Photoelectrochemical (PEC) water splitting is a promising technology for converting solar energy into clean and storable chemical energy. In PEC cells, semiconductors play a key role in absorbing photons from the light source to create mobile charge carriers. Various semiconductor materials have been studied for the high performance PEC cells, including metal oxides1–4, nitrides5, 6, Si7–13, Ⅲ-Ⅴ compound semiconductor materials14–16, and others17, 18. Among these, Si-based photoelectrodes have attracted substantial interest due to silicon’s moderate bandgap (1.12 eV), high charge mobility and diffusion lengths, and well established technological infrastructure19. However, Si-based photoanodes for the oxygen evolution reaction (OER) remain challenging to engineer due to the complex four-electron reaction mechanism which requires a large overpotential, and poor chemical stability in alkaline solutions. To improve the OER performance of Si-based photoanodes, metal-insulator-semiconductor (MIS) structures have been widely used for Si-based photoanodes due to their high efficiency and improved stability9–11, 20. An efficient metal catalyst at the surface improves the reaction kinetics of photoanodes, reducing the overpotential for OER. Corrosion of Si in aqueous electrolytes is suppressed in MIS photoelectrode structures by protecting the Si surface using thin layers of insulators such as TiO213, 20, 21, NiOX9, 11, 12, SrTiO38, SiNX22, and SiOX9–11.
For a Si-based MIS photoanode, minority carriers are generated by illumination of the semiconductor and, typically, extracted to the metal layer by tunneling through the insulator, as shown in Fig. 1a. The tunneling current density decreases exponentially with increasing insulator layer thickness so that, with few exceptions10, ultrathin insulators under 5 nm are needed for efficient MIS photoanodes23. However, it has also been reported that MIS photoelectrodes with thin insulator layers can be susceptible to corrosion of the semiconductor in alkaline solutions, so that insulator layer thicknesses above 50 nm are required for long-term stability in alkaline solutions24. Due to this trade-off between efficiency and stability, optimizing the insulating layer thickness is key to the OER performance of MIS photoanodes.
Our previous work demonstrated a new approach to increasing the effective conductivity of a thick insulating layer in MIS photoelectrodes by electrically inducing localized dielectric breakdown of the insulating region beneath metal catalyst islands10. Since the breakdown process resulted in formation of localized conductive paths through the thick oxide layer, photogenerated minority carriers are easily transported via the conductive path instead of by tunneling. Using this approach, improved photocurrents were observed with good long-term stability due to the thick insulating layers. However, this method requires complex and time-consuming processing techniques for the formation of localized breakdown paths, along with lithographic patterning of the top metal layer. Recently, Loget et al.11 and Zhao et al.12 reported on the deposition of Ni catalysts for the metal layer of MIS photoanodes using electrodeposition and electroless deposition, respectively. These methods yielded well defined Ni deposition for the MIS photoanode and enabled efficient OER performance without using complex lithographic techniques. However, stability in high pH aqueous solutions was limited to 40 hours or considerably less, potentially due to the absence of an insulating protective layer of sufficient quality and thickness.
Herein, we demonstrate a low cost and highly scalable method for producing high-performance and very stable Si-based MIS photoanodes by exploiting a thin-film reaction of Al through an insulating oxide layer followed by Ni-electrodeposition, and not requiring any lithographic patterning. The thin-film reaction of Al with SiO2 or Si, leading to localized penetration of Al “spikes” into the underlying material, has been studied extensively since it can cause an electrical short in silicon pn-junction structures with Al contact metallization. Al spiking can also occur through an insulating SiO2 layer, and has been exploited to form Ohmic contacts through an oxide passivation layer on Si25, 26. In an MIS photoelectrode structure with Al as the metal layer, annealing of the Al/SiO2/Si structure above 300 ℃ causes Al to penetrate the underlying SiO2 and induces formation of localized metal spikes in the MIS structure. As shown in Fig. 1b, each metal spike provides a local conductive path through the thick SiO2 layer. Typical spike densities are ~ 108–109 cm− 2, enabling very efficient collection of photogenerated carriers. The surrounding regions of oxide remain electrically insulating and retain their protective functionality. In this work, after the formation of localized Al spikes through the SiO2 layer, the Al is etched and replaced, via electrodeposition, by Ni, which serves as the OER catalyst. During the electrodeposition process, Ni covers the exposed Si surface resulting in growth of dispersed Ni nano-islands on the SiO2 surface at the corresponding locations. The remaining exposed thick SiO2 and the electrodeposited Ni have excellent corrosion resistance in alkaline aqueous solutions. This process results in formation of Si-based MIS photoanodes with high efficiency, good long-term stability, low cost, and high manufacturability without using any complex and costly lithographic patterning techniques.
