Catalytic Activity and Stability of Non-Platinum Group Metal Oxides for the Oxygen Evolution Reaction in Anion Exchange Membrane Electrolyzers

Catalyst materials were obtained from commercial suppliers and used without further treatment. PGM baseline: IrO₂ (Alfa Aesar, 99.99%); non-PGM monometallic oxides: Mn₂O₃ (US Research Nanomaterials Inc., 99.2%), γ-Fe₂O₃ (US Research Nanomaterials Inc., 99%), Co₃O₄ from two suppliers (US Research Nanomaterials Inc., 99%; Particular Materials, no purity provided), NiO (US Research Nanomaterials Inc., 99%); non-PGM bimetallic oxides: NiFe₂O₄ from three suppliers (US Research Nanomaterials Inc., 98%; Nanografi, 98.99%; Sigma Aldrich, ≥98%), CoNiO₂ (US Research Nanomaterials Inc., 99.9%); cathode catalyst: Pt on C support (47% Pt, Tanaka Kikinzoku Kogyo, TEC10E50E).

X-ray diffractograms (XRD) were obtained (2θ = 13.5°–88°) using a Bruker D8 Discover with Cu K-α radiation (λ = 1.5406) and GADDS XRD system. Crystalline sizes were determined from the XRD patterns using Bragg's Law and the Scherrer equation. ³² The Brunauer-Emmet-Teller (BET) surface area ³³ of each screened catalyst was determined from N₂ physisorption isotherms (at 77 K) measured using a Quantachrome Autosorb iQ. The double layer capacitance was estimated from cyclic voltammograms collected in the non-Faradaic region (0–0.6 or 0–0.3 V) at scan rates of 100, 50, 20, and 10 mV s⁻¹ and used as a proxy for the electrochemically active surface area ³⁴ (details on this method are included in Section S1 of the Supporting Information (SI)).

Ink solutions for rotating disc electrode (RDE) testing were formulated to target a catalyst loading of 17.8 μg_M cm⁻² on the electrode and an ionomer loading of 0.1 μg_Nafion^® mg_cat ⁻², following procedures reported in literature. ^35,36 Specifically, measured quantities of the oxide catalysts were dispersed in a solution consisting of 76 vol% deionized (DI) water (Milli-Q; ≥18.2 mΩ resistance and <5 ppb organic carbon content) and 24 vol% of n-propanol (Sigma Aldrich, OmniSolv, HPLC Grade); tables including full ink RDE formulations are included in the SI (Tables SI–SIII, SI). This solution was placed in an ice bath for at least 300 s before the ionomer was added. This mixture was then bath sonicated (900 s), horn sonicated (45 s), and bath sonicated again (2100 s), all in ice. Additional horn sonication was performed as needed before deposition onto the electrode. Prepared inks (10 μL) were deposited onto hand-polished (Alumina MicroPolish, 0.3 μm, Buehler, 40–6363–006) Au RDE working electrodes rotating at 100 RPM. The rotation speed was increased to 500–700 RPM for drying; the optimal rotation speed for drying varied amongst metal oxides and was selected based on the most-uniform coverage achieved, determined visually.

Anode inks for MEA testing were formulated to target 0.5 mg_cat cm⁻² and 30 wt% ionomer/catalyst (Versogen, PiperION-A TP-85, 5 wt% in ethanol) in a solution consisting of 10 vol% DI water and 90 vol% n-propanol. Cathode inks were formulated to target 0.3 mg_cat cm⁻² and 30 wt% ionomer/catalyst (also PiperION-A TP-85 from Versogen) in a solution containing 57 vol% DI water, and 43 vol% n-propanol; full ink formulations are listed in Tables SIV and SV, SI. Differences in ink formulations between RDE and MEA are related to differences in deposition methods, as described in previous works. ^12,35 Anode inks were airbrush sprayed directly onto the Ni porous transport layers (PTL; mounted to a vacuum hot plate, 80 °C). Cathode inks were sprayed directly onto the membranes (mounted to a vacuum hot plate, 80 °C) using a Sonotek spray station with Accumist nozzle. Anode and cathode catalyst loadings were confirmed with X-ray fluorescence (XRF; measurements taken 3x at 30 s exposures on a Fischer XDV-SDD XRF).

All RDE tests were conducted at ambient temperature (23 °C) in 170 ml of 0.1 M sodium hydroxide (NaOH) electrolyte (Sigma-Aldrich, TraceSELECT grade, 99.9995%), bubbled with N₂ (≥300 s) to remove dissolved O₂. A glass, three-electrode cell was used (Fig. 1a), which included a catalyst-coated Au RDE working electrode (0.196 cm², Pine Research Instrumentation, AFE5T050AU), Au wire/mesh counter electrode, and reversible hydrogen reference electrode (RHE). The working electrode speed was controlled by a modulated speed rotator (Pine Research Instrumentation, AFMSRCE) and electrochemical measurements were performed using an Autolab PGSTAT302N potentiostat (Eco Chemie, Metrohm Autolab). Linear sweep voltammetry was used to determine catalyst activity; these tests were preceded by five cycles of cyclic voltammetry from 1.2 to 1.8 V (to condition and break-in the catalyst) and were obtained between 1.2 and 2.0 V (scan rate: 20 mV s⁻¹, RDE rotation rate: 2500 revolutions per minute (RPM)). All RDE results were corrected for ohmic (iR) drop through the electrolyte using values obtained from a current interrupter (1.65 V) which was built into the Autolab potentiostat; this correction was performed during the measurement and not during post-processing. All reported voltages are vs RHE. Atmospheric pressure at the experiment site (Golden CO) was assumed to be 82.2 kPa based on an elevation of 1730 m; changes to the thermodynamic cell potential based on this pressure difference from standard conditions (101.3 kPa) were determined and are discussed in Section S4 of the SI.

