Catalyst deactivation during water electrolysis: Understanding and mitigation

The rapid expansion of renewable energy sources such as solar and wind power has dramatically reduced electricity generation costs and amplified the need for energy storage.¹ The water electrolysis process, which dissociates water molecules into oxygen and hydrogen using surplus electricity, not only meets this increased demand for energy storage and alleviates the pressure on power grids but also sets the stage for a sustainable and eco-friendly hydrogen economy.² Currently, water electrolysis has evolved into three technologies operating under ambient conditions.³ Alkaline water electrolysis (AWE) utilizes 25–40 wt. % aqueous solutions of KOH or NaOH as the electrolyte, where hydrogen evolution reactions (HERs) and oxygen evolution reactions (OERs) are conducted. Proton exchange membrane water electrolysis (PEMWE) and anion exchange membrane water electrolysis (AEMWE) both adopt non-porous polymer membranes to function as the electrolyte and separator, with the primary distinction being the ions transported by the membrane: H⁺ for PEMWE and OH⁻ for AEMWE. At the present time, AWE and PEMWE are considered mature technologies, offering stable operation for over 60 000 h in commercial stacks.^4,5

Yet, the industrial progression of these technologies is constrained by AWE’s low efficiency and PEMWE’s high cost (due to the use of noble metals), prompting the search for effective, stable, and cost-effective electrocatalysts. While substantial research focuses on enhancing the activity of electrocatalysts or reducing their costs, discussions concerning deactivation and stability have only recently gained attention.^6–8 Insufficient stability often obstructs or even negates the viability of some “highly efficient catalysts,” impacting the maintenance/replacement cost and the long-term energy efficiency.³ Even catalysts reported as stable in lab studies with overpotential drafting of hundreds of μV h⁻¹ are far from satisfactory, as the overpotentials can accumulate after thousands of hours of operation.⁵ Therefore, a comprehensive understanding of catalyst deactivation during water electrolysis is crucial to guiding stable catalyst design. A few reviews focusing on specific topics of water electrolysis have emerged recently, from OER^9–11 to the molecular catalyst,¹² but no introductive analysis includes OER and HER.

This contribution aims to provide an introductive overview of electrocatalyst deactivation during water electrolysis. Following the deactivation evaluation methods and mechanisms reinforced by case studies, we delve into the intricacies of assessment and characterization techniques used in deactivation pathway analysis. We underscore the preliminary assessment with the Pourbaix diagram and elucidate the benefits of employing in situ methods. Furthermore, we explore recent advancements in mitigating deactivation, ranging from catalyst design to sophisticated electrode engineering. This discussion will serve as a general guide for researchers conducting deactivation analysis in water electrolysis.

No catalyst can realistically be expected to last forever. Quantitative evaluation methods, including experimental set-up and electrochemical methods, are required to compare the extension of deactivation across different catalysts.

Two commonly employed set-ups for electrochemical stability measurement are the rotating disk electrode (RDE) and membrane electrode assembly (MEA).¹³ Although RDE is more frequently used due to its convenience and lower catalyst loading, it can underestimate catalysts’ stability.¹⁴ The RDE-based test typically requires catalysts at the milligram level with a current density of ∼100 mA cm⁻² for a few hours. Such current density is significantly lower than in real-world conditions, likely due to bubble coverage.¹³ In addition, the volumes of different RDE set-ups can influence the stability assessment.¹⁵ Thus, while RDE can be used for initial catalyst screening, MEA is suggested for accurate stability evaluation as it better simulates industrial conditions.¹⁶ 

