General Considerations for Improving Photovoltage in Metal-Insulator-Semiconductor Photoanodes.

Metal-insulator-semiconductor (MIS) photoelectrodes offer a simple alternative to the traditional semiconductor-liquid junction and the conventional p-n junction electrode. Highly efficient MIS photoanodes require interfacial surface passivating oxides and high workfunction metals to produce a high photovoltage. Herein, we investigate and analyze the effect of interfacial oxides and metal workfunctions on the barrier height and the photovoltage of a c-Si photoanode. We use two metal components in a bimetal contact configuration and observe the modulation of the effective barrier height and the resulting photovoltage as a function of the secondary outer metal. The photovoltage shows a strong linear dependence by increasing the inner metal workfunction, with the highest photovoltage achieved by a MIS photoanode using a platinum inner metal. We also found that coupling a thin aluminium oxide with an interfacial silicon oxide and controlling the oxide thickness can significantly improve the photovoltage of an MIS junction photoanode.

Introduction

Sunlight-driven photoelectrochemical (PEC) n class="Chemical">water splitting is a promising approach to store intermittent solar energy by enabling a renewable and sustainable production of hydrogen, which can be later directly consumed as a transportation fuel or utilized as a feedstock in industrial and stationary power sectors.[1] At the heart of PEC systems are semiconducting photoelectrodes that absorb and transform light into photoexcited charge carriers, which are then directly used to drive nonspontaneous electrochemical reactions such as electrolysis of water. One important step in the PEC processes is charge separation driven by an electrostatic field at semiconductor interfaces which can be formed by bringing a semiconductor into contact with an electrolyte solution (semiconductor–liquid junction), with an opposing-type semiconductor (p–n homo- or heterojunction), or with a metal (Schottky junction, e.g., metal–semiconductor junction or metal–insulator–semiconductor junction).

Metal–insulator–semin class="Chemical">conductor (MIS) structures have been the focus of great interest for solar energy conversion devices because of their simple fabrication, low-temperature processing, and the potential of achieving higher efficiencies than the traditional p–n junction. For diffused p–n silicon junctions, the photovoltage and thus the device efficiency is limited by Auger recombination in the emitter region with an upper limit of 720 mV for devices thicker than the minority carrier diffusion length.[2,3] The MIS junction concept offers an advantage over the conventional p–n junction by utilizing a metal carrier-selective contact to form a rectifying junction with the photoabsorber, eliminating the need for highly doped emitter, thereby allowing the device to achieve a photovoltage of greater than 720 mV.

A significant development in MIS photoelectrodes has set a ren class="Chemical">cord photovoltage of 630 mV,[4] close to that obtained from the p–n junction electrode.[5] The keys to achieving a high photovoltage MIS photoelectrode include the use of thin tunnel oxides and application of appropriate metal contacts with suitable workfunctions.[6] For n-type semiconductor photoanodes, the metal workfunction should be sufficiently high to ensure a strong inversion near the semiconductor surface,[6] which is a prerequisite for a large generation of photovoltage. However, most high workfunction metals such as platinum (Pt) and iridium (Ir) are scarce and expensive. Conversely, p-type semiconductor photocathodes benefit from the need for low workfunction metals such as titanium (Ti) and aluminum (Al) to create a strong inversion. Although the low workfunction metals are relatively cheap, these metals exhibit poor catalytic activity and thus require an additional catalytic overlayer such as Pt to improve the reaction kinetics on the photocathode surface.[7,8]

Another important factor that determines a high photovoltage in an MIS photoelectrode is the thin tunnel n class="Chemical">oxide at the metal–semiconductor interface. The interfacial oxide should have a surface passivation ability to minimize carrier recombination, and minimum charge extraction barriers to allow for facile electronic transport.[5] In addition, the oxide layer should have dielectric properties that maintain the workfunction mismatch between the metal and the semiconductor to enable a large photovoltage generation across the MIS junction. Native silicon oxides (SiO) or chemically oxidized SiO are commonly used as interfacial insulating layers in MIS photoelectrode structures.[9,10] Apart from simplicity, the motivation for using such oxides is often the assumption that they have the same passivation properties as that of the high-quality thermal silicon dioxide (SiO2), a surface passivation material that has been used extensively in photovoltaic research and industry.

In our previous work, we have incorporated an n class="Chemical">ultrathin layer of aluminum oxide (Al2O3) in conjunction with the chemically grown SiO at the metal–semiconductor interface and observed significant improvements of photovoltage which was attributed to the excellent surface passivation induced by Al2O3 and the simultaneous increase of the barrier height.[10] Although this is partially true, in this work, we aim to expand upon the previous analysis and seek to identify the additional contribution that results in photovoltage enhancement upon the introduction of an additional oxide. Additionally, we explore the application of less-precious metal contacts and analyze their impact on the performance and trade-offs associated with the use of lower workfunction metals. We first investigate the junction properties of MIS structures using various metal contacts with various workfunctions by performing dark current–voltage measurements. Using the diode models for MIS junctions, we extract the important parameters such as the dark saturation current and the ideality factor that will determine the photovoltage of an MIS photoanode. We additionally determine the barrier heights using capacitance–voltage method and Mott-Schottky analysis, and compare the experimental values with the theoretical models for MIS junctions. Furthermore, we demonstrate how the interfacial oxides and the workfunction of each metal affect the Schottky barrier height within the MIS junctions. In separate experiments, we fabricate MIS photoanodes by depositing ultrathin bimetal contacts with various inner metals and show how the outer metal modulates the effective barrier height of the MIS photoanodes. Using the extracted parameters from the Schottky diode modeling, we then compare the experimentally measured photovoltages with the theoretically determined values. Finally, we demonstrate how tuning the thickness of the Al2O3 interface layer and using the appropriate inner metal contact can improve the photovoltage of an MIS photoanode.