Al thin-film reaction on Si-based MIS photoanode
Figure 1c shows the fabrication process for a Si MIS photoanode using the Al/SiO2 thin-film reaction to form localized Al spikes. As shown schematically in Fig. 1b, to help ensure long-term stability of the photoanode, 90 nm thick SiO2 layers were formed by thermal oxidation to serve as the insulator of the MIS structure24. A 90 nm thick SiO2 layer also provides low optical surface reflectance in water in the 400 ~ 600 nm wavelength range. To form localized Al paths through the SiO2 insulating layer, 100 nm Al was deposited by DC sputtering, followed by annealing. During the annealing process, the Al penetrates locally through the SiO2 layer and contacts the Si substrate. Since this Al spiking thin-film reaction occurs in localized areas across the whole surface, an array of localized contacts between the metal and semiconductor layers is formed within the Al/SiO2/Si structure. Since the Al film is easily corroded in solution, however, it is not suitable for the metal layer of an MIS photoanode. Therefore, the Al was replaced by Ni, which shows good stability and acts as an OER catalyst. The Al layer was first etched by Al etchant, leaving exposed the unreacted SiO2 surface and the Si surface underlying the areas where the spiking Al reaction occurred. After etching the Al, Ni was deposited on the resulting surface by electrodeposition. During this process, Ni fills the exposed Si areas first, since higher electric fields are present at the exposed Si surface, and then forms Ni nano-islands. As a result, a Ni/SiO2/Si MIS photoanode with localized Ni conductive paths through the oxide can be fabricated using the Al thin-film reaction followed by etching and Ni electrodeposition.
To optimize the Al thin-film reaction and spiking process, a series of Al/SiO2/Si/SiO2/Al samples was prepared by 90 nm SiO2 growth and 100 nm Al deposition on both sides of Si substrates, and their series resistance from top to bottom (top-bottom resistance) was measured before and after different annealing processes. It has been reported that, in an Al/SiO2/Si structure, Al reduces the underlying SiO2 layer and penetrates to the Si layer during annealing, the phenomenon referred to as Al spiking25–27. The formation of Al spikes requires around 2.56 eV of activation energy for the SiO2 reduction process, and can occur at temperature above 300 °C27. The top-bottom resistance changes were measured for annealing temperatures of 450 ~ 600 °C and durations of 0 ~ 24 hours as shown in Fig. 2a. Before the annealing process, the top-bottom resistances of the Al/SiO2/Si/SiO2/Al substrates were relatively high (20 ~ 30 Ωcm2) due to the thick SiO2 insulating layers. In the red-shaded region of Fig. 2a, the high resistance was maintained for the first several hours for all annealing temperatures. However, as the temperature and duration of annealing increased, there was a sharp transition in resistance to ~ 10% of its initial value (green region in Fig. 2a). After this transition, the resistance reaches ~ 0.5% of its initial value (blue region in Fig. 2a) for longer annealing time and/or higher annealing temperature. During the annealing process, Al penetrates locally through the SiO2 layers and reaches the underlying Si substrate, forming a metallic path and Ohmic contact between the Al and Si layers as shown in Fig. 2b. Formation of these structures causes the abrupt drop in top-bottom resistance at the transition points. Before the annealing process, the Al layer exhibits uniform morphology on the surface (Fig. 2c). After annealing of an Al/SiO2/Si structure at 550 °C for 24 hours, localized Al spikes were observed on the surface (Fig. 2d). After etching the Al layer from the Al/SiO2/Si substrate, SiO2/Si surfaces before and after the annealing process were characterized again using plan-view scanning electron microscopy (SEM). Without any annealing, a uniform SiO2 surface was observed after removal of the Al layer. After annealing and subsequent etching away of the Al, random localized voids in the SiO2 layer, corresponding to areas where Al spiking occurred, were observed on the Al-etched SiO2/Si surface with ~ 1.7% surface coverage by the void regions and with void diameters ranging from 50 to 100 nm, corresponding to an areal density of voids of ~ 2–8 × 108/cm2. (Supplementary Fig. 1) The corresponding average distance between adjacent voids is considerably smaller than the typical diffusion length of minority carriers in c-Si, enabling efficient collection of photogenerated holes during photoanode operation. This is a natural consequence of the nature of the Al spiking reaction into thermally grown SiO2, and enables the desired spatial distribution of conductive paths to metal catalyst layers to be realized without deliberate lithographic patterning.