Three electrochemical methods are generally employed for stability evaluation, including chronopotentiometry (CP), chronoamperometry (CA), and cyclic voltammetry (CV) (Fig. 1).¹⁷ The CP method measures potential changes over time at a constant current density (often 10 or 100 mA cm⁻²), while the CA method monitors current response at a fixed potential. The potential increment in CP and the current decline in CA can be used as the stability descriptor. Both methods have pros and cons. The CP method represents a more industrial-relevant approach that prioritizes current density (i.e., the production rate). Still, the selection of current density by many, such as 10 mA cm⁻², is often impractical to provide meaningful insights. For instance, intense bubble evolution at high current density can expose the mechanical weakness of catalysts masked at low current density. Meanwhile, the CA method is more problematic since the electrochemical behavior of catalysts is potential-dependent, posing the challenge of choosing a representative potential value.¹⁷ Regarding the CV method, researchers often compare the voltammograms before and after thousands of cycles to justify the stability of catalysts. It is claimed that CV can simulate the on–off cycles of water electrolyzers during long-term operation. Still, despite its popularity, such an approach may confuse electrochemical stability with electrocatalytic stability. Electrochemical stability can be claimed by CV sweeping within a selected potential range, while electrocatalytic stability is only justifiable under operational conditions (constant potential or current). Overall, we suggest the CP method at a large current density to mimic industrial electrolysis and the CV method to demonstrate stability during on–off cycles. The Mullins group recently proposed a standard protocol for evaluating electrocatalytic stability in lab-scale flow alkaline water electrolyzers. It combines electrochemical impedance analysis, multistep current load, and on–off cycles.¹⁸ Readers are encouraged to follow these recommendations to evaluate the potential of the catalysts properly. It is also worth mentioning that the application of iR compensation in the post-analysis of polarization and CV data needs to be evaluated carefully. The reason is that, during electrolysis, the resistance change can reflect possible structural and compositional changes in the electrochemical system. Careless compensation can conceal useful information and mislead the conclusion.^19,20

A recent development in the field is the proposed “stability number” (or S-number) to describe catalyst stability. It is defined as the ratio between the amounts of evolved oxygen and dissolved iridium.²¹ This indicator is independent of factors such as loading, surface area, measurement duration, applied current density, molecule binder content, or substrate materials, and shows promise for comparing the stability of various materials.^15,21 However, this theory assumes dissolution as the primary deactivation mode, necessitating the development of additional methods to assess other deactivation modes such as aggregation and phase transition.

Generally, the deactivation of electrocatalysts under operational conditions refers to a decrease in current and/or loss of active sites. To understand the deactivation, we need to revisit the reaction mechanisms of water electrolysis. Water splitting encompasses cathodic HER and anodic OER, with their reversible potentials specified as 0 and 1.229 V, respectively. When the reaction deviates from equilibrium, an overpotential is necessary to generate a driving force for the reaction. This force correlates with the adsorption energy of intermediates on the electrocatalyst surface; overly strong or weak adsorption inhibits the transition from intermediates to products or the formation of intermediates, respectively. The intermediates for HER are relatively simple: adsorbed H (*H) in acidic conditions and *H along with adsorbed OH (*OH) in alkaline conditions.^22,23 Norskov et al. computed the adsorption energy of various metals and compounds, identifying a “volcano” relationship between adsorption energy and the exchange current density under acidic conditions.^23,24 This relationship was later extended to alkaline electrolytes by Markovic et al. to incorporate OH adsorption into the model.²⁵ The OER features two most prominent pathways [Fig. 2(a)]: the adsorbate evolution mechanism (AEM) and the lattice oxygen-mediated mechanism (LOM).²⁶ AEM is completed by combining two adsorbed oxygens, and the formation of OOH* is the rate-determining step, while LOM involves a combination of lattice and adsorbed oxygen, whereas the energetically unfavored step is deprotonation.²⁷ Factors that can alter the adsorption energy of intermediates or reaction pathways will influence the catalyst’s activity. These factors include surface structure,²⁸ crystalline phase,²⁹ particle size, and morphology.³⁰ Changes in these factors serve as the origin of catalyst deactivation.

In this regard, deactivation can be categorized mainly into the following pathways, as illustrated in Fig. 2(b). It should be noted that various electrocatalysts often suffer from more than one deactivation pathway.

The migration of clusters and nanoparticles during water electrolysis often forms larger particles to lower their surface energy. Such behavior not only reduces the number of sites exposed to the electrolyte but also lowers the intrinsic activity of individual sites due to surface energy reduction. One well-known example is the Pt/C catalyst (carbon-supported Pt nanoparticles), which undergoes Pt migration, coalescence, and Ostwald ripening under industrial HER conditions (large current and high temperature), resulting in a lower electrochemically active surface area (ECSA), causing performance decline.^31,32 Similar behavior is also evident in IrRu oxides during prolonged OER tests, where the small crystalline domains with high defect concentrations evolve into larger ones free of defects, leading to intrinsic activity loss.³³ Non-precious metal oxides, such as Co₃O₄ and Co(OH)₂, share a similar problem.^34,35

