Insights into the Self-Inhibiting Photoreduction of Cu₂O in Bicarbonate Electrolytes

Abstract

Cuprous oxide (Cu₂O) has long been regarded as the quintessential photoelectrochemical cathode due to its narrow band gap, native p-type behavior, and low cost. However, it is not stable in aqueous electrolytes upon irradiation, limiting its potential application for producing hydrogen or converting wastewater or CO₂ by using direct sunlight. Although Cu₂O’s instability is well-documented, the mechanism and factors governing its photocurrent decay over time are not. In this work, electrodeposited Cu₂O layers with increasing thickness were synthesized to investigate their transient structure–photocurrent density relationships in bicarbonate electrolytes. The photocurrent decay for all the thicknesses could be modeled by an exponential function with an initial photocurrent j _ph,0 and a decay rate constant c. It was found that j _ph,0 follows the Gärtner equation, while the contraction of the space charge layer and the light-dampening effect of the metallic Cu overlayer were evaluated to explain c. Finally, the use of a fast reduction kinetics species (NO₃ ^–) and the comparison with another electrolyte provided critical insights into the effect of KHCO₃ in the self-reduction of Cu₂O.

1. Introduction

Photoelectrochemistry (PEC) is the research field that merges photochemistry and electrochemistry with the aim of increasing the efficiency of electrochemical processes by exploiting the renewable and inexpensive energy of sunlight. It does so by integrating semiconductorsmaterials that convert photons in free chargesinto so-called photoelectrodes. The most relevant processes for which the PEC alternative is explored are the hydrogen evolution reaction (HER) and CO₂ reduction reaction (CO₂RR), , given the role that they are envisaged to play in a future net-zero carbon emission scenario. Even though some examples of reactor studies and applications in the early stage of technological development have been published, , PEC research focuses mostly on material development, with transition metal-based semiconductors (oxides and chalcogenides) being the most explored ones, and bench-scale studies. Even at that basic level, the challenges are multiple, considering that an ideal photoelectrode material should be able to absorb most of the light spectrum, generate high amounts of free charges, separate electrons and holes efficiently, and show high activity and selectivity toward the desired reaction. In HER and CO₂RR, photoelectrodes can be either employed as an anode (photoanode) or as a cathode (photocathode), with the first requiring n-type semiconductors and the second requiring p-type semiconductors. Photoanodes like TiO₂, WO₃, Fe₂O₃, and BiVO₄ have been demonstrated to be active toward the oxygen evolution reaction (OER) with good stability. − On the photocathode side, Cu₂O is one of the few oxides that shows a native p-type conductivity due to the Cu vacancies that form during synthesis. Its direct band gap (E _g ) is in the range of 1.9–2.2 eV, and its corresponding maximum single-junction current density is ∼15 mA cm^–2 under AM1.5 irradiation, which makes it appealing for photovoltaic and PEC applications.

One often overlooked aspect is the stability of the semiconductor in its initial chemical form. It is arguably one of the most important features since permanent changes in the oxidation state usually result in the loss of photoactivity. Whether a semiconductor is stable or not when illuminated is a matter of thermodynamics and kinetics: if the self-reduction/oxidation potential is within the band gap, then the material is prone to reduction/oxidation. However, if the kinetics of the process are slow enough, then the material can be stable. This is especially true if the charges involved in the self-reduction/oxidation process are extracted from the material by a faster reaction or by an external circuit.

Despite a couple of conflicting studies, there is a broad consensus regarding the photocathodic instability of Cu₂O in aqueous electrolytes. , Although it was previously observed that there is a link between the prevalent orientation of the crystals and the stability, with the (100) facets being more susceptible to proton-promoted reduction than the (111) facets, follow-up studies reported that the (111) facet is still being reduced. In all cases, a photocurrent decay was observed. However, a model linking the optoelectronic properties of Cu₂O photocathodes and the photocurrent decay has not been proposed so far. In this work, a correlation between film thickness, optoelectronic properties, and photocurrent decay for electrodeposited Cu₂O is reported. Moreover, a decay model based on the Gärtner equation is proposed to explain the photocurrent decrease with time that is observed during PEC reduction over electrodeposited Cu₂O. In perspective, the model can be rationally extended to explain the photocurrent response of instable photo electrodes in general and support the development of protection techniques.

2. Experimental Methods

2.1. Photoelectrode Preparation

The photoelectrodes were prepared by electrodepositing Cu₂O on FTO glass substrates (7 Ω/sq, 2.2 mm thick, Sigma-Aldrich). The electrodeposition was carried out in a three electrode setup using 2.5 cm × 2 cm FTO glass as the working electrode (2 cm × 2 cm as deposition area), Ag/AgCl (3 M KCl, Metrohm) as the reference electrode, and a platinum gauze as the counter electrode. Before deposition, the FTO substrates were sonicated in acetone, isopropanol, and Milli-Q water for 15 min each and then dried with pressurized air. For 100 mL of electrodeposition solution, 10 g of CuSO₄·5H₂O (>98.0%, Sigma-Aldrich) and 31.7 g of lactic acid (natural, ≥85%, Sigma-Aldrich) were dissolved in Milli-Q water to obtain concentrations of 0.4 and 3 M. Then, 2 M KOH (85–100%, VWR) was added to the solution until a stable pH 12 was reached. The final solution volume was ∼400 mL. The electrodeposition was performed by applying −0.6 V vs Ag/AgCl at 60 °C while slightly stirring the solution (200 rpm). The deposition was automatically stopped when the theoretical charge required for the desired Cu₂O layer thickness was achieved, according to Faraday’s law of electrolysis (eq ).

