Resistance-based analysis of limiting interfaces in multilayer water splitting photocathodes by impedance spectroscopy

Experimental part Device fabrication Growth of polycrystalline Cu2O substrate by thermal oxidation: High-purity Cu foil (Jiarun Material, 99.9999%) with a thickness of 0.05 mm was oxidized following the method of Minami.1,2 The Cu foil was cut into small pieces of about 15 mm*15 mm. The Cu substrates were heated in a quartz tube furnace under argon from room temperature to 1030oC over 1 hour and kept for 1 hour at this temperature. Afterwards, air was introduced into the tube and the foils were oxidized at 1030 oC for 2 hours. Argon was again introduced into the tube and the sheets were further annealed at 1030 oC for 2 hours. Finally, the temperature was reduced to 500 oC in 1 hour and then allowed to cool naturally to room temperature under argon. The final thickness of the Cu2O film is 75 m.3


Introduction
Several promising photocathode absorber materials for PEC applications-normally prone to (photo)corrosion in aqueous electrolyte solution-can be stabilized by using protective overlayers such as TiO 2 , generating "buried junctions" where the photovoltage is generated at a solid-state junction underneath the semiconductor-electrolyte junction. Amongst these promising semiconductors are Cu 2 O, 1,2 CuO, 3 Si, 4-7 III-V semiconductors, 8,9 Cu 2 BaSnS 4 , 10 Cu 2 ZnSnS 4 , 11 and BiVO 4 . 12,13 Additionally, semiconducting buffer layers such as Ga 2 O 3 (ref. 2 and 14) or CdS 3 can be inserted between the photoabsorber and protection layers to increase the built-in voltage, the charge separation and the overall device efficiency. A scheme of these multi-layer structures is presented in Fig. 1.
Due to the high complexity of these photoelectrodes along with the involved charge carrier processes, in-depth routine analysis is challenging. So far, there are few reports on the systematic study on interface and charge transport limitations through the multilayered structure under operational conditions. A scheme of the most common photophysical and electrochemical processes taking place in PEC devices are depicted in Fig. S1. † Electrochemical impedance spectroscopy (EIS) is a powerful non-destructive characterization method widely applied in different elds such as solar cells [15][16][17][18] and PEC systems. [19][20][21] Devices can be investigated under operando conditions both in the dark and under illumination. In the dark, measurements are normally performed to facilitate a Mott-Schottky analysis, which gives access to the band alignment of the semiconductors as well as the doping density and type of the majority carriers (electrons or holes). [22][23][24] In contrast, under illumination, the light induced processes, which are normally dominated by the minority charge carriers, can be evaluated and their implications for the device efficiency understood.
In PEC research, EIS under illumination has been applied to a variety of photoelectrodes, though interpretation of the data is not trivial, since an equivalent circuit (EC) model is required that sufficiently mimics the underlying photophysical and electrochemical processes in the system under study. 19,21 Even for such a relatively simple and well investigated system as Fe 2 O 3 , several EC models have been developed, as surface treatments, nanostructuring and/or fabrication process conditions alter the properties of the material. 20,25-28 Furthermore, the EI spectra can be complicated to analyze if two or more processes take place in the same time domain, as they can usually not be separated during the tting procedure. On the other hand, the overall resistance of the two processes can be determined, which enables one to draw conclusions on their nature, especially if their potential dependence is investigated. Normally the capacitive elements distinguish the EC models, as the detected resistances are normally all in series (see Fig. S2 †). Here, we apply a resistance-based analysis using EIS as a facile and generally applicable method to quickly identify the problematic layers and/or interfaces in these complex device architectures.
