Understanding the Performance of NiO Photocathodes with Alkyl-Derivatized Cobalt Catalysts and a Push-Pull Dye.

Mesoporous NiO photocathodes containing the push-pull dye PB6 and alkyl-derivatized cobaloxime catalysts were prepared using surface amide couplings and analyzed for photocatalytic proton reduction catalysis. The length of the alkyl linker used to derivatize the cobalt catalysts was found to correlate to the photocurrent with the highest photocurrent observed using shorter alkyl linkers but the lowest one for samples without linker. The alkyl linkers were also helpful in slowing dye-NiO charge recombination. Photoelectrochemical measurements and femtosecond transient absorption spectroscopic measurements suggested electron transfer to the surface-immobilized catalysts occurred; however, H2 evolution was not observed. Based on UV-vis, X-ray fluorescence spectroscopy (XRF), and X-ray photoelectron spectroscopy (XPS) measurements, the cobalt catalyst appeared to be limiting the photocathode performance mainly via cobalt demetallation from the oxime ligand. This study highlights the need for a deeper understanding of the effect of catalyst molecular design on photocathode performance.

Introduction

Since climate change is actively occurring, fossil fuels must be replaced by clean, renewable fuels.[1] Water-splitting dye-sensitized photoelectrochemical cells (WS-DSPECs) have been proposed as one way to meet this demand for clean fuels.[2−6] WS-DSPECs work by mimicking the processes occurring in natural photosynthesis by performing water oxidation catalysis (on the photoanode) in tandem with fuel forming reactions (on the photocathode). Here, we will focus on the photocathode component with the goal to produce H2 gas as a renewable fuel.

Photocathodes in WS-DSPECs contain both molecular dyes to harvest sunlight and molecular proton reduction catalysts to perform hydrogen evolution catalysis.[7−9] Both molecules are immobilized on a semiconductor electrode (usually NiO), commonly by a covalent linkage using surface anchoring groups (carboxylic acids, phosphonic acids, hydroxamic acids, silatranes), which are already attached to the molecules. To operate, the photocathode is illuminated, which causes the dye to become photoexcited and inject a hole into the NiO valence band, generating the reduced dye. The reduced dye can then reduce a nearby catalyst molecule, initiating proton reduction catalysis once the catalyst gains enough reducing equivalents and protons. This is quite challenging, as all of these processes must occur faster than unwanted charge recombination to function efficiently.[7,8]

To date, tandem WS-DSPECs have exhibited poor overall performance primarily because the photocathodes have performed poorly compared to the photoanode despite a multitude of tested photocathode strategies and architectures.[7−11] Some strategies for photocathode assemblies have included codeposition of dye and catalyst,[10,12−16] covalently linking dye and catalyst,[17−20] layer-by-layer strategies,[21−23] adding hydrophobic or electron shuttling layers,[24,25] atomic layer deposition,[26] and supramolecular assemblies.[27] In addition, we have recently developed a surface amide coupling procedure for photocathode assembly where dye or catalyst molecules containing carboxylic acid functional groups can be amide-coupled to amine-derivatized surface-immobilized groups to easily tune and screen a variety of photocathode architectures.[14]

In many cases, efficient catalysis on photocathodes is difficult because charge recombination processes are faster than the charge accumulation at the catalyst needed for turnover to occur.[12,22,23,28−34] To help slow down charge recombination, some strategies can be employed to increase the catalytic efficiency. A strategy that has shown promise involves putting the surface-bound dye on the electrode in a hydrophobic environment such as by surrounding the dye with long alkyl chains either codeposited or included as a part of the molecular structure.[24,35] This has been shown to slow down dye–NiO charge recombination with ruthenium dyes[35] and has also been applied in photocathodes with a Mo–S cluster proton reduction catalyst and a push–pull dye.[24] In addition, covering the NiO and/or encasing the dye with an insulating material, such as Al2O3, via atomic layer deposition (ALD) can aid in slowing dye–NiO charge recombination; this strategy has been applied in both the photocathode and photoanode field.[26,36−40] Using these strategies, one can produce a photocathode with a greater chance of performing catalysis.

Recently, we have studied the dye PB6 (Figure ) for use in NiO-based photocathodes (Figure ).[14,25,41,42]PB6 is known to perform ultrafast hole injection (<300 fs) upon photoexcitation and generate the reduced state PB6– (−0.93 V vs normal hydrogen electrode (NHE) for PB6/PB6–),[41] which is at a potential capable of reducing many proton reduction catalysts. We have also studied the photoelectrochemical performance of NiO-PB6 photocathodes, which show excellent long-term photoelectrochemical stability in aqueous conditions.[14] In separate studies, we found that PB6 is stable under ALD preparation conditions and demonstrated that using ALD of Al2O3 and/or TiO2 can slow charge recombination on NiO-PB6 electrodes in a variety of assemblies.[41−43] Furthermore, ALD is a useful technique as it can be used to build a platform for immobilizing a catalyst further from the metal oxide surface or dye.[44] As an added benefit, ALD has been shown to increase a dye’s photostability and decrease dye desorption from the electrode.[26,36,39,45]

Figure 1

Molecular structures of PB6 dye; Co proton reduction catalyst; and L4, L8, and L11 alkyl linkers to be used in the study. The amine of the linkers was amide-coupled to the carboxylic acid of Co during electrode preparation.