$$Q=\frac{d\rho zFA}{MM}$$ 1

where d is the layer thickness, ρ is Cu₂O bulk density (6 g cm^–3), z is the number of electrons exchanged during electrodeposition, F is the Faraday constant, A is the area of the electrode, and MM is the Cu₂O molar mass (143.1 g mol^–1). The Faradaic efficiency was calculated by weighing the FTO before and after deposition, converting it to charge, and dividing it by the theoretically expected value. It was ∼98% for all samples.

The samples were named as “S-x”, where x is the targeted thickness of the cuprous oxide layer in μm (e.g., the sample with a thickness of 0.25 μm was named S-0.25).

Cu thin films for UV–vis analysis were deposited on FTO using a magnetron sputter coater (MiniLab 125, Moorfield Nanotechnology) from a Cu target (99.99%, NanoVision). The deposition was performed in Ar at 6 × 10^–3 mbar with a growth rate of 5.4 Å s^–1. The deposition was stopped after reaching the targeted thickness (adjusted by a previously determined correction factor).

2.2. Photoelectrochemical Experiments

PEC experiments were performed in a single-compartment three-electrode cell (Figure S1). A platinum wire and a mini reversible hydrogen electrode (RHE, Gaskatel Mini-HydroFlex) were used as counter and reference electrodes, respectively. The photoelectrode was front-illuminated with a 150 W Xe–Hg lamp (SLS401, Thorlabs) with a measured irradiance of 130 mW cm^–2. The area of the illuminated working electrode in contact with the electrolyte was 1.5 cm × 1.5 cm, and the electrical contact was made via Cu tape on the FTO (Figure S2). The 0.5 M KHCO₃ (99.0%, VWR) electrolyte was purged with CO₂ for at least 30 min before each experiment (pH 7.5). During chopped light chronoamperometry (CL-CA), the light path was chopped with a frequency of 0.1 Hz (duty cycle 50%) and the tests were run for 30 min under a constant CO₂ flow of 5 mL min^–1 bubbling in the cell. A GC (Shimadzu 2014) equipped with a Restek Shicarbon ST column and a thermal conductivity detector (TCD) with He (99.999%, Air Liquide) as carrier gas was connected in-line for gaseous product quantification (H₂, CO, CH₄, and C₂H₄), while samples of the electrolytes were injected in an HPLC (Alliance 2695) equipped with an IC-column and a 210–2996 PDA detector (Waters) for formate quantification. Chopped light linear sweep voltammetry (CL-LSV) was performed under the same conditions. All potentials are reported vs RHE unless explicitly mentioned.

For the nitrate reduction experiments, a 0.1 M KNO₃ (>99%, Chm-lab)/0.5 M KHCO₃ electrolyte was used. The electrolyte was purged with Ar for at least 30 min before pumping it into the cell, and it was recirculated from a gas–liquid separator at a flow rate of 2.5 mL min^–1. The solution reached pH ∼7 after purging. During reaction, Ar was bubbled through the electrolyte in the gas–liquid separator. A total electrolyte volume of ∼9 mL was used in every test to fill the lines and the cell and keep ∼4 mL of liquid in the gas–liquid separator. At the end of each experiment, the electrolyte was collected and weighed to ensure correct estimation of the product concentration.

For nitrate reduction experiments in the absence of KHCO₃, 0.1 M K₂SO₄ (>99%, Chem-lab) and 0.1 M KNO₃ (>99%, Chm-lab) were used as the supporting electrolyte and NO₃ ^– source, respectively. The procedure for these tests was analogous to that for nitrate reduction in KHCO₃.

NO₂ ^– analysis was performed by ion chromatography (IC, 930 Compact IC Flex, Metrohm) equipped with an anion separation column (Metrosep A Supp 19–100/4.0, Metrohm) and an electrical conductivity detector (ProfIC Conductivity Detector MF). The eluent, a solution of 8.00 mM Na₂CO₃ and 0.25 mM NaHCO₃, was pumped at a flow rate of 0.7 mL min^–1.

Electrochemical impedance spectroscopy (EIS) was carried out in the same cell. Yet, the electrolyte for this procedure was N₂-saturated 1 M Na₂SO₄ (pH 6.7). In order to keep the solution quiescent inside the cell, no gas was purged during this procedure.

2.3. Photoelectrode Characterization

Scanning electron microscopy (SEM) analysis was carried out using a Thermo Fisher Scientific Quanta FEG 250 instrument operated at 30 kV and at different magnifications.

The crystal structures of the samples were studied by employing X-ray diffraction (XRD) with an X-ray diffractometer (D8 Advance Eco, Bruker) equipped with Cu K_α radiation (λ = 0.15406 nm). The detector rotation speed was set at 0.04°/s in the 2θ range from 20° to 80°. The spectra were analyzed with DIFFRAC.EVA (Bruker) implemented with the COD database.