Zoom In Zoom Out Reset image size

To assess catalytic stabilities, catalyst activities at 1.65 V were measured before and after a 13.5 h hold at 1.8 V. Prior to post-test data collection, (i) the used electrolyte was stirred, and 15 ml were extracted for inductively coupled plasma mass spectrometry (ICP-MS) analysis (Thermo Scientific, iCAP-Q ICP-MS), (ii) the electrolyte was refreshed and saturated with N₂, and (iii) the electrodes were gently rinsed with DI water and allowed to dry to remove any trapped O₂. ³⁷

IrO₂, Mn₂O₃, Co₃O₄, and NiFe₂O₄ catalysts were tested in single-cell, membrane electrode assemblies (MEA) which each had an active area of 5 cm² (Fig. 1b). Specifically, the MEAs used consisted of carbon paper gas diffusion layers (Fuel Cell Earth, MGL280, 80280–40), Ni PTLs (Bekaert, BEKIPOR 2NI 18–0.25), PTFE gaskets (254 μm; thickness selected to facilitate compression of the MEA of approximately 20%) and PiperION-A TP-85 (Versogen) membranes (thickness = 80 μm). Hardware for these tests consisted of Al endplates (Fuel Cell Technologies), Au current collectors, and Ni triple serpentine flow fields (Fig. 1b). Before testing, membranes were soaked in 3 M potassium hydroxide (KOH; Millipore, Emsure grade) for 48 h (with the solution refreshed once at 24 h), to facilitate ion exchange from carbonate to hydroxide form; membranes were then stored in 0.5 M KOH until use.

All MEA tests were performed in a 1 M KOH supporting electrolyte (at 80 °C) with a flow rate of 50 ml min⁻¹ to the anode and no flow to the cathode (to reflect PEM commercial operating conditions where H₂ backpressure is applied). ⁴ All MEA tests were performed using a two-electrode configuration and reported potentials are cell potentials. The equilibrium cell potential was corrected for the lower atmospheric pressure of the test site (82.2 kPa) and the non-standard temperature (80 °C) using the Nernst equation (Section S4, SI). ^38,39 Electrochemical measurements were taken using an Autolab PGSTAT302N potentiostat (Eco Chemie, Metrohm Autolab) and a 20 A booster (Eco Chemie, Metrohm Autolab). For the initial screening, all materials were tested with the following protocol: (1) potentiostatic polarization curve (20 s per voltage) at 1.40 V, 1.45 V, 1.50 V, 1.55 V, 1.60 V, 1.65 V, 1.70 V, 1.80 V, 1.90 V, and 2.00 V, (2) potentiostatic hold at 2 V (for cell conditioning), (3) potentiostatic polarization curve (120 s per voltage) at identical intervals to step 1, (4) Cyclic voltammetry (100 mV s⁻¹, 50 mV s⁻¹, and 20 mV s⁻¹); (5) Electrochemical impedance spectroscopy at 1.25 V and at intervals identical to step 1.

RDE measurements were used to screen the OER activities of non-PGM monometallic (Mn₂O₃, Fe₂O₃, Co₃O₄, NiO) and bimetallic (NiFe₂O₄, CoNiO₂) oxides and compare them to IrO₂ (Fig. 2). The physical and electrochemical properties of the screened oxide catalysts are listed in Table I. RDE measurements minimize interferences from transport limitations of reactants (OH⁻) and enable rigorous comparisons among catalysts by separating kinetic (V < 1.7 V) and evolved O₂(g) transport limited (V > 1.8 V) regimes. Furthermore, due to the low loadings of the catalysts on the RDE tips, we expected minimal interference of catalyst layer microstructure on the transport of O₂ (g) products; thus, we approximated measured current densities as being directly proportional to O₂ (g) evolution rates. Measured current densities were normalized by four different methods, (i) the geometric surface area of the electrode (where trends are equivalent to those for raw current, as data for all catalysts were divided by the same geometric factor), (ii) the physical surface area (estimated from BET methods), (iii) the double-layer capacitance (determined from cyclic voltammetry), and (iv) the mass loading of metal in each catalyst. The geometric surface area of the electrode is commonly used in literature to normalize current densities, ^15,27 although this method does not provide any information on site accessibility. Normalization by the BET surface area accounts for differences in the physically accessible surface area, but does not account for differences in electrochemically active surface areas, nor the type and number of active sites. Rigorous quantification of the electrochemically active surface area requires appropriate probe molecules that adsorb on specific active sites with known stoichiometry; for example, CO stripping methods are widely used for metals such as Pt. ³⁴ However, an appropriate probe molecule has yet to be identified for non-PGM oxides due to difficulties in finding a molecule that selectively and strongly binds to electrochemically active sites for these oxides. Alternative methods include measurement of the double layer capacitance, which approximates the electro-active surface area based on the quantity of charge involved in forming the double layer. ³⁴ This method, however, can have large inherent errors due to the fact that the double-layer capacitance varies with both the catalyst material and the composition of the electrolyte, and cannot be translated to an electrochemically active surface area without the knowledge of the specific capacitance of an ideal, smooth surface of the material of interest obtained in identical electrolyte conditions. ³⁴ As a result, this method can only provide order-of-magnitude comparisons amongst different materials. ¹⁵ While developing a more accurate method to quantify electrochemically active sites within metal oxides for OER catalysis is desired, it is beyond the scope of this study. Consequently, performance trends were determined from the obtained data normalized by multiple methods.

