Operando Identification of the Reversible Skin Layer on Co3O4 as a Three-Dimensional Reaction Zone for Oxygen Evolution.

Co oxides and oxyhydroxides have been studied extensively in the past as promising electrocatalysts for the oxygen evolution reaction (OER) in neutral to alkaline media. Earlier studies showed the formation of an ultrathin CoO x (OH) y skin layer on Co3O4 at potentials above 1.15 V vs reversible hydrogen electrode (RHE), but the precise influence of this skin layer on the OER reactivity is still under debate. We present here a systematic study of epitaxial spinel-type Co3O4 films with defined (111) orientation, prepared on different substrates by electrodeposition or physical vapor deposition. The OER overpotential of these samples may vary up to 120 mV, corresponding to two orders of magnitude differences in current density, which cannot be accounted for by differences in the electrochemically active surface area. We demonstrate by a careful analysis of operando surface X-ray diffraction measurements that these differences are clearly correlated with the average thickness of the skin layer. The OER reactivity increases with the amount of formed skin layer, indicating that the entire three-dimensional skin layer is an OER-active interphase. Furthermore, a scaling relationship between the reaction centers in the skin layer and the OER activity is established. It suggests that two lattice sites are involved in the OER mechanism.

Introduction

The search for commercially viable catalysts for the oxygen evolution reaction (OER), which is the bottleneck for electrochemical water splitting, is a key challenge in the worldwide transition to a renewable-based energy system. Earth-abundant catalysts in general and cobalt (hydro-)oxides in particular are of great interest. The latter show promising catalytic properties and are stable in alkaline and neutral electrolytes under ambient conditions. Moreover, they may be synthesized with a great variety of morphologies (nanosheets, particles, or thin films) and with a crystalline or amorphous structure. Preparation methods include solvothermal,[1−3] photochemical,[4] and electrochemical[5−7] syntheses as well as deposition under vacuum conditions.[8−11] Especially the mixed-valence Co3O4 spinel has been extensively studied and may be considered as a prototypical OER oxide catalyst.[2,3,5−10,12−48] In many cases, the precise surface structure and electrochemically active surface area (ECSA) of the catalysts are unknown or poorly defined, however. Thus, the comparison of the electrocatalytic activity of different catalysts is difficult.

A further challenge in understanding the OER on earth-abundant catalysts is characterizing the influence of the surface structure on the catalyst’s reactivity. Many of these materials undergo surface restructuring at the strongly oxidizing potentials where the OER occurs (for recent reviews, see refs (49−52)). Here, the surface of the catalyst is typically converted into an oxide in which the metal cations shift to a higher oxidation state, which is considered to be more electrochemically active. In the case of Co3O4, the OER occurs on a subnanometer skin layer composed of X-ray-amorphous CoO(OH) (i.e., a phase that does not contain sufficiently crystalline regions to produce measurable Bragg peaks) and not on the surface of the bulk Co3O4 spinel structure.[2,3,7] In contrast to many other systems, this skin layer is converted back into crystalline Co3O4 by reversing the electrode potential. This reversible skin layer formation was first observed on polycrystalline samples by Bergmann et al. and was believed to be promoted by the OER.[2,3] However, our group showed that the skin layer forms gradually at potentials above 1.15 VRHE, i.e., close to the thermodynamic equilibrium potential of the Co3O4/CoOOH phase transition, and well before the onset of the OER, by operando surface X-ray diffraction (SXRD) studies of epitaxial Co3O4 thin-film electrodes.[7] The skin layer thus has to be attributed to the oxide electrochemistry itself, and this explains its reversible formation and back-crystallization with potential.

The structural complexity of the oxide/electrolyte interface structure probably explains why the OER mechanism on cobalt oxide is still under debate despite extensive atomic-scale studies, including in situ spectroscopic studies[8,24,41,42,45,53] and theoretical studies by density functional theory (DFT).[20,21,29,32−35,54−56] Studies by X-ray absorption spectroscopy (XAS), X-ray photoelectron spectroscopy, and Raman spectroscopy indicated an overall increase in the Co oxidation state under OER conditions and attributed the high electrocatalytic activity to those species.[24,41,42,45,53] However, these measurements do not enable determination of the precise location and amount of these OER-active Co species. Most studies assume that they are restricted to the oxide surface, but in some cases, it was suggested that all Co centers in the oxide contribute.[57,58] Furthermore, OER mechanisms involving one as well as two sites have been proposed.

In the above experimental studies, the OER activity of Co3O4 is qualitatively linked with the increased Co oxidation state and the associated coordination changes at the oxide surface, from which the identification of the active site was attempted.[2,3] However, the dependence of the electrocatalyst properties on the spatial extension of the transformed surface region, i.e., the effective surface density of active sites, has not been determined yet. Also, in ab initio theory studies of Co3O4 catalysts, only crystalline surfaces were considered up to now and thus did not allow conclusions on the role of the skin layer.[20,29,32−35,54]

In this work, we derive a quantitative relationship between the amount of formed skin layer and the OER activity. To achieve this, we performed systematic comparative studies by atomic force microscopy (AFM), operando SXRD, and electrochemical measurements of well-defined spinel-type Co3O4 epitaxial films with an identical (111) surface orientation and similar microscopic oxide surface area. These include films deposited on Ir(100) by physical vapor deposition (PVD) under ultrahigh vacuum (UHV) conditions and electrodeposited films on Au(111), Au(100), or CoOOH(001). We present simultaneously obtained electrocatalytic and operando structural data of these samples and correlate those quantitatively. All samples exhibit reversible potential-dependent changes in grain size and strain in the pre-OER regime, which are assigned to the formation of the skin layer. Our data clearly show that the average thickness of this skin layer, which varies strongly with the sample type, has a pronounced impact on the OER activity. A quantitative relationship between the skin layer volume and the electrocatalytic properties is established, which implies that the entire skin layer is a three-dimensional OER-active region. In addition, we discuss this structure–reactivity relationship to estimate a turn-over frequency (TOF) and provide insight into the number of sites involved in the reaction.