Electrodeposition of Ni catalyst
For MIS photoanodes, the metal catalyst layer is essential in enhancing overall reaction rate, reducing the onset voltage, and increasing the current density. Ni is frequently used as an OER catalyst due to its good electrical conductivity, efficient catalytic effect, and corrosion resistance at high pH11. In this study, Ni was incorporated as a catalyst for the photoanode using electrodeposition, following the formation of localized voids in the SiO2 protective layer by Al thin-film reaction and etching. Other catalyst materials could readily be employed using this approach. The morphology of Ni deposited on the surface can be controlled by varying the applied bias and deposition time for electrodeposition. When negative bias is applied to the substrate, Ni deposition begins by filling the voids in the SiO2 layer, since the highest electric fields are present at the interface between the aqueous solution and the exposed Si surface. Ni is nucleated at these locations and eventually forms larger islands on the surface. Figure 3a-c shows SEM images of Ni electrodeposited on the SiO2/Si surface for different electrodeposition conditions. When low electrodeposition bias (-0.5 V) was applied, Ni nano-islands were observed on the surface and their number and size increased with increasing applied bias and electrodeposition time. Shown in Fig. 3d is a box plot of the diameters of the Ni nano-islands after 80 min of electrodeposition at applied bias voltages of -0.5 V, -1.0 V, and − 2.0 V. The mean diameter of the Ni nano-islands increased from 0.26 µm to 0.83 µm as the applied bias increased from − 0.5 V to -2.0 V. In addition, a broader diameter distribution was observed for the Ni nano-islands deposited at more negative applied bias. Growth and eventually coalescence of the Ni nano-islands occurs with increasing Ni coverage. Figure 3e shows the Ni coverage on the SiO2 layer as a function of electrodeposition time. At -0.5 V applied bias, the Ni coverage increased very slowly, not exceeding 5% after 120 min electrodeposition time. When higher bias was applied, clearer increases in Ni coverage were observed with increasing electrodeposition time and, after 120 min, ~ 78% and ~ 100% Ni coverage was obtained at -1.0 V and − 2.0 V, respectively. As shown in Fig. 3c, after 120 min electrodeposition at -2.0 V, the Ni aggregates form a continuous Ni film with a thickness of ~ 0.3 µm that covers almost the whole surface of the SiO2 layer. We refer to photoanodes fabricated in this manner, with Ni metal catalysts penetrating the SiO2 protective layer via voids created by the Al spiking and etching processes, as spiked Ni/SiO2/Si photoanodes.