Zoom In Zoom Out Reset image size

For each normalization method, the overpotentials were compared at relevant current densities. For the geometric surface area, 10 mA cm⁻² was selected based on reporting conventions in the literature. For all other methods, the appropriate current density was selected such that the region being compared was similar to that for 10 mA cm⁻² in the geometric surface area assessment, and such that the current was within the Tafel region for the majority of the non-PGMs tested. Additionally, the activity of the Au RDE working electrode was obtained without a catalyst layer to ensure the Au substrate was not significantly contributing to measured activities; these results are included in Fig. S2 of the SI.

IrO₂, the state-of-the-art OER catalyst for PEM electrolyzers, was used as a benchmark in this study (shown in black in Fig. 2). Note that IrO₂ is likely not a feasible candidate for use in commercial AEM systems due to the tendency for Ir to oxidize into soluble forms at high pH and due to its high capital cost; ⁶ however, as this material is the current standard for PEM electrolysis and is widely used in many device-level AEM publications, it is studiedhere to provide context and establish general performance targets for the non-PGM oxides tested. For the data normalized by the geometric surface area (Fig. 2a), IrO₂ showed the lowest overpotential (<300 mV) at the onset of OER, an exchange current density on the order of 10⁻⁶ mA/μg_M, and a moderate Tafel slope below 1.55 V (70 mV/dec) which increased to 120 mV/dec at approximately 1.6 V, consistent with results reported in previous studies. ^16,40,41 These changes to the Tafel slope can be directly related to changes in kinetic mechanism, specifically in either the identity of the rate determining step or in the surface coverage; in later sections, we discuss the differences in Tafel slope as it relates to possible differences in kinetic mechanism for IrO₂ and non-PGM oxides. Note that above ∼1.8 V the Tafel slope continues to increase for IrO₂, (and, in fact, for all samples); while such changes may be indicative of additional changes in kinetic mechanism, this cannot be easily distinguished from changes related to transport limitations of evolved O₂(g) leaving the catalyst surface and, therefore, changes to the Tafel slope above 1.8 V are not considered in discussions of changes in the kinetic mechanism.

The composition of the IrO₂ sample was evaluated via XRD. The diffractograms of this sample (Fig. S3, SI) confirmed the presence of IrO₂ (2Θ = 30°–40°, amorphous ⁴² ); however, these diffractograms also indicated that the sample contained large Ir nanoparticles (2Θ = 40°, 47.5°, 69°, and 84°, closely matching JCPDS card 46–1044; crystalline size 20 nm). Ir metal is expected to be more active than IrO₂; previous works have reported 2-times higher site-specific activity for Ir vs IrO₂ ^16,40,41 and 10-times higher activity per geometric surface area of the electrode for Ir vs IrO₂. ¹⁶ Despite the possible inhomogeneity of the as-received IrO₂ sample, and subsequent differences in Ir/IrO₂ ratio loaded on the RDE, the activities of seven replicates were found to be similar (Fig. S4, SI). We hypothesize that the contribution of Ir metal on measured activities was compensated by the larger particle sizes of Ir (and thus lower surface areas) relative to those of IrO₂, resulting in negligible changes in activity with different Ir/IrO₂ ratios. However, the degree of deactivation of four replicates varied significantly; two samples showed an improvement in activity (by 33%–43%) and two samples showed deactivation (by 69%–76%). Ir and IrO₂ have been shown to have different deactivation rates in acidic media due to differences in their dissolution kinetics; Ir has been found to dissolve at a faster rate than IrO₂, which is suggested to be the result of oxidation into soluble forms. ^6,41 Oxidation kinetics are expected to increase with increasing pH (higher OH⁻ concentration), and thus we expect that Ir deactivation via this Ir oxidation and dissolution mechanism would be exacerbated in alkaline media. Therefore, we hypothesize that samples with higher deactivation rates were those with higher Ir metal contents. This is discussed further in the section on stability.

Both Mn₂O₃ and Fe₂O₃ exhibited high overpotentials (>500 mV) across all current density regimes, regardless of the normalization method (shown in magenta and red, respectively, in Fig. 2). Poor OER activity of Mn-based materials has been well-reported in literature; Mn content has been shown to increase the overpotential for OER in ferrous (MFe₂O₄) ⁴³ and cobalt-based spinel (M_xCo_3-xO₄) ⁴⁴ materials. Interestingly, Mn₂O₃ had a significantly higher exchange current density than Fe₂O₃ (on the order of 10⁻⁶ mA/μg_M for Mn₂O₃ vs 10⁻¹¹ mA/μg_M for Fe₂O₃) suggesting that Mn₂O₃ had faster inherent kinetics than Fe₂O₃. However, Mn₂O₃ also exhibited a higher Tafel slope than Fe₂O₃ (120 mV dec⁻¹ for Mn₂O₃ vs 60 mV dec⁻¹ for Fe₂O₃ for data normalized by the geometric surface area, Fig. 2a). A higher Tafel slope translates to a larger overpotential required to increase the rate by an order of magnitude, and, consequently, high Tafel slopes are often associated with poor performing catalysts. Poor OER activity of Fe₂O₃ has been reported previously and has been attributed to the low electrical conductivity of Fe ^45–47 and the non-optimal binding strength of O intermediates on Fe. ^17,48 Burke et al. found in their study of ultra-thin film oxy(hydroxides) that FeO_xH_y had the highest turnover frequency (defined as the number of O₂ molecules evolved per second per metal site), when compared to monometallic (oxy)hydroxides of Mn, Co, and Ni, ²⁴ suggesting that Fe sites can have high activity for OER when the conductivity limitation is mitigated. Furthermore, combing Fe with Ni can lead to better O-binding energetics and improved activity. ⁴⁸ In fact, Fe sites are often used as dopants in multimetallic OER catalysts such as NiFe₂O₄, which will be discussed in detail later.