Diffuse reflectance UV–vis spectroscopy was performed with a Shimadzu UV 2600i instrument equipped with an integrating sphere. The wavelength was scanned from 800 to 200 nm. The Tauc plots were drawn following a previously reported method.

3. Results and Discussion

3.1. Morphological and Surface Characterization

Figure A shows the relative Cu₂O peak intensities ((A _hkl ) _i /Σ­(A_hkl ) _n ), which were calculated from the corresponding XRD spectrum for each layer thickness (Figure S3). A maximum (111) relative peak intensity (36.5°, COD 1010941) was found for S-3.3 along with a monotonic increase of the (220) peak (61.5°, COD 1010941) intensity as a function of thickness. SEM images showed that the crystals for S-0.25 to S-3.3 (Figure S4 and Figure B) were mainly cubic with the {100} facets exposed; for S-6.7 (Figure C), they were truncated tetrahedrons exposing mainly the {111} facets parallel to the substrate, and for S-10 (Figure D), they appeared as truncated pyramids exposing the {111} and {220} facets. Those changes in orientation may be explained by the deposition procedure. It goes without saying that for thicker samples, the deposition time was longer. However, it should be noted that under potentiostatic conditions, the electrodeposition current decreased from ∼4 to 2 mA cm^–2 with time due to the increasing resistivity of the electrode. A decrease in current density may cause the change in prevalent growth rate from the faster <111> direction (S-0.25 to S-3.3), forming prevalently cubes with exposed {100} facets, to the slower <110> direction (S-10), forming octahedrons with exposed {111} and {220}.This was accompanied by an intermediate growth phase that formed truncated tetrahedrons with the {111} facets parallel to the substrate surface (S-6.7). , Although the current-dependent growth was already observed by Paracchino et al., it occurred in a lower current range (∼0.1 to 1.5 mA cm^–2), possibly due to the phosphate buffer used in their deposition solution. Another possible explanation involves the change of local deposition conditions as the time increases (e.g., Cu²⁺ concentration, temperature, and pH). For example, shifts in pH are known to influence the preferred crystal growth direction. In fact, Wang et al. suggested that relatively abrupt changes in preferred crystal orientation happen at specific pHs, namely, (100) forms at pH 8 to 9, (110) forms at pH 9 to 10, and (111) forms between 10 and 12, although a small increase in the (110) contribution could be observed going from pH 12 to 11. This dependency was recently used by Pan et al. to fabricate polycrystalline Cu₂O layers with a specific prevalent orientation. Moreover, differences in local [OH^–] and [Cu²⁺] influence the deposition current, the morphology of the surface, and even N _A within the range of 10¹⁶–10¹⁷ cm^–3.

1.

(A) Relative XRD peak intensity as a function of the Cu₂O layer thickness (d). SEM images of (B) S-3.3, (C) S-6.7, and (D) S-10.

In addition to XRD, X-ray photoelectron spectroscopy (XPS) analysis was also performed for S-3.3 and is reported in Figure S5. Since the electrodeposition was carried out in potentiostatic mode, we did not expect the surface oxidation state of the films to be markedly different among samples with different thicknesses. It can be observed from the Cu 2p3 spectra that apart from the main Cu­(I) peak at 932 eV, a Cu­(II) peak (934 eV) and a broad satellite peak (944 eV) are detected (Figure S5A), indicating the presence of Cu²⁺ species at the surface. However, the spectra taken after removing ∼2 nm by Ar sputtering show that the layer was only composed of Cu­(I) (Figure S5C), which leads us to conclude that the presence of Cu²⁺ could be attributed to traces of precursor CuSO₄ or oxidation in contact with air. Moreover, the CuLMM spectra (Figure S5B) showed only the characteristic Cu₂O peak (570 eV), which proves the absence of Cu⁰ (characteristic peak at 568 eV).

3.2. Photoelectrochemical Experiments in Bicarbonate

Deriving photocurrent values from CL-LSV is not advised for unstable semiconductors since they may already degrade at potentials close to the flat band potential (E _FB ). Nevertheless, the technique is still useful to identify the range of potentials in which the material shows photoactivity and to confirm the type of semiconductor. As depicted by the voltammetric curves in Figure S6, all Cu₂O layers show p-type behavior and photoactivity below 0.6 V. Considering the activity and the range of potentials used in CO₂ photoelectrochemical reduction, − a working electrode potential E of 0.1 V was selected for the following CL-CA tests.

The experimental photocurrent density was measured from CL-CA (Figure A and Figure S7A–F). This technique allows calculating the net photocurrent density (j _ph ) as a function of time (as reported in Figure B) by subtracting the dark current density (j _D ), caused by the applied bias, from the current density under illumination (j _L ). The photocurrent transients were excluded from the calculation, as graphically explained in Figure S7G.

2.

(A) CL-CA at 0.1 V in CO₂-saturated 0.5 M KHCO₃. SEM images of (B) S-0.25 and (C) S-3.3 after 30 min of CL-CA. (D) Photocurrent density over time for S-1.7 fitted with the exponential model (black line). XPS analysis of S-3.3 after CL-CA at 0.1 V in CO₂-saturated 0.5 M KHCO₃. (E) Atomic percentages of Cu (from Cu 2p spectra) and O (from O 1s spectra) as a function of increasing Ar sputtering depth. (F) CuLMM spectra at depths of 2 and 200 nm. The first spectra were taken at 2 nm to obtain a surface without CuO and CuCO₃ contaminations due to air and electrolyte exposure, respectively.