Molecular catalysts based on cobalt, nickel, or iron are the most commonly studied earth-abundant compounds for electrocatalytic proton reduction and have also been applied as catalysts in photoelectrocatalysis by immobilization on photocathodes.[7,8,43,46] For example, cobaloxime catalysts have shown promise in electrocatalysis[43,46−49] and have been immobilized onto photocathodes for photocatalytic proton reduction catalysis using different types of assemblies.[13,17,50−53] Recently, a new cobaloxime catalyst, Co (Figure ), has been shown to be active for electrocatalytic proton reduction (−0.45 V vs RHE) when incorporated into a surface-immobilized metal–organic framework (MOF).[54] In this MOF, Co performed more than 20000 turnovers under aqueous conditions with no loss in current over 18 h of operation. Since it had good stability on the electrode, it has the potential to be applied in a photocathode.

Thus, here we report the assembly of photocathodes with PB6 incorporating ALD of Al2O3 in an attempt to slow charge recombination and provide a platform for surface-immobilized catalysts. Alkyl-linked Co catalysts are used to provide a hydrophobic environment around PB6 with the aim to further slow charge recombination, and surface amide couplings are used to prepare alkyl Co catalysts for easy tunability.

Results and Discussion

Assembly Rationale

In the photocathodes studied herein, we use PB6 as the dye molecule and Co as the catalyst molecule on the electrodes (Figure ). As stated above, PB6 is capable of hole injection into the NiO valence band upon photoexcitation with the resulting PB6– possessing enough driving force to reduce the Co catalyst.[41] Electrodes were assembled in the architecture depicted in Figure . In this architecture, PB6 is partially covered by Al2O3 using ALD. ALD of Al2O3 was chosen for several reasons: (1) it may slow NiO–dye charge recombination;[26,37,40] (2) it can be used as an anchoring platform to subsequently immobilize Co on the electrode and so separate Co further from NiO, which theoretically will slow Co–NiO charge recombination;[44] and (3) it could also enhance dye photostability and hinder desorption.[26,39,45] Alkyl linkers (L4, L8, or L11; Figure ) with varying lengths are attached to the Al2O3 via their carboxylic acid anchoring groups, providing a hydrophobic environment around PB6, which also aims to slow dye–NiO charge recombination.[35] Co is then attached to the linkers via a surface amide coupling procedure we recently developed via coupling of the carboxylic acids of Co with the amine of the linkers to create a surface amide bond.[14] The linkers and Al2O3 layer help separate Co further from the NiO surface (which could help slow Co–NiO and Co–PB6 charge recombination). The different linkers also place Co at varying distances from PB6. Distance between dye and catalyst could change performance as electron transfer kinetics are distance-dependent;[22,23] thus, it is important to investigate which linker gives the best performance.

Figure 2

Photocathode architectures prepared using alkyl-linked catalysts. PB6 is first bound to the NiO substrate, followed by 0.2 nm of Al2O3 applied by atomic layer deposition (ALD). The linkers are then bound to the Al2O3 layer and Co subsequently attached via a surface amide coupling to the amine on the surface-bound linker.

Electrode Preparation

NiO thin films were prepared by screen-printing NiO paste onto fluorine-doped tin oxide (FTO)-coated glass electrodes (see the Supporting Information, SI, for more details). The NiO electrodes were next soaked in a 0.2 mM 1:1 dichloromethane/methanol solution of PB6 for 50 min, rinsed, and air-dried, resulting in NiO-PB6 electrodes; this gave a PB6 surface coverage of ∼35 nmol/cm2 (76% of a monolayer, Figure S4) as found by UV–vis spectroscopy, providing free space on the surface for the linker molecules to bind later (Scheme , step 1). Next, ALD of Al2O3 was performed at 300 °C in the presence of trimethylaluminum, giving a ∼0.2 nm layer of Al2O3 over the PB6, producing NiO-PB6-Al2O3 electrodes (Scheme , step 2). Next, the linkers (L4, L8, or L11) were bound to the Al2O3 surface by soaking NiO-PB6-Al2O3 in a 4 mM solution of either L4, L8, or L11 overnight (Scheme , step 3). Post sensitization, the electrodes were rinsed and air-dried, resulting in NiO-PB6-Al2O3-L4, NiO-PB6-Al2O3-L8, or NiO-PB6-Al2O3-L11 electrodes. Finally, a surface amide coupling procedure was performed to attach Co to the electrodes by soaking the linker-functionalized electrodes overnight in 0.1 mM Co methanol solution containing 1 mM 1-ethyl-3-(3-dimethylaminopropyl)carbodiimide (EDC) as an amide coupling agent (Scheme , step 4).[14] Post sensitization, electrodes were rinsed and air-dried affording NiO-PB6-Al2O3-L4-Co, NiO-PB6-Al2O3-L8-Co, or NiO-PB6-Al2O3-L11-Co electrodes. A control sample without linker was also prepared by soaking NiO-PB6-Al2O3 in a 0.1 mM methanol Co solution overnight, affording NiO-PB6-Al2O3-Co electrodes. Further details on the electrode preparation method can be found in the Supporting Information (SI).