Such transitions are frequently due to the inherent instability of electrocatalysts in chemical and electrochemical environments, where the metastable phase converts to less active but more stable phases. For example, we previously reported that a metal–organic framework with Co nodes could evolve from highly active α-Co(OH)₂ to less active β-Co(OH)₂ during extended alkaline OER, explaining its rapid deactivation.³⁶ Lee et al. reported the degradation pathway for Ni(OH)₂ during alkaline OER and identified the conversion from γ-NiOOH to β-NiOOH that deactivated the catalyst.³⁷ Tan et al. identified a deactivation pathway of Ir/IrOx during acidic OER as the phase transition from hydrous Ir oxide/hydroxide to anhydrous Ir oxide, in addition to the well-known Ir dissolution pathway.³⁸ 

The occupation of active sites by species with stronger interactions with them prevents further utilization of the sites for electrocatalysis, thus poisoning the active sites. Such species can be impurities in the electrolyte and by-products or intermediates from side reactions. An infamous example is sulfur poisoning, which not only deactivates Pt-based catalysts during HER via Pt–S bond formation but also negatively impacts non-precious metal-based catalysts, such as MoS₂.³⁹ 

Site dissolution represents the corrosion of electrocatalysts under chemical (i.e., without potential) and electrochemical (i.e., with potential) environments, leading to metal leaching into the electrolyte. This phenomenon is the most common cause of catalyst deactivation during water electrolysis. For example, many metal oxides and hydroxides are chemically unstable in acidic conditions, as represented by Ni(OH)₂,⁴⁰ often making them unsuitable catalysts for acidic OER. The alkaline condition is not safe either, and metal oxides still exhibit certain dissolution behaviors during OER operation, as shown by Co_1+δFe_2−δO₄ and CoO catalysts, which lose iron and cobalt species during the reaction.^34,41 The dissolution behavior is typically electrolyte-dependent, as demonstrated by Hofmann and co-workers using a Co₂P electrocatalyst, which exhibits stoichiometric dissolution of Co and P during acidic HER but overwhelming P leaching during alkaline HER.⁴² Such dissolution happens readily once the catalyst is immersed in the electrolyte and accelerates with sweeping potential. The most recognizable example of site dissolution is iridium dissolution in PEMWE devices. The anodic oxidation of Ir leads to the production of the highly active hydrous IrO_x. Still, such species suffer 100-fold lower dissolution stability than IrO₂, leading to significant performance decline and Ir dissolution during long-term electrolysis.⁴³ 

The detachment of the catalyst refers to the weakening of the interaction between the catalyst layer/particles and substrate, leading to catalyst loss and an increase in contact resistance, contributing to performance decline. Such a pathway is an important contributor to the degradation of industrial water electrolyzers. For example, Zaccarine’s group demonstrated that the surface property of the substrate significantly impacts the adhesive force between the Nafion^TM binder in the catalyst layer and the substrate, where the metallic surface often results in delamination.⁴⁴ Meanwhile, vigorous gas bubble production can also compromise the structural stability of the composite by applying internal stress within the catalyst layer, promoting the detachment of active materials, as shown by Todoroki et al. in their analysis of NiCo₂O₄ during alkaline OER.⁴⁵ 

Gas bubbles adhering to the catalyst surface can temporarily deactivate the enclosed sites, leading to a performance decline. Such a problem, strictly speaking, is not catalyst deactivation but a natural part of the gas evolution reaction. Once the bubble detaches, the active sites are available again. Therefore, rapid removal of gas bubbles can minimize the impact of bubble blockage.

After outlining the common pathways to deactivation, we provide a general overview of both HER and OER catalysts below, highlighting some representative examples. Tables I and II summarize the deactivation pathways of common HER and OER catalysts.

HER catalysts are generally less unstable than their OER counterparts due to their lower operating potential.⁴⁶ For noble metal-based HER catalysts, current research is directed toward reducing the size and optimizing the atomic utilization of precious metal atoms, with agglomeration and detachment being the primary deactivation pathways.^47,48 Non-precious metal catalysts, particularly transition metal-based compounds such as sulfides,⁴⁹ phosphides,⁵⁰ carbides,⁵¹ and multi-element compounds,⁵² are identified as promising HER candidates. However, Ledendecker et al. showed that, while they exhibit satisfactory stability at low HER overpotentials, most non-precious catalysts commence dissolution into acidic solutions when the potential is halted [Fig. 3(a)].⁵³ This behavioral pattern is evident when cycling the catalyst between the operating voltage and the open circuit voltage. Moreover, despite the HER catalyst functioning at a reducing potential, oxygen permeating from the anode can still oxidize certain species on the cathode, such as oxidation from S²⁻ to SO₃²⁻,⁵⁴ P^x− to PO₄²⁻,^42,55 and Mo to MoO₄²⁻ (MoS₂ + O₂ + OH⁻ → MoO₄²⁻ + SO₃²⁻ + H₂O) [Figs. 3(b-I) and 3(b-III)].⁵⁴ These occurrences can adversely affect the catalyst’s effectiveness.

