In Operando Deconvolution of Photovoltaic and Electrocatalytic Performance in ALD TiO2 Protected Water Splitting Photocathodes

In this work, we demonstrate that buried junction photocathodes featuring an ALD TiO2 protective overlayer can be readily characterized using a variation of the dual working electrode (DWE) technique, where the second working electrode (WE2) is spatially isolated from the hydrogen-evolving active area. The measurement of the surface potential during operation enables the operando deconvolution of the photovoltaic and electrocatalytic performance of these photocathodes, by reconstructing J–DV curves (reminiscent of photovoltaic J–V curves) from the 3-electrode water splitting data. Our method provides a clearer understanding of the photocathode degradation mechanism during stability tests, including loss of the catalyst from the surface, which is only possible in our isolated WE2 configuration. A pnSi/TiO2 photocathode was first investigated as a well behaved model system, and then the technique was applied to an emerging material system based on Cu2O/Ga2O3, where we uncovered an intrinsic instability of the Cu2O/Ga2O3 junction (loss of photovoltage) during long term stability measurements.


MAIN TEXT
Photoelectrochemical (PEC) water splitting has been recognized as a promising avenue for harvesting renewable hydrogen fuel from inexhaustible solar energy and water. [1][2][3][4] One approach to achieve larger open-circuit voltages (V oc ) is using so-called "buried junctions", which can be modeled as a series combination of a p-n junction photoabsorber, a protective layer, and surface catalyst (pn/cat). 5,6 In such a situation, the V oc is decoupled from the semiconductor-electrolyte interface and the increased band bending of the p-n junction can significantly enhance electron-hole pair separation. 7 The efficiency of a pn/cat photocathode is largely determined by the intrinsic properties of the buried p-n junction. 8 The semiconductor-catalyst and catalystelectrolyte interfaces also play a critical role in the overall performance of the photelectrodes. Issues such as charge transport in the protective layer, the nature of the semiconductor-catalyst contact (Ohmic or Schottky-type), as well as the chemical reactions occurring at the catalyst-electrolyte interface are intertwined. 9,10 Therefore, it is highly desirable to develop an experimental technique that can not only evaluate the PEC performance but also simultaneously provide an understanding of these different interfaces during PEC operation.
The dual working electrode (DWE) technique was first reported in the 1970s.
Photoanodes coated with thin gold films showed a photovoltaic effect, representing early examples of in situ measurements of the surface potential. [11][12][13] Recently, the Boettcher group has used the DWE technique to study a photoanode-catalyst interface. 14 However, the second working electrode in all of the previous works has either been a transparent conducting oxide (TCO) or a thin metal film that covers the entire active area. 15,16 For systems that do not employ TCOs as part of the buried junction structure, it has thus far not been possible to carry out DWE studies without introducing a metallic film, which influences the measurement through partial light absorption and by affecting the catalyst binding to the photoelectrode surface. It is therefore essential and advisable to measure the device in the same configuration as would be in the actual PEC cell for unambiguous interpretation of the results.
Here, we demonstrate that ALD TiO 2 , a common protective layer for water splitting photocathodes, enables the surface potential to be measured at a position that is remote from the active area, due to the high doping and conductivity of the film. 17,18 . A pn + -Si photocathode with 100 nm-thick ALD-TiO 2 protective layer was chosen as a model system for the development of the DWE technique. A schematic diagram of the DWE setup is depicted in Fig. 1a. WE1 is used for controlling the back contact potential (V1) of the photocathode. WE2 is connected to the photocathode surface and kept at open circuit during PEC measurements to directly probe the surface potential (or in other words to probe the energy level of surface-accumulated electrons) in relation to the reference electrode (V2). To avoid direct contact of WE2 and the active area, WE2 was contacted a small distance away from the illuminated area and separated by a thin coating of opaque epoxy, as shown in Fig. 1b. When illuminated, a photovoltage is created across the pn + -Si homojunction. Considering that both the n + -Si/TiO 2 and TiO 2 /Pt are ohmic contacts, 17,19 the measured potentials of V1 and V2 directly give the energetic positions of the quasi-Fermi level of holes and electrons, respectively. The difference between V1 and V2, denoted ΔV, is the output voltage of the pn + Si junction. It is worth noting that as the photoelectrons diffuse away from the illuminated area (as shown in Fig. 1b), the electrons cannot enter the electrolyte and will ultimately recombine. This leads to a lower electron density than in 5 the illuminated area and a slightly reduced V oc is recorded. The small drop in the measured surface potential is robust and reproducible, and does not complicate the analysis herein.