Scheme 1

Electrode Assembly Process: (1) PB6 Sensitization, (2) ALD of Al2O3, (3) Alkyl Linker, (L4, L8, or L11) Electrode Sensitization, and (4) Co Immobilization via a Surface Amide Coupling Procedure

Electrode Characterization

As we have previously observed, UV–vis spectroscopic measurements of the NiO-PB6 electrodes show a signature peak at 530 nm, indicative of PB6 binding to the electrode (Figure A).[14,25,41,42] This feature also matches well with the UV–vis of PB6, suggesting that the molecular structure is retained upon binding to the electrode surface (Figure S3). Post ALD of Al2O3, the UV–vis spectrum of NiO-PB6-Al2O3 does not change compared to NiO-PB6, strongly supporting the good stability of PB6 at the high-temperature ALD conditions (Figure A). The electrodes with linker and catalyst molecules have UV–vis spectra similar to NiO-PB6-Al2O3 (Figures S6 and S7). This is expected as the UV signatures of the linkers are screened by NiO scattering below 400 nm (Figure S8) and the relatively large extinction coefficient of PB6 masks the expected features of Co (Figures S3 and S8).

Figure 3

(A) UV–vis spectra of NiO-PB6 (pink) and NiO-PB6-Al2O3 (black). (B) Attenuated total reflection Fourier-transform infrared (ATR-FTIR) spectra of Co catalyst powder (black) and NiO-L11-Co (pink).

To gain evidence for Co immobilization onto the electrode via amide coupling, X-ray fluorescence (XRF) spectroscopy was used to assess if cobalt was present on the electrodes. Indeed, all samples with Co (NiO-PB6-Al2O3-L4-Co, NiO-PB6-Al2O3-L8-Co, NiO-PB6-Al2O3-L11-Co, or NiO-PB6-Al2O3-Co) were found by XRF to have cobalt, confirming immobilization of Co onto the electrodes, while control samples not exposed to Co lacked cobalt (Table S1). To obtain Co loadings, an XRF calibration curve was constructed using electrodes with known Co loadings prepared by drop-casting Co on the electrodes (Figure S2). Using this calibration curve, the Co loadings on all photocathodes were found to be ∼15–16 nmol/cm2.

To determine if the linker/Co amide couplings had occurred, octanoic acid, a CH3-terminated alkyl linker (Figure S1), was bound to the NiO-PB6-Al2O3 electrodes (see the SI for preparation details). Since there is no amine present in octanoic acid, no amide coupling should occur, and thus no Co should bind to the electrode. In addition, if the octanoic acid linkers have saturated the surface, then Co should not bind since there will not be any free surface sites available on the Al2O3 layer. To test this, an amide coupling procedure was performed by exposing the NiO-PB6-Al2O3–octanoic acid electrode to a Co/EDC solution overnight, as described above. Insignificant loadings of Co were found on the electrode (3 nmol/cm2 compared the ca. 15–16 nmol/cm2 observed for electrodes with amines present), which strongly suggests that the amide coupling occurs only when the amine-derivatized linkers are bound to the electrode surface; the small quantity of Co found on the NiO-PB6-Al2O3–octanoic acid electrodes likely results from Co binding to the remaining free surface sites not blocked by PB6 or the alkyl linkers. This control also indirectly suggests that the linkers are bound to Al2O3, blocking the majority of remaining surface sites. In addition, this control further supports that Co does not displace the linkers on the surface during the amide coupling procedure. Contrastingly, when no linkers are present on the surface (i.e., NiO-PB6-Al2O3-Co), Co can bind to the electrode in higher amounts (∼19 nmol/cm2), presumably in empty surface sites not occupied by PB6 molecules. Finally, ATR-FTIR measurements showed a diagnostic feature at 1225 cm–1 for the N–O– stretch of the oxime ligand (Figures B and S13).[55] This was observed for both Co catalyst powder and surface-immobilized Co strongly supporting that the molecular structure is retained upon binding to the electrode. In addition, X-ray photoelectron spectroscopy (XPS) measurements displayed a clear peak in the N 1s region, also supporting the presence of the oxime ligand on the surface (Figure S14). One should note that PB6 also contains N and thus has a peak in the N 1s region; however, when Co is added to the electrodes, the N 1s signal significantly grows, which further suggests the binding of Co to the electrodes. Taken together, the octanoic acid control experiments as well as XRF, ATR-FTIR, and XPS measurements support that Co is successfully immobilized on the electrode surface via an amide coupling to the preimmobilized linkers L4, L8, and L11.