PEC performance for the Ni/SiO2/Si photoanodes
PEC performance was measured for spiked Ni/SiO2/Si photoanodes fabricated using various Ni electrodeposition processes in 1M KOH (pH = 14) aqueous solution with a standard three-electrode system under standard AM 1.5G 1sun illumination. Figure 4a shows linear sweep voltammetry (LSV) measurements with chopped illumination for Ni/SiO2/Si photoanodes fabricated with and without the Al thin-film reaction process. Both photoanodes were fabricated with the same Ni coverage on the surface. For the spiked structures, the Ni layer was deposited by electrodeposition at -2.0 V applied bias for 60 min yielding ~ 75% surface coverage with electrodeposited Ni. For the photoanode fabricated without Al spiking, the Ni layer was deposited using e-beam evaporation with the same 75% coverage as the electrodeposited Ni layer using a lithographically defined pattern consisting of 60 µm diameter dots in a square array with 60 µm pitch. As shown in Fig. 4a, the Ni/SiO2/Si photoanode without spiking showed very low current density, under 10 µA/cm2. Clear OER activity was observed with the spiked structure, which showed a low onset potential of 1.0 V versus reversible hydrogen electrode (RHE) and a high saturation current density of 25 mA/cm2. This result confirms that the spiked structure dramatically improves the performance in aqueous solution of the Ni/SiO2/Si photoanode by providing conductive paths through the SiO2 layer. All results reported below are for spiked Ni/SiO2/Si photoanode structures fabricating using the Al thin-film reaction process.
Figure 4b shows the PEC performance of spiked Ni/SiO2/Si photoanodes for different electrodeposited Ni morphologies on the surface. We prepared three spiked Ni/SiO2/Si photoanodes with different electrodeposition recipes. The measured Ni coverages of the photoanodes were 3.9%, 34.4%, and 80.2% for 80 min of Ni electrodeposition at -0.5 V, -1.0 V, and − 2.0 V applied biases, respectively. The spiked Ni/SiO2/Si photoanode with 3.9% Ni coverage showed poor PEC performance due to insufficient Ni coverage on the surface. The highest saturation photocurrent density (27 mA/cm2) was observed for 34.4 % Ni coverage, achieved by Ni electrodeposition at-1.0 V bias. For the spiked Ni/SiO2/Si photoanode with 80.2% Ni coverage, a lower saturation photocurrent density (16 mA/cm2) was observed compared to the photoanode with 34.4% of Ni coverage, a reduction we attribute to blocking of incident light by the Ni layer. However, an enhancement in onset potential for OER was observed, from 1.1 V to 0.9 V versus RHE, for 80.2% Ni coverage on the surface compared to that for 34.4%. This result indicates that, unsurprisingly, the catalytic effect of Ni was improved with increasing Ni coverage, leading to lower onset potential. However, excessive Ni coverage limits light absorption in the underlying Si absorber, reducing the photocurrent density.
The corrosion resistance of the spiked Ni/SiO2/Si photoanodes at high pH was assessed by chronoamperometry (CA). As shown in Fig. 4c, CA tests for spiked Ni/SiO2/Si photoanodes with different Ni electrodepositions at -1.0 and − 2.0 V were performed in 1M KOH (pH = 14) aqueous solution at 1.3 V versus RHE for 48 hours. Due to the 90 nm thick SiO2 layer and the high corrosion resistance of Ni for high pH solutions, both spiked Ni/SiO2/Si photoanodes showed excellent stability over the entire 48 hour measurement duration. Due to the higher rate of gas evolution for the photoanode with Ni electrodeposition at -2.0 V, more fluctuations in photocurrent density were observed during the stability test. A small decrease in photocurrent density was observed for both photoanodes during the stability test, since some of the produced O2 bubbles remain attached to the surface and suppress the OER reaction. After drying the surface using Ar dry gas and changing to a new 1M KOH aqueous solution, the photocurrent density recovered to the initial values, indicating that there was no chemical corrosion on the surface.