We investigated two systems, a more traditional photoabsorber material with relatively simple architecture (p-Si/TiO 2 / RuO x ) as a well-behaved standard, and one based on the emerging material cuprous oxide (Cu 2 O/Ga 2 O 3 /TiO 2 /RuO x ), which features a buffer layer and therefore contains additional interfaces. The heterojunction with Ga 2 O 3 is of particular interest due to the high photovoltage that can be obtained (1.2 V). 29 The n-type layer in the investigated photocathodes that is in contact with the electrolyte solution serves as both a protective layer and as an electron accepting layer. Commonly, TiO 2 is used due to its good (photo)electrochemical stability over a wide range of pH. On the other hand, the built-in voltage between Si or Cu 2 O with TiO 2 is relatively low. By implementing ZnO or AZO, 30 ZnS, 31 or Ga 2 O 3 (ref. 14) as an interlayer, the built-in voltage and therefore also the photovoltage can be greatly increased for the Cu 2 O-based devices. 29 The increased photovoltage is reected in an earlier onset potential in J-V curves. The shi in onset potential of the photocurrent by the Ga 2 O 3 interlayer is between 500 to 600 mV as compared to the devices with a direct Cu 2 O/TiO 2 contact, reecting the increased built-in voltage in the semiconductor heterojunction.

Photocathodes in the dark
We rst performed a Mott-Schottky analysis of the two photocathode types and of the ALD Ga 2 O 3 to understand the energetic alignment of the different materials. The results on the built-in voltage (combined with the optical bandgap and the calculated distance of the E F to the valence and conduction band) are presented in Fig. 1. A more detailed description of the Mott-Schottky plots and the processing of the resulting data can be found in the ESI (Fig. S3, S4  ) have been implemented based on literature values and our own measurements (see Fig. S5 † and main text below). [32][33][34] By evaluating the band diagrams in Fig. 1, potential barriers for photogenerated charge carriers are revealed. Such unfavorable band bending can lead to a limited charge transfer for electrons at the junction if the space charge forming the barrier is not thin enough (preventing tunneling). Indeed, we can conclude on the basis of Fig. 1a (for silicon) and Fig. 1b (for Cu 2 O) that there is one limiting junction for the charge transfer of electrons from the TiO 2 to the catalyst, and in case of the Cu 2 O-based devices a second one between the Ga 2 O 3 and the TiO 2 (see red circles in Fig. 1b). Especially this latter one between the two n-type ALD layers could cause resistive limitations for the electron ow. We emphasize at this point that Mott-Schottky plot analysis must be taken as a rst approximation of the actual band alignments in the PEC photocathodes. For example, the conduction band edges of the Ga 2 O 3 with the photoabsorber and the TiO 2 will most probably not be symmetrical, but will still show the two potential barriers between Cu 2 O/Ga 2 O 3 and Ga 2 O 3 /TiO 2 as indicated in Fig. 1b.

Photocathodes under illumination
Typical J-V curves of the two systems are presented in Fig. 2. The photocurrent onset is shied $0.6 V for the Cu 2 O-based device, showing the higher built-in voltage as compared to the silicon devices. On the other hand, the J-V curve of the Cu 2 O-based devices is more resistive, showing a lower ll factor (FF). When plotting the photocurrent logarithmically, three distinct regions are observed, which are also found in the resistances extracted from the EIS spectra (Fig. 3). In region I, at most positive potentials, a small (photo)anodic current is observed (nA to mA cm À2 ). This phenomenon can also be observed for other photocathodes [35][36][37][38] and originates from the leakage current of the majority carriers. In region II, the electron quasi Fermi level is negative enough (on the RHE scale) so that the catalyst is reduced from RuO x to RuO x À1 by injecting electrons into the Gaussian shaped density of states (DOS) of the catalyst, showing photocathodic current (mA cm À2 ). 32 Still most of the photogenerated charge carriers are subject to recombination as the electrochemical potential of the H 2 evolution has not yet been reached. In region III, the hydrogen evolution current can be observed as the electron quasi Fermi level has reached the second reduction step of the catalyst and the hydrogen evolution potential. The protons present in the electrolyte solution then act as electron acceptor, now enabling the charge transfer and the photocurrent generation.