It is of upmost importance to highlight that no CO₂ reduction products nor H₂ were detected during PEC tests, possibly due to the product concentration being below the limit of detection of our equipment and the current density being almost entirely a self-reduction current density according to eq .

$${Cu}_{2}O+2e^{-}+H_{2}O\to 2Cu+2{OH}^{-}$$ 2

It has to be noted that 0.1 V is higher than some of the standard CO₂RR reduction potentials (e.g., to CO and HCOO^–) and the HER, which implies that during the dark periods, neither of them should take place.

The appearance of the photoelectrodes after CL-CA (for example, see Figure S8 for S-1.7 after PEC reduction) strongly indicates that Cu₂O self-reduction took place. Postreaction SEM showed increased roughening of the surface caused by the formation of small Cu crystals on top of the Cu₂O ones (Figure C,D and Figure S9). Due to the different currents generated by the films, samples S-0.25 to S-1.7 displayed a rougher surface (Figure C) than the thicker samples S-3.3 to S-10 (Figure D). XRD showed the appearance of Cu (111), (200), and (220) reflections at 43.5, 50.6, and 74.5° (COD 10000063), respectively, and the decrease of the Cu₂O signal (Figure S10). Importantly, the Cu peaks are more visible for S-0.25 to S-1.7 and are practically absent for the thicker samples. For the latter samples, this could stem from the larger Cu₂O amount still present in the bulk at the end of the experiments and the fact that metallic Cu was confined to the surface. To probe this, XPS measurements were carried out as a function of depth via Ar sputtering. The atomic percentages of Cu and O were derived from the Cu 2p and O 1s spectra, respectively. At the maximum experimentally possible sputtering depth of 200 nm, the atomic percentage of Cu was still 90% (Figure E). The CuLMM spectra recorded after sputtering off 2 and 200 nm from the surface (Figure F) show the typical pattern of Cu⁰ (with the main peak at 568 eV) with a similar intensity. From this analysis, we concluded that a metallic Cu overlayer is indeed formed and that it is at least 200 nm in thickness. Although the exact depth at which the transition to Cu₂O takes place could not be determined, the combined information from XPS and XRD suggests the stratified Cu/Cu₂O nature of the thicker electrodes after CL-CA. It is worth mentioning that when an AZO/TiO₂ protection layer was applied to a Cu₂O photoelectrode (see details in Section 5 of the Supporting Information), the exponential decay was not observed anymore (Figure S11). While that confirmed the effectiveness of such layers in preventing photoreduction, − it also meant that the buried Cu₂O was rendered inactive, as the photocurrent achieved by these protected photoelectrodes was 3 orders of magnitude lower than the unprotected one.

3.3. Modeling the Initial Photocurrent Density

The curves of j _ph as a function of time (see Figure B and Figure S12) for unprotected Cu₂O were fitted with good agreement (R ² ≥ 0.95) to an exponential decay as in eq

$$j_{ph}=a+j_{ph,0}·e^{ct}$$ 3

where j _ph,0 expresses the initial photocurrent density. Because for the thicker samples, j _ph did not decay to zero within the experiment time, constant a was introduced to better fit those curves. However, since its numerical value accounted for a maximum of just 2% of the j _ph,0 value, it was not considered as a significant parameter, in contrast to c, the decay rate constant. The reported j _ph,0 and c values were averaged from the fitting of three replicas with the corresponding standard deviation. Based on the analysis, we concluded that there is no explicit correlation between Cu₂O thickness, j _ph,0 , or c (Figure A,B).

3.

Parameters derived from the exponential model as a function of the Cu₂O layer thickness: (A) initial photocurrent density (j _ph,0 ) and (B) decay rate constant (c).

The optoelectronic properties of the semiconductor are the ones affecting the photocurrent density, and they may not be directly linked to its physical thickness. For example, the Gärtner model correlates the photocurrent density with the semiconductor’s space charge layer thickness and absorption coefficient. Even though this model is the most popular one in the literature, more advanced models have been developed over the years since its publication. A discussion of those models and the rationale behind the choice of Gärtner’s model in our work is presented in Section 6 of the Supporting Information. The Gärtner equation in eq is defined at each wavelength as

$$j_{G(\lambda )}={\varphi }_{0(\lambda )}q[1-\frac{e^{(-{\alpha }_{(\lambda )}W)}}{1+{\alpha }_{(\lambda )}L}]$$ 4

where j _G is the net photocurrent density (mA cm^–2), ϕ ₀ is the photon flux (cm^–2 s^–1), q is the elementary charge (1.602 · 10^–19 C), α is the absorption coefficient (cm^–2), W is the thickness of the space charge layer (cm), and L is the electron diffusion length (cm). The dependency of the variables on the wavelength (λ) is made explicit by the subscript. Under white light irradiation, j _G must be integrated over the entire wavelength range to obtain the photocurrent value. The equation can be further simplified to eq considering that αL ≪ 1. Integrating the equation over the wavelength gives the total j _G .

$$j_{G(\lambda )}={\varphi }_{0(\lambda )}q[1-e^{(-{\alpha }_{(\lambda )}W)}]$$ 5

The absorption coefficient was derived from UV–vis spectroscopy. Reflectance (R)total reflectanceand transmittance (T) were measured to obtain the absorption coefficient α, according to eq .

$$\alpha =-\frac{\ln \left(\frac{T}{{(1-R)}^2}\right)}{d}$$ 6

where d is the sample thickness in cm.