Experimental Section

Chemicals

All chemicals were used as received: potassium hydroxide pellets (KOH, Alfa Aesar, 85%), hydrogen peroxide (H2O2, 30% (w/w) in H2O, contains stabilizer, Sigma-Aldrich), sulfuric acid (H2SO4, 99.999%, Sigma-Aldrich), hydrofluoric acid (HF, ACS reagent 48%, Sigma-Aldrich), hydrochloric acid (HCl, reagent grade, 37%, Sigma-Aldrich), potassium hexacyanoferrate(II) trihydrate (K4Fe(CN)6·3H2O, ≥99% puriss. p.a., ACS reagent, Sigma-Aldrich), potassium hexacyanoferrate(III) (K3Fe(CN)6, ≥99% puriss. p.a., ACS reagent, Sigma-Aldrich). Water with resistivity of 18.2 MΩ cm from Milli-Q integral ultrapure water (Merck Millipore).

Preparation of Substrates

Phosporus-doped (n-type, (100)-oriented, single-side polished, resistivity 0.1–0.3 Ω cm, 525 μm) and degenerately boron-doped (p+-type, (100)-oriented, singe-side polished, resistivity < 0.005 Ω cm) Si wafers were purchased from Si-Mat. The n-type Si wafers were first cleaned in a piranha solution containing a mixture of H2SO4 and H2O2 (3:1 volume ratio) at 120 °C for 20 min to remove the organic contaminants. The n-type Si wafers were then dipped into a buffered HF etchant (2%) for 2 min at room temperature to strip the native oxide on the Si wafer surface. Next, the Si wafers were immersed in a RCA SC-2 solution consisting of H2O, HCl, and H2O2 (5:1:1 by volume ratio) at 75 °C for 10 min to regrow the oxide layer (SiO). The same procedure was applied on the p+-type Si wafers.

Atomic-layer deposition of aluminum oxides

Atomic-layer deposin class="Chemical">tion (ALD) of aluminum oxides (Al2O3) was conducted in a home-built thermal ALD system (developed at AMOLF) at 250 °C at a base pressure of 0.01−0.05 mbar. The ALD cycle consisted of a 10 ms pulse of H2O, a 18 s N2 purge, a 10 ms pulse of trimethylaluminum (TMA), and another 18 s N2 purge to complete the cycle. Eight ALD cycles were used to deposit 1 nm thick Al2O3. In selected experiments, 24, 32 and 80 ALD cycles were used to deposit 2.7 nm, 3.9 nm and 8 nm layers of Al2O3, respectively. The thickness of the Al2O3 was estimated by ellipsometer (J.A. Wollam) using dielectric models for Al2O3 and Si native oxide on a Si substrate.

Sputter Deposition of Metals

Platinum (n class="Chemical">Pt), nickel (Ni), and cobalt (Co) were deposited using Prevac radio frequency (rf) magnetron sputtering from a Pt, Ni, and Co target, respectively (Mateck, 99.95%, 2 in. diameter, 5 mm thickness). The Ar flow was kept at 15 sccm, and the working pressure was held at 3 μbar. For Pt deposition, the rf power was kept at 25 W, and the deposition rate was approximately, 0.138 Å s–1. For Ni and Co depositions, the rf power was 100 W, and the deposition rate was 0.2 and 0.16 Å s–1, respectively. Titanium (Ti) was deposited in the AJA sputter chamber from a Ti target (Mateck, 99.95%, 2 in. diameter) using a dc power. The Ar flow was maintained at 15 sccm, and the working pressure was 3 μbar. The dc power for Ti deposition was 10 W.

Fabrication of MIS Devices

MIS devices were fabricated by deposin class="Chemical">ting an Al2O3 layer onto n-Si wafers using atomic layer deposition (ALD). The front Schottky contact of each MIS device was formed by depositing different metals, such as Pt, Ni, Co, and Ti, each with a thickness of 60 nm via sputtering. In separate experiments, a set of photoanodes was fabricated by depositing 2 nm thick inner metals (Pt, Ni, Co, or Ti), followed by depositing 4 nm of Ni-capping layer using low-power sputtering to ensure uniform deposition and homogeneous coverage. The back sides of the n-Si samples were scratched using a sand paper to remove the oxide layer, followed by cleaning the residue using ethanol. Next, the Ohmic back contacts were formed by rubbing the back side surfaces of the Si samples with a Ga–In alloy (75.5:24.5 wt %, 99.9% metal basis, Alfa Aesar). The Ohmic back contact of the p+-Si sample was formed by sputtering Pt. The schematic structure of the MIS photoanode is shown in Figure a.

Figure 1

(a) Schematic of planar MIS photoanodes for water oxidation. Representative energy band diagram of MIS photoanodes: (b) in the dark and (c) under illumination.

(a) Schematic of planar n class="Chemical">MIS photoanodes for water oxidation. Representative energy band diagram of MIS photoanodes: (b) in the dark and (c) under illumination.

(Photo)electrochemical Measurements

PEC measurement of the photoanode was n class="Chemical">conducted in a three-electrode configuration in 1 M KOH electrolyte solution under simulated AM1.5 solar irradiation (100 mW cm–2) using a Newport Sol3A Class AAA solar simulator (type 94023A-SR3) with a 450 W xenon short arc lamp. A mercury/mercury oxide (Hg/HgO in 1 M KOH, Hach Lange) electrode was used as the reference electrode, and an Ni coil was used as the counter electrode. The Hg/HgO electrode had a potential of 0.9222 V versus the reversible hydrogen electrode and was calibrated using silver/silver chloride (Ag/AgCl, in saturated KCl, Hach Lange). The exposed area of the working electrode was 0.2826 cm2. During the measurement, the electrolyte was continuously stirred using a magnetic stir bar. Cyclic voltammetry, electrochemical open-circuit, and impedance spectroscopy were performed using a potentiostat PARSTAT MC (Princeton Applied Research, AMETEK). The cyclic voltammetry data were recorded at a constant scan rate of 50 mV s–1.