Photoelectrochemistry

To test if the photocathodes could produce photocurrent upon illumination, photoelectrochemical measurements were performed. First, chronoamperometric (CA) measurements were performed on the PB6-only samples in 0.1 M 2-(N-morpholino)ethanesulfonic acid (MES) buffer at pH 5 with a 0.1 V vs NHE bias. Similar to what we previously observed with PB6-based photocathodes, all samples showed a response to light resulting in stable photocurrent (Figure S17).[25,29] This indicates that upon photoexcitation, PB6 subsequently injects a hole into the NiO valence band.

When Co is immobilized on the electrode, all samples showed an increase in photocurrent upon illumination, which was significantly larger compared to the samples without Co (Figures A, S17, and S18). Since PB6– (−0.93 V vs NHE for PB6/PB6–) should theoretically be able to reduce Co (−0.45 V vs RHE) on the electrode, the larger photocurrent with the catalyst present suggests that PB6– can reduce Co at least once (Figures , S17, and S18). In addition, control samples of bare NiO, NiO-L4, NiO-L8, NiO-L11, NiO-Co, NiO-L4-Co, NiO-L8-Co, and NiO-L11-Co did not generate significant photocurrent, supporting that the photocurrent generated from electrodes with both PB6 and Co should be from Co reduction and/or catalysis (Figure S19).

Figure 4

(A) Chopped light chronoamperograms of NiO-PB6-Al2O3-L4 (gray) and NiO-PB6-Al2O3-L4-Co (pink) in 0.1 M MES buffer at pH 5 with an applied potential of 0.1 V vs NHE. The photocathodes were illuminated from 30 to 40, 50 to 60, 70 to 80, and 90 to 150 s. (B) Photocurrent trend with linker length. The linker length was estimated by measuring the distance from the carbon of the carboxylic acid to the nitrogen atom of the amine in Chem3D. The photocurrent density after 100 s of illumination is shown for NiO-PB6-Al2O3-Co (no linker, labeled “none”), NiO-PB6-Al2O3-L4-Co (L4), NiO-PB6-Al2O3-L8-Co (L8), and NiO-PB6-Al2O3-L11-Co (L11).

Interestingly, a reproducible trend in photocurrent was observed in Co containing electrodes depending on the linker used (Figure B). Electrodes with the L4 linker (NiO-PB6-Al2O3-L4-Co) generated the highest photocurrents, but when the longer L8 and L11 linkers were used, the photocurrents decreased as the linker length increased. Presumably, as the Co catalyst gets further from PB6, it is harder to transfer an electron to the Co catalyst, and thus the photocurrent decreases. Interestingly, without a linker (NiO-PB6-Al2O3-Co), the currents were the lowest of all samples. This may suggest that Co–NiO charge recombination is too fast for the charge to build up at the Co center. Charge recombination could be fast if Co is too close to the NiO surface without the benefit of the surface alkyl linkers, which would slow charge recombination when present. Therefore, the use of surface alkyl linkers on these photocathodes is necessary to provide higher photocurrents by possibly slowing PB6–NiO charge recombination. The L4 linker gave an optimal architecture where Co is slightly distanced from PB6 and NiO, allowing it to receive an electron from PB6– while possibly slowing Co–NiO charge recombination. Clearly, small molecular changes on an electrode surface can provide noticeable changes in photocurrents. One should note that due to the inherent heterogeneity of the films, some differences in photocurrent magnitudes occurred between batches prepared on separate occasions. However, the overall trend in photocurrent with linker length was reproducible for each batch of samples prepared at the same time. To minimize errors when comparing samples, care was taken to prepare a full set of samples using the same sensitization conditions starting on the same day. Since the use of the L4 linker provided samples with the highest photocurrents, we will only use these electrodes hereafter.