After the Pt catalyst deposition, the conventional current density-back contact potential (J-V1) curve of the pn + Si/TiO 2 /Pt(ed) photocathode was obtained in 0.5 M H 2 SO 4 with a linear sweep from positive to negative potential (Fig. 1c). Under one sun illumination, the pn + Si/TiO 2 /Pt(ed) photocathode exhibits an onset potential for water reduction of ~0.5 V RHE . As V1 becomes more negative, the photocurrent density increases and eventually saturates to 25 mA cm -2 at V1 = -0. A hidden J-ΔV curve, analogous to the current-voltage characteristic of a PV cell, can then be extracted and is plotted in Fig. 1c. The V oc and J sc are 475 mV and 24.6 mA cm -2 , respectively (the characteristics are also listed in Supplementary Table S1). A significant loss of fill factor is observed when comparing the J-V1 and J-ΔV curves, which derives from the additional series resistances in a PEC cell versus a PV cell, namely the TiO 2 /catalyst junction resistance, the overpotential of the catalyst required for driving a chemical reaction, and the solution resistance. 20 In essence, the J-ΔV curve shows the best possible fill factor that can be achieved by the J-V1 curve. In practice, a real PEC J-V1 curve will always have a smaller fill factor due to the catalyst overpotential as well as the series resistances mentioned above. Moreover, because the solution resistance does not change in the strongly acidic media, the difference in fill factor between J-V1 and J-ΔV curves represents only a change in the electron transfer kinetics through the TiO 2 /Pt/electrolyte interface.
In order to more clearly visualize the effect of the surface potential on the current, a stepwise chronoamperometry measurement was carried out whereby the potential of V1 was stepped every 30 s and both V2 and the photocurrent were recorded (Fig. 1d). When gives a detailed discussion of the relationship between photovoltage and the onset potential.
A critical issue for photoelectrodes is the long-term stability. In order to characterize the degradation mechanism in the Si photocathode, we performed a 2 h stability test by holding V1 at 0 V RHE , a typical value for these types of test in the literature. 8 Fig. 2a shows the J-V1 and J-ΔV curves before and after the 2 h stability test. Compared with the initial J-V1 scan, the scan after the 2 h shows similar onset potential and slightly decreased saturation photocurrent, but a remarkably poorer fill factor. As the J-ΔV curves remain the same, it is immediately apparent that the problem relates to the catalyst and not to the buried junction. Fig. 2b depicts how the surface potential V2 and photocurrent density change over time under a static back contact potential of 0 V RHE . Over 2 h, the photocurrent density drops from ~23 to ~20 mA cm -2 , while V2 steadily shifts to more negative values. Fig. 2c clearly shows the negative shift of the surface potential following the stability test, which indicates that higher overpotential is needed in order to achieve a similar current density. A poor contact between the surface and the catalyst (TiO 2 /catalyst) as well as worsening kinetics at the catalyst/electrolyte interface (e.g. surface poisoning) will result in a higher overpotential for the catalytic interface. 21 Pt was then re-deposited onto the electrode surface (Fig. 2c,d). Due to the fact that the fill factor is completely restored upon re-platinization, we can confirm that neither a degradation in the p-n junction of the silicon nor resistive losses through e.g. formation of a silicon oxide layer are responsible for the change in the J-V1 curve. The degradation likely results from desorption of the Pt nanoparticles, as has been previously observed for electrodeposited platinum on ALD TiO 2 . 22 When the ALD TiO 2 was replaced by a thin metallic Ti film, the Pt catalyst binding was much more robust over a 2 h stability measurement ( Supplementary Fig. S7).
