M, Burk DE (1991) IEEE Electron Device Lett 12:401CrossRefGoogle Scholar
  53. 53.
    Shockley W, Read WT Jr (1952) Phys Rev 87:835CrossRefGoogle Scholar
  54. 54.
    Hall RN (1952) Phys Rev 87:387CrossRefGoogle Scholar
  55. 55.
    Auger P (1952) C R A S 177:169Google Scholar
  56. 56.
    Zhang Z, Yates JT Jr (2012) Chem Rev 112:5520CrossRefGoogle Scholar
  57. 57.
    Yoneyama H (1993) Crit Rev Solid State Mater Sci 18:69CrossRefGoogle Scholar
  58. 58.
    Grätzel M (2001) Nature 414:338CrossRefGoogle Scholar
  59. 59.
    Gelderman K, Lee L, Donne SW (2007) J Chem Educ 84:685CrossRefGoogle Scholar
  60. 60.
    van de Krol R, Grätzel M (2012) Photoelectrochemical hydrogen production. Springer, New YorkCrossRefGoogle Scholar
  61. 61.
    Sato N (1998) Electrochemistry at metal and semiconductor electrodes. Elsevier, AmsterdamGoogle Scholar
  62. 62.
    Tung RT (2014) Appl Phys Rev 1:011304CrossRefGoogle Scholar
  63. 63.
    Cohen ML (1979) J Vac Sci Technol 16:1135CrossRefGoogle Scholar
  64. 64.
    Cendula P, Tilley SD, Gimenez S, Bisquert J, Schmid M, Grätzel M, Schumacher JO (2014) J Phys Chem C 118:29599CrossRefGoogle Scholar
  65. 65.
    Mills TJ, Lin F, Boettcher SW (2014) Phys Rev Lett 112:148304CrossRefGoogle Scholar
  66. 66.
    Lin F, Boettcher SW (2014) Nat Mater 13:81CrossRefGoogle Scholar
  67. 67.
    Kamat PV (2002) Pure Appl Chem 74:1693CrossRefGoogle Scholar
  68. 68.
    Jakob M, Levanon H, Kamat PV (2003) Nano Lett 3:353CrossRefGoogle Scholar
  69. 69.
    Subramanian V, Wolf EE, Kamat PV (2004) J Am Chem Soc 126:4943CrossRefGoogle Scholar
  70. 70.
    Yoshida M, Yamakata A, Takanabe K, Kubota J, Osawa M, Domen K (2009) J Am Chem Soc 131:13218CrossRefGoogle Scholar
  71. 71.
    Lu X, Bandara A, Katayama M, Yamakata A, Kubota J, Domen K (2011) J Phys Chem C 115:23902CrossRefGoogle Scholar
  72. 72.
    Chen Z, Jaramillo TF, Deutsch TG, Kleiman-Shwarsctein A, Forman AJ, Gaillard N, Garland R, Takanabe K, Heske C, Sunkara M, McFarland EW, Domen K, Miller EL, Turner JA, Dinh HN (2010) J Mater Res 25:3CrossRefGoogle Scholar
  73. 73.
    Bard AJ, Faulkner LR (2001) Electrochemical methods, 2nd edn. Wiley, New York, pp 736–768Google Scholar
  74. 74.
    Fukasawa Y, Takanabe K, Shimojima A, Antonietti M, Domen K, Okubo T (2011) Chem Asian J 6:103CrossRefGoogle Scholar
  75. 75.
    Albery WJ, Bartlett PN (1984) J Electrochem Soc 131:315CrossRefGoogle Scholar
  76. 76.
    Chamousis RL, Osterloh FE (2014) Energy Environ Sci 7:736CrossRefGoogle Scholar
  77. 77.
    Butler MA, Ginley DS (1978) J Electrochem Soc 125:228–232CrossRefGoogle Scholar
  78. 78.
    Paracchino A, Laporte V, Sivula K, Grätzel M, Thimsen E (2011) Nat Mater 10:456CrossRefGoogle Scholar
  79. 79.
    Esposito DV, Levin I, Moffat TP, Talin AA (2013) Nat Mater 12:562CrossRefGoogle Scholar
  80. 80.
    Hu S, Shaner MR, Beardslee JA, Lichterman M, Brunschwig BS, Lewis NS (2014) Science 344:1005CrossRefGoogle Scholar
  81. 81.
    Shinagawa T, Garcia-Esparza AT, Takanabe K (2014) ChemElectroChem 1:1497CrossRefGoogle Scholar
  82. 82.
    Hamann CH, Hamnett A, Vielstich W (eds) (2007) Electrochemistry, 2nd edn. Wiley, WeinheimGoogle Scholar
  83. 83.
    Trasatti S (1972) J Electroanal Chem 32:163CrossRefGoogle Scholar
  84. 84.
    Greeley J, Jaramillo TF, Bonde J, Chorkendorff I, Nørskov JK (2006) Nat Mater 5:909CrossRefGoogle Scholar
  85. 85.
    Matsumoto Y, Sato E (1986) Mater Chem Phys 14:397CrossRefGoogle Scholar