Ultrafast Electron Transfer Dynamics

Since the NiO-PB6-Al2O3-L4-Co electrodes showed promising photocurrent responses, we wanted to further understand the charge transfer dynamics on the surface. To do this, femtosecond transient absorption spectroscopy (fsTAS) was performed on dry photocathodes in air. The simplest electrode, NiO-PB6, was first analyzed. NiO-PB6 was photoexcited at 560 nm, resulting in transient spectra displaying a negative ground-state bleach at 530 nm, a positive absorption at > 650 nm, and stimulated emission at 620 nm (Figure A) similar to what we have previously observed with NiO-PB6 films.[14,41,42] Global analysis was performed by fitting the data to a triple exponential function resulting in the following lifetimes: τ1 = 0.58 ps, τ2 = 10 ps, and τ3 = 3500 ps (Table ). Hole injection from PB6* into the NiO valence band likely corresponds to the first time component (τ1), which generates the PB6– reduced state.[14,41] The second (τ2) and third (τ3) slower time components are likely attributed to charge recombination between PB6– and NiO(+).[14,41]

Figure 5

(A) Transient absorption spectra of NiO-PB6 with photoexcitation at 560 nm. (B) Decay associated spectra of NiO-PB6; time constants, (pink) τ1, (orange) τ2, and (yellow) τ3, have units of ps. (C) Kinetic traces at 620 nm for NiO-PB6-Al2O3-L4 (black) and NiO-PB6-Al2O3-L4-Co (pink).

Table 1

Results from Global Analysis from Fitting to a Triple Exponential Functiona

sample	τ0 (ps)	IRF (ps)	τ1 (ps)	τ2 (ps)	τ3 (ps)
NiO-PB6	0.09	0.19	0.58	10	3500
NiO-PB6-Al2O3	0.10	0.16	0.50	9.3	3900
NiO-PB6-Al2O3-L4	0.10	0.19	0.47	11	5500
NiO-PB6-Al2O3-L4-Co	0.09	0.17	0.48	9.1	4600

IRF is the width of instrument response function (full width at half-maximum), τ0 is time zero, and τ1–3 is the first, second, and third time components, respectively.

Stimulated emission from PB6* can be followed in the kinetic traces at 620 nm showing an instrument-limited bleach followed by a growth of a positive signal (τ1 = 0.58 ps) and decay to baseline (∼1 ns, represented by τ2 and τ3). This trace suggests the formation of at least one intermediate upon excitation and is in good agreement with the formation of the charge-separated state, NiO(+) and PB6–.[14,41,42] Furthermore, the formation of the charge-separated state can be followed through the 620 nm trace, where we see the formation of PB6* (initial negative signal), formation of PB6– (positive signal), and decay to the ground state via charge recombination (positive signal returns to baseline). The positive signal in the 620 nm kinetic trace is less pronounced than our previous work[14] likely due to the lower loading of PB6 on the electrodes in this study.

Furthermore, decay associated spectra support the generation of PB6* and PB6– on the electrodes (Figure B); the first time component (τ1) matches the characteristic difference spectra of PB6* (minus the ground state), while the second (τ2) and third (τ3) time components show spectra that correspond to the difference spectrum of PB6– consistent with our prior work.[14,41,42] One should note that since photocathode surfaces have an inherent heterogeneity, injection and recombination processes may overlap on different timescales and the time components could include multiple overlapping processes.[32,56]

Next, the effect of ALD, alkyl linker, and catalyst on the photocathode architecture was analyzed. When ALD of Al2O3 was added to the photocathode in NiO-PB6-Al2O3, similar transient absorption spectra and decay associated spectra were observed compared to the photocathode lacking Al2O3 (Figures S23 and S24). With Al2O3, a slower NiO(+)–PB6– charge recombination, represented by τ2 and τ3, could be expected as ALD of Al2O3 has been shown to be a means of slowing charge recombination in photoelectrodes, including our previously reported PB6-derivatized electrodes.[26,37,41] However, we do not observe a significant difference in the time components between the NiO-PB6 and NiO-PB6-Al2O3 electrodes (Table ). Since the Al2O3 layer is thinner (0.2 nm) than the optimized photocathodes in our prior work (1 nm),[42,46] it is likely the Al2O3 layer is too thin to significantly slow charge recombination. Regardless, as is seen in the decay associated spectra, we are still able to generate PB6– on the electrodes, which shows the potential to be used to reduce Co in the photocathode. Thus, in the photocathodes discussed here, Al2O3 simply serves as a surface to immobilize Co further from the NiO surface.

When the L4 linker was added to the photocathodes in NiO-PB6-Al2O3-L4, transient absorption spectra similar to NiO-PB6 were again observed (Figure S23). The decay associated spectra also clearly show the presence of the PB6* and PB6– species (Figure S24). With the alkyl chains surrounding PB6 creating a hydrophobic environment, a slower NiO(+)–PB6– charge recombination would be expected.[24,35] Indeed, the third time component is lengthened from τ3 = 3900 ps for NiO-PB6-Al2O3 to τ3 = 5500 ps for NiO-PB6-Al2O3-L4 (Table ), suggesting that the alkyl linker helps slow charge recombination as hypothesized.