Comparing the different resistances determined for the Si and Cu 2 O devices by EIS analysis (see Nyquist and Bode plots in Fig. S6 and S7 † as well as Table S1 † for associated frequencies), we can also observe three distinct potential regions, which coincide with those of the photocurrent plots ( Fig. 3c and d). Furthermore, clear similarities in the potential dependent behavior of the resistances can be noted between the different photocathode types. The low frequency resistance R CT (charge transfer resistance) has a strong potential dependence for both photocathode types. Also, R TiO 2 and R SC (related to the TiO 2 layer and the recombination resistance inside the semiconductor junction, respectively) show clear changes with applied potential. R inter (discussed in detail below), in contrast, is only present in the Cu 2 O-based devices and is rather constant with potential, though showing a valley at $0.6 V vs. RHE and another one at The similarities between the resistances of the different photocathodes are even more evident when accounting for the different photovoltages in each system: aer shiing the data to give the same quasi-Fermi level of electrons (set by the onset of the hydrogen evolution current as determined by R CT at the beginning of region III), the J-V curves and extracted resistances of the Si and Cu 2 O devices overlap with one another, enabling a direct comparison (Fig. S8 †). The detection of the onset potential via R CT is more sensitive as compared to the extrapolation of the photocurrent to the voltage axis or by setting an arbitrary (low) photocurrent as a measure.
The low frequency resistance, R CT , mirrors the shape of the J-V curves in potential region III ( Fig. 3c and d) and represents the hydrogen-generating charge transfer over the hydrogen evolving catalyst (HEC) into the electrolyte solution (R CT -H 2 in Fig. 4). Prior to the onset of hydrogen evolution in potential region II, R TiO 2 shows a deviation from exponential behavior representing the increased current ow for the HEC reduction (Fig. S9 †). In part of this potential region, R CT is convoluted with R TiO 2 . For the Si-based devices, the separation of R CT and R TiO 2 could be achieved by extrapolating an exponential decrease of R TiO 2 and subtracting it in the potential region of the rst HEC reduction (R CT -RuO x in Fig. 4). In the case of the Cu 2 O devices, the separation did not yield reliable results due to the interference with the additional resistance R inter . R CT -RuO x itself mirrors the current ow for the rst reduction step of RuO x to form RuO x À1 (from Ru(IV) to Ru(III)), which is a prerequisite for hydrogen evolution. [32][33][34]39,40 Notably, the shape of R CT -RuO x reects the DOS of the catalyst. RuO x as the HEC can be seen as a protonic electrochemical capacitor due to H + exchange with the electrolyte solution: RuO x (OH) y + dH + + de À 4 RuO xÀd (OH) y+d (see also ESI and Fig. S5 † for further discussion on the redox states of the HEC). 39,40 The charge inside the HEC can be visualized by its storage and release process during transient photocurrent decay measurements (TPC). 32,41 In the TPC experiments, a steady state white light bias was applied (10% sun). The wavelengths of the light pulse to generate the photocurrent transients was either 447 nm ( Fig. 5a) or 530 nm (Fig. 5b), which led to an excitation of charge carriers near the surface of the photoabsorber. The possible excitation at 627 nm (red) is near the absorption onset of the Cu 2 O and would therefore probe the bulk of the semiconductor with the deeper penetration depth of this wavelength (see Niu et al.). 2 As the focus in the TPC experiments was the detection of the charge inside the HEC, we have used a perturbation pulse with 5 to 10% of a sun as light intensity (i.e. 50 to 100% of the actual white light bias). Therefore, our TPC measurements are not "small" perturbation experiments near equilibrium but rather photocurrent transients similar to the investigations by Laskowski et al. with a strong perturbation. 41 The TPC measurements were performed similar to the EIS measurements at different bias potentials to further observe the potential dependent changes of the accumulated charge. During the light pulse, normally a double-exponential decay of the photocurrent was observed. The fast component of the decay is associated with the recombination of photogenerated charge carriers while the slow one is related to the formation of

RuO x
1À or RuO x 2À species. 32 Aer the light pulse is turned off, the photocurrent steeply drops and changes sign before returning to the steady state white light bias current. The photocurrent with inverted sign represents the release of stored charge inside the HEC (the decay back to steady state white light bias current was normally mono-exponential). Examples of the TPC transients can be found in Fig. S10 † together with a table of the exponential ttings (Tables S3 and S4 †).
The charge in a current transient was determined from the integral of the exponential decay of the current density over pulse duration (Q ¼ Ð Jdt; with Q as the charge per cm 2 , J as the current density and t as time). The amount of charge determined in the TPC experiments for the storage (from the slow component of the photocurrent decay during the light pulse) and release (aer the light pulse has been turned off) process are very similar and show the capacitive loading of the HEC at its different redox states (Fig. 5, see also Table S5 †). As the release of stored charge is competing with the hydrogen evolution reaction, normally the amount of released charge detected is lower as compared to the initially stored charge, especially near the RuO x 2À redox potential.