OER catalysts share similarities with those used in HER, including noble metals, alloys, chalcogenides, sulfides, phosphides, oxides, and hydroxides.^56,57 However, there are distinct differences. First, many metals are more prone to leaching during OER, especially in high-overpotential and acidic environments.⁵⁸ In particular, Fe-based and Ni-based compounds, highly active and relatively stable in alkaline conditions, can transform into Fe³⁺ and Ni²⁺ in acidic environments.⁵⁹ Similar instability in Co-based and Mn-based compounds has been identified.^60,61 Second, structural reconstruction during OER is more pronounced and widespread. In situ experiments have revealed the transformation from pristine form to (oxy)hydroxides, such as FeOOH, CoOOH, and eventually inactive phases.^36,37,62,63 For example, transition metal carbide, pnictide, sulfide, and chalcogenide are converted to metal (oxy)hydroxides shortly after applying anodic potential.^64,65 Third, the OER mechanism can involve lattice oxygen, increasing the risk of structural collapse and metal dissolution. RuO₂, for instance, is a highly active OER catalyst with a lower cost than IrO₂. Yet, its activity declines rapidly due to the dynamic dissolution of Ru sites via the LOM path [Fig. 3(c)].⁶⁶ Such observations underscore the need for ongoing research and innovation to understand these stability challenges.

To start evaluating the possible deactivation pathways of the electrocatalysts, the Pourbaix diagram (also known as the E-pH diagram) is a valuable tool for prediction. This diagram is based on the principles of electrochemical thermodynamics and offers a graphical representation of the stable phases of an element or compound as a function of potential and pH. The potential axis of the Pourbaix diagram reflects the energy of electrons, and the pH axis corresponds to the activity of protons in the solution, effectively mapping out the thermodynamically stable species at various levels of acidity and redox potential. Therefore, the diagram can predict whether a substance will be in its elemental form, ionized, or precipitated as a solid under different conditions, providing insights into potential corrosion or dissolution processes that could affect catalyst stability. Online databases, such as the Materials Project,^79,80 offer readily accessible Pourbaix diagrams for a wide range of materials, making them a convenient tool for initial catalyst evaluations. An example of Ni, Fe, and Ni + Fe (1:1 ratio) is shown in Fig. 4 to showcase the informative diagrams where both solid-state materials and soluble ion species are illustrated.

However, it is essential to recognize the limitations of Pourbaix diagrams. They provide thermodynamic predictions that may not accurately reflect the behavior of materials under real-world conditions, which also involve kinetic considerations. For instance, a predicted stable phase may not be observed if the kinetics of its formation are too slow. Similarly, a predicted unstable system is not necessarily reactive under given conditions.⁸¹ Therefore, while Pourbaix diagrams are a helpful starting point for assessing catalyst stability in HER and OER, their predictions should be further validated with experimental characterization.

There are two ways to analyze electrocatalysts experimentally: ex situ and in situ analysis. Comparing structural information before and after long-term electrolysis can offer some insights into deactivation, as a few researchers have adopted. Techniques such as x-ray diffraction (XRD) for phase changes, x-ray photoelectron spectroscopy (XPS) for valence changes, scanning electron microscopy (SEM) and transmission electron microscopy (TEM) for morphology changes, electrochemical tests for electrochemically active surface area (ECSA) changes, and inductively coupled plasma optical emission spectroscopy (ICP-OES) for metal dissolution have all been influential in studying catalyst deactivation.^82–84 However, these ex situ characterization methods provide limited insights or misleading information into the catalyst deactivation mechanism. Once leaving the electrocatalytic environment, electrochemically active species on the electrocatalyst, especially those potential-dependent and intermediate species, will undergo rapid changes (e.g., decomposition, desorption, and migration).⁸⁵ Even worse, these techniques generally require sample preparation, including liquid removal, air exposure, surface modification, and mechanical damage, which wrecks the desired information. Therefore, in situ analysis is generally preferred for deactivation analysis.