  86. 86.
    Man IC, Su H-Y, Calle-Vallejo F, Hansen HA, Martinez JI, Inoglu NG, Kitchin J, Jaramillo TF, Nørskov JK, Rossmeisl J (2011) ChemCatChem 3:1159CrossRefGoogle Scholar
  87. 87.
    Grimaud A, May KJ, Carlton CE, Lee YL, Risch M, Hong WT, Zhou J, Shao-Horn Y (2013) Nat Commun 4:3439CrossRefGoogle Scholar
  88. 88.
    Subbaraman R, Tripkovic D, Chang KC, Strmcnik D, Paulikas AP, Hirunsit P, Chan M, Greeley J, Stamenkovic V, Markovic NM (2012) Nat Mater 11:550CrossRefGoogle Scholar
  89. 89.
    Suntivich J, May KJ, Gasteiger HA, Goodenough JB, Shao-Horn Y (2011) Science 334:1383CrossRefGoogle Scholar
  90. 90.
    Smith RDL, Prévot MS, Fagan RD, Zhang Z, Sedach PA, Siu JMK, Trudel S, Berlinguette CP (2013) Science 340:60CrossRefGoogle Scholar
  91. 91.
    Gong M, Li Y, Wang H, Liang Y, Wu JZ, Zhou J, Wang J, Regier T, Wei F, Dai H (2013) J Am Chem Soc 135:8452CrossRefGoogle Scholar
  92. 92.
    Gong M, Zhou W, Tsai MC, Zhou J, Guan M, Lin MC, Zhang B, Hu Y, Wang DY, Yang J, Pennycook SJ, Hwang BJ, Dai H (2014) Nat Commun 5:5695CrossRefGoogle Scholar
  93. 93.
    Muller BR, Majoni S, Memming R, Meissner D (1997) J Phys Chem B 101:2501CrossRefGoogle Scholar
  94. 94.
    Yoshida M, Takanabe K, Maeda K, Ishikawa A, Kubota J, Sakata Y, Ikezawa Y, Domen K (2009) J Phys Chem C 113:10151CrossRefGoogle Scholar
  95. 95.
    Yoshida M, Maeda K, Lu D, Kubota J, Domen K (2013) J Phys Chem C 117:14000CrossRefGoogle Scholar
  96. 96.
    Townsend TK, Browning ND, Osterloh FE (2012) Environ Sci 5:9543Google Scholar
  97. 97.
    Jin J, Walczak K, Singh MR, Karp C, Lewis NS, Xiang C (2014) Energy Environ Sci 7:3371CrossRefGoogle Scholar
  98. 98.
    Popczun EJ, McKone JR, Read CG, Biacchi AJ, Wiltrout AM, Lewis NS, Schaak RE (2013) J Am Chem Soc 135:9267CrossRefGoogle Scholar
  99. 99.
    Jiang P, Liu Q, Liang Y, Tian J, Asiri AM, Sun X (2014) Angew Chem Int Ed 53:12855CrossRefGoogle Scholar
  100. 100.
    Popczun EJ, Read CG, Roske CW, Lewis NS, Schaak RE (2014) Angew Chem Int Ed 53:5427CrossRefGoogle Scholar
  101. 101.
    Dionigi F, Vesborg PCK, Pedersen T, Hansen O, Dahl S, Xiong A, Maeda K, Domen K, Chorkendorff I (2011) Energy Environ Sci 4:2937CrossRefGoogle Scholar
  102. 102.
    Kisch H (2010) Angew Chem Int Ed 49:9588CrossRefGoogle Scholar
  103. 103.
    Mills A, Wang J (1999) J Photochem Photobiol A 127:123CrossRefGoogle Scholar
  104. 104.
    Yang X, Ohno T, Nishijima K, Abe R, Ohtani B (2006) Chem Phys Lett 429:606CrossRefGoogle Scholar
  105. 105.
    Mills A (2012) Appl Catal B 128:144CrossRefGoogle Scholar
  106. 106.
    Kato H, Asakura K, Kudo A (2003) J Am Chem Soc 125:3082CrossRefGoogle Scholar
  107. 107.
    Sakata Y, Matsuda Y, Nakagawa T, Yasunaga R, Imamura H, Teramura K (2011) ChemSusChem 4:181Google Scholar
  108. 108.
    Maeda K, Teramura K, Lu D, Takata T, Saito N, Inoue Y, Domen K (2006) Nature 440:295CrossRefGoogle Scholar
  109. 109.
    Maeda K, Xiong A, Yoshinaga T, Ikeda T, Sakamoto N, Hisatomi T, Takashima M, Lu D, Kanehara M, Setoyama T, Teranishi T, Domen K (2010) Angew Chem Int Ed 49:4096CrossRefGoogle Scholar

Copyright information

© Springer International Publishing Switzerland 2015

Authors and Affiliations

  1. 1.Division of Physical Sciences and Engineering, KAUST Catalysis Center (KCC)King Abdullah University of Science and Technology (KAUST)ThuwalSaudi Arabia

Personalised recommendations