Impedance Spectroscopy

Solid-state impedance spectroscopy of MIS devices was performed by n class="Chemical">connecting of the front metal contact with the back metal contact. Impedance spectroscopy of MIS photoanodes with thin bimetal contacts was performed electrochemically in solution containing 50 mM K3Fe(CN)6, 350 mM K4Fe(CN)6, and 1 M KCl in a three-electrode measurement using a Pt wire placed in a fritted glass tube as the reference electrode and a Pt coil as the counter electrode. The experimental setup was kept in the dark during the measurement.

Results and Discussion

MIS Schottky Junction

The rectifying behavior of an MIS juncn class="Chemical">tion arises from the workfunction mismatch between the metal (Φm) and the semiconductor (Φs) which forms an electrostatic potential barrier (analogous to a Schottky barrier) between the materials (Figure b). In the absence of defect states at the interfaces and potential drop in the oxide layer, the barrier height of an n-type semiconductor is given by the difference between the metal workfunction and the electron affinity of the semiconductor (χs). When designing an MIS structure, it is generally assumed that the effective workfunction of the metal in contact with a semiconductor and/or an insulator is the same as that in vacuum. In practice, considerable defect states are present and exist both at the semiconductor–insulator interface and at the insulator–metal interface. The existence of these interfacial states may alter the effective workfunction of the metal (Φm,eff) relative to the underlying surface and affect the junction properties, in particular the barrier height of the MIS junction.

The barrier height is an important property of the junction that determines the photovoltage of an MIS photoelectrode. Under illuminan class="Chemical">tion, the upper limit of the photovoltage of an MIS junction is reduced by the amount of free energy losses arising recombination events in various regions of the semiconductor. This is indicated by the upward movement of the hole quasiFermi level near the interface as a result of the excess carrier recombination (Figure c). The photovoltage (Vph) of an illuminated MIS junction electrode is then given by the difference between the electron (EF,n) and hole (EF,p) quasiFermi levels at the interface.

Schottky Barrier Height and Effective Pinning Factor of MIS Junction

The junction propern class="Chemical">ties at the metal–insulator–semiconductor interfaces can be examined by modeling the current–potential (j–V) curves in the dark. A simple and realistic dark j–V characteristic can be described by an implicit diode equation that takes into account the parallel resistance (Rp) and the series resistance (Rs) in eq (11)where j is the dark current density, j0 is the dark saturation current density, n is the diode ideality factor, q is the elementary charge (1.6 × 10–19 C), k is the Boltzmann’s constant (1.38 × 10–23 J K–1), T is the temperature (293 K), and V is the applied voltage. For an MIS junction with a large barrier height, the generation and recombination current in the space-charge region[12] can be more accurately modeled by adding a second diode in parallel connection with the first in an equivalent circuit (Figure S1, Supporting Information), which is represented by the double-diode equation in eq (11)Here, j01 is the dark saturation current density from the thermionic emission, n1 is the diode ideality factor at high forward bias, j02 is the dark saturation current density from recombination and generation processes in the space-charge region, and n2 is the diode ideality factor for recombination and generation current at low forward bias.

Figure a shows the semilogarithmic dark j–V characterisn class="Chemical">tics of the n-Si/SiO/Al2O3/’metal’ systems. The symbols (open circles) indicate the measured data, and the straight lines represent the fit results derived from eqs and 2. Positive applied voltages (V > 0) represent the forward bias region, and negative voltages (V < 0) represent the reverse bias region. Four different metal contacts were investigated in our MIS structures: Pt, Ni, Co, and Ti with corresponding workfunctions of 5.6, 5.2, 5.0, and 4.3 eV, respectively.[13] The thickness of each metal was 60 nm, and the ellipsometrically measured thickness of the oxide layers was approximately 1.8 nm for chemical SiO and 1 nm for ALD–Al2O3. All the investigated samples exhibited rectifying behavior, as indicated by the weak voltage dependence of the reverse current, and the exponential increase of forward current with increasing applied voltage. The n-Si/SiO/Al2O3 samples with Pt and Ni contacts showed four distinct regions in each corresponding dark j–V curve: a reverse bias region, two linear regions with different slopes at low and high forward biases, and a current limited region at very high forward bias. The four regions in the dark j–V were simultaneously fitted using a two-diode model described in eq . The n-Si/SiO/Al2O3 samples with Co and Ti contacts displayed only one linear curve in the intermediate region of the forward bias, and thus the fitting procedures were performed using a single-diode model in eq . Figure b depicts representative dark j–V characteristics of the n-Si/SiO/‘metal’ systems. All the samples demonstrated a linear line in the intermediate forward bias region and thus were fitted using a single-diode model. The n-Si/SiO with Pt, Co, and Ni contacts behaved as rectifying junctions, whereas the n-Si/SiO/Ti behaved as an Ohmic contact, as indicated by the symmetry of j–V curve in the forward and the reverse bias region. From the fit results, the mean ideality factors were n = 1.1 for the n-Si/SiO/Al2O3/‘metal’ and n = 1 for the n-Si/SiO/‘metal’ systems.

Figure 2

Dark j–V curves of (a) n-Si/SiO/Al2O3/‘metal’ and (b) n-Si/SiO/‘metal’ systems. The thickness of each metal contact is 60 nm. Symbols represent the measured data, and lines represent the fit results. Tabulated values of the extracted parameters are provided in Table S1 and S2, Supporting Information.

Dark j–V curves of (a) n-Si/n class="Chemical">SiO/Al2O3/‘metal’ and (b) n-Si/SiO/‘metal’ systems. The thickness of each metal contact is 60 nm. Symbols represent the measured data, and lines represent the fit results. Tabulated values of the extracted parameters are provided in Table S1 and S2, Supporting Information.