The measured transmittance of S-0.25 decreased to zero already at λ < ∼480 nm (Figure S13), which in practical terms means that at higher energy, the light beam is completely attenuated before exiting the sample. In the equation proposed by Gärtner, the carriers generated beyond the space charge layer undergo fast recombination and, therefore, do not participate in the surface reaction. The exponential term in eq is in fact the transmittance at a depth equal to W. If α was considered infinite, then the transmittance at those wavelengths would be zero even for extremely thin layers (as in the case of W). However, that assumption may not hold up and can cause large overestimations of the photocurrent in the model. On the other hand, if an arbitrary value that is small enough would be substituted instead of zero in the transmittance data set, another problem would arise when dividing the logarithmic term (corresponding to the absorption with Napierian base) by thickness. The maximum absorption coefficient would saturate at a value that is defined mainly by the thickness in an artificial way, therefore potentially resulting in underestimations of j _G with increasing thickness. Since direct calculation of eq may not be reliable at shorter wavelengths, α was derived as the slope of the absorption as a function of d, at each λ. Since the slope below 500 nm could not be defined, α was considered constant. It has to be noted that S-6.7 and S-10 were excluded from the calculation of α because their absorption near the bandgap energy did not follow the same trend as the other five samples. A possible explanation of this behavior is given in Section 7 of the Supporting Information. Nevertheless, it was assumed that all of the samples have the same absorption coefficient shown in Figure A.

4.

(A) Derived absorption coefficient and (B) Tauc plot with bandgap energy (E _g ).

The absorption coefficient was then used to construct the Tauc plot (Figure B), considering only the direct allowed transition to derive the band gap. E _g defines the upper limit of the j _G(λ) integral, while the lower limit is set to 320 nm, i.e., the absorption onset of the PEC cell’s window (fused silica).

The space charge layer thickness W was calculated from eq .

$$W=\sqrt{\frac{2{\epsilon }_0{\epsilon }_r}{{qN}_A}(E-E_{FB}-\frac{kT}{q})}$$ 7

where ε ₀ is the vacuum permittivity (8.854 × 10^–14 F cm^–2), ε _r is the relative permittivity (6.6), N _A is the acceptor density (cm^–3), E is the applied potential (V), E _FB is the flat band potential (V), k is the Boltzmann constant (1.381 × 10^–23 J K^–1) and T is the temperature (K). N _A and E _FB were derived from Mott–Schottky analysis. A simplified R-(RQ)-(RQ) equivalent circuit (with Q being a constant phase element) was applied to estimate the capacitance of the space charge layer (C _sc ). Out of at least three convoluted semicircles appearing in the EIS spectra, only the two semicircles at higher frequencies (100 kHz to 100 Hz) were fitted. Figure S14A shows an example of an EIS spectrum (S-1.7 at 0.52 V) and the corresponding equivalent circuit. In all cases, C _sc was calculated at the frequency of the maximum Z″. E _FB was extracted from the intercept of the C _sc curve with the E-axis, while N _A was calculated from the slope (Figure S14B–H). All samples showed a negative slope, which is characteristic of a p-type semiconductor. S-0.25 was the only sample showing a positive slope at E < 0.56 V, which is attributed to exposed FTO support (SnO₂ being an n-type semiconductor). Finally, the space charge layer width W was calculated at 0.1 V. The parameters obtained from Mott–Schottky analysis are reported in Table . E _FB and N _A values are similar to the values already reported in the literature. ,,, However, the values found for S-0.25 and S-10 should be carefully considered. Pan et al. recently demonstrated that N _A and E _FB are isotropic properties; therefore, they should not be influenced by differences in crystal orientation. Considering N _A , the value of 10¹⁹ cm^–3 seems higher than the most often reported value, ∼10¹⁷ cm^–3. Meanwhile, for S-0.25, it is reasonable to think that the substrate may influence the capacitance estimation via Mott–Schottky analysis for the thin 250 nm layer; however, the same cannot be true for S-10. In that case, it is possible that surface states and traps influenced the measurement of the capacitance, although it is difficult to demonstrate this. The presence of transient photocurrent throughout almost the entire CL-CA for S-10 suggests that there is accumulation of charges (often caused by surface states and traps) and that these are relatively more abundant than for the other thicknesses. The same rationale can be applied for E _FB , since its calculation derives from the same experimental data.