NiO exhibited much lower overpotentials (∼200 mV lower) than Fe₂O₃ and Mn₂O₃ across current density regimes and normalization methods (shown in yellow in Fig. 2). Additionally, NiO performance approached that of IrO₂ above 10 mA cm⁻² and 1 mA/μg_M when the current densities were normalized by the geometric surface area (Fig. 2a) and metal loading (Fig. 2d), respectively. In fact, NiO showed higher activity than IrO₂ across current density regimes (0.05 to 1 mA μF⁻¹) when the current densities were normalized by the double layer capacitance (Fig. 2c). Furthermore, NiO had an exchange current density on the order or 10⁻⁶ mA μg_M ⁻¹, similar to that of IrO₂. Note that a curved, vertical feature was observed for NiO between η = 200–300 mV; such a feature was likely caused by NiO oxidation to NiOOH. ^49–51 This feature was also present for the other normalization methods but at lower current densities than those plotted in Figs. 2a, 2b, and 2d. Although NiO has been previously reported to exhibit high OER activity, ^20–22 this has often been attributed to a trace amount of Fe impurities present either in the electrolyte ^23,52 or in the oxide lattice. ^23,52,53 To exclude any effects of Fe contamination in the electrolyte, a Pt electrode was held in the electrolyte solution at −0.5 V for 900 s to electrochemically plate out possible contaminants before testing; any Fe ions in solution should reduce and deposit onto the Pt electrode as metal at such potentials. No change in NiO performance before and after the contaminant removal procedure was observed (Fig. S5, SI), indicating that the electrolyte was free of Fe contaminants. It is possible, however, that Fe was present in the as-received NiO sample due to contamination during synthesis.

Co₃O₄ showed the highest activity of all tested non-PGM oxide catalysts, exhibiting the lowest overpotential for OER (∼500 mV, Figs. 2e–2h) across normalization methods and across the current density regimes tested (shown in blue in Fig. 2). Additionally, Co₃O₄ surpassed IrO₂ performance at high current densities (>10 mA cm⁻²) across all normalization methods except by the BET surface area, for which Co₃O₄ instead approached IrO₂ performance (Figs 2b, 2f). This sample had an exchange current density on the order of 10⁻⁸ mA/μg_M and a Tafel slope of 60 mV dec⁻¹. While Co-containing materials have been reported in the literature to have high OER performance—a result which has been attributed to high electrical conductivity, ⁵⁴ highly mobile oxygen vacancies, ^55,56 and the presence of multiple oxidation states in Co spinel materials ¹⁹ —few studies have reported Co₃O₄ performance exceeding that of IrO₂. A Co₃O₄ sample from a different supplier was also evaluated to confirm that Co₃O₄ performance was not sample-specific (Fig. 3). The two samples had the same spinel structure, as confirmed by XRD diffractograms (Fig. S6, SI) and nearly identical double-layer capacitances (approximately 10 μF for each). The samples, however, had different BET surface areas (40 vs 90 m² g⁻¹ for Co₃O₄ A and B samples, respectively; Table I) due to the difference in their particle sizes (15.5 vs 9.6 nm from XRD for Co₃O₄ A and B samples, respectively; Table I). These samples had nearly identical performance across normalization methods (Fig. 3), except by the BET surface area (Fig. 3b), and both samples surpassed IrO₂ activity at moderate current densities (∼5 mA cm⁻²; Fig. 3a). These results show that although a general correlation is expected between BET and electrochemical surface areas, there are differences in the number of sites that are physically accessible to N₂ (g) vs electrochemically accessible to OH⁻, posing questions about the relevance of using BET surface area as a normalization method in electrochemical systems. It is also worthwhile to note that the curved feature for Co₃O₄ B is the result of a phase change from Co₃O₄ to CoOOH, ⁵⁷ which is also present for Co₃O₄ A but at slightly lower current densities.

Zoom In Zoom Out Reset image size

An assessment of the Tafel slopes for all monometallic oxide materials revealed a possible potential dependence of the mechanism for IrO₂ vs non-PGMs; the Tafel slope for IrO₂ changed at lower current densities (5 mA cm⁻² _geo) than non-PGM oxides (>10 mA cm⁻² _geo). This change is not expected to be related to transport limitations, as the trend persisted when the scan rate was changed (Fig. S7, SI) and no phase change is predicted for this region in Pourbaix diagrams. ⁶ Moreover, if the deviation was caused by transport limitations of evolved O₂ (g), all catalysts should deviate from their linear Tafel regions at the same O₂ (g) generation rates, and, therefore, at the same current. Instead, this deviation was more likely the result of a mechanism change, either in the kinetically relevant step or surface coverage of bound intermediates. ⁸ Conversely, the kinetic region for all non-PGMs extended into the moderate current density regimes; IrO₂ is the only material that exhibited a change in Tafel slope at current densities lower than 10 mA cm⁻² _geo, suggesting that there may be a different kinetic mechanism governing IrO₂ vs non-PGMs. Furthermore, for data normalized by the double-layer capacitance, the change in the Tafel slope for IrO₂ versus non-PGMs was more dramatic. This is in part due to the large double-layer capacitance of IrO₂ obtained from the cyclic voltammetry method (discussed in Section S1, SI); this high double-layer capacitance may be related to the accessibility of "inner" reacting surface sites, as has been suggested for RuO₂. ⁵⁸ It is also interesting to note slight differences in Tafel slope even among non-PGMs, suggesting that there may be a difference in mechanism (or surface coverages) among these materials. In particular, for current normalized by the geometric surface area, NiO initially had a Tafel slope dissimilar to the other testing Ni-containing materials (120 mV dec⁻¹ for NiO vs 60 mV dec⁻¹ for other Ni-containing materials) for current densities below 1 mA cm⁻², abovewhich all Ni-containing materials had a similar Tafel slope of 120 mV dec⁻¹. This would suggest that Fe and Co dopants change the mechanism on a pure Ni surface, either by changing surface coverage or kinetically relevant steps.