For instance, although Ni/NiO rendered the higher activity for alkaline HER than Ni and NiO, all samples exhibited identical ex situ Raman spectra with characteristic peaks of Ni(OH)₂ after 10 000 s of electrolysis.⁸⁶ Only by using in situ Raman spectra were the activity contributions from the β-Ni(OH)₂ intermediate in Ni/NiO and their decline under HER conditions (reduction to metallic Ni) during long-term electrolysis identified as the cause of deactivation. The discrepancies between in situ and ex situ results were attributed to air exposure during sample preparation. Moreover, in situ techniques simultaneously enable a more comprehensive analysis of electrocatalysts on multiple scales. For example, Povia et al. analyzed IrO₂ deactivation during accelerated stability tests using in situ small angle x-ray scattering (SAXS) and x-ray absorption spectroscopy (XAS).⁸⁷ Combining the morphologic and structural evidence, they deduced the contributions of mass reduction, active site loss, and intrinsic activity decline to be ∼15%, ∼20%, and ∼30%, respectively, which could not be reliably evaluated using ex situ methods.

Figure 5 illustrates several in situ methods that can be integrated with long-term electrolysis. These methods offer insights into the deactivation mechanism from various perspectives and scales, from the bulk electrolyte interface to the bulk catalyst. This section outlines these methodologies, underscoring their potential to address existing challenges and inherent limitations.

This involves tracking species released by the electrocatalyst into the bulk electrolyte, which can offer valuable information regarding the dissolution of metal and non-metal components. Dissolved species can be identified and quantified using ICP-OES, ion chromatography (IC), and mass spectrometry (MS).^42,88 If the species exhibits adsorption within the UV-visible range—as is the case with most transition metal species and some anions—ultraviolet-visible spectroscopy (UV-Vis) becomes a valuable tool for monitoring changes in species concentration.⁸⁹ 

Monitoring the weight of a catalyst during water electrolysis can illuminate site dissolution and catalyst detachment. The quartz crystal microbalance (QCM) is frequently used for this purpose.⁹⁰ As a sensitive device operating on piezoelectricity, it measures changes in resonant frequency caused by mass alterations on the crystal surface, such as catalyst particle detachment or dissolution. This allows real-time detection of micro-to-nanoscale weight changes. Furthermore, by varying environmental conditions, QCM can differentiate between rigid and soft changes on the surface, helping identify whether shifts are due to solid particle detachment or adsorption/desorption of gaseous or liquid species.

For studying surface poisoning phenomena, Fourier transform infrared spectroscopy (FTIR) and surface-enhanced Raman spectroscopy (SERS) are vital tools. They offer insights into the adsorption of inhibitory species that might impede catalytic activity. We have previously provided a beginner’s guide on Raman spectroelectrochemistry.⁹¹ In addition, the reconfiguration of the catalyst surface, a critical aspect affecting the catalyst’s stability and activity, can be analyzed using various x-ray methods [such as x-ray absorption spectroscopy (XAS) and XPS], Mössbauer spectroscopy, atomic force microscopy (AFM), scanning tunneling microscopy (STM), TEM, and Raman spectroscopy.^41,92–96 These techniques enable direct observation of surface changes such as atomic rearrangement and phase transformation during the electrochemical process.

During long-term electrolysis, the structural reconfiguration typically extends beyond the interface and deep into the bulk structure. Therefore, monitoring the structural evolution of the bulk catalyst can be beneficial for understanding the underlying mechanisms. Common methods for this purpose include XRD, XAS, and TEM. XRD offers valuable insights into the crystal structure, phase, and strain within the catalyst, enabling researchers to detect changes in the bulk structure over time. XAS, encompassing x-ray absorption near edge structure (XANES) and extended x-ray absorption fine structure (EXAFS), provides a more profound understanding of the bulk structure by revealing the oxidation state, coordination number, and bond distances of the elements within the catalyst.³⁷ While typically employed for surface structure analysis, in situ TEM can also be used to study the bulk structure, particularly for nanoscale catalysts, by providing high-resolution images of the catalyst and revealing changes in particle size, morphology, and internal structure.⁹⁷ An excellent example of using in situ TEM to analyze the morphological and structural evolution of Co₃O₄ during OER is provided by a group led by Portehault and Ersen.⁹³ 