Finally, the effect of the Co catalyst was examined in NiO-PB6-Al2O3-L4-Co electrodes. Similar to the other photocathodes, the transient absorption spectra show the same features as NiO-PB6 and the decay associated spectra suggest the formation of PB6– and PB6* on the electrode (Figures S23 and S24). The generation of PB6– on the catalyst-containing electrodes further supports that the electrodes function as expected. With Co present on the electrode, a faster τ2 and/or τ3 would be expected if PB6– can transfer an electron to Co on the surface, thereby bringing PB6– back to the ground state. In fact, we observe a shortening of τ3 = 5500 ps for NiO-PB6-Al2O3-L4 to τ3 = 4600 ps for NiO-PB6-Al2O3-L4-Co (Table ), suggesting that PB6– may reduce Co on the electrodes. In addition, kinetic traces at 620 nm show a smaller maximum signal in the presence of a catalyst. Recall from above that the traces at 620 nm show an initial negative signal (formation of PB6*), which rises to a positive signal (formation of PB6–) and decays to baseline (PB6– returns to the ground state); for further discussion of this, please see our previous work.[41] Since the positive max signal is smaller with the catalyst, this suggests faster decay of PB6– to the ground state, which could occur more rapidly if PB6– reduces a neighboring Co on the NiO surface. Thus, the traces at 620 nm also support that electron transfer to Co occurs on the ps time scale (Figure C). These results are similar to our previous study where we saw evidence for electron transfer from PB6– to a nickel complex on coimmobilized photocathodes using an identical analysis.[14]

H2 Evolution

Furthermore, since NiO-PB6-Al2O3-L4-Co showed the best photocurrents and because we had evidence for electron transfer to Co by both photoelectrochemical measurements and fsTAS, H2 evolution measurements were performed to see if catalytic proton reduction occurred. To do this, a Clark electrode was used to detect H2; however, no significant H2 was detected. If Co is not able to gain two reducing equivalents and two protons before charge recombination occurs, then this could result in a poor activity. However, we wondered if there could be other explanations for the inactivity. Recall that Co is active for electrocatalytic proton reduction in a MOF-immobilized electrode, hence we would expect that Co could be active for photoelectrocatalytic proton reduction when immobilized on the photocathodes.[54] Since we were not able to detect any H2 when immobilized for photoelectrocatalysis, we wondered what may be happening on the electrode to limit performance.

Rationale for Electrode Performance

To understand the limitations of the electrode, stability measurements were performed. First, short-term stability was tested. To do this, chopped light CA was performed on the NiO-PB6-Al2O3-L4-Co electrodes over a 3 min period at 0.1 V vs NHE in 0.1 M MES buffer at pH 5. The electrodes were removed from the buffer, allowed to sit for 2 weeks in the dark, and then the CA was repeated (Figure S20). Remarkably, the majority of the photocurrent was retained, highlighting good short-term photostability. In addition, Co was still detected on the electrodes after CA by XRF measurements.

Since the electrodes appeared stable over the short term, longer 1 h CAs were performed. During these measurements, a large spike in photocurrent (hundreds of μAs) was observed that slowly decayed over the course of an hour, which reached less than 10 μA/cm2 in 60 min (Figure S21). Clearly, the long-term stability of the electrodes is a limiting factor. UV–vis spectra of the electrode before and after the 1 h CA and of the electrolyte post CA show that the dye remains on the surface and does not appear to photodegrade or desorb, consistent with our prior results (Figure S9).[14] Since PB6 was stable and did not limit the photocurrent, XRF measurements were performed to analyze the cobalt content on the electrode. To do this, XRF measurements were performed over 1 h, which display cobalt loss from the electrode over time, showing that about 80% of the initial cobalt was lost from the surface in 1 h (Figure ). UV–vis spectra of the electrolyte during this time course also showed a new species growing over time, confirming that a species from the surface was desorbing into the electrolyte (Figure S10); the spectra does not match that of Co but could perhaps be a cobalt-containing salt or cobalt nanoparticles that form if the cobalt demetallates from the oxime ligand.[57] In addition, this new species is not present in the electrolyte if an identical CA experiment is performed with NiO-PB6-Al2O3-L4 over 1 h, suggesting that the new UV–vis peak is likely from a cobalt species (Figure S11). Furthermore, Co remains bound to the electrode when it is soaked in 0.1 M MES buffer for 1 h, suggesting that the cobalt loss is only initiated when the samples are exposed to photoelectrochemical conditions. Cobaloxime degradation is not an unusual phenomenon, and specifically, cobalt demetallation in cobaloxime complexes often occurs, as the reduced cobaloxime is known under certain conditions to be unstable when reduced in protic environments.[13,16,18,49,57−60] Finally, since cobalt is only lost under photoelectrochemical conditions on the photocathodes, this further supports that Co is reduced at least once on the electrode.