The utility of Figure a,b is that the barrier height can be extracted from the dark j–V curve. In an inn class="Chemical">timate metal–semiconductor structure where thermionic emission is the dominant current transport mechanism under moderate forward bias,[14] the barrier height can be directly derived from the dark saturation current using the thermionic emission theory. However, in the presence of an oxide layer at the interface, the current transport across the MIS junction is also governed by carrier tunneling through the oxide. The dark saturation current of an MIS junction is therefore given by the thermionic emission equation corrected for tunneling[15,16]where A* is the Richardson’s constant (120 A cm–2 K–2 for Si), ϕb is the barrier height, χ (in eV) is the mean barrier height presented by the tunnel oxide (i.e., the conduction band offset between the semiconductor and the oxide), and δ is the thickness of the oxide (in Å). The term is the transmission coefficient for electron tunneling through a one-dimensional rectangular barrier, and is the tunnel exponent (there is a constant (not shown) with a value ≈1 and units eV–1/2 Å–1 to make the exponent dimensionless).[15,16] Rearranging eq , the barrier height of an MIS junction is given by

From eq , it is clear that the barrier height has a strong dependence on the . In real n class="Chemical">situations where the interfacial oxide is very thin (δ < 30 Å), the experimental value of is appreciably different and typically can be much lower than the theoretical quantity. This discrepancy has been largely attributed to the nonuniformities of the oxide layer and the inhomogeneous distribution of the oxide thickness.[17,18] Additionally, barrier lowering due to image charges (i.e., image-force lowering) as well as the gradual disappearance of the band structure of a few atomic layers can also reduce the .[19]

For an MIS junction n class="Chemical">consisting of multiple interfacial oxides such as in the devices shown in this work, the effective tunnel exponent is not accurately known, and the barrier height cannot be explicitly derived from the dark saturation current. Alternatively, the barrier height can be estimated from Mott−Schottky analysis of inverse square of the space-charge capacitance−voltage (Csc−2−V) of the electrode, measured using impedance spectroscopy. (Figure S3, Supporting Information). Independent determination of barrier height allows the approximation of for multiple interfacial oxides using eq . Using the experimentally determined barrier heights from the Mott-Schottky analysis (Table S4, Supporting Information) and j0 values from the fit results of Figure b, and putting them into eq , the for MIS devices with an SiO (1.8 nm) interlayer was estimated to be 1.2, close to the value reported previously for wet chemical SiO with the same thickness.[19] Using the same approach, the calculated values for were approximately 1.6 for the n-Si/SiO (1.8 nm)/Al2O3 (1 nm) in contact with Pt, Ni, and Co, and 4 for the n-Si/SiO (1.8 nm)/Al2O3 (1 nm) with the Ti metal contact. The higher for the n-Si/SiO/Al2O3/Ti is presumably due to the formation of an additional oxide layer as a result of oxidation at the Al2O3/Ti interface. High workfunction metals such as Pt are resistant to oxidation but low workfunction metals such as Ti are extremely reactive and their surfaces may readily oxidize upon contact with an underlying oxide. Previous studies have indicated that the tunnel barrier of the n-Si/SiO2/metal system increases when a lower workfunction metal such as Al is used in comparison to the higher workfunction Au.[19] This may also be the case for the n-Si/SiO/Al2O3/Ti, considering both Ti and Al have similarly low workfunctions.

Figure a,b show the experimentally determined barrier height as a funcn class="Chemical">tion of the metal workfunction for the n-Si/SiO/Al2O3/‘metal’ (solid circles) and n-Si/SiO/‘metal’ (solid diamonds) systems, respectively. For comparison, the barrier heights calculated using the previously determined and j0 values in eq are also presented as hollow stars. The barrier height of the n-Si/SiO/Ti was not obtained from the space-charge capacitance measurement using impedance spectroscopy due to a high leakage current but a value close to 0.37 ± 0.02 eV is expected based on the calculation using eq . An excellent linearity between the barrier height and the metal workfunction was observed on the n-Si/SiO/Al2O3/“metal” systems with a slope of 0.32. In the absence of the Al2O3 interlayer, a linear relation with nearly the same slope was only apparent on the n-Si/SiO with Pt, Co and Ti contacts, while a large deviation was observed when using a Ni contact which exhibited a lower barrier height than that expected from the linear trend with the metal workfunction. In general, with the exception of the n-Si/SiO/Ni, the incorporation of 1 nm-thick Al2O3 increased the barrier height by 0.05 ± 0.01 eV.

Figure 3

Effective barrier height as a function of the metal workfunction of (a) n-Si/SiO/Al2O3/‘metal’ and (b) n-Si/SiO/‘metal’ systems. The thickness of each metal contact is 60 nm. Solid symbols are determined from Mott−Schottky analysis of inverse square of the space-charge capacitance−voltage (Csc−2−V) of the electrode, measured using impedance spectroscopy., and open symbols are extracted using eq from the fit results of Figure . Mott–Schottky plots are shown in Figure S4, Supporting Information.

Effective barrier height as a funcn class="Chemical">tion of the metal workfunction of (a) n-Si/SiO/Al2O3/‘metal’ and (b) n-Si/SiO/‘metal’ systems. The thickness of each metal contact is 60 nm. Solid symbols are determined from Mott−Schottky analysis of inverse square of the space-charge capacitance−voltage (Csc−2−V) of the electrode, measured using impedance spectroscopy., and open symbols are extracted using eq from the fit results of Figure . Mott–Schottky plots are shown in Figure S4, Supporting Information.