Previous reports have suggested that bimetallic systems, such as Ni-Fe oxides, have higher OER activities than their monometallic counterparts due to the synergy between the two metals. ^23–25 NiFe₂O₄, however, showed similar overpotentials across all normalization methods (shown in orange and yellow, respectively, in Fig. 2) and similar exchange current densities (on the order of 10⁻⁶ mA/μg_M for NiO and 10⁻⁷ mA/μg_M for NiFe₂O₄) compared to those of NiO. Such results contradict previous reports that have shown that even a small amount of Fe enhances Ni performance when incorporated into the NiO crystalline structures ^23,52,53,59 or in the electrolyte solution. ^23,52 Optimal Fe contents proposed by these studies range from 15 to 40 wt% Fe in the bimetallic catalyst, ^23,53,59 whereas the NiFe₂O₄ spinel material studied in this work contained 66 wt% Fe (by metal content). For NiO, the electrolyte was rigorously cleaned prior to testing (see the section on NiO above and Section S7, SI), ruling out the possibility of Fe contamination in the electrolyte impacting NiO results. Instead, we hypothesize that NiFe₂O₄ samples had a prohibitively high Fe content which inhibited its activity. Stability testing of NiFe₂O₄ revealed an improvement of performance over time-on-stream concurrent with the increased dissolution of Fe (confirmed from ICP-MS methods), suggesting that an increased Ni/Fe ratio led to the increased performance (discussed in detail in the later section on stability testing).

Figure 4 compares the OER activities of three NiFe₂O₄ samples from different commercial suppliers (Table I) and NiO. The OER activities of all NiFe₂O₄ samples and NiO were found to be similar across normalization methods; these samples had similar onset potentials (350 to 400 mV) with polarization curves that met at moderate current densities (∼1 mA cm⁻² _geo; Fig. 4a). The Tafel behavior of NiO and NiFe₂O₄ C closely matched across normalization methods (Figs. 4a–4d) and differed slightly from that of NiFe₂O₄ A and B; for current normalized by the geometric surface area, NiO and NiFe₂O₄ C had Tafel slopes of 120 mV across current densities whereas NiFe₂O₄ A and B had Tafel slopes of 60 mV dec⁻¹ at overpotentials below ∼500 mV which increased to 120 mV dec⁻¹ above η = 500 mV. Again, Tafel slopes are compared only in the kinetic regime (below ∼1.8 V) to avoid possible interferences from evolved O₂(g) transport in the analysis. To evaluate whether this behavior was caused by structural differences, X-ray diffractograms were obtained for all samples (Fig. 5a). These diffractograms show that all NiFe₂O₄ samples contained peaks corresponding to NiFe₂O₄ spinel (2Θ = 30.2°, 35.6°, 37.2°, 43.4°, 57.4°, 63.0°, 75.2°, and 80.0°; closely matching JCPDS card 86–2267) with differing amounts of α-Fe₂O₃ contaminants (2Θ = 33.1°, 40.8°, 53.8°, and 64.2°; closely matching JCPDS card 39–1346); these phases are indicated by black dashed and grey dotted lines, respectively, in Fig. 5a. NiFe₂O₄ A, NiFe₂O₄ B, and NiFe₂O₄ C, were composed of 5.0%, 23.1%, and 3.4% of the α-Fe₂O₃ phase, respectively. These values were estimated from the ratio of the area of the most prominent peak assigned to the α-Fe₂O₃ phase (2ϴ = 33.1°) vs the sum of this α-Fe₂O₃ peak area and the area of the most prominent peak assigned to the NiFe₂O₄ phase (2ϴ = 35.6°). The differences in Tafel behavior for materials with different quantities of the α-Fe₂O₃ phase pose questions about the relative activities and kinetic mechanisms of these different phases of Fe. Furthermore, NiFe₂O₄ C, which had the greatest amount of α-Fe₂O₃, showed the largest double-layer capacitance, while NiFe₂O₄ A, which had the least amount of α-Fe₂O₃, showed the smallest double-layer capacitance (Figs. 5a, 5b). The α-Fe₂O₃ and NiFe₂O₄ phases have different magnetic properties (α-Fe₂O₃ is weakly ferromagnetic at room temperature whereas NiFe₂O₄ is ferrimagnetic) ^60,61 which have been linked to differences in catalytic activities. ^62,63 Differences in physical surface area and crystalline size were also evaluated for all NiFe₂O₄ samples to investigate a possible particle size effect. Samples A and C had comparable crystalline sizes and BET surface areas (10 nm and 75 m² g⁻¹, respectively), while sample B had a slightly larger crystalline size and thus slightly lower surface area (∼13 nm and ∼55 m² g⁻¹; Table I). These differences, however, did not impact performance; sample B showed similar overpotentials to sample A across current density regimes and normalization methods, excluding by BET surface area (Figs. 4a–4d). This suggests that for the NiFe₂O₄ samples evaluated, crystalline size was a less significant factor in determining performance, although the samples tested may not have provided enough range to observe significant effects of particle sizes.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