Microscopy methods, including optical microscopy and TEM, can directly visualize particle detachment and bubble formation. Notably, Panchenko et al. demonstrated an in situ synchrotron radiography method that enables the observation of Ir catalyst detachment and bubble formation within the catalyst layer.⁹⁸ 

In situ methods for electrocatalyst deactivation studies do have limitations. One such issue is time resolution, as sampling may be too slow to track rapid changes. For example, collecting a Raman spectrum normally takes seconds to a few minutes, which may not be suitable to study rapid surface reconfiguration. Spatial resolution is another challenge when high-resolution data are needed to understand catalyst behaviors. Scanning electrochemical microscopy, for instance, cannot provide atomic-level resolution on the catalyst surface. Sampling depth also varies significantly,⁹⁹ affecting the ability to investigate interfacial species or go deep into the catalyst structure. Furthermore, specific technical principles, such as XRD’s inability to detect disorder features, can limit the scope of these methods. Another critical limitation is the specialized equipment and expertise required to perform these characterizations. Meanwhile, specific techniques may only provide qualitative data, limiting the extent to which results can be quantitatively analyzed. In addition, the experimental conditions required for some of these methods might not perfectly mimic real-world operating conditions, which can affect the relevance and applicability of the results. Finally, while in situ methods provide valuable real-time insights, they often only focus on isolated parameters, potentially missing the interplay of multiple factors influencing catalyst deactivation.

As a general suggestion, when selecting an appropriate in situ method for a specific study, it is crucial to understand its limitations and avoid overinterpretation. To gain a more comprehensive understanding of the deactivation process, it is highly recommended to combine pieces of evidence from multiple in situ methods.

Building on an understanding of the deactivation mechanism, mitigation efforts focus mainly on catalyst design and electrode engineering. Catalyst design aims to develop more stable and cost-effective catalysts, while electrode engineering works by regulating the external environment of the catalytically active species for better durability.

Highly ordered materials generally show higher stability than disordered lattices. For instance, aged Ni(OH)₂ with higher crystallinity is much more stable than disordered ones in acidic conditions.⁴⁰ Therefore, improving the crystallinity of catalysts can reduce the deactivation rate. Secanell and co-workers reported that IrO_x with enhanced crystallinity showed greater resistance to dissolution than disordered ones during acidic OER.³⁸ However, such crystallinity improvement often decreases the intrinsic activity of individual sites, trading the activity for stability.

Encapsulating catalysts with a more stable material has proven to be an effective countermeasure for site dissolution and catalyst detachment. For instance, Huang et al. have developed a CoS₂ layer on the exterior surface of Co-doped MoS₂ for acidic HER and monitored the dissolution of Mo and S using in situ UV-Vis spectral analysis [Figs. 3(b-II) and 3(b-IV)].⁵⁴ They noted that constructing an outer layer mitigated the dissolution of the two species, improving the electrocatalytic duration concurrently. The in situ XAS indicated that while the core remained stable, the outer CoS₂ layer, despite an increase in Co valence under open circuit potential, showed reversible oxidation under HER working conditions, which enables the outer CoS₂ layer to function as a protective barrier against oxidants such as O₂ and OH⁻. A carbon shell delivered a similar protection effect, as demonstrated by Yoo et al.¹⁰⁰ Engaging a protective layer is especially effective for OER catalysts. Xu et al. recently reported that La and Mn doping can enhance the structural stability of spinel cobalt oxide for acidic OER under near-industrial conditions, attributing to a La/Mn-based protective layer.¹⁰¹ Bhardwaj et al. demonstrated that an ultrathin SiO_x overlayer can significantly reduce the Pt dissolution caused by side-reaction with Cl⁻ during OER.¹⁰² However, one should be cautious that the effectiveness of the overlayer depends on the catalyst, as demonstrated by Boettcher’s group.¹⁰³ They analyzed the role of a TiO_x layer in preventing the Ir dissolution from IrO_x, concluding that a fully covered TiO_x could accelerate the Ir dissolution.