The slope of the linear trend between the barrier height and the metal workfuncn class="Chemical">tion (S = dϕb/dΦm) is indicative of the effective pinning factor for the MIS contact that ranges from 0 for perfect pinning to 1 for no pinning.[20] In an MIS structure, Fermi level pinning effects typically arise from two distinct contributions: (i) the existence of interfacial trap states at the semiconductor–oxide interface and (ii) movement of the metal Fermi level due to occupation of intrinsic states at the surface of the oxide layer by metal electrons (i.e., metal Fermi level pinning). While the former tends to pin the semiconductor Fermi level to the interface states, the latter tends to alter the effective value of metal workfunction which, in most cases, differs appreciably from the value in vacuum. It should be noted that the S parameter of a dielectric material (i.e., the oxide layer) should be distinguished from the effective S parameter of an MIS structure as a whole because interface states exist both at metal–insulator and insulator–semiconductor interfaces.[21] For example, Al2O3 has been reported to have a dielectric S parameter between 0.63 and 0.69[22] but the effective S parameter of an MIS junction employing an Al2O3 interfacial dielectric would be considerably different.

Previous empiricn class="Chemical">al and experimental studies have shown that intimate contact between an n-Si and a metal results in a nearly complete pinning of the Fermi level with an effective S parameter as small as 0.075.[23,24] The effective S parameter of 0.32 for the n-Si/SiO/Al2O3/‘metal’ systems, therefore, underlines the ability of the oxide layers to alleviate the Fermi level pinning within the n-Si-based MIS junction. In addition to intrinsic states, extrinsic defect states may also arise from the interfacial reaction between the metal and the oxide and may contribute to modification of the effective workfunction of the metal.[22] The existence of extrinsic defect states within an MIS contact is typically indicated by a large deviation of barrier height from the linear trend with varying metal workfunction.[21,22] We note however, that of all the samples investigated herein, such a deviation only occurs on the n-Si/SiO/Ni. As the barrier heights of other n-Si/SiO/‘metal’ systems exhibit a positive linear scaling with increasing metal workfunction, it may suggest that the distribution of the extrinsic defect states arising from the interfacial chemical reaction is metal-dependent. By introducing an Al2O3 interfacial layer, such a deviation can be eliminated, thus suggesting that the formation of defect-related states is considerably hindered.

Schottky Barrier Height and Effective Pinning Factor of MIS Photoanodes with Thin Bimetal Contacts

In an effort to directly evn class="Chemical">aluate the effect of the Al2O3 interlayer on the n-Si photoanode, a series of bimetallic MIS structures was fabricated by depositing 2 nm of various inner metals to serve as the Schottky contact, and 4 nm of Ni outer metals to serve as the catalyst and protection overlayer. The front metal contacts were designed very thin to avoid significant optical losses for high photocurrent generation without compromising the protection ability to prevent corrosion and to promote catalytic reaction of water oxidation. Because of the inability to directly probe the ultrathin metal contact (i.e., total thickness of 6 nm), impedance measurements for barrier height determination were performed electrochemically in a solution containing a reversible, one-electron ferri/ferrocyanide (Fe(CN)63–/4–) redox couple (Figure S5, Supporting Information). Afterward, all samples were subjected to contact with 1 M of potassium hydroxide (KOH) solution for at least 24 h (from hereafter called the aging process). Impedance measurements were then repeated to estimate the barrier heights of all the samples.

Figure a shows the effective barrier height of n class="Chemical">n-Si/SiO/Al2O3/‘inner metal’/Ni photoanodes as a function of the inner metal workfunction. Barrier height shifts were observed when using thin bimetal contacts in comparison to structures with a thick single layer metal contact. For example, the barrier height of the n-Si/SiO/Al2O3/Pt (2 nm)/Ni (4 nm) was slightly lower than the n-Si/SiO/Al2O3/Pt (60 nm) (Figure S6, Supporting Information). In bimetal Schottky structures, the shift of barrier height is closely related to the ineffective metal screening by an ultrathin inner metal, which is largely affected by the outer metal workfunction. The theoretical Thomas–Fermi screening length is 0.7 Å for most metals[25] but previous experimental studies have shown that the extent of the inner metal band bending in a bimetal Schottky diode can exceed 1.5 nm.[26] Considering the thickness of our inner metal is only 2 nm, the bimetal effective workfunction is expected to shift toward the value of the outer metal workfunction. Additionally, we also cannot rule out the possibility that interfacial diffusion of the inner metal may occur either to the the underlying oxide or to the outer metal (i.e., metal interdiffusion), which can potentially alter the effective workfunction of the bimetal structure.

Figure 4

Effective barrier height as a function of the inner metal workfunction of n-Si/‘oxide’/bimetal photoanodes with (a) both SiO/Al2O3 and (b) only SiO interfacial layers. Barrier heights are determined from Mott−Schottky analysis of inverse square of the space-charge capacitance−voltage (Csc−2−V) of the electrode, measured using impedance spectroscopy. The thickness of each inner metal is 2 nm and the thickness of the Ni outer metal is 4 nm. Mott–Schottky plots are shown in Figure S5, Supporting Information.

Effective barrier height as a funcn class="Chemical">tion of the inner metal workfunction of n-Si/‘oxide’/bimetal photoanodes with (a) both SiO/Al2O3 and (b) only SiO interfacial layers. Barrier heights are determined from Mott−Schottky analysis of inverse square of the space-charge capacitance−voltage (Csc−2−V) of the electrode, measured using impedance spectroscopy. The thickness of each inner metal is 2 nm and the thickness of the Ni outer metal is 4 nm. Mott–Schottky plots are shown in Figure S5, Supporting Information.