Experimental Section

Preparation of Co3O4(111) Films on Ir(100)

Deposition of Co3O4 on Ir(100) was performed by physical vapor deposition (PVD) under UHV conditions,[8,9] using a procedure adapted from the method of Heinz and co-workers.[59,60] Initially, the disc-shaped Ir(100) single crystal (Surface Preparation Laboratory, 99.995%, depth of roughness <0.03 μm, accuracy of orientation <0.1°) was cleaned by Ar+ bombardment (Linde 6.0, 9 × 10–5 mbar, 1.8 keV, 300 K) and several annealing cycles (UHV, 1123 K, 5 min; followed by annealing in O2, Linde 5.0, 5 × 10–8 mbar, 1123 K). Flashing the sample to 1123 K in UHV yielded the Ir(100)-(5×1) reconstructed surface. Following this, successive flashing of the surface to 873 K in an O2 atmosphere (5 × 10–8 mbar) and cooling to 393 K in O2 yielded the Ir(100)-(2×1)O reconstructed surface, which was confirmed by LEED. On this substrate, cobalt (Alfa Aesar, 99.95%, 2 mm rod) was then deposited in a reactive O2 atmosphere (8 × 10–6 mbar) at temperatures between 243 and 300 K to avoid the initial adsorption of water, with the evaporation rate calibrated by a quartz crystal microbalance. The evaporation time was adjusted to yield films of 15 nm thickness. An ordered film was formed by annealing in O2 (5 × 10–7 mbar) at 523 K for 3 min, at 698 K for 5 min (1 × 10–7 mbar), and finally in UHV at 698 K for 3 min. The successful preparation of (111)-oriented Co3O4 was confirmed by LEED, which revealed the typical diffraction pattern of the spinel structure.

Preparation of Co3O4(111) films on Au(100), Au(111), and CoOOH(001)

Electrodeposition (ED) of Co3O4 on Au(100) and Au(111) was performed on hat-shaped single crystals (MaTecK, accuracy of orientation <0.1°), which were initially cleaned for 1 min in a hot 1:2 mixture of 30% H2O2 and 96% H2SO4 (both Carlo Erba, RSE) and then flame-annealed for 5 min using a butane torch. Oxide deposition was performed in an aqueous, oxygen-free solution of 1 mM Co(NO3)2 + 1.2 mM sodium tartrate (tart) + x M NaOH, with either x = 1 or x = 2 prepared from high-purity Co nitrate (Merck, >99.0%), Na tartrate (Sigma-Aldrich, ACS reagent, >99.5%), NaOH (Merck, ACS reagent, Fe content <0.0005%), and Milli-Q water. Oxide deposition was performed at reflux temperature (∼103 °C) at a constant potential of -0.55 V vs a mercury sulfate reference electrode, where the Co(II) complex is oxidized and Co(III) precipitates on the electrode surface.[5,6] A charge density of 8 mC cm–2 was passed to grow Co3O4 films with (111) orientation of 15–25 nm thickness. After deposition, the samples were removed from the reflux cell, rinsed with ultrapure water, and dried with Ar. For Co3O4 deposition on CoOOH(001), an atomically smooth CoOOH(001) layer was first electrodeposited on Au(111) in a 5 M NaOH Co tartrate solution at reflux temperature as described in ref (7). The sample was removed from the reflux cell, rinsed with ultrapure water, and dried with Ar before use for the deposition of the Co3O4 film on top.

Ex Situ AFM Characterization of the Samples

The morphology of the samples was characterized with atomic force microscopy (Agilent PicoPlus) in a N2 atmosphere using the tapping mode. Silicon AFM tips with a cantilever oscillating frequency of ∼190 kHz (μmasch) were used. For each sample, different regions of the samples (at distances of ∼1 mm) were imaged to ensure that the observed morphology was representative of the sample morphology. The measured images allowed us to obtain the Co oxide coverage and roughness. Since the films are composed of tightly packed islands, the film roughness obtained by AFM depends on the tip sharpness and may be underestimated.

Operando SXRD Experiments

Operando SXRD studies employed the same methodology as in our previous study[7] (see Supporting Information, Section S1 for details) and were performed at two different surface diffraction beamlines: (i) at ESRF beamline ID03 (photon energy 22.5 keV, photon flux 4 × 1011 counts/s, beam size 250 μm width × 30 μm height) and (ii) at PETRA III beamline P23 (photon energy 18.7 or 22.5 keV, photon flux 5 × 1011 counts/s, beam size 200 μm width × 30 μm height). All measurements were performed in a six-circle geometry and at a fixed grazing incidence angle of 0.34°.

During the measurements, the samples were kept in an electrochemical cell specifically designed for operando SXRD studies.[7] For the samples on Au substrates, the cell was made from PEEK, with a PTFE seal fitted tightly around the hat-shaped sample to expose only the polished top part with the epitaxial Co3O4 film to the electrolyte. Imperfect sealing can result in partial penetration of the electrolyte into the gap between the PTFE seal and the crystal wall, resulting in leakage currents that manifest as a small slope in the voltammograms. This affects the pseudocapacitive charges but not the OER current, which is severely limited by mass transport restrictions within the gap. The Ir(100) samples were measured in a hanging meniscus cell[61] due to the different geometric shapes of these crystals (disc-shaped instead of hat-shaped). In this cell, the electrode is in contact with a free-standing electrolyte meniscus, which provides a better-defined geometric sample area, but limits the accessible potential range to that of low OER currents (≤2 mA cm–2). In all experiments, the samples were mounted in air and brought in contact with the electrolyte under open-circuit conditions. Then, potential control was established with a potentiostat (Ivium CompactStat) using a Ag/AgCl reference electrode (3.4 M) connected via a glass capillary and either a glassy carbon rod or a Pt wire installed in the outflow as a counter electrode. For comparison with the literature, all potentials in this work are referred with respect to the reversible hydrogen electrode (RHE).