Figure 6

Percent cobalt retained on NiO-PB6-Al2O3-L4-Co during a 1 h CA under illumination with an applied bias of 0.1 V vs NHE in 0.1 M MES buffer at pH 5. XRF spectroscopic measurements were performed before and after CA to determine the percent cobalt retained on the electrode.

To obtain evidence for either cobalt demetallation or Co desorption, several experiments were performed. First, electrodes were prepared with a version of L4, but instead of a carboxylic acid anchor, a silatrane anchor was used (L4Sil, Figure S1). Silatranes are known to be stable anchor groups when bound to metal oxide surfaces under acidic, basic, and neutral conditions.[14,61−63] If Co was desorbing in the NiO-PB6-Al2O3-L4-Co electrode, the silatrane anchor should ensure that the whole catalyst remains bound to the surface in the NiO-PB6-Al2O3-L4Sil-Co electrodes. To test this, CA was performed for 1 h on the NiO-PB6-Al2O3-L4Sil-Co electrodes (see the SI for preparation details; Figure S22) and cobalt content analyzed before and after CA. Cobalt loss was still observed (50%), which suggests that cobalt was likely demetallating from the oxime ligand as Co desorption is less likely with a strong surface anchor. The amount of cobalt loss observed was slightly lower than with the carboxylic acid anchor (80% loss for the carboxylic acid anchor), which suggests that some Co from the NiO-PB6-Al2O3-L4-Co electrode may desorb, and the silatrane anchor can help stabilize Co desorption in the NiO-PB6-Al2O3-L4Sil-Co electrode.

Second, electrodes were prepared using ALD to further stabilize Co. ALD of TiO2 has also been applied in various photoelectrode assemblies as a method to stabilize electrode constructs.[36,64−66] Specifically, ALD of TiO2 has been shown to increase and stabilize photocurrents when used in GaInP2-based photocathodes with cobaloxime catalysts.[67] In these electrodes, TiO2 was applied over the cobalt catalyst under milder conditions at 150 °C.[67] By partially covering Co with TiO2 on the electrode, Co desorption should not occur, similar to the electrodes containing silatrane-anchored Co. NiO-PB6-Al2O3-L4-Co-TiO2 electrodes were prepared (see the SI for details) and CA performed for 1 h (Figure S22). Again, cobalt loss was observed by XRF (79% loss), suggesting that cobalt was demetallating from the oxime ligand rather than by simple desorption of the entire catalyst.

Furthermore, N 1s region XPS spectra of the NiO-PB6-Al2O3-L4-Co samples over time support that the majority of the oxime ligand (400.4 eV, dashed line in Figure ) may remain on the surface after 1 h of CA (Figure ). A small decrease in the N 1s intensity is initially observed, suggesting that some Co desorption may occur, consistent with the above results using the silatrane anchor (NiO-PB6-Al2O3-L4Sil-Co). Alternatively, if cobalt demetallates from the oxime ligand, one of the two oxime ligands coordinated to cobalt should be lost, while the other remains surface-immobilized; this could also lower the N 1s intensity. During 1 h of CA, an additional peak appears over time (402.1 eV) and some broadening of the oxime ligand peak is observed, which is likely due to the MES buffer adsorbing on the electrode (Figure S15). MES buffer shows two signals in the N 1s region at 402.1 and 399.4 eV (Figure S15). When the XPS spectra are overlaid and examined, the spectrum of NiO-PB6-Al2O3-L4-Co at 60 min appears to have both oxime ligand and MES components (Figure S15); additional XPS analysis of this can be found in the SI. In addition, cobalt loss is still observed when no linker is present on the surface post CA (NiO-PB6-Al2O3-Co) with the majority of the oxime ligand remaining on the electrode; this further suggests that cobalt loss does not result from the use of alkyl linkers but is an inherent instability when surface-bound to the photocathode under photoelectrochemical conditions (Figure S16). Therefore, based on the XPS data, it appears that the majority of the oxime ligand remains on the photocathode after 1 h of electrolysis, supporting that mainly cobalt demetallation occurs on the electrode. Additional XPS data can be found in the SI. Taken together, the UV–vis data, XRF data, silatrane alkyl linker control experiments, ALD of TiO2 over Co, and the XPS data support that cobalt demetallation from the oxime ligand is likely the main deactivation pathway of these electrodes.

Figure 7

XPS spectra of the N 1s region of NiO-PB6-Al2O3-L4-Co (pink, 0 min). Photocathodes were illuminated for 3 min (orange) and 60 min (yellow) with a 0.1 V vs NHE bias in 0.1 M MES buffer at pH 5. A control sample with just the oxime ligand of Co bound to NiO is shown in gray (NiO-oxime). The NiO-oxime ligand electrode was prepared by soaking a NiO electrode in a 0.1 mM oxime ligand (structure shown in Figure S14) methanol solution for 2 h, rinsed with methanol, and air-dried.