Numerical analysis of photoanode potential distributions
The effects of localized contacts in the spiked Ni/SiO2/Si photoanode were also analyzed computationally using a commercial numerical solver, COMSOL Multiphysics, to help explain the favorable onset potentials observed. Three different models were simulated, consisting of different Ni/SiO2/Si structures with metal back contacts (Fig. 5a). Models 1 and 2 were designed as typical MIS structures without metal spikes, with SiO2 thicknesses of 5 nm and 90 nm, respectively. Model 3 had a similar structure to Model 2, but also included a cylindrical metal spike with a diameter of 60 nm and a Schottky contact between Ni and Si at the bottom of the spike. Figure 5b, c shows the simulated band structures and hole concentrations of Models 1, 2, and 3. The electrical behavior of Models 1 and 2 is readily explained by the MIS capacitor model28. The band structure of a typical MIS contact with a very thin insulating layer is nearly Schottky-like, as shown in Fig. 5b. Due to the high work function of Ni (5.0 eV) and small potential difference across the thin SiO2 layer, there is weak surface inversion at the Si surface and a favorable onset potential would be expected. With increasing thickness of the SiO2 layer, even weaker surface inversion is observed with less band bending in the Si layer. Therefore, a lower concentration of holes accumulates at the Si surface for Model 2, which has a thicker SiO2 layer, causing decreased photovoltage28. As shown in Fig. 5c, two types of interfaces are present in Model 3 depending on radial distance (R) from the spiked area: Si/Ni at R = 0 nm and Si/SiO2/Ni at R = 200 nm. For R = 200 nm in Model 3, the simulated band structure is almost same as for Model 2. In the case of R = 0 nm, Ni is in direct contact with Si, with an 0.67 eV Schottky barrier, leading to stronger surface inversion and higher local hole concentration at the Si surface. Figure 5d shows the simulated conduction-band edge energy (EC) profile as a function of the radial distance (R) from the center of the spiked area and the depth below Si surface. The Si conduction-band edge energy decreases by ~ 0.5 eV as R increases from 0 to 200 nm, so that higher hole concentrations are present in the Si/SiO2/Ni structures primarily near the spiked area. In the MIS photoanode with a localized spike structure, the localized high hole concentration on the Si surface is sufficient to lead to a photovoltage improvement of the photoanode compared to that expected for structures with no spiking.
Spiked Ni/SiO2/Si photoanode with p+n-Si
To further improve the photochemical onset potential, we used a p+n-Si substrate for the spiked Ni/SiO2/Si photoanode. It has been reported that thin p+ doping on the n-Si surface improves the OER performance of MIS photoanodes due to the higher hole density at the Si surface28, 29. The p+ doping on the surface was performed by boron diffusion with an expected junction depth of ~ 300 nm, since a junction depth above 1 µm is expected to increase interface recombination28. Figure 6a shows LSV measurements with chopped illumination for spiked Ni/SiO2/Si photoanodes fabricated using p+n-Si and n-Si substrates with similar Ni coverage on the surface (~ 35%). The spiked Ni/SiO2/p+n-Si photoanode showed enhanced OER performance compared to the spiked Ni/SiO2/n-Si photoanode, with a lower onset potential of 0.7 V versus RHE and higher saturation current density of 32 mA/cm2. In addition, as shown in Fig. 6c, highly stable photocurrent density was observed in a CA measurement for the spiked Ni/SiO2/p+n-Si photoanode in 1M KOH aqueous solution at 1.3 V versus RHE, with a constant photocurrent density of 21–22 A/cm2 maintained over the entire duration of a 7-day stability test. For the calculation of Faradaic efficiency, the evolved H2 and O2 gases were measured for the spiked Ni/SiO2/p+n-Si photoanode at 1.23 V versus RHE. The generation rates of H2 and O2 gases exhibited the expected stoichiometric ratio (2:1). The calculated Faradaic efficiency started at 72.7% and saturated near 86% at 20 min, indicating good stability for the 120 min measurement (Fig. 6b).
The improvement in OER performance for the spiked Ni/SiO2/p+n-Si photoanode was analyzed by simulating its band structure for a gaussian acceptor doping profile corresponding to a boron diffusion process from the surface with junction depth of 300 nm, referred to as Model 4 (Supplementary Fig. 3). Figure 6d shows the simulated band structures of Model 4 with R = 0 nm and R = 200 nm. The spiked (R = 0 nm) and non-spiked (R = 200 nm) areas in Model 4 showed almost the same band structures, since the p+-Si and Ni form an Ohmic contact and band bending in the Si is dominated by the p+n junction. The p+ doping at the n-Si surface enables a higher hole density to accumulate at the Si surface, as shown in Fig. 6e, leading to improved onset potential compared to the spiked Ni/SiO2/n-Si photoanode.