All measurements were performed in 0.1 M NaOH (pH = 13), made from NaOH (Sigma-Aldrich, suprapure) and Milli-Q water. The electrolyte was continuously exchanged at a rate of 5 μL/s, using a remote-controlled pump system. This prevents the accumulation of radicals generated by the X-ray beam.

Results

Structure and Morphology of the Co3O4 Films

Prior to studies in an electrochemical environment, the prepared thin films were characterized ex situ using SXRD and AFM. Figure a, which shows crystal truncation rods (CTRs) of the samples at an in-plane scattering vector qinplane = 2.5385 Å–1, demonstrates that all deposited films consist of spinel-type cobalt oxide, i.e., Co3O4, with the (111) orientation. On all surfaces, including those with a square symmetry (Au(100) and Ir(100)), only the (111) orientation is observed. This is in agreement with previous work on electrodeposited or PVD-grown Co3O4 films.[5−10] In addition, all deposits have a well-defined epitaxial arrangement with respect to the substrate lattice. In the accessible q-range 0 < qz < 5 Å–1, six bulk diffraction peaks for Co3O4 are observed (marked by solid lines), which can be identified as the (1̅1̅3), (22̅2), (004), (115), (404), and (226) peaks of a (111)-oriented Co3O4 film. These (HKL) indices are given with respect to the simple cubic Co3O4 unit cell to facilitate comparison with the literature.

Figure 1

SXRD results on the structure and epitaxial arrangement of the five samples. (a) Crystal truncation rods at qinplane = 2.5385 Å–1, the in-plane scattering vector of the Co3O4(404) peak. Intensities are displayed on alogarithmic scale and with a vertical offset between different samples. Positions of Co3O4 peaks along the rods are marked with solid lines and indexed according to the Co3O4 simple cubic unit cell. Positions of CoOOH peaks are marked with dotted lines and indexed according to the CoOOH hexagonal unit cell. (b) Schematic in-plane diffraction patterns of the five samples, derived from reciprocal space surveys, that illustrate the epitaxial arrangements. The positions of Co3O4 and substrate CTRs are indicated by gray and colored circles, respectively.

Table gives an overview of selected samples, their respective deposition conditions, and the resulting film properties. In the following, samples will be denoted as “Co3O4/substrate-1M or -2M” to indicate on which substrate the Co3O4 film is deposited and, for the films on Au(111), whether the films were prepared in 1 M or 2 M NaOH. For the sake of clarity, only operando SXRD and electrochemical characterization of the samples listed in Table are shown in the main text. However, three further samples, obtained by the same preparation methods, were also characterized by operando SXRD. Corresponding data, given in the supplementary information (Table S1), will be used in the global discussion.

Table 1

Overview of the Co3O4 Samples Presented in this Study and Their Respective Deposition Conditionsa

sample	substrate	method	d⊥ (nm)	d|| (nm)	disland (nm)	σ
Co3O4/Ir(100)	Ir(100)	PVD in UHV	15	18	22	1.07
Co3O4/CoOOH-1M	CoOOH(001)/Au(111)	ED in 1 M NaOH	15	19	62	1.08
Co3O4/Au(111)-1M	Au(111)		18	18	52	1.04
Co3O4/Au(100)-2M	Au(100)	ED in 2 M NaOH	12	14	54	1.5
Co3O4/Au(111)-2M	Au(111)		24	35	31	1.15

The average film thickness d⊥ and in-plane grain size d|| were obtained by XRD, whereas the lateral island size disland and the roughness factor σ were obtained from the AFM measurements.

Because the Ir(100) and Au(100) surfaces have a square and the Au(111) and CoOOH(001) surfaces have hexagonal symmetry, different in-plane arrangements are expected for the Co3O4(111) deposits. We observe that for Co3O4/Ir(100) and Co3O4/Au(100)-2M, the Co3O4 [112̅] direction is oriented along the [100] and [010] directions of the substrate lattice (Figure b). The latter is in agreement with previous results.[6,8−10,59] The resulting lattice mismatch is quite large, 17% for Co3O4/Au(100)-2M and 10% for Co3O4/Ir(100). In contrast, for the Co3O4 films on Au(111) and CoOOH(001)/Au(111), both the film and the substrate have a hexagonal in-plane arrangement where the [112̅] directions of the Co3O4 deposit and the Au(111) lattice are aligned. Here, the lattice mismatch is only 0.9%.

For Co3O4/CoOOH-1M, the CTR shows three additional peaks (Figure a, dashed lines), which are identified as the (012), (017), and (018) peaks of the underlying CoOOH(001) film. They appear on the CTR because they are located at almost exactly the same in-plane position. These CoOOH peaks are also weakly present on the CTR of Co3O4/Au(100)-2M, revealing that this sample is not completely pure-phase Co3O4 and contains a minor CoOOH component.

Atomic force microscopy (Figure ) observations indicate that all deposits exhibit a granular morphology, are highly homogeneous, and cover the substrate completely. In the case of the Au(100) substrate, the darker areas of the AFM image are also covered by Co3O4 islands with smaller heights. The islands typically have a smooth top, terminated by a Co3O4(111) surface, and a triangular or hexagonal shape, with edges that are oriented at angles of 120° with respect to each other and which are parallel to the main lattice directions of the underlying single crystalline substrate. This morphology is in accordance with the epitaxial nature of the Co3O4 films and suggests that the films are fully crystalline.

Figure 2

AFM images of the five samples obtained before immersion in the electrolyte. The bottom row shows horizontal cross-sections through the image (marked by white lines in the images).