The OER reactivity of CoNiO₂ was compared to that of Co₃O₄ and NiO (shown in green, blue, and yellow in Fig. 2, respectively, and separately in Fig. S8 in SI) to assess any possible synergy between Ni and Co. Compared to Co₃O₄, CoNiO₂ showed higher overpotentials across current densities and normalization methods (∼50 mV larger) and a similar exchange current density (on the order of 10⁻⁸ mA/μg_M for both CoNiO₂ and Co₃O₄). In contrast, when compared to NiO, CoNiO₂ had a similar overpotential but a larger exchange current density (10⁻⁸ mA/μg_M for CoNiO₂ and 10⁻⁶ mA/μg_M for NiO). CoNiO₂ had a comparable double-layer capacitance (10.3 vs 11.5 μF for CoNiO₂ and Co₃O₄, respectively; Table I) to Co3O4, but a smaller BET surface area (20.3 vs 36.6 m² g⁻¹ for CoNiO₂ and Co₃O₄, respectively; Table I). X-ray diffractograms for CoNiO₂ (Fig. 5b) showed peaks matching both CoNiO₂ (2Θ = 37.2°, 43.2°, 62.8°, 75.2°, and 79.2°; closely matching JCPDS Card 10–0188) and Co₃O₄ (2Θ = 31.2°, 36.8°, 38.5°, 44.8°, 55.5°, 59.3°, 65.2°, 77.4°, and 78.5°; closely matching JCPDS card 43–1003) with different crystalline sizes (11.3 vs 17.6 nm for CoNiO₂ and Co₃O₄, respectively; Table I). These phases are indicated by black dashed and grey dotted lines, respectively, in Fig. 5b. The crystalline size of Co₃O₄ in the mixed oxide was comparable to that of monometallic Co₃O₄ (17.6 vs 15.5 nm for Co₃O₄ in CoNiO₂ and Co₃O₄ (monometallic), respectively; Table I), pointing away from a particle size effect being the cause of decreased activity for CoNiO₂ vs Co₃O₄. The fact that CoNiO₂ had lower activity than Co₃O₄ alone may be due to a difference in the Co oxidation state between these two materials; the Co oxidation state in CoNiO₂ is 2+, vs 2.67+ (average) in Co₃O₄ and 3+ in CoOOH. The CoOOH phase is likely formed before OER onset based on the curved feature in Co₃O₄ Tafel plots and reported phase diagrams for Co. ^6,57 It is further possible that differences in catalyst synthesis procedures, including the possibility of surfactants left on the surface and blocking sites, may be masking potential Ni-Co synergy advantages.

In summary, our RDE activity screening results suggest that Fe₂O₃ and Mn₂O₃ are less-feasible candidates as OER catalysts. In contrast, Co₃O₄ was found to be the highest-performing catalyst across normalization methods, and NiO and NiFe₂O₄ had similar and relatively high OER activities. These RDE results show that non-PGMs, such as Co₃O₄ and Ni-containing materials, are therefore viable OER catalysts for integration into AEM devices based on their high OER activities. Next, we discuss their stability and their performance at the device level to further evaluate their possible replacement of IrO₂ catalysts.

A selection of materials (IrO₂, Mn₂O₃, Co₃O₄, and NiFe₂O₄) was evaluated for stability at operating conditions by comparing the current (i at 1.65 V) before and after stressing at 1.8 V for 13.5 h. A potential of 1.65 V was selected for the activity comparison because it lies in the kinetic region for all four catalysts tested; previous works have used 1.55 V to compare kinetic-region activity for OER catalysts, ⁴¹ but the late onset potential of Mn₂O₃ necessitated the use of higher voltage. A combination of ICP-MS analyses and assessment of the change in the active surface area via determination of the double-layer capacitance before and after testing was used to assess possible deactivation/activation mechanisms. While the double-layer capacitance has inherent errors when used to compare materials of different identities, it is considered accurate for evaluations between materials of the same identity. ³⁴

Four independent tests were performed for the IrO₂ sample; these tests showed high variability among experiments, which led to a large error bar in Fig. 6a. Two runs resulted in a significant loss in activity at 1.65 V (by −69% to −76%), while the other two runs exhibited an increase in activity (by +33% to +43%); the activity changes for each sample are shown in Fig. S9 (SI). Such a dramatic activity loss was not related to the dissolution of Ir into the solution; ICP-MS analysis of the used electrolyte revealed that all samples showed only a small loss of Ir that was similar among the four runs (1%–4%; Table SVI, SI). Instead, we hypothesize that the discrepancy among replicates is related to the inhomogeneity of the as-received IrO₂ sample. As discussed in the previous section, initial activities for all runs were similar; this is likely attributed to the largercrystalline size (and, relatedly, smaller surface area) of Ir metal compensating for its predicted higher activity, ^16,41 thereby resulting in the similar initial activity regardless of Ir metal content. Upon exposure to elevated potential, those large Ir metal nanoparticles would be oxidized into large IrO₂ particles; it is expected that these large IrO₂ particles would be essentially inactive due to the inherently lower activity of IrO₂ vs Ir metal and large particle sizes. Indeed, the errors among four different IrO₂ runs diminished when the results were normalized by the double-layer capacitance determined before and after stability testing (Fig. 6b).

Zoom In Zoom Out Reset image size

Mn₂O₃ samples did not deactivate within experimental error, and results were consistent between samples both for measured current and for current normalized by the double-layer capacitance. ICP-MS results indicated a mass loss of ∼1% for all Mn samples (Table SVII, SI). The activity of Mn₂O₃ remained lower relative to the other tested samples.