1. Collected Parameters in Order to Calculate the Photocurrent Density According to the Gärtner Equation.

sample	E FB (V)	N A (cm –3 )	E app (V)	W (nm)
S-0.25	0.60	3.7 × 1019	0.1	3
S-0.50	0.50	3.2 × 1017	0.1	29
S-0.75	0.58	6.5 × 1016	0.1	71
S-1.7	0.55	9.5 × 1016	0.1	57
S-3.3	0.55	2.7 × 1017	0.1	34
S-6.7	0.55	3.2 × 1017	0.1	30
S-10	0.71	2.4 × 1019	0.1	4

Then, the photon flux ϕ ₀ was calculated by integrating the power spectrum of the lamp, considering a total power output of 130 mW cm^–2 (Figure S15). However, the photon flux needs to be further corrected for the losses caused by reflection at the surface of the sample, as in eq . It has to be noted that under real conditions, R may differ from the measured ones, since R always depends on the refractive index of two materials in contact, i.e., air in the case of the measured materials and the electrolyte inside the photoelectrochemical cell. Nevertheless, it was assumed that R would not change significantly.

$$j_{G(\lambda )}={\varphi }_{0(\lambda )}(1-R_{(\lambda )})q[1-e^{(-{\alpha }_{(\lambda )}W)}]$$ 8

A comparison between the experimental photocurrent density and the predicted Gärtner photocurrent density is presented in Figure A. It shows that even though j _G did not completely match the experimental data, it still followed the same trend as that of j _ph,0 . Comparing the standard deviation of j _ph,0 and the difference between j _G and experimental j _ph,0 , S-6.7 stands out significantly (Figure B). The mismatch may be attributed to the assumptions made while deriving α and W and to the simplification for αL. Furthermore, the contribution due to the resistance of the Cu₂O layer itself could be a dampening factor that is not considered. It is known that the resistivity of Cu₂O is anisotropic since the mobility in the [111] direction is ∼15 cm² V^–1 s^–1 (1 order of magnitude higher than the mobilities of [110] and [100]), and it is expected that the resistance would be higher for the thicker layer than for the thinner one. Nevertheless, the model captures phenomenologically the major aspects of the initial photocurrent density response.

5.

(A) Comparison between the experimental initial photocurrent, j _ph,0 , and the predicted Gärtner photocurrent, j _G . (B) Comparison between experimental j _ph,0 standard deviation (in red) and deviation of j _G from the experimental data (in blue).

3.4. Modeling the Photocurrent Density Decay

Ideally, the photocurrent density would decay until Cu₂O had completely reacted. However, two features were noticed in the CL-CA graphs (Figure S7): (i) The dark current increased from zero to significant values after the first illumination period (5 s), suggesting that even though Cu₂O was stable at 0.1 V in the dark, the first period under illumination induced changes in the surface that allowed the reduction to continue in the dark. (ii) The photocurrent density disappears before the dark one, suggesting that Cu₂O does not need to be completely reduced before the photoelectrode deactivates. This implies that the decay in the photocurrent density could not be attributed to the loss of Cu₂O alone. It is worth mentioning that for S-0.25, S-050, and S-0.75, the dark current density fell to zero when the charge exchanged during the reaction was enough to fully reduce Cu₂O (Figure S7A–C). This further confirmed that the only process taking place was reduction of the photoelectrode. The first hypothesis formulated to explain the exponential decay involved the space charge layer contraction. Specifically, we argued that since the oxide reduction starts from the surface, part of the Cu₂O that formed the space charge layer could be lost with time and the space charge layer thickness could progressively decrease as a direct consequence. In this case, a time-dependent Gärtner model would be expressed by eq .

$$j_{G(t)}={\varphi }_0(1-R)q[1-e^{(-\alpha (W-s_{(t)}))}]$$ 9

where s(t) is the thickness corresponding to the amount of reduced Cu₂O and is calculated by integrating the CL-CA curves and converting the charge density found in this way into thickness using eq . However, that hypothesis was promptly discarded, since (i) the resulting decays were practically linear and (ii) the condition of s > W, for which j _G(t) = 0, was reached when the photocurrent was still clearly visible in the experimental curves (Figure S16).

Hence, it was hypothesized that the Cu overlayer, which constantly forms on the surface of the photoelectrode, could decrease the photon flux that effectively reaches the semiconductor. To test that hypothesis, a series of time-dependent j _G(t) were calculated while taking into account the dampening effect of the Cu layer formed after fixed time intervals (10).

$$j_{G(t)}={\varphi }_{0}q(1-R_{Cu2O})[1-e^{(-{\alpha }_{Cu2O}W)}](1-R_{Cu})e^{(-{\alpha }_{Cu}{x}_{(t)})}$$ 10

where R _Cu is the measured reflectance of the Cu layer, α _Cu is the absorption coefficient of Cu, and x is the thickness of the Cu layer. To focus on the time-dependent parameters, the dependence on λ was not explicitly shown in eq , but it is still implicitly present.

Similar to s _(t), x _(t) was calculated from the experimental CL-CA data, integrating the current density curves (dark and light) and converting the charge density to thickness with eq , but this time considering Cu’s density (ρ _Cu = 8.96 g cm^–3) and molar weight (MW = 63.55 g mol^–1). The absorption coefficient was calculated following the procedure shown previously, after measuring the T and R of FTO sputter-coated with increasing Cu thickness (Figure S17A). Since R was not thickness-dependent, R _Cu was averaged from the four samples (Figure S17B). The decay rate constant (c _G ) was derived by fitting the plotted j _G(t) value (Figure S18) with (8). From its derivation, it is possible to observe that c _G is in physical terms directly related to the reaction (eq ) rate constant. However, the exponential fit reached an R ² between 0.90 and 0.95, suggesting that the Cu overlayer may induce a photoelectrode inhibition that is different from the experimental one.