R TiO 2 is a strongly potential dependent resistance at intermediate frequencies present in potential regions I and II (Fig. 3). At very positive potentials R TiO 2 dominates the photoanodic current ow. In region II R TiO 2 decreases exponentially but shows a deviation in the form of a valley as discussed above. Aer the formation of RuO x À1 , R TiO 2 further decreases exponentially. As both the Si and Cu 2 O-based systems possess the TiO 2 protection layer and the associated interfaces (TiO 2 /catalyst or the TiO 2 /electrolyte solution), one possibility is that R TiO 2 is related to a resistive transport inside the TiO 2 , or a second possibility is that it is related to a potential barrier at the aforementioned interfaces (for both of these cases, R TiO 2 manifests as a series resistance). To differentiate these options, we varied the protection layer thickness between 50, 100 and 150 nm. A transport related resistance should show a signicant increase for the thicker protection layers while a resistance related to a potential barrier at the TiO 2 /RuO x interface would not be inuenced by the different thicknesses of the TiO 2 layer.
The results for Cu 2 O and silicon (Fig. S11 †) did not show any signicant change in the resistance with increasing thickness of the ALD TiO 2 layer. A charge transport resistance within the TiO 2 can therefore be excluded. Considering the observation in Fig. 1a and b of the unfavorable band bending at the TiO 2 contact to the catalyst we can conclude that R TiO 2 represents the resistive limitation at this interface. The implications, however, for the overall device efficiency of this series resistance are limited to the photocurrent rise, as it is low or vanished when the surface potential approaches the thermodynamic HER potential. An additional observation is that the thickness of the TiO 2 layer did not have any signicant inuence on the V OC nor on the photocurrent. The resistance at the highest frequencies, R SC , is mainly constant in the potential regions I and II and increases in region III. It is related to the recombination current of the photogenerated charge carriers and the steeper its increase in potential region III the better the FF of the J-V curves. The increase of R SC with increasing reverse bias aer passing the H 2 evolution potential represents the reduced recombination of photogenerated electron hole pairs (i.e. higher recombination resistance) as now the band bending in the space charge of the photoabsorber increases and acceptor states are available inside the electrolyte solution to enable the ow of photocurrent. It should be noted that in PEC photoelectrodes, the system is effectively in open-circuit until the surface potential is in a region where charge transfer can occur. Thus, if the quasi Fermi level of the electrons has not reached the second reduction step of the catalyst and the energetic position of the hydrogen evolution potential, the photogenerated charge must recombine (i.e. low recombination resistance, potential region I and II). 42 The observation that R SC increases more slowly in the case of the Cu 2 O as compared to the Si samples (see Fig. 3 and S8 †) originates from an additional series resistance that is present in the metal oxide-based devices, namely R inter . So far, the three different resistances described have been observed in both photocathode types. In the case of the fourth resistance, R inter , observed only in the cuprous oxide devices, it stands to reason that it is related to the Ga 2 O 3 layer or the Cu 2 O surface (or interface). Though R inter shows as mentioned one valley around the second reduction of the catalyst, the valley at the rst reduction step is not always clearly visible and is therefore disregarded in the analysis.
Generally, one would expect a reduction of R inter similar to R TiO 2 with increasing negative potential if it is purely related to a potential barrier at the Ga 2 O 3 /TiO 2 interface. Compared to R TiO 2 this reduction of the barrier would appear only aer the electron quasi Fermi level reaches the redox potential of the H 2 evolution potential, as only then the barrier at the Ga 2 O 3 /TiO 2 interface would begin to reduce. This behavior is what we can observe in the descent into the valley starting from 0.8 V vs. RHE in Fig. 3d and 6 (for devices with Ga 2 O 3 layer).