Overall, based on evidence from the photoelectrochemical results, fsTAS results, and the observed cobalt loss from the photocathodes, Co can be reduced at least once on the photocathodes. Since cobaloxime reduction is known to cause catalyst instability, the reduction of Co likely contributes to limiting the photocathode performance by causing cobalt demetallation.

Furthermore, Co has also been shown to be electrocatalytically active for homogeneous proton reduction catalysis under aqueous conditions at pH 5.3, producing 15.2 μmol H2 with a 90% Faradaic yield.[54] However, the stability of Co under homogenous electrocatalytic conditions has not been analyzed. Specifically, it remains unknown whether the homogenous analogue degrades, and if it does, how—possibly via demetallation, ligand degradation, or nanoparticle formation. As degradation under homogenous conditions has been observed with many other cobaloxime catalysts, this possibility remains likely.[57] As we have explored cobalt demetallation and ligand changes above, this leaves nanoparticle formation to still be considered. Since nanoparticles are usually active for proton reduction catalysis and we do not see any H2 evolved from our system, we likely do not have significant nanoparticle formation on the photocathodes studied here. Thus, cobalt demetallation remains the most likely reason for low photocathode performance.

Clearly, the catalyst’s surrounding environment plays a role in its stability, as Co was stable and catalytically active when immobilized in a MOF.[54] The MOF structure may act as a protecting environment by keeping the cobalt metal in a secure binding site upon reduction, where the oxime-coordinated cobalt is held in the MOF by all four carboxylic acids on the oxime ligands in a rigid, locked, enclosed architecture. Conversely, on the photocathodes, Co is in a less rigid or contained environment, being attached to the metal oxide surface via carboxylic acid anchoring groups on the oxime ligand or alkyl linkers. In addition, in this architecture, Co likely has greater exposure to the surrounding electrolyte than in the enclosed MOF structure, providing an easier way for the cobalt to demetallate. Without alkyl linkers, Co is likely bound to the photocathode surface by one or two of the carboxylic acid anchors on the oxime ligands, which results in a less rigid surface binding than in the MOF, which had all four carboxylic acids securing the catalyst. The less rigid environment likely contributes to instability of these electrodes, promoting cobalt loss by having more conformational freedom. In the photocathodes with alkyl linkers, the cobalt is coordinated to oxime ligands with flexible alkyl linkers, and in this architecture, Co has more conformational freedom due to the flexible alkyl chains, which could upon photoreduction encourage the release of the cobalt center, which then travels into the bulk electrolyte.

Electrochemically driven catalysis in the MOF is perhaps simpler than photoelectrocatalysis on a photocathode as the reducing equivalents are provided via electron hopping from the conductive electrode to Co, whereas, in the photocathode, charge must accumulate at Co after PB6 excitation, hole injection, and electron transfer to Co twice. Since there are more steps to accumulate charge at Co, which takes more time, there are more opportunities for charge recombination, which makes accumulating two reducing equivalents at Co more challenging. In the photocathodes studied here, there is an additional challenge as the reduction of Co likely causes cobalt demetallation and photocathode deactivation. Therefore, if the cobalt metal center is removed from the ligand, charge accumulation will not occur at the catalyst since it no longer remains intact, which then lowers the overall performance of these electrodes.

Conclusions

Here, we have prepared photocathodes containing PB6 dye and alkyl-linked cobalt catalysts. The use of linkers led to higher photocurrents compared to when catalysts were linked directly on the electrode. An intriguing photocurrent dependence on the linker length of Co was found via photoelectrochemical measurements with the L4 linker showing highest photocurrents followed by a current drop-off with longer linkers, which separate Co further from PB6. In our system, the L4 linker provided the optimal height for the Co catalyst with respect to PB6 and NiO, showing the largest photocurrents reproducibly. Ultrafast electron transfer dynamics suggested that the alkyl linkers slowed NiO–PB6 charge recombination, while ALD had little effect, likely due to the layer being too thin. Evidence for electron transfer to Co by PB6– was supported by photoelectrochemical experiments and the fs-transient absorption studies. Since Co was likely reduced by PB6– at least once on the electrodes, the best samples were tested for H2 evolution, but no H2 was detected. To understand this, several control experiments were performed that suggested the dominating pathway for photocathode deactivation was via cobalt demetallation from the oxime ligand, leading to a loss in photocurrent and catalysis. Based on these results, to make improvements in photocathode performance, molecular design of catalysts specifically for surface-immobilized photoelectrocatalysis must absolutely be considered in photocathode design. Of course, slowing charge recombination is also important to consider, but if the catalyst becomes deactivated by not having an optimal molecular structure, then there is no chance for the charge to build up at the catalyst. Thus, one should always consider that catalysts used for electrocatalysis may not always be a suitable choice for photoelectrocatalysis as the mechanisms and timescales for charge accumulation are different.