The main difference between the different types of deposits is the in-plane island size and the surface roughness, which depend on the preparation conditions. In accordance with the literature,[8,9,59,60] the PVD-prepared Co3O4/Ir(100) samples are composed of a high density of (111) oriented tightly packed islands of similar height. Co3O4 electrodeposition in 1 M NaOH (Co3O4/Au(111)-1M and Co3O4/CoOOH-1M) results in films that consist of large, tightly packed islands with extended (111) top surfaces. Films with more disconnected three-dimensional islands are deposited from electrolytes containing 2 M NaOH (Co3O4/Au(111)-2M and Co3O4/Au(100)-2M). In this case, the side walls of the islands constitute a considerable fraction of the total oxide surface that is exposed to the electrolyte. For Co3O4/Au(111)-2M, these side walls are in addition oriented only loosely along well-defined lattice directions and, therefore, will be composed at least partly of surface orientations that differ from Co3O4(111).

A more quantitative analysis of the morphology was performed by determining the average in-plane size of Co3O4 islands in the AFM images (disland) as well as the in-plane grain size (d||), obtained from the in-plane width of Co3O4 Bragg reflections in the SXRD data (see Table ). Notably, disland considerably exceeds d|| in many of the samples. In particular, this is found in Co3O4/Au(111)-1M and Co3O4/CoOOH-1M films, which consist of triangular islands with long straight edges that reflect the epitaxial growth and (111) orientation of the Co3O4 film. Here, disland is ≈3 times larger than d|| (Table ), indicating that despite their flat-top surface, the islands consist of several grains separated by narrow grain boundaries, which may not be resolvable by AFM. Previous UHV-STM studies of Co3O4/Ir(100) films grown by PVD[59] suggest the presence of such narrow grain boundaries within the oxide film islands.

If we assume that the grain boundaries are too narrow for major electrolyte penetration, only a fraction of the vertical edges of the grains is in full contact with the bulk electrolyte in Co3O4 films electrodeposited in 1 M NaOH. In contrast, for electrodeposits formed in 2M NaOH, where disland ≈ d||, the oxide film consists of free-standing three-dimensional islands consisting of typically just one grain. Thus, all of the edges of each grain are exposed to the electrolyte. This morphological difference will have to be kept in mind to understand the difference in the potential-dependent changes of these samples.

The electrochemically active surface area (ECSA) of the films is determined from the roughness factor σ obtained from the AFM images (Table ). For the films deposited in 1 M NaOH and the PVD films, σ is close to 1. The highest σ value is found for Co3O4/Au(100)-2M, followed by Co3O4/Au(111)-2M. For the latter, the true roughness most likely is somewhat underestimated by σ because of the small size of the three-dimensional (3D) oxide islands and resulting convolution effects with the finite AFM tip size. A conservative estimation yields an upper limit of 3.2–4.5 for the Co3O4/Au(111)-2M samples (see Supporting Information, Section S4). Still, this is much smaller than the differences in electrocatalytic activity discussed in the following. Thus, simple geometric effects can be ruled out for explaining the different properties of the samples.

Electrochemical Behavior

The catalytic properties of the different samples were characterized in a solution of 0.1 M NaOH, where the Co3O4 films are very stable. The samples were found to withstand many potential cycles into the OER regime, performed over several hours, without irreversible structural changes or changes in the cyclic voltammograms (CVs) as long as the potential was kept positive of 0.77 V (see Supporting Information, Figures S2 and S3). This also indicates the absence of significant Fe impurities under the employed experimental conditions, which are known to result in progressive shifts in the CVs with time.[62] The CVs presented in Figure were acquired in the SXRD cells described above and are corrected for the IR drop in solution.

Figure 3

Cyclic voltammograms of the five samples in 0.1 M NaOH (pH = 13). Dashed lines show the pre-OER region with j multiplied by a scaling factor and offset by 5 mA cm–2 for clarity. All potentials are IR-corrected; current densities are corrected by roughness factor σ of the oxide film.

The OER overpotential η, determined for a current density of j = 1 mA cm–2, differs for the different families of samples (Tables and S1), with Co3O4/Ir(100) showing the highest value of η, i.e., the lowest activity. The oxides electrodeposited from 1 M NaOH (Co3O4/Au(111)-1M and Co3O4/CoOOH-1M) have intermediate OER activity, and those deposited from 2 M NaOH (Co3O4/Au(111)-2M and Co3O4/Au(100)-2M) are the most active catalysts. Overall, the overpotentials vary by up to 120 mV, corresponding to 2 orders of magnitude difference in the OER current densities at a fixed potential (Table S1). The Tafel slopes (Tables and S1) are similar for all samples, suggesting a similar rate-determining step of the OER. They are comparable to those found in previous studies of the OER on Co oxide catalysts at pH 13.[63]

Table 2

Electrocatalytic and Structural Data of Co3O4 Samples Presented in This Studya

sample	Tafel slope (mV dec–1)	η (V)	Δd⊥ (nm)	Δd|| (nm)	Δε⊥ (%)	Δε|| (%)
Co3O4/Ir(100)	67	0.474	0	0.1	0.01	0.01
Co3O4/CoOOH-1M	56	0.411	0.3	0.2	0.15	0.01
Co3O4/Au(111)-1M	57	0.422	0.2	0.1	0.17	0.05
Co3O4/Au(100)-2M	64	0.396	0.3	0.9	0.43	0.24
Co3O4/Au(111)-2M	67	0.351	0.7	2.2	0.53	0.19

Given are the Tafel slope and the OER overpotential η for a current density of 1 mA cm–2, the changes in in-plane (Δd||) and out-of-plane (Δd⊥) grain size between 1.00 and 1.65 V, and the relative changes in-plane (Δε||) and out-of-plane (Δε⊥) strain over this potential range.