However, R inter starts to increase again aer passing 0.6 V vs. RHE, indicating that it is a superposition of different contributions. To better understand R inter , Cu 2 O-based devices without the ALD Ga 2 O 3 layer were fabricated and evaluated. In Fig. 6, the J-V curves and R inter of devices without and with 20 and 70 nm ALD Ga 2 O 3 layer are plotted before (a and c) and aer (b and d) adjusting to the H 2 evolution onset potential by R CT . As expected, the built-in voltage (see Fig. S3b †) and the photovoltage for devices without Ga 2 O 3 layer are lower but interestingly R inter is still present, albeit with a different potential dependence. It has a constant and low value in potential region I and II at positive potentials before the onset of the photocurrent ($90 ohm cm 2 as compared to $1000 ohm cm 2 for the case with 20 nm Ga 2 O 3 interlayer, Fig. 6), and then increases at more negative potentials.
This observation shows that the Cu 2 O surface properties are of crucial importance for high device efficiencies, as surface traps or an unwanted overlayer are limiting the charge transfer to the n-type semiconductor (discussed in more detail below). The magnitude of the R inter (for devices with Ga 2 O 3 layer) in potential region II correlates with the increased V OC observed in the devices. And in potential region III R inter represents the series resistance that dominates the ll factor of the J-V curves (Fig. S12 †), pinpointing the main limitation for the FF of the Cu 2 O photocathodes.
These observations on R inter (series resistance in Region III, V OC correlation in region II, and existence even without the Ga 2 O 3 interlayer (though in latter case with a different potential dependence)) imply that it is composed of at least two components (see Fig. S13 †). One component, as visible for the case with 0 nm of Ga 2 O 3 , must be related to the Cu 2 O surface properties (R surface ). The second contribution must be related to the ALD Ga 2 O 3 interlayer (R Ga ) at the Ga 2 O 3 /TiO 2 interface. The height of R Ga in potential region II is accompanied with a higher built-in voltage and V OC (see also Mott-Schottky plots in Fig. S3b †). With 20 nm Ga 2 O 3 still no bulk of the buffer layer can be formed as the layer is to thin and the full heterojunction potential is therefore lower as compared to 70 nm of buffer layer (in which the bulk can be already formed, see Fig. S14a-c †). Higher built-in voltage in turn increases the barrier height at the Ga 2 O 3 /TiO 2 interface and the associated resistance.
To better understand the potential dependence of R Ga we have to consider the interplay of the redox potential in solution, the applied potential and the position of the conduction bands of Ga 2 O 3 and TiO 2 . Before the H 2 onset potential, the barrier between Ga 2 O 3 and TiO 2 remains unchanged (Fig. S14d †). At the onset of the H 2 evolution the potential barrier at the interface between the Ga 2 O 3 and the TiO 2 will start to decrease as the Fermi level inside the TiO 2 layer is now pinned by the redox potential of the hydrogen evolution potential inside the electrolyte solution. On the other hand, the conduction band of Ga 2 O 3 will continue to move to higher energies with scanning to more negative potentials ( Fig. S14e and f †), reducing the barrier between the two n-type metal oxides. This will manifest as an exponential decrease of the R Ga if it is related to a potential barrier (as for the case of R TiO 2 ).
Indeed, R Ga is, like R TiO 2 , reducing exponentially with increasing reverse bias, limiting its inuence on the J-V curve mainly to the region of the hydrogen evolution current onset (see Fig. 6d). The relation between the built-in voltage and the change of R Ga further supports our interpretation that it represents the potential barrier predicted by the Mott-Schottky plots (in Fig. 1d).
It is known that Cu 2 O in ambient conditions undergoes surface oxidation to Cu 2+ species such as Cu(OH) 2 or CuO, which can limit the performance of the devices. 43 According to Lee et al. 29 a reduction of the Cu 2+ species by the gallium ALD precursor is possible as the authors have shown by XPS measurements which should improve the surface quality. Still we have implemented an additional etching of the Cu 2 O surface as a rst cleaning process. 43 Therefore, the as-prepared Cu 2 O sample was immersed in nitric acid (1 : 1 of H 2 O and 65% HNO 3 ) for 10 seconds, followed by rinsing with DI water and sonication in isopropanol for 30 seconds. Finally, the foil was dried under nitrogen ow and immediately transferred into the ALD chamber for the ALD buffer layer deposition.