When lattice oxygen participates in the OER process (LOM), the structural stability of the catalyst is challenged. Therefore, modulating active sites for stable LOM or alternative AEM pathways can reduce deactivation. Shi et al. demonstrated such OER-dependent stability using a RuO₂ model catalyst.⁷⁴ The Ru charge (oxidation states) was engaged as an indicator of the Ru–O bonding structure (ionic or covalent) and strength, where the bond covalency of Ru–O is related to the difficulty in breaking the Ru–O bond, a critical step in OER and catalyst dissolution. As depicted in Fig. 6(a), the activities of the AEM pathway exhibit a volcano-like trend as a function of Ru charge. Simultaneously, the formation energies of Ru vacancy (ΔG_VRu) and oxygen vacancy (ΔG_VO) decrease with the rise in Ru charge within the LOM scope, thus being less stable [Figs. 6(b) and 6(c)]. This trend suggests a higher likelihood of LOM at a lower overpotential. Experimentally, the team synthesized an MRuO_x (M = Ce⁴⁺, Sn⁴⁺, Ru⁴⁺, Cr⁴⁺) oxide with tailored Ru charges. In situ ¹⁸O-labeled differential electrochemical mass spectrometry (DEMS) was used to quantify the contribution from AEM and LOM. Activity and stability tests mirrored the calculated trend, while XAS and in situ Raman spectroscopy indicated the superior structural stability of SnRuO_x compared to RuO_x. Such an example demonstrates the effectiveness of Ru site modulation via other metal cations in changing the OER mechanism and improving its stability. This strategy is mainly adopted for metal oxides. For example, Co₃O₄ with low crystallinity is not a stable OER catalyst in acidic conditions due to the formation of a hydrous Co oxide layer.⁷⁵ However, recent developments in binary catalysts based on Co and Mn oxides have shown significant stability under near-industrial conditions.^101,104 For instance, Co₂MnO₄ synthesized by Li et al. demonstrated an impressive lifetime of over two months (1500 h) at 200 mA cm⁻² at a pH of 1, ∼60-fold longer than that of Co₃O₄.¹⁰⁴ Co₂MnO₄ features a cubic spinel structure, with Mn predominantly occupying octahedral sites and Co occupying both octahedral and tetrahedral sites. Theoretical calculations showed that the dissolution of lattice oxygen in Co₂MnO₄ is least favored thermodynamically. Further projected density of states (PDOS) analysis revealed that the electronic overlapping between the 2p orbitals of O and the 3d orbitals is positioned at a lower energy level for Co₂MnO₄ compared to Co₃O₄, explaining its higher stability. In another report, Wang et al. incorporated Sb into the CoFe oxides and proved that the strong Sb–O bond could stabilize the lattice oxygen, avoiding Co and Fe dissolution during acidic OER.¹⁰⁵ Such results prove that the OER pathway and stability can be modulated by controlling the electronic configuration of metal sites. A similar strategy is also effective for the HER catalyst. Benson et al. modified the electronic structure of 1T-MoS₂ using interfacial electron-donating functional groups, which not only greatly improved the activity but also prevented the phase transition from 1T to the less active 2H phase.¹⁰⁶ 

Notably, a self-healing strategy for OER catalysts proposed by Nocera et al. emerged recently.¹⁰⁷ They discovered that specific catalytic metal ions (Co, Ni, Fe, and Mn) could redeposit onto the anode, creating a dynamic balance between dissolution and deposition and maintaining performance stability. With active sites hosted by an inert matrix, the self-healing CoFePbO_x demonstrated stability in acidic electrolytes at temperatures up to 80 °C.¹⁰⁸ However, this catalytic system requires additional metal species in the electrolyte, making it less practical for commercial implementation.

A common method to increase catalyst utilization efficiency involves reducing the particle size to increase the specific surface area, even to single atoms. However, smaller sizes can promote catalyst dissolution and agglomeration.^45,109 Meanwhile, particle detachment is also a major cause of the deactivation of modified electrodes. One solution is to enhance the interaction between the active materials and the substrate using an underlayer between the substrate and the catalyst layer. This can be achieved by utilizing the strong binding interaction between the catalyst precursor and specific sites within the support during the catalyst preparation. For instance, Dai et al. deposited platinum clusters onto graphene defects by mixing defective graphene with an aqueous K₂PtCl₄ solution.¹¹⁰ By controlling the dosage of the K₂PtCl₄ solution, the size of the platinum cluster and the distribution of Pt on defect sites can be modulated. Accelerated durability tests of HER showed that the activity decay of Pt clusters on defects is slower than conventional Pt/C catalysts due to the strong interaction between Pt and defects. Another example was demonstrated by Lončar and co-workers, showing that IrO₂ dispersed on titanium oxynitride (TiON_x) exhibits strong metal–support interaction and, therefore, high stability.¹¹¹ Smiljanić also employed TiON_x for Pt nanoparticle stabilization during acidic HER, avoiding particle agglomeration and detachment deactivation pathways in the Pt/C catalyst.¹¹² 