In the pre-OER regime, the CVs of the four samples (Figure ) prepared by electrodeposition exhibit a comparable shape, which is in accordance with CVs reported in the literature.[2,12] The current in the range of 30 to 100 μA cm–2 is much larger than what would be expected for simple double-layer charging and is attributed to pseudocapacitive processes.[7,16,45] Near the onset of the OER, a pair of redox waves is found at about 1.45 V, which is close to the CoOOH/CoO2 equilibrium potential.[2,12,64] In the CV of the PVD-deposited sample, Co3O4/Ir(100), similar albeit much weaker redox peaks are found, and the pseudocapacitive current is only a few μA cm–2 in the potential range below 1.4 V. The low current density of the latter sample is in accordance with previous electrochemical measurements.[9]

Operando SXRD Measurements

Structural changes in the film were monitored by operando surface X-ray diffraction measurements. The same method as in our previous studies of Co oxide catalysts was employed.[7] Together with the electrochemical current, four structural parameters were obtained simultaneously from two-dimensional (2D) detector images of a Co3O4 Bragg peak as a function of potential. The in-plane (d||) and out-of-plane (d⊥) grain size, which was obtained from the horizontal and vertical full width at half maximum (FWHM) of the Bragg peak via d = 2π/σFWHM, and the in-plane (ε||) and out-of-plane (ε⊥) strain, calculated from the positional shift of the Bragg peak’s in-plane and out-of-plane scattering vector via ε = qbulk/q – 1 (see Supporting Information, Section S1 and Figure S5 for characteristic raw data). Measurements were performed at two different peaks, namely, HKL = (404) and (131), which gave comparable results. To ease the comparison between the different samples, the changes in grain size and relative strain changes are plotted using 1 V as the reference state of the sample. The SXRD data obtained in an inert gas atmosphere prior to immersion and after immersion at 1 V provide an almost identical height d⊥ of the grains, i.e., of the film thickness. This indicates that no major electrolyte-induced restructuring of the Co3O4(111) surface occurs at this potential, making it a suitable reference point.

Operando SXRD results on the structural changes during the potential cycles in 0.1 M NaOH are summarized in Figure and Table . The PVD-prepared sample (Co3O4/Ir(100)) stays unchanged in the full potential range. There are no changes in grain size and strain within the experimental detection limits (<0.01% in Δε; <0.1 nm in Δd). In contrast, all electrodeposited samples exhibit reversible grain size and strain changes as a function of potential, which are highly reproducible in successive cycles (see data for Co3O4/Au(111)-1M and Co3O4/Au(111)-2M for examples). As discussed previously,[7] the hysteresis between the positive and negative potential sweep is attributed to the finite potential sweep rate. In potential step experiments (Figure S7), the skin formation occurred on timescales of a couple of seconds and the resulting steady-state changes in grain size were similar as in the potential sweep measurements. This is in accordance with our previous work, where similar structural changes were found in SXRD measurements performed at scan rates of 10 and 50 mV/s and under stationary conditions.[7]

Figure 4

Operando surface X-ray diffraction of the five samples performed during cyclic voltammetry in 0.1 M NaOH, showing from top to bottom the change in in-plane grain size, the change in out-of-plane grain size, the change in in-plane strain, and the change in out-of-plane strain. Potentials were IR-corrected. As a reference point for the strain and grain size changes, E = 1.00 V was chosen.

The amplitude of the structural changes differs substantially for the differently prepared Co3O4(111) films. In general, the largest changes are found for the samples prepared in the 2 M NaOH electrolyte (Co3O4/Au(100)-2M and, in particular, Co3O4/Au(111)-2M) and the smallest for Co3O4/Ir(100). We mention in passing that for Co3O4/Au(111)-1M and Co3O4/Au(111)-2M, a decrease in grain size is also observed at potentials <1 V. It can be attributed to the initial stage of Co3O4 conversion to Co(OH)2 and will not be discussed here further.[7]

For a more detailed analysis, we first discuss the horizontal and vertical strain (Figure , bottom two rows). For all electrodeposited samples, both Δε|| and Δε are negative and increase in magnitude with the increasing potential, indicating a lattice contraction, i.e., a shrinking of the Co3O4 unit cell volume. Moreover, the magnitude of the Δε change is roughly twice that of Δε||. The larger Δε value may be expected, as the in-plane expansion and contraction of the Co3O4 islands are constricted by epitaxial clamping to the substrate lattice and neighboring islands. The variations of Δε|| and Δε are rather monotonous over the entire pre-OER range, but in some cases seem to become less pronounced near the onset of the OER (especially for Δε⊥). For the samples grown in 2 M NaOH (Co3O4/Au(100)-2M and Co3O4/Au(111)-2M), the potential dependence is 4–10 times larger than for samples grown in 1 M NaOH (Co3O4/Au(111)-1M and Co3O4/CoOOH-1M).

A more complex behavior is found for the horizontal and vertical change in average grain size (Δd|| and Δd⊥) (Figure , top two rows). As mentioned in the introduction, these changes are related to the reversible structural transformation of the oxide surface region at potentials >1.15 V, resulting in the formation of a skin layer (Figure a). In agreement with the previous studies of this phenomenon, we observe no diffraction peaks of other Co oxide or oxyhydroxide phases in this potential regime and thus likewise attribute the skin layer to an X-ray amorphous CoO(OH) phase with an increased Co oxidation state.[2,3,7] This increase in oxidation state was also confirmed for the electrodeposited Co3O4(111) films in preliminary in situ XAS experiments (to be published). The X-ray amorphous nature of the skin layer implies that the sublattice of the Co ions, which dominates the X-ray diffraction, has to be disordered. This is in full agreement with assumptions based on the coordination of the cations in the spinel lattice, where Co2+ occupy the tetrahedral (Th) sites and Co3+ the octahedral (Oh) sites, respectively. Oxidation of the Co2+ will thus require a change in their coordination, resulting in the disordering of the metal sublattice. On the basis of X-ray absorption spectroscopy results, Bergmann et al. proposed that the Co oxidation state increases within the pseudocubic close-packed O2– lattice under retention of the Co coordination but that at more positive potentials amorphization occurs, accompanied by a partial change in Co coordination from tetrahedral to octahedral and a slight rearrangement of the O2– lattice.[2,3] This rather local structural rearrangement would facilitate the reversible transformation of the Co3O4 in the skin layer region.