For the pn + -Si photocathodes described above, the Pt catalyst was electrodeposited onto either the TiO 2 or Ti metal surface as nanoparticles with a size range of ~10-30 nm ( Supplementary Fig. S8a). This non-continuous catalyst morphology may be unfavorable for efficient extraction of the surface electrons, resulting in a poor fill factor of the J-V1 curve. 23 Therefore, we investigated a nominally 2 nm-thick Pt film with nearly full coverage on the TiO 2 layer by sputter coating (denoted pn + Si/TiO 2 /Pt(sp)). The Pt deposited in this way makes the surface slightly rough ( Supplementary Fig. S8b). Fig. 3a compares the J-V1 curves of pn + Si/TiO 2 /Pt(sp) and pn + Si/TiO 2 /Pt(ed) photocathodes under one sun illumination. Sputtered Pt exhibits a similar onset potential and improved fill factor, but much reduced photocurrent densities due to the optical transmission loss through the 2 nm-thick Pt film (Supplementary Fig S9). For better comparison between the sputtered and electrodeposited samples, we also measured the J-V1 curve of the pn + Si/TiO 2 /Pt(sp) at an increased light intensity to achieve a similar photocurrent density, plotted in green. The green curve exhibits an earlier onset potential despite having the same V oc as the Pt(ed) curve, suggesting a better catalytic activity of sputtered Pt over electrodeposited Pt. Support for this hypothesis is shown by comparing their individual catalytic activities towards H 2 generation when deposited on FTO slides ( Fig S10).
Additionally, pn + Si/TiO 2 /Pt(sp) always shows an enhancement in the fill factor the in J-V1 curves, reflecting the smaller TiO 2 /Pt/electrolyte interfacial resistance for the TiO 2 /Pt(sp) as compared to the TiO 2 /Pt(ed). In the case of a conformal Pt film, electron transfer is more probable as the catalyst surface area is increased, which is also indicated by the much more positive V2 value in pn + Si/TiO 2 /Pt(sp), shown in Supplementary Fig.   S11. For example, to reach the same saturation photocurrent, a ~130 mV overpotential is required for pn + Si/TiO 2 /Pt(sp) but ~200 mV for pn + Si/TiO 2 /Pt(ed).
What we have already hypothesized by the performance of the different Pt on FTO is confirmed by the determination of the J-ΔV curves. At similar saturation photocurrents the J-ΔV of pn + Si/TiO 2 /Pt(sp) is essentially identical with the pn + Si/TiO 2 /Pt(ed) (see the green and red dashed curves) while the J-V1 curve shows a clearly higher FF for the device with sputtered Pt.
Thus far, we have developed the DWE technique with a model pn + Si/TiO 2 /Pt(ed) photocathode, with which we can gain a deeper insight into the PEC process and the photocathode stability. Next, we applied this technique to the emerging material ALD TiO 2 -protected Cu 2 O to demonstrate the generality of the technique. 22,24 An n-type Ga 2 O 3 was used as a buffer layer between the Cu 2 O and TiO 2 overlayer because this interlayer ensures a positively shifted onset potential, compared to that of n-Al:ZnO (AZO) ( Supplementary Fig. S12). 25,26 Fig. 4a schematically depicts the multilayer structure of the Cu 2 O/Ga 2 O 3 /TiO2 photocathode. In a similar fashion as for the silicon photocathodes described previously, a second working electrode was introduced to probe the surface potential V2, prior to deposition of the Pt catalyst by sputtering.
PEC measurements were performed in a pH 5 phosphate/sulfate solution. Fig. 4b displays the J-V1 and J-ΔV curves. A positive onset potential of ~0.9 V RHE is observed in the J-V1 curve, and at V1 = 0 V RHE , the photocurrent density is 3.9 mA cm -2 . The onset potential is much more positive than that from both Cu 2 O/ZnO and Cu 2 O/AZO photocathodes, reflecting the larger photovoltage generated by the Cu 2 O/Ga 2 O 3 junction.
In the case of the J-ΔV curve, the V oc , J sc and fill factor are 836 mV, 4.0 mA cm -2 and 36.1%, respectively. Compared with reported Cu 2 O/Ga 2 O 3 solar cells in the literature, the V oc and fill factor values are comparable, but the J sc is lower due to light absorption by the Pt catalyst. 26,27 Still, resistance at the TiO 2 /Pt/electrolyte interfaces contribute to the fill factor loss between the J-V1 and J-ΔV curves. The fill factor loss is not very significant.
As the J-ΔV is basically mirroring the J-V1 curve, this indicates that the buried Cu 2 O/Ga 2 O 3 junction is responsible for the shape of the J-V1 curve of Cu 2 O/Ga 2 O 3 /TiO 2 /Pt(sp) and not the catalytic activity of the Pt catalyst. When using an electrodeposited Pt the PEC system exhibits a lower fill factor, showing that the TiO 2 /Pt/electrolyte resistance indeed can also limit J-V1 performance, as shown in Supplementary Fig. S15. We further carried out a stepwise test on the Cu 2 O/Ga 2 O 3 /TiO 2 /Pt(sp) photocathode under illumination. Results and discussion are provided in Supplementary Fig. S16 and S17 and confirm our statements above.