A comparison between experimental c and predicted c _G , reported in Figure A, showed that even in this case, the trend between experimental and modeled data was similar, but the values did not match. Specifically, c _G underestimates the photocurrent decay rates of S-0.25 to S-0.75 (e.g., c = −0.048 s^–1 versus c _G = −0.021 s^–1 for S-0.50) while overestimating the S-3.3 to S-10 ones (e.g., c = −0.004 s^–1 versus c _G = −0.013 s^–1 for S-6.7). This was also confirmed by the difference between the parameters σ­(c)/c and the percentage error |c – c _G |/c, used to evaluate the goodness of the fitting model (Figure B). Thus, this confirms that the light-dampening effect of the Cu layer over the Cu₂O surface cannot be regarded as the sole factor determining the photocurrent density decay, as explained below.

6.

(A) Comparison between experimental rate constant, c, and the decay rate modeled applying the modified Gärtner model, c _G . (B) Comparison between the percentual standard deviation of c (in red) and percentage error of the modified Gärtner model (in blue).

For example, if Cu grows in a nonhomogeneous way, i.e., by a series of nucleation and growth events that result only in partial coverage of the Cu₂O underneath, the decay would be slower at shorter reaction times. To probe this, digital videos were recorded during CL-CA from behind the FTO electrode for S-0.25 and S3.3. Significant frames showing light passing through the photoelectrode over time are reported in Figure S19. The videos allowed us to qualitatively assert that (i) at least macroscopically, the darkening due to the formation of Cu appeared over the surface in an inhomogeneous way (areas that darken first are circled in Figure S19) and (ii) the darkening timespan appeared to be much longer than the timespan in which the photocurrent density went to zero. On the other hand, Cu islands could already be observed after the first illumination step for S-0.25, as shown by Figure S20.

At this point, further analysis of the CL-CA and j _ph –t curves was conducted. It was noticed that for S-0.25, S-0.50, and S-0.75, the j _ph decay could be divided into two main regions (Figure S21A). Whereas the first decay reaches a photocurrent plateau (P1), in the second decay, the photocurrent goes actually to zero (P2), which is associated with the full reduction of the Cu₂O layer. The two plateaus are shown in Figure S21C,D. Interestingly, the photocurrent in P1 seems to be reached with all of the samples (Figure S21D,E). By comparing the j _ph –t and the CL-CA curves, it is possible to notice that during the decay, the photocurrent transients decrease while completely disappearing in P1. The photocurrent transients are usually composed of (i) cuspids, resulting from the accumulation of charges in trap states and/or surface states when the light is turned on; (ii) a decay phase, which is explained by recombination of the accumulated charges; and (iii) a final plateau, which is explained by charge transfer to the electrolyte only. Since transients are indicative of those phenomena, their study is a way to investigate recombination. , In the literature, these studies show that (i) the decay part usually lasts tenths of seconds and (ii) the transients do not change significantly with time. On the contrary, in our case, the transients did not reach a clear plateau in 5 s of illumination. This is longer than the characteristic recombination time; therefore, this may suggest that at least part of the charges accumulated at the surface and trap states did not recombine with holes during the decay phase of the transient but drove the self-reduction of Cu₂O instead. Moreover, the transient cuspids decrease in magnitude over time until they disappear in P1, implying that the capability of the surface and trap states to accumulate charges declines as well. We therefore argue that the fast photocurrent density decay may be linked to the specific reduction of the surface states. Surface states are physically atoms of the crystal lattice that are under- or overcoordinated and may form when electrolyte ions adsorb to the surface. , If the adsorbates are ions that can promote the reduction of Cu₂O, e.g., H⁺, , it is reasonable to think that the surface states themselves will be the first to reduce. Consequently, their reduction may cause the loss of the accumulation capability of the surface and the establishment of a new equilibrium with the electrolyte or Cu formed on the surface. The latter would result in a change in band bending and consequently in a change of W. Those specific changes could not be directly analyzed in this work, since it would require in depth charge dynamics analysis techniques like photoelectrochemical impedance spectroscopy (PEIS) and intensity modulated photocurrent spectroscopy (IMPS), , which were unfortunately unavailable to the authors at the time of writing. However, the effect of a redox species with fast reaction kinetics could indirectly substantiate our hypothesis, as discussed in the next section.

Finally, we would like to highlight the differences between our model and a recent model for the photocurrent decay of BiVO₄ photoanodes proposed by the Haussener group. , They identified time regions during which the decay proceeds at different rates. However, the BiVO₄ photoanode did not show the dark current contribution to the current under illumination, as shown by Cu₂O. Moreover, BiVO₄ degradation causes it to dissolve in the electrolyte until a stable photocurrent is achieved, without the formation of a layer that potentially hinders the absorption of light as for the formation of Cu. Therefore, the photocurrent decay could be expressed as a piecewise function of the instantaneous loss of BiVO₄ thickness and three periods over which the Gärtner equation could be defined by different space parameters (hole diffusion length, L _p ; W and the film thickness L _film ). Even though their approach is theoretically close to ours, the aforementioned differences in photocurrent density and dark current density behavior brought us to develop eqs and instead.