Figure 5

Schematic models of the crystal structure of (a) electrodeposited Co3O4(111) films prior to skin layer formation and (b) PVD-prepared Co3O4(111) films. (c,d) Scheme of the Co3O4 film morphology formed by deposition in (c) 2 M NaOH and (d) 1 M NaOH, illustrating the structure at potentials <1.15 V (top) and after the formation of the skin layer (indicated in red) at more oxidative potentials in the OER range (bottom). Grain boundaries in the islands are depicted by black lines.

The skin layer starts to form at potentials of roughly 1.15 V, which is several hundred mV negative of the OER regime. As already discussed in our previous work,[7] this onset potential is in good agreement with the thermodynamic equilibrium potential for the transition between Co3O4 and CoOOH[64]Thus, the reversible skin layer formation has to be attributed to the oxide redox chemistry rather than the OER. The charge transfer associated with this reaction contributes to the pseudocapacitive current in the pre-OER regime.

The amount of skin layer formation differs for the different samples. Between 1.15 and 1.55 V, Δd|| and Δd⊥ are largest for Co3O4/Au(111)-2M. For the other electrodeposited samples, the Δd⊥ changes have similar values. The situation is different for Δd||, where the decrease is still ≈1 nm for Co3O4/Au(100)-2M but <0.2 nm for Co3O4/Au(111)-1M and Co3O4/CoOOH-1M. This difference can be partly explained by a geometric effect. In the case of Co3O4/Au(100)-2M and Co3O4/Au(111)-2M, in which a large fraction of the grains’ side walls is exposed to the electrolyte, significant changes in horizontal and vertical grain size are observed, with Δd|| > Δd⊥. These indicate that the skin layer forms everywhere on the top and sides of the oxide grain, as already proposed in our previous study (Figure c).[7] For flat-top films electrodeposited in 1 M NaOH (Figure d), the vertical changes Δd⊥ are only 0.3 ± 0.1 nm, which suggests that on planar Co3O4(111) surfaces, only a single Co layer is transformed during skin layer formation. Here, the Δd|| changes are much smaller. Taking into account that d|| ≈ disland/3, the islands in these islands have to contain internal domain boundaries, which are narrow (see above). It is likely that the skin layer forms at the islands’ external facets only and not at the internal grain boundaries, resulting in an effective reduction of Δd|| (Figure d). However, as we show below, the differences in the amount of formed skin layer are substantially larger than those that can be accounted for by the differences in the sample morphology.

In contrast to the electrodeposited oxides, the PVD-deposited Co3O4/Ir(100) samples remain structurally unchanged. For these films, the potential-dependent changes in grain size and strain are negligible, with Δd|| and Δd⊥ being both <1 Å and Δε|| and Δε⊥ being less than 0.03%. These observations indicate that no skin layer forms on these samples, especially also not on the islands’ top surfaces. This is surprising, taking into account the identical crystal structure and (111) surface orientation of the oxide film as compared to the electrodeposited films. This behavior differs from that of all electrodeposited films, where even on extended flat Co3O4(111) islands, a thin skin layer forms.

The structural stability of PVD-deposited Co3O4/Ir(100) has to be attributed to the different preparation method, which is a high-temperature process in a water-free environment. Previously detailed STM/LEED-IV studies of PVD-prepared oxide layers in UHV revealed that the top of the islands is atomically smooth and exhibits a characteristic surface structure.[59] Specifically, the Co3O4(111) surface is terminated by a sublayer of Co2+ cations (half-filled) residing on an O2– layer, followed by a Co3+ cation layer (Figure b). The layer spacing between the top Co2+ and subsequent O2– layer is strongly contracted as compared to the Co3O4 bulk spinel lattice, and the O2– ions in the surface layer exhibit almost no vertical buckling but slight in-plane relaxations. It was shown that this structure survives the transfer to the alkaline electrolyte,[9] although slight changes in the terminal atoms may occur, e.g., due to surface hydroxylation. Therefore, oxidation of the topmost Co2+ cations in the pre-OER regime is expected above 1.15 V, which leads to a surface region that contains two planes of Co3+ cations at the onset potential of the OER. This likely suppresses further surface restructuring in the vertical direction, taking into account that also for planar Co3O4(111) surfaces of electrodeposited films, only the 2 topmost Co atomic layers are transformed (Δd|| < 0.5 nm). From a structural viewpoint, oxidation of the topmost Co2+ cation plane, which already deviates from the bulk lattice structure, should not be detectable by SXRD, even if this would lead to a relaxation of its atomic positions. Further support for this scenario comes from recent SXRD observations for UHV-prepared Fe3O4(100) single crystals in NaOH solution.[65] For this oxide, the presence of a reconstructed surface layer, which extends over three cationic planes and exclusively contains Fe3+ ions, was found to be sufficient to stabilize the structure of Fe3O4 bulk and no skin was observed including in the OER regime.[65]

Discussion

Reversible Potential-Induced Structural Changes

Our study reveals two major differences in the potential-dependent structure of the electrodeposited samples with respect to those prepared by PVD. For the electrodeposited samples, we observe the formation of a skin layer (indicated by a decrease of d⊥ and d||). Second, we find a simultaneous shrinkage of the Co3O4 bulk lattice, i.e., a reversible decrease of the unit cell volume ΔV/V = Δε⊥ + 2Δε|| (Δε⊥ and Δε|| are <0) (see Table S1). These two effects are strongly correlated; samples with thicker skin layers also exhibit larger bulk lattice changes. In the case of the PVD-prepared samples, no skin layer is formed, ΔV/V does not change significantly, and the sample structure remains unchanged over the entire studied potential range (see Figure ).