To study the stability of the Cu 2 O/Ga 2 O 3 /TiO 2 /Pt(sp) photocathode, a 2 h chronoamperometric measurement was performed under illumination at V1 = 0 V RHE . Fig. 4b shows the comparison of the J-V1 curves before and after a 2 h stability test. The onset potential shows a negative shift of nearly 120 mV, although the photocurrent density remains similar. Fig. 4b also provides the initial J-ΔV curve and the one after the an important role. Furthermore, the PEC performance degradation mechanism was investigated and discussed. We have demonstrated that the stability of underlying p-n junctions in buried junction photocathodes can be characterized in operando. As new material combinations are synthesized for PEC measurements, the DWE electrode technique enables a rapid diagnosis of the cause of degradation in these systems, while also obtaining the PV characteristics of these newly developed junctions without the need to construct separate PV cells.   When V1 is more positive than 0 V RHE , V2 is similar to V1, thus ΔV is 0, indicating that no voltage difference is generated under dark conditions in this potential region. When V1 is scanned to more negative potentials than 0 V RHE , V2 stays mainly unchanged with a value slightly more negative than 0 V RHE , contributing to the tiny dark current. This region of very negative V1 corresponds to a reverse bias across the p-n Si homojunction, as is evident from the ΔV-V1 curve.
Under illumination, ΔV maintains a constant value of 475 mV until V1 reaches ~0.47 V RHE . In this positive potential region, ΔV equals the output V oc as there is no current flow across the interface. The potential where ΔV begins to shrink indicates the real onset potential of hydrogen production, which appears in this case at ~0.47 V RHE . As the photocurrent increases while sweeping the potential negatively, ΔV decreases until a saturation photocurrent density is obtained, and then continues to shrink as the photocurrent remains saturated.

J-ΔV
where P in of one sun illumination is the incident illumination power density (100 mW cm -2 ). For the pn + Si/TiO 2 /Pt(sp) sample, an increased light intensity (P in = ~137 mW cm -2 ) was also used. Photoexcited electrons cannot be transferred into solution since the potential of the photoexcited electrons is still not high enough to drive proton reduction. In this region of V1 (before the onset of hydrogen generation), ΔV values remain constant and equal to V oc . The flat band potential of TiO 2 itself is located near E(H + /H 2 ). As soon as V2 is slightly more negative than 0 V RHE , hydrogen evolution is thermodynamically allowed, and electrons flow into the electrolyte by reducing protons into hydrogen gas. The V1 potential at this point is defined as the onset potential.

Supplementary
As V1 becomes more negative, the ΔV value shrinks as the photocurrent corresponding to hydrogen generation increases. When V1 catches up to V2, ΔV is 0 and the buried p-n junction is at the short-circuit condition. Finally, as soon as the photocurrent becomes saturated due to the photon flux and recombination, V2 reaches a steady value independent of V1. The reverse bias across the p-n junction becomes stronger as V1 level continues to move to more negative potential.
Supplementary Figure S5 Zoom-in view of J-V1 (solid) and J-ΔV (dashed) curves of a pn + Si/TiO 2 /Pt(ed) photocathode with a scan rate of 10 mV s -1 .
The V oc appears at a more negative potential than the onset of photocurrent under certain sweeping conditions, as shown in Supplementary Fig. S5. This means that V2 is more positive than 0 V RHE while current is flowing. In order to confirm that these small photocurrents 13 correspond to hydrogen evolution and not to e.g. proton intercalation, we carried out faradaic efficiency measurements (See Supplementary Fig. S6). Hydrogen was indeed observed at very small cathodic photocurrent densities, such as -0.08 mA cm -2 , at potentials more positive than we would expect from the J-ΔV analysis. This phenomenon likely arises from the fact that the slightly lower electron density than in the illuminated area and consequently an actually slightly reduced V oc is detected compared to the illuminated area (as shown in Fig. 1b). Thus, the onset potential appears earlier than the V oc in Supplementary Fig. S5. This V oc loss is highly dependent on the distance between the Au contact and the illumination area, represented by d in this scheme. For all the samples we measured, d is ~1 mm. When intentionally lengthening d, V oc further decreases, yielding ~75 mV loss at d = 3 mm. From our experience with more than fifty samples, the offset between the measured V oc and the apparent onset from the J-V1 curves is typically a few tens of mV with a platinum catalyst.