Figure a also indicates that the effecn class="Chemical">tive barrier height of all the samples increased after the aging process. Such an increase of the barrier height is ascribed to the transition of the workfunction of the surface layer due to incidental oxidation of Ni to nickel oxide/nickel dihydroxide (NiO/Ni(OH)2) after prolonged contact with the KOH electrolyte.[9,10] The vacuum workfunction of NiO is typically between 5.2 and 5.6 eV and depending on the processing conditions and/or posttreatments, the NiO workfunction can be as high as 6.7 eV.[27,28] Furthermore, the thickness of the surface layer should increase as a result of Ni lattice expansion upon transformation to NiO.[29,30] Such an increase of the surface layer thickness is then expected to extend the modulation of the bimetal effective workfunction and thus shift the barrier height of the Schottky junction.

Figure b shows the effective barrier height of n class="Chemical">n-Si/SiO/‘inner metal’/Ni without an Al2O3 interlayer. No substantial variation of barrier height was observed between the fresh and the aged sample. The effective barrier height of the n-Si/SiO with thin bilayer metal contacts exhibited a nonideal scaling when changing the inner metal workfunction in a fashion similar to those observed with a single thick metal contact (Figure b), regardless of the presence of the outer metal, even after the oxidation of Ni surface layer in KOH. Impedance characterization techniques failed to measure the capacitance of the n-Si/SiO/Ti/Ni for barrier height determination due to the nonrectifying behavior of the junction which resulted in a high leakage current, but a barrier height of close to 0.37 eV is expected based on the assumption of a similar trend in Figures b and 4b. Given the fact that the effective barrier heights of all n-Si/SiO/‘inner metal’/Ni samples remain constant despite the oxidation of the outer Ni, which should increase due to increasing effective workfunction of the surface layer, the inner metal Fermi level seems to strongly pin to the interface states that are most likely to be metal-dependent.

PEC Performance of MIS Photoanodes

The PEC activity of the n class="Chemical">MIS photoanodes was examined by performing cyclic voltammetry in 1 M KOH solution under simulated solar irradiation. Figure a shows the j–V curves of the illuminated n-Si/SiO/Al2O3/‘inner metal’/Ni photoanodes after the aging process in 1 M KOH electrolyte. Among all inner metals investigated in this work, the most negative photocurrent onset potential (defined as the potential required to achieve an anodic current of 100 μA cm–2) was achieved by the sample with the Pt inner metal (−270 mV relative to the formal potential for water oxidation, E0′ (O2/OH–) = 1.23 V versus the reversible hydrogen electrode) and the most positive onset potential was observed on the sample with inner Ti (+140 mV relative to E0′ (O2/OH–)). In the absence of light, the samples with inner Pt, Ni, and Co exhibited nearly zero current, whereas a nonnegligible dark current was observed on the sample with Ti as the inner metal. This nonzero dark current is consistent with the lower barrier height of the n-Si/SiO/Ti/Ni which increases the leakage current under reverse bias condition in a solid-state measurement or under anodic potential bias in an electrochemical measurement.

Figure 5

Representative current–potential (j–V) characteristics of (a) n-Si/SiO/Al2O3/‘inner metal’/Ni and (b) n-Si/SiO/‘inner metal’/Ni photoanodes after aging, measured in 1 M KOH solution under simulated solar illumination (solid lines). Dark currents are indicated by the dotted lines of the same color. The vertical dashed line indicates the formal potential for water oxidation, E0′ (O2/OH–). The j–V behavior of the nonphotoactive p+-Si/SiO/Ni is also shown. The measured photovoltage as a function of inner metal workfunction of (c) n-Si/SiO/Al2O3/‘inner metal’/Ni and (d) n-Si/SiO/‘inner metal’/Ni photoanodes. The thickness of the inner metal and the outer Ni are 2 and 4 nm, respectively.

Representative current–potenn class="Chemical">tial (j–V) characteristics of (a) n-Si/SiO/Al2O3/‘inner metal’/Ni and (b) n-Si/SiO/‘inner metal’/Ni photoanodes after aging, measured in 1 M KOH solution under simulated solar illumination (solid lines). Dark currents are indicated by the dotted lines of the same color. The vertical dashed line indicates the formal potential for water oxidation, E0′ (O2/OH–). The j–V behavior of the nonphotoactive p+-Si/SiO/Ni is also shown. The measured photovoltage as a function of inner metal workfunction of (c) n-Si/SiO/Al2O3/‘inner metal’/Ni and (d) n-Si/SiO/‘inner metal’/Ni photoanodes. The thickness of the inner metal and the outer Ni are 2 and 4 nm, respectively.

Figure b depicts j–V curves of the n-Si/n class="Chemical">SiO/‘inner metal’/Ni photoanodes. Without Al2O3, all the MIS photoanodes showed significant positive shifts of onset potential for water oxidation. Even with Pt as the inner metal, the onset potential was only −50 mV relative to E0′ (O2/OH–). A considerable dark leakage current was observed on the sample with Ni contact due to the low barrier height of the n-Si/SiO/Ni, in agreement with the thermionic emission theory described in eq . The photoactivity of the photoanode, generally indicated by the photocurrent plateau at sufficiently positive potentials, was not apparent on the n-Si/SiO/Ti/Ni. Both illuminated and dark j–V responses were identical and showed a behavior typically displayed by a nonphotoactive electrode for water oxidation. These results clearly indicate that the photovoltage of the n-Si/SiO/Ti/Ni was indeed zero.

Figure c summarizes the photovoltage of n-Si/n class="Chemical">SiO/Al2O3/‘inner metal’/Ni photoanodes as a function of the inner metal workfunction. The photovoltage was determined from the difference in anodic current onset potential between the illuminated MIS photoanode and the nonphotoactive p+-Si/SiO/Ni electrode in Figure a,b. Electrochemical open-circuit measurements in the dark and under illumination were also performed before and after aging in KOH to confirm the photovoltage (Figure S7, Supporting Information) and revealed values identical to those measured using the aforementioned approach. The photovoltage of n-Si/SiO/Al2O3/‘inner metal’/Ni photoanodes showed a positive linear scaling with increasing inner metal workfunction. Consistent with the increase of barrier height, the photovoltage of all samples increased upon oxidation of Ni surface layer in KOH. The highest photovoltage was achieved by the n-Si/SiO/Al2O3/Pt/Ni after aging, reaching a value of 520 mV. In the absence of an Al2O3 interlayer, all photoanodes exhibited lower photovoltages and showed almost no photovoltage shift after prolonged contact in KOH (Figure d), in close agreement with their barrier heights that remained constant upon aging (Figure b).