The potential-induced strain in the horizontal and vertical direction, Δε|| and Δε⊥, is strongly correlated with the thickness of the formed skin layer (Figure ). This suggests that the strain is at least partly induced by the skin layer formation. Because of the increase in the Co oxidation state in the CoO(OH) phase, a decrease in the Co–O bond length is expected, corresponding to a compressive strain. This strain is transmitted to the Co3O4 bulk lattice. Because the oxide grain sizes are much smaller than the typical decay length for lattice strain, which is in the μm range within oxide single crystals,[66] the strain should be rather uniformly distributed across the oxide grain. Within this scenario, we expect that Δε|| and Δε⊥ and thus also ΔV/V should decrease approximately proportionally to the volume fraction of the grain that is converted into the CoO(OH) skin layer, Vskin/V (see Supporting Information, Section S2 for the determination of Vskin/V). This is indeed observed in the SXRD data (Figure ). A closer look at these data further shows that the amplitude of ΔV/V is a function of the oxide morphology. It is largest for oxides grown in 2 M NaOH, where the islands consist of single grains, which expose all of their lateral facets to the electrolyte. The magnitude of the overall strain changes that would be expected due to the oxidation state change in the skin layer can be estimated to be in the range of 5 × 10–4 to 5 × 10–3 for the studied samples (see Supporting Information, Section S3), which is in good agreement with the SXRD experiments.

Figure 6

Plot of the relative volume change ΔV/V as a function of the skin volume fraction Vskin/V. Circles correspond to the samples shown in Figures –4 and diamonds correspond to auxiliary measurements.

In a similar way, the strain changes negative of 1 V may be explained by the gradual transformation of Co3O4 into Co(OH)2, i.e., a material with an even lower mean Co oxidation state. The observed direction of the strain changes over the full potential range is in full accordance with this scenario.

Impact of the Skin Layer on OER Activity

For the electrodeposited films, the Co3O4 is covered in the OER regime by the skin layer (Figure a), which does not result from the OER but from the oxide chemistry[7] and which is the OER-active phase. Figure shows that the OER activity of the Co3O4 films increases with the skin volume per electrode surface area. This quantity corresponds to a length that we denote in the following as ⟨dskin⟩ (calculated as described in Section S2 of the Supporting Information). The true meaning of this length is discussed below. At a fixed potential of 1.65 V (Figure a), the current is clearly correlated with ⟨dskin⟩. The current reaches 10 mA cm–2 for the samples grown in 2 M NaOH, which exhibits the largest ⟨dskin⟩, while it is about two-decade smaller for the Co3O4/Ir(100) film, onto which nearly no skin develops (see Table S1). Correspondingly, at a fixed current density of 1 mA cm–2 (Figure b), the OER overpotential decreases with increasing ⟨dskin⟩ from 360 mV for Co3O4/Au(111)-2M to 480 mV for Co3O4/Ir(100). As already discussed in Section , this difference cannot be caused solely by the differences in the microscopic roughness of the oxide films, which is at most a factor of 4.5. Thus, it has to be attributed to the actual differences in the average skin layer thickness of the studied samples. Most probably, these are related to the different surface terminations of the Co3O4 islands. Our observations indicate that the skin layer is the smallest on the (surface reconstructed) PVD-prepared Co3O4(111) film. For electrodeposited films, the skin layer involves about 2 Co layers for the Co3O4(111) surface of films grown in 1 M NaOH and can be thicker than 4 Co layers on the side walls of the 3D islands of Co3O4/Au(111)-2M films, where also significant amounts other facets than Co3O4(111) are exposed to the electrolyte. In addition, also differences in the surface defect densities may play a role.

Figure 7

Plots of (a) the current density at an overpotential of 420 mV and (b) the overpotential η necessary to reach an OER current density of 1 mA cmECSA–2 as a function of the average skin layer thickness ⟨dskin⟩. Circles correspond to the samples shown in Figures –4 and diamonds correspond to auxiliary measurements.

The observed correlation between ⟨dskin⟩ and the electrocatalytic activity provides strong evidence that not only the oxide surface but the entire skin layer contributes to the OER. In other words, the skin is a three-dimensional OER zone that involves Co centers located within the very first atomic planes of the oxide catalyst. This conclusion explains the results by Jiao and Frei, who found that the photocatalytic water splitting efficient on Co3O4 nanoparticles does not scale with the particle surface.[58] It is also consistent with the interpretation of operando mass spectroscopy experiments by the Baltruschat group, which showed that lattice oxygen is involved in the reaction but that only oxygen sites close enough to the surface are exchanged with O from the solution.[44]

For a deeper microscopic understanding of the correlation between ⟨dskin⟩ and the OER activity, we again consider the atomic-scale structure of the CoO(OH) skin layer phase at the oxide surface. As discussed above as well as in many previous studies, the change in the oxidation state will lead to a Co coordination change and pronounced lattice disordering. This should result in defects such as pores and nanoscale surface roughness, which will make sites below the surface accessible and thus increase the effective density of active reaction sites. We propose that these defects are of subnanometer dimensions and form by highly local restructuring. This is supported by the very high reversibility of the structural changes. Even many skin formation/recrystallization cycles do not lead to changes in the film morphology or quantitative grain sizes, indicating that long-range restructuring is unlikely (see AFM images in Figure S2). In addition, the presence of defects will increase oxygen mobility within the skin layer. This will enhance reaction pathways that involve lattice oxygen, which can be more easily supplied than in the spinel structure.

Our SXRD data provide the average thickness ⟨dskin⟩ of Co3O4 that is converted into the skin layer. In the literature, the TOF is calculated very often with respect to the total number of Co centers within the entire sample. Here, we estimate the TOF by assuming that all of the Co centers within the skin are contributing equally to the OER. Using the known Co ion density in Co3O4 (24 per (8.0837 Å)3 unit cell), we can estimate the number of sites involved in the OER as ⟨nskin⟩ =4.54 × 1015 cm–2 × ⟨dskin⟩/nm. The OER turn-over frequency (TOF) of our Co3O4(111) films is therefore TOF = jOER/(4 × e0 × ⟨nskin⟩), This approach yields a TOF at 1.65 V that ranges between 0.12 s–1 for Co3O4/Ir(100), the least active catalyst, and 0.64 s–1 for Co3O4/Au(111)-2M, the most active catalyst. We will address this discrepancy further in the next section.