Co₃O₄ samples significantly deactivated over time-on-stream (by 45%–71%; Fig. 6a). Such a deactivation percentage decreased to 29%–54% when the current was normalized by double-layer capacitance measured before and after stressing (Fig. 6b), suggesting that Co₃O₄ deactivation was partially due to a loss in active surface area. A loss in active surface area can be caused by delamination, dissolution, catalyst oxidation, or particle growth. ICP-MS results indicated minimal Co dissolution (mass loss: <0.1%; Table SVIII, SI), and Co delamination losses are expected to be minimized by the RDE coating technique utilized. ^35,41 Furthermore, if Co delamination had occurred, the activity drop would be expected to be larger; therefore, Co loss (dissolution or delamination) was not responsible for deactivation. Instead, we hypothesize that the mechanism of catalyst deactivation was caused by oxide growth and, relatedly, particle growth, which led to a decrease in active surface area; this is supported by an observed decrease in the double-layer capacitance (∼30%) for the Co₃O₄ samples (Fig. S10, SI). Additional catalyst deactivation may have been caused by decreased electrical conductivity with increasing oxide content. At the device level, the implications of Co₃O₄ deactivation via oxide and particle growth would require the development of conditioning procedures that could minimize and stabilize this deactivation method.

The activity of NiFe₂O₄ increased after stress testing (+60% to +87%; shown in orange, Fig. 6). As discussed above, the synergy between Ni and Fe for OER is well-discussed in literature, where bimetallic systems of these two metals have often shown higher activity compared to their monometallic counterparts. ^23,52,53 Initial activities of the three NiFe₂O₄ samples evaluated in this study, however, matched the activity of NiO (Fig. 4), possibly because the as-received NiFe₂O₄ samples had too-high Fe content (66 wt% Fe by metals content); the most active compositions reported in literature ranged between 15 to 40 wt% Fe. ^23,53 We hypothesize that the observed activity increase of NiFe₂O₄ after stability testing was due to Fe dissolution and a decrease in Fe content. In fact, ICP-MS results showed that the three tested samples had mass losses between 2 to 8% (Table SIX, SI), where increased Fe dissolution correlated to increased activity. Additionally, the extent of improvement increased from 60%–87% to 147%–203% when the current was normalized by the double-layer capacitance. Normalization by the double-layer capacitance was performed to capture changes in the active surface area upon stressing; the double-layer capacitance for these samples decreased by 24%–47%. These results seem to suggest that the improvement, at least in part, was due to a change in per-site activity as optimal Fe content was approached. We also note that the improvement in activity for NiFe₂O₄ may be related to the development of NiFeOOH surface phases, as NiFeOOH has been shown in the literature to have high activity for OER. ²⁴ No notable changes in redox features were observed in the cyclic voltammograms of this sample before and after testing, suggesting either that there was no phase change, or that such a phase change from NiFe₂O₄ to NiFeOOH was immediate upon contact with either air or the NaOH electrolyte. When considering scale-up to the device level, methods to stabilize Fe content at an optimal value while preventing further Fe dissolution are necessary. Specifically, further elucidation of the extent of Fe dissolution over time, the effect of more aggressive stress testing (i.e., cycling), and methods of stabilizing NiFe oxide catalysts with optimal Fe content will be needed to facilitate the scale-up of this material to the device level. Further exploration of the activation of NiFe₂O₄ and its possible integration at the device level is the subject of our current work, which will be published at a later time.

Figure 7 shows the MEA results for IrO₂, Mn₂O₃, Co₃O₄, and NiFe₂O₄. While RDE testing provides fundamental insights into thekinetics of thesematerials, MEA testing is required to assess the integration feasibility of these materials at the device level; this method addresses the effects of incorporating the catalyst layer with the anion exchange membrane and gas diffusion and porous transport layers. In fact, RDE and MEA results may vary due to the complex interplay between catalyst, ionomer, and electrolyte, or the effects of device hardware. ^40,64–66 Thus, a parallel assessment of RDE and MEA testing is needed to show how integration variables—including ink rheology, catalyst-ionomer interaction, and catalyst layer structure and uniformity—influence the catalytic performance of PGM and non-PGM oxides.

Zoom In Zoom Out Reset image size

Mn₂O₃ exhibited the highest overpotential across current density regimes, followed by IrO₂ > NiFe₂O₄ > Co₃O₄ (Fig. 7a). At 1 A cm⁻², Co₃O₄ and NiFe₂O₄ both achieved lower overpotentials than IrO₂ (η = 744 mV, 689 mV, and 718 mV for IrO₂, Co₃O_4, and NiFe₂O₄, respectively; Fig. 7b), with the overpotential gap increasing with increasing current density (at 1.5 A cm⁻², η = 858 mV, 778 mV, 807 mV or IrO₂, Co₃O_4, and NiFe₂O₄, respectively). This high activity of NiFe₂O₄ at the device level was confirmed to be consistent across multiple experiments; three independent replicates showed similar performance across current density regimes (Fig. S11, SI).

Figure 7c shows the HFR-free potential as a function of mass-specific current; trends between materials obtained in the Tafel region (here, up to about 1.6 V) of HFR-free potentials (Fig. 7c) are assumed to be related, primarily, to kinetic differences, and, therefore, these results can be reasonably compared to RDE results. HFR is related to ohmic losses from the membrane and cell hardware, contact resistances, and bulk resistances of the diffusion media, and can be calculated from electrochemical impedance spectroscopy (EIS) measurements at each potential of the polarization curve (this method is discussed in detail in Section S16 of the SI). In Fig. 7c, IrO₂, NiFe₂O₄, and Co₃O₄ all had similar activity for OER up to 1.6 V, which was consistent with the EIS results (1.60 V; Fig. 7d), which indicated that these materials have similar charge transfer resistances (related to the diameter of the hemispherical curve). The polarization results from this region (Fig. 7c) loosely followed trends from RDE testing, where Mn₂O₃ had the highest overpotential of all tested materials, Co₃O₄ had a similar overpotential to IrO₂ at 10 mA cm⁻² _geo (500 mV), and NiFe₂O₄ had a slightly higher overpotential than Co₃O₄ and IrO₂ across current density regimes. Additionally, as in RDE testing, IrO₂ deviated from Tafel kinetics at lower current densities than non-PGM materials in MEA testing; this may again be indicative of differences in mechanism for IrO₂ vs non-PGM materials. Slight differences in kinetic activity trends (i.e., in RDE there was a separation between activity for IrO₂, Co₃O₄, and NiFe₂O₄ whereas in MEA testing these materials had similar activity in the Tafel region, Fig. 7c) likely resulted from differences in catalyst layer microstructure and properties (resistances, ionomer distribution, and interfacial contacts) which are not applicable at the RDE level. Furthermore, differences observed between materials at current densities beyond those associated with Tafel kinetics (potentials > 1.6 V, Figs. 7a and 7c) were likely influenced by device integration parameters (i.e., between anode and cathode catalyst layers and the anion exchange membrane andporous transport layers) and emphasize the importance of testing at the RDE and MEA levels in parallel at the catalyst screening and development stages.