During the aging process in KOH solun class="Chemical">tion, all samples remained stable and the total metal thickness of 6 nm did not seem to allow for ionic diffusion to the underlying oxide layers. Previous report has shown that the as-deposited ALD–Al2O3 is chemically unstable in 1 M KOH.[31] The dissolution of Al2O3 may lead to the delamination of the metal overlayers, which may result in the degradation of photovoltage and subsequently in the deactivation of the photoelectrode. The XPS depth profiling equipped with ion etching on the device consisting of n-Si/SiO (1.8 nm)/Al2O3 (1 nm)/Pt (2 nm)/Ni (4 nm) has indicated that the Pt/Ni overlayers had transformed into Pt/Ni/NiO/Ni(OH)2 and that the Al2O3 remained intact after 18 h in 1 M KOH.[10] Without metal overlayers, no XPS peak corresponding to Al2O3 was observed after 1 h in 1 M KOH, indicating that the Al2O3 had rapidly dissolved in an alkaline solution.[10]

The theoreticn class="Chemical">al photovoltage (Vph) of an MIS photoanode can be calculated using the following equationwhere jph is the light-limited photocurrent density. The relation between the photovoltage and the barrier height can be obtained by substituting eq with eq The mean parameters were assumed to be the same as previously obtained for MIS devices with thick metal contacts: n = 1 and = 1.2 for the n-Si/SiO/‘inner metal’/Ni and n = 1.1 and = 1.6 for the n-Si/SiO/Al2O3/‘inner metal’/Ni photoanodes. As indicated in Figure , the theoretical model (eq ) predicts the photovoltages accurately for the given barrier heights, except for the n-Si/SiO/Al2O3/Ti/Ni (i.e., the two red circles at the bottom left corner because the effective tunnel exponent for this particular structure is slightly higher ( = 4), but using this value will actually give accurate results, see Figure S8, Supporting Information). Clearly, the increase of photovoltage upon the addition of Al2O3 onto the SiO interface layer is related to the lower dark saturation current, which depends not only on the barrier height but also on the oxide thickness. In the absence of an oxide interlayer, the photovoltage will be substantially lower (hypothetical relation between the photovoltage and the barrier height of metal–semiconductor photoanodes without an interfacial oxide is indicated by the dashed blue line).

Figure 6

Measured photovoltage as a function of the effective barrier height of MIS photoanodes using only SiO (black circles) and SiO/Al2O3 (red circles) interfacial layers. For comparison, the calculated photovoltage as a function of the barrier height using eq is shown (black and red lines). Parameters used for MIS systems using SiO only include an ideality factor n = 1 and a tunnel exponent = 1.2. For MIS devices with SiO/Al2O3, the n = 1.1 and the = 1.6. The hypothetical photovoltage of MS photoanode without an interfacial oxide is shown (blue dashed line).

Measured photovoltage as a function of the effecn class="Chemical">tive barrier height of MIS photoanodes using only SiO (black circles) and SiO/Al2O3 (red circles) interfacial layers. For comparison, the calculated photovoltage as a function of the barrier height using eq is shown (black and red lines). Parameters used for MIS systems using SiO only include an ideality factor n = 1 and a tunnel exponent = 1.2. For MIS devices with SiO/Al2O3, the n = 1.1 and the = 1.6. The hypothetical photovoltage of MS photoanode without an interfacial oxide is shown (blue dashed line).

Equation n class="Chemical">also suggests that the photovoltage will increase with a higher ideality factor. From the fit results of dark j–V curves in Figure , the mean ideality factor of the n-Si/SiO/Al2O3/‘metal’ systems is indeed higher than that of the same structure without Al2O3 (i.e., nSiO = 1.1, nSiO = 1). This is because the ideality factor increases with increasing oxide thickness, as expressed by[11,15]where εi is the dielectric permittivity of the interfacial oxide, εs is the dielectric permittivity of the semiconductor, and W is the space-charge width of the semiconductor. The Dit,s and Dit,m are the density of interface states in equilibrium with the semiconductor and the metal, respectively. For an MIS structure with a very thin oxide, such as the n-Si/SiO/‘metal’, the charge exchange between the interface states and the metal is relatively easier than with the semiconductor, and thus the population of interface states is controlled by the metal, so that Dit,s is almost negligible. In a situation where Dit,m is very large, the ideality factor will be close to unity. For thicker oxides, such as the n-Si/SiO/Al2O3/‘metal’, charge exchange between the interface states and the metal decreases by the total oxide thickness, and thus the interface states will partially equilibrate with the semiconductor. In a simple formalism where Dit equals to the sum of Dit,m and Dit,s,[32] introducing an Al2O3 to the existing SiO will increase Dit,s by the same amount that reduces Dit,m. As a result, the ideality factor will depart from unity.

So far, the analyn class="Chemical">sis presented above has assumed that the barrier height and the ideality factor remain constant under illumination. In reality, their values in the dark can be different from those under illumination. In the former case, the potential drop in the oxide layer is partly controlled by charging/discharging of interface states with applied bias, whereas in the latter case the potential drop is influenced by the injection of minority carriers into the interface states. Considering the excellent agreement between the experimental and theoretically calculated values of photovoltage, the change of barrier height and ideality factor should be negligibly small under illumination. Further complications that are not discussed here arise from the fact that the interfacial oxide can either increase or decrease the barrier height, depending on the polarity of the charge induced by the oxide. The Al2O3 is known to exhibit a fixed negative charge,[33−35] and if used with an n-type semiconductor can lead to a field-effect passivation of the underlying surface. This can also be the origin of the barrier height shift (Figure S9, Supporting Information) and thus the photovoltage improvement of the n-Si/SiO/Al2O3/‘inner metal’/Ni photoanodes.