How Many Sites does the OER Require?

The quantitative trends shown in Figure provide additional mechanistic insight into the OER mechanism on Co3O4. The double logarithmic plot in Figure a gives a scaling relationship j ∝ ⟨nskin⟩ with an exponent κ ≈ 1.6 (dashed line) because ⟨nskin⟩ is proportional to ⟨dskin⟩. To understand the trends for the overpotential η at a fixed value of jOER (Figure b), we consider the OER Tafel slope b and write jOER (η) = e0zk0⟨nskin⟩κ × 10η/, where z is the charge transfer per evolved O2 and k0 is the rate constant. Using this expression, η should decrease linearly with log⟨nskin⟩, with the slope being κ·b. Indeed, such a linear dependence is observed for the data shown in Figure b and the observed slope of −93 mV dec–1 is in good agreement with the experimental value of the Tafel slope (≈60 mV dec–1) times the exponent κ ≈ 1.6. If we take this exponent into account in the calculation of the TOF, the factor of 5.3 between the values for Co3O4/Ir(100) and Co3O4/Au(111)-2M can be explained.

In principle, two explanations may account for the fact that the exponent κ is clearly larger than one. First, the surface density of the OER sites may scale nonlinearly with that of the Co ions in the skin layer and, second, the OER may require more than one reaction site. The former seems less plausible, taking into account that quasi in situ EXAFS of Co3O4 did not evidence significant changes of the Co coordination shells between the pre-OER and OER conditions.[3] The second interpretation appears very likely, even though the OER is most commonly described as a four-step reaction mechanism involving a single site as assumed for the OER on Co oxide catalysts.[21] The question as to whether the OER requires one single or several sites has been debated for decades, in particular to account for the experimental value of the Tafel slope b.[67−69] More recent operando spectroscopic studies provide experimental evidence that the OER on Co oxides is a two-site reaction. Time-resolved IR studies of photocatalytic water oxidation on Co3O4 by Zhang et al. found a mechanism involving two neighboring lattice oxygen.[24] A similar mechanism was reported in operando X-ray absorption and Raman spectroscopy studies of CoOOH by Moysiadou et al., where the desorption of the adjacent oxygen atoms as O2 was identified as the rate-determining step.[53] All of these observations are consistent with the high OER activity of molecular OER catalysts containing two Co centers (Co-OECs).[70] Furthermore, recent DFT studies of CoOOH likewise indicated that two-site OER mechanisms should be energetically preferred.[56,71]

For two-site mechanisms, the reaction rate should be proportional to nskin2 and an exponent κ = 2 would be expected. The lower value observed in our experiments could be rationalized by decreasing accessibility of the OER sites that are deeper within the skin layer. This seems reasonable, considering that for subsurface sites, mass transport within the CoO(OH) phase is required.

Conclusions

Our results provide clear evidence that the disordered near-surface region forming reversibly on Co3O4 in the pre-OER regime, i.e., the X-ray amorphous skin layer, is a three-dimensional oxygen evolution reaction zone. This interphase region depends on the oxide surface orientation. For Co3O4(111) surfaces of electrodeposited samples, the transformation seems to be limited to the topmost two Co layers. Somewhat thicker interphase layers apparently form on other Co3O4 surface orientations, which are present on the side walls of the epitaxial oxide islands. For PVD-prepared Co3O4(111) films, the amount of reversibly formed skin layer is insignificant. The different amounts of the OER-reactive interphases explain why the OER activity of Co3O4(111) epitaxial films with the same bulk spinel structures may span over two decades, while the geometrical roughness of all samples spans a much narrower range. Using true operando SXRD, we measured the average thickness of the oxide that is converted into the skin layer and estimated the number of active sites within the skin layer on all samples. Moreover, the scaling law between the reaction rate and the site density indicates that two surface sites are involved in the OER.

These observations are relevant for the fundamental understanding of the oxide’s reactivity as well as for developing strategies to activate the surface for the OER. Most studies implicitly assume that the catalytic reaction proceeds exclusively on the catalyst surface and involves adsorbates and surface lattice oxygen. If this would be the case, the transformation of the outmost atomic layer of the oxide into a phase with a higher oxidation state would be sufficient. The formation of thicker skin layers would not lead to increased activity and thus qualitative knowledge about the presence of a restructured surface would be sufficient for understanding the catalytic activity. Our results indicate that the OER activity directly scales with the thickness of the restructured skin layer over a range spanning several nanometers, i.e., length scales that strongly exceed those of an oxide monolayer. Fundamental understanding of the oxide’s reactivity therefore requires quantitative data on the spatial extension of the restructured layer under OER conditions. Furthermore, the activity of such catalysts may be substantially boosted by activation strategies that promote the formation of thicker skin layers.

In fact, our results suggest that rather subtle differences in the surface morphology can have a large influence on the OER activity of Co3O4 catalysts. This may not only be the case for pure and doped Co3O4 but also for other spinel-type transition metal oxide catalysts. Thus, great care has to be taken in comparing the OER activity of differently prepared catalysts. The described surface morphology effects can make it difficult to derive a clear interpretation of the influence of catalyst preparation conditions or chemical modifications on its catalytic properties as long as the spatial extension of the reactive skin layer is not determined. On the other hand, these observations pave the way to a controlled surface engineering of such oxide catalysts. As illustrated in this work, the oxide morphology can be tuned by the deposition method. This allows preparation of highly stable oxide surfaces with a well-defined but low number of reaction sites, which are highly suitable for fundamental studies of the OER mechanism, as well as more reactive catalysts with surface orientations or chemical modifications that promote the formation of thicker OER-reactive interphase layers. These ideas, gained from model catalysts, may provide a basis for the targeted design of highly active real oxide catalysts.