3.5. Nitrate Reduction and the Role of Bicarbonate in Self-Reduction

NO₃ ^– PEC reduction was recently reported over Cu₂O surfaces in K₂SO₄, and its fast reduction kinetics was proposed as the reason for negligible photoelectrode self-reduction. , Therefore, we added 0.1 M KNO₃ to the KHCO₃ electrolyte to observe if the fast reduction of NO₃ ^– would stop the self-reduction and the current from decaying. The reduction was performed with S-3.3 photoelectrodes, resulting in a 78 ± 4% FE for NO₂ ^–. The NO₃ ^– was not completely effective in stopping the decay, but it slowed it down (Figure ). A closer analysis of the CL-CA plot shows transients that could be assigned to deep depletion of the space charge layer in the first 40 s (Figure S22), proving extremely fast charge transfer toward the electrolyte and the surface. After that, the well-known accumulation-recombination transients started to appear. After ∼1000 s, the transients disappeared again. The time at which the transient disappear in CO₂-saturated KHCO₃ is ∼700 s, suggesting that the phenomena causing the disappearing transient is common, but the rates are of course different.

7.

(A) Comparison between CL-CA using S-3.3 at 0.1 V in CO₂-saturated 0.5 M KHCO₃ (blue), Ar-saturated 0.5 KHCO₃ + 0.1 M KNO₃ (red), and Ar-saturated 0.1 K₂SO₄ + 0.1 M KNO₃ (green). (B) Comparison between photocurrent densities achieved using S-3.3 at 0.1 V in CO₂-saturated 0.5 M KHCO₃ (blue), Ar-saturated 0.5 KHCO₃ + 0.1 M KNO₃ (red), and Ar-saturated 0.1 K₂SO₄ + 0.1 M KNO₃ (green).

To investigate whether KHCO₃ can promote the self-reduction of Cu₂O, NO₃ ^– reduction was repeated, but in 0.1 M K₂SO₄ as the supporting electrolyte instead. In this case, no decay was observed, and the photocurrent was stable for the entire measurement, implying that even if a Cu layer formed, it did not interfere with the photocurrent generation nor with NO₃ ^– reduction, since a FE of 100% was achieved for NO₂ ^–. Interestingly, of highlight is that the CL-CA curve showed that the transients were constantly present throughout the measurement, indicating that the surface states are preserved under these conditions (Figure S22). Besides, this clearly indicates that KHCO₃ has an active role in the phenomena that cause the photocurrent decay.

From the literature and eq , it is known that the presence of H⁺ promotes the reduction of Cu₂O, especially on the {111} facets. Hence, since the pH of the bicarbonate electrolyte (pH 8.3 for Ar-saturated and 7.5 for CO₂-saturated KHCO₃) is higher than that for Ar-saturated K₂SO₄ (pH 7), Cu₂O should be more stable in KHCO₃. However, it is possible that the H⁺-donor character of HCO₃ ^– may be the cause of Cu₂O instability. On a deeper level, that would imply that PEC CO₂ reduction over bare Cu₂O is not possible in aqueous media, since the equilibrium of the carbonates in water dictates that a part of the CO₂ is always in HCO₃ ^– form. Furthermore, that would be true regardless of the chosen supporting electrolyte. In situ and in operando studies and possibly computational reaction modeling may be implemented in the future to understand the mechanism of the HCO₃ ^–-promoted Cu₂O self-reduction, i.e., confirm if the reduction of specific sites, like surface states, is the cause of the exponential decay.

4. Conclusions

In this work, electrodeposited Cu₂O photoelectrodes were employed to study the deactivation of Cu₂O in KHCO₃, the main electrolyte used in CO₂RR studies. CL-CA was performed to extract the photocurrent density decrease over time, j _ph(t) , and an exponential decay model was formulated to fit the experimental data. The initial photocurrent (j _ph,0 ) and the rate constant c were defined as fitting parameters, but there was no explicit correlation between them and the film thickness. This lack of correlation was attributed to the thickness-dependent changes in optoelectronic properties. Consequently, those properties were measured and used as parameters in the Gärtner equation to calculate the theoretical photocurrent (j _G ). Overall, the trends in j _G and j _ph,0 were qualitatively similar, but the individual values showed pronounced deviations.

First, the decay was linked to a contraction of the space charge layer width, although the resulting decay was too fast. A similar j _G(t) equation was used to account for the optical dampening effect of the Cu layer that forms on the photoelectrode surface. Hence, a decay rate constant c _G was estimated by fitting j _G(t) with the exponential decay. It was found that the two decay constants differ significantly in the single values, even though following qualitatively similar trends. Therefore, it was concluded that the light dampening of the Cu overlayer plays a role in the decay, although not being the only factor contributing to it. The possibility that surface state reduction may be the origin of such fast decay was also discussed. Finally, introducing a species with fast reduction kinetics (NO₃ ^–) and testing the photoelectrodes in K₂SO₄ demonstrated an active role of HCO₃ ^– in the Cu₂O instability. Further in operando studies, charge dynamics analysis, and computational surface modeling could help in understanding the specific mechanism of the HCO₃ ^– promoted self-reduction.

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acsomega.5c06018.

• PEC cell scheme, SEM images, XRD spectra, XPS spectra, UV–vis spectra, ALD procedure, chronoamperometric and voltammetry graphs, and digital pictures (PDF)

The authors declare no competing financial interest.