Figure a,b show that increasing the n class="Chemical">oxide thickness improved the MIS junction photovoltage. For the n-Si/SiO/Al2O3/Pt/Ni photoanode, the photovoltage reached a maximum value of 538 mV as the Al2O3 thickness increased to 2.7 nm. Further increase of Al2O3 thickness diminished the photovoltage due to the decrease of barrier height (Figure S9, Supporting Information). As the Al2O3 became much thicker, the tunneling barrier also increased to such an extent that it suppressed the photocurrent. We note that the behavior of our device is distinct as compared to the previously reported n-Si/SiO2/Al2O3/Ir photoanode whose photovoltage rapidly drops with a small increase of Al2O3 thickness.[4,36] A relatively high photocurrent observed with an Al2O3 thickness of 3.9 nm (measured by ellipsometer) indicates that our Al2O3 is rather conductive. A high leakage current has been frequently observed in Al2O3 films deposited by thermal ALD, and has been experimentally attributed to the hydrogen incorporation during the deposition in a system involving hydrogen-containing oxygen precursor such as H2O.[37−40] According to density function theory calculations, interstitial hydrogen in Al2O3 can introduce a defect level near the midgap that enables electronic conduction through the oxide.[41] The carrier lifetime in this energy level is brief, and therefore the tunneling conductivity is sensitive to the oxide thickness. Since our Al2O3 films were deposited by thermal ALD and using H2O oxygen precursor, therefore, it is reasonable that the same mechanism of defect-mediated conduction within the hydrogen-contaminated Al2O3 film controls the carrier transport, which allows for the passage of relatively high photocurrent within thick tunnel oxide interlayers.

Figure 7

(a) Representatives j–V curves and (b) photovoltages of n-Si/SiO/Al2O3/Pt/Ni with various Al2O3 thicknesses. The photovoltage was determined by comparing the onset potential of the MIS photoanodes under illumination and the nonphotoactive p+-Si/SiO/Ni electrode in the dark.

(a) Representatives j–V curves and (b) photovoltages of n class="Chemical">n-Si/SiO/Al2O3/Pt/Ni with various Al2O3 thicknesses. The photovoltage was determined by comparing the onset potential of the MIS photoanodes under illumination and the nonphotoactive p+-Si/SiO/Ni electrode in the dark.

Improvements of photovoltage by controlling the interfacin class="Chemical">al SiO2 thickness have been observed in Schottky barrier solar cells[42−44] and MIS photoelectrodes.[29,45] However, many of these devices require high precision of interfacial oxide thickness with subnanometer accuracy to improve the photovoltage without compromising the tunneling current in the oxide layer. For highly resistive oxides such as SiO2, a small thickness increase of a few angstroms can lead to significant carrier recombination and photocurrent suppression, which will result in a reduced fill factor and a drop of photovoltage. On the other hand, the leaky characteristics of our Al2O3 offer the flexibility and a better control to optimize the oxide thickness without losing too much photocurrent. Other interfacial oxide materials that show similar tunneling properties include titanium dioxide (TiO2),[4,5,36,46] graphene oxide (GO),[47] and cobalt oxide (CoO).[48] This highlights the importance of using a conductive tunnel dielectric to achieve a highly efficient MIS photoelectrode.

Conclusion

In summary, we have demonstrated the role of interfacial n class="Chemical">oxides and metal workfunction for improving the photovoltage in MIS junction photoanodes. The use of a thin chemically grown SiO alone is evidently not sufficient to favorably shift the Schottky barrier height. Depending on the metal used, the thin SiO can facilitate the formation of extrinsic defect states that reduce the metal effective workfunction and subsequently decrease the barrier height. Adding a supplementary oxide such as Al2O3 to the interfacial SiO can eliminate the development of extrinsic defect states and simultaneously increase the barrier height which is advantageous for increasing the photovoltage. The improvement of photovoltage upon the addition of an interfacial oxide is also found to be partly a result of the reduced dark saturation current which is not solely determined by the barrier height but also by the oxide thickness. With an optimum thickness, the interfacial oxides reduce the majority-carrier thermionic emission current without significantly affecting the minority carrier-injection, thereby improving the photovoltage of the MIS photoanode.

The results and analyn class="Chemical">sis presented herein also attempt to serve as guidelines for designing bimetal–insulator–semiconductor photoanodes with realistically achievable high photovoltages. The photovoltage dependence on the inner metal workfunction indicates that a high workfunction metal such as Pt is fundamentally required to achieve a highly efficient MIS photoanode with bilayer metal contacts. Using a lower workfunction inner metal substantially sacrifices the photovoltage in a trade-off for a lower cost inherent to the natural abundance of the low workfunction metals. Most high workfunction metals such as Pt and Ir are indeed scarce, but the search for Schottky contact candidates should not be limited to materials in the metallic form. Transition metal oxides such as molybdenum trioxide (MoO3), chromium trioxide (CrO3), vanadium pentoxide (V2O5), tungsten trioxide (WO3), and nickel oxide (NiO) are relatively cheap and known to have high workfunctions, exceeding the workfunction of noble metals.[28,49] Therefore, research efforts should be guided toward the development and the use of these metal oxides as carrier-selective contacts for buried, rectifying junction photoanodes. Some of these oxides may have poor chemical stability and inferior catalytic activity, but with the rapid advancements in the protection and catalytic layer research and their successful implementations in a wide variety of photoelectrodes, a cheap, chemically stable, and highly efficient photoanodes can be achieved with all earth-abundant materials.