In addition to the stabilization effect, enhancing catalyst–support interaction may promote electronic transfer and regulate valence state variation during catalysis. For example, the strong interaction between Co₃O₄ and CeO₂ allows the Co³⁺ species to be easily oxidized into catalytically active Co⁴⁺ species, bypassing the potential-induced surface reconstruction process and, thus, increasing stability.¹¹³ Li et al. constructed a MoS₂ monolayer on Ti substrates, which showed better HER activity than Pt and no degradation in 2 months of continuous testing due to the n-doping effect of Ti.¹¹⁴ 

Bubbles generated during electrolysis can block active sites and cause repeated mechanical stress, leading to catalyst detachment.^115–117 Effective bubble removal is key to maintaining catalyst stability. A few strategies have been proven effective (Fig. 7), including morphology, pattern, and surface engineering. For example, bubble evolution on microneedle CoS₂ is faster than on a planar film,¹¹⁵ and a Pt electrode with striped-pattern superlattices alleviates the bubble stretch force.¹¹⁶ These designs increase the catalyst’s lifetime under high current conditions by 2–8 times. Fabricating an aerophobic overlayer can also be beneficial. Rye et al. demonstrated the effectiveness of a polyallylamine hydrochloride coating for accelerating bubble removal.¹¹⁷ By adjusting the coating method, the pores of the polymer coating could be regulated to minimize the possible negative impact on mass transfer and surface area. This method provides the possibility of constructing large-scale aerophobic surfaces for different catalysts.

This review has offered an overview of catalyst deactivation during water electrolysis, emphasizing the deactivation pathways, assessment techniques, and current countermeasures. Despite significant advancements in reducing catalyst costs and enhancing activity in the past decade, we emphasize the importance of comprehensive studies on catalyst stability and deactivation during long-term electrolysis.

One challenge in analyzing catalyst deactivation is the extensive time commitment required for full stability assessments. Given that water electrolyzers can operate for tens of thousands of hours in working conditions, accelerated stability tests that simulate real scenarios and evaluate catalyst stability under different deactivation modes are critically needed.^118,119 Furthermore, considering the variable nature of catalyst deactivation, we propose a wider adoption of stability indicators, such as the stability number, to provide a more comprehensive assessment of catalyst stability in addition to activity benchmarking.

The intricacy of in situ/operando characterization may discourage in-depth research on catalyst deactivation. Since most in situ cells are specifically designed to collect certain types of information, their structures might differ from those found in actual operating conditions. Therefore, the disparity between laboratory measurements and industrial instruments warrants additional attention. Given the limitations inherent to each in situ/operando characterization method, we recommend cross-referencing results from diverse measurements for a more accurate analysis.

Further complicating matters is the incomplete understanding of the relationship between catalyst structure and deactivation. It is commonly accepted that catalyst structure impacts the OER pathway and, subsequently, catalyst stability. However, the specific nature of these influences remains a subject of debate.^120,121 Some indicators have been proposed to predict catalyst stability, but their universal applicability needs further validation. For the HER, particularly in acidic conditions, we suggest emphasizing catalyst dissolution under open circuit potential and the relationships between structure and stability at this potential. Developing indicators that can bridge the gap between catalyst structure and stability could significantly improve catalyst design optimization.

Weiran Zheng is grateful for the support of the Guangdong Basic and Applied Basic Research Foundation (Grant No. 2023A1515012277) and the Guangdong Technion-Israel Institute of Technology (Grant No. ST2200002).

The authors have no conflicts to disclose.

L.D. and W.Z. drafted the manuscript together.

Lijie Du (杜礼杰): Formal analysis (lead); Writing – original draft (lead). Weiran Zheng (鄭蔚然): Conceptualization (lead); Funding acquisition (lead); Resources (lead); Writing – review & editing (lead).

Data sharing is not applicable to this article as no new data were created or analyzed in this study.