In Fig. 7, the results for devices with and without etching are presented. Devices were measured without the Ga 2 O 3 layer to avoid superposition with the R Ga component of R inter . The performance of the pre-etched devices showed an increased FF due to the reduced R surface at the Cu 2 O/TiO 2 contact (which could also be observed for the case with ALD Ga 2 O 3 buffer layer, Fig. S15 †). Nevertheless, R surface is still present aer the etching, implying that the surface treatment is not sufficient to achieve a defect free interface. This resistance is probably induced by an energetically low-lying energy state at the surface through which the electrons must ow in order to cross over to the buffer layer. In contrast to R Ga , the Cu 2 O surface resistance, R surface , would not reduce with further reverse bias. It increases as the conduction band of Ga 2 O 3 shis up at more negative applied potential, increasing the barrier for the electrons that have to be transferred through the limiting Cu 2 O surface layer (see Fig. S14d-f †).
Combining the knowledge about the potential dependence of R inter with and without the Ga 2 O 3 layer, we can conclude that R inter is the source of the main limitations observed in the J-V curves. We can achieve high photovoltages, but the ll factor is limited by the resistive components of R inter . Our results can explain now e.g. why Lee et al. observed for their Cu 2 O/Ga 2 O 3based solar cell devices high photovoltages of 1.2 V but a hampered FF of only 45%. 29 The resistances observed in the EIS spectra under illumination have been assigned to the relevant charge carrier processes, and we can understand their inuence on the J-V curves, a precondition to make sophisticated routine analysis and to facilitate the development of a generally applicable EC. Further investigations are being conducted to tackle the limitations observed in the Cu 2 O devices. On the one hand, we expect that the decrease of the space charge layer width through increased doping of the Ga 2 O 3 layer would reduce the potential barrier and therefore the associated resistance for the electrons at the Ga 2 O 3 /TiO 2 interface, and may enable both high V OC and FF. Furthermore, the higher doping would enable to achieve a higher built-in voltage at lower buffer layer thickness. On the other hand, with the presented analysis approach, we can evaluate the effect of different surface treatments on R surface and optimize the surface cleaning process, which should nally lead to a better FF and overall device efficiency.

Conclusion
We have investigated silicon-and Cu 2 O-based multilayer photocathode devices by EIS in the dark and under illumination to pinpoint the limitations for the J-V curves and device efficiencies. Based on the dark measurements we have identied potential barriers at the TiO 2 /RuO x and Ga 2 O 3 /TiO 2 interfaces. Under illumination conditions we have analyzed the EIS data in terms of the potential dependence of the determined resistances. We have shown that shiing the data to align the hydrogen onset potential in the different samples (determined by the onset of the hydrogen evolution via the charge transfer resistance) enables the comparison of similar device types with different photovoltage and even different photocathode absorber materials. The resistances were assigned to the related photophysical or electrochemical processes, namely, the recombination of photogenerated charge carriers in the bulk or space charge region of the semiconductor, the charge transfer into the electrolyte solution, and the limited charge transfer due to a potential barrier between the TiO 2 and the electrolyte solution. In the case of the Cu 2 O-based devices, an additional resistance appears, convoluting several different processes such as charge transfer from the absorber to the n-type semiconductor and a potential barrier at the Ga 2 O 3 /TiO 2 interface. Based on our analysis we can conclude that this emerging photocathode material is hampered by its surface quality (i.e. interface defects), limiting the charge transfer between the Cu 2 O and the buffer layer and therefore the ll factor for these devices. The series resistance associated with the Cu 2 O surface could be reduced by a commonly applied etching treatment of the Cu 2 O surface, but our analysis shows directly that such a pre-treatment is not sufficient to achieve an ideal or nearly ideal surface that enables reproducibly high efficiency devices. We further identied a potential barrier for the photogenerated electrons at the Ga 2 O 3 /TiO 2 interface which limits the rise of the photocurrent at the onset of the hydrogen evolution. By this analysis scheme we have opened the way for a fast and better understanding of the interplay of the different materials and interfaces and how to evaluate optimization treatments in complete multilayer device structures. Finally, the knowledge gained on the determined resistances and their associated photophysical and electrochemical processes will advance the development of a generally applicable EC of these multilayer photocathode systems.

Conflicts of interest
The authors declare no competing nancial interest.