Measured overpotentials were separated into ohmic, kinetic, and residual contributions via a voltage breakdown analysis (Figs. 7e–7g; analysis details in Section S16 of the SI). Among all catalysts, the ohmic contributions to overpotential were similar (see Fig. 7e), matching expectations as all materials were tested using the same membranes and cell hardware. Kinetic contributions to overpotential (Fig. 7f) were also similar between samples, with Tafel slopes found to be between 90 and 100 mV dec⁻¹. The most significant differences in performance, however, were found to be attributable to the residual contributions to overpotential (Fig. 7g); this factor is expected to be inclusive of the transport of the KOH electrolyte through the catalyst layer, transport of O₂ (g) products away from the catalyst layer, and of catalyst layer resistances (including catalyst-ionomer interactions).

A combination of EIS analysis, loading-change experiments, and conditioning time experiments was utilized to probe the origin of residual contributions to overpotential in MEA testing. Anode catalyst loading was changed to assess possible differences in site accessibility related to the thickness of the catalyst layer. For Co₃O₄, minimal performance differences were found between loadings for Co₃O₄ B (0.2 to 0.6 mg cm⁻²), suggesting that all sites were fully accessible for this material in the catalyst loading ranges tested (Fig. S12, SI). Note that there may be additional trends observed at higher loading intervals; however, this is outside the scope of this publication. For NiFe₂O₄, thicker catalyst layers (2.0 mg cm⁻² vs 0.5 mg cm⁻²) resulted in higher ohmic losses and lower activity (Fig. S13, SI). Changes to catalyst conditioning time for Co₃O₄ B and IrO₂ were performed to assess differences in catalyst layer stabilization (Figs. S14 and S15, SI), and minimal differences were found between conditioning results for samples with different catalyst loadings; all samples exhibited a stabilization of performance at 30 ks in a test lasting 100 ks. For assessments of conditioning time for IrO₂, stabilization of performance was similarly achieved at 30 ks up to the tested duration of 100 ks.

EIS in the non-Faradaic region (1.25 V; Fig. 7h) was obtained to study catalyst layer resistances and their role in residual contributions to overpotential. The study of catalyst layer resistances and their features in EIS has been discussed for electrochemical devices generally; ^67–70 in short, catalyst layer resistances are inclusive of electronic and ionic resistances in the catalyst layer, which, in turn, consists of the catalyst (electron-conducting phase) and ionomer (ion conducting phase). High catalyst layer resistances can lead to an uneven distribution of overpotential, poor catalyst utilization, and high voltage losses as current density increases; therefore, an understanding of catalyst layer resistances is key to minimizing overpotential losses. In this work, catalyst layer resistances were found using transmission line theory for porous electrodes, ⁶⁷ where the catalyst layer resistance was assumed to be equal to three times the width of the 45° region at high-frequencies in non-Faradaic EIS. This method is discussed in Section S20, SI. The results show that catalyst layer resistances increased with Co₃O₄ < Mn₂O₃ < NiFe₂O₄ < IrO₂ (Table SX, SI). These resistances, in part, account for the differences in residual contributions to overpotential and the deviation of Tafel slopes at moderate current density (Fig. 7c). More importantly, these results provide an avenue for further optimization of AEM device-level performance; catalyst layer resistances can be decreased through novel PTL design and through optimization of catalyst layer architecture such as catalyst layer thickness, the ratio of catalyst/ionomer, and catalyst-ionomer integration. While a full analysis and discussion of catalyst layer resistances in AEM systems and their features in EIS is beyond the scope of this study, this is a topic of our future work to be published at a later time.

Thus far, we have discussed the catalytic activity, stability, and device-level integrability of non-PGM catalysts for OER in AEM systems. When determining the scale-up viabilityof OER catalysts, material criticality is also an important consideration, specifically asconcerns over material availability grow for technologies essential to a clean energy transition, including Li-ion batteries ⁷¹ and electrolyzers. ⁵ While a range of methods have been used for determining material criticality, all methods generally include a metric of the supply risk of the material (based on the likelihood of supply disruption, ranging from projected material costs, the political or economic stability of countries of origin, and recyclability of material) and of the vulnerability of a system (i.e. the global economy, a given country, a company, or a given technology) to that supply disruption. ^72,73 These two metrics are compared and a threshold is established based on the context of the study being performed. Several recent works have evaluated materials relevant to electrolyzers using such methods; for example, in a recent review by Schrijvers et al. ⁷² which considered criticality assessments across a variety of methods from a variety of stakeholders, it was found that in all reviewed criticality assessments, Co had "high" material criticality more than 2-times as often as Ni or Fe. ⁷² Furthermore, Co is currently a key component in Li-ion batteries and has appeared on critical materials lists alongside PGMs such as Ir globally. ^5,28,31 Such a high criticality of Co, in turn, poses questions about the validity of its use as IrO₂ replacement, despite its high initial OER reactivity as shown in this work.