Single-Atom Catalysts: A Perspective toward Application in Electrochemical Energy Conversion

Abstract

Single-atom catalysts (SACs) hold great promise for maximized metal utilization, exceptional tunability of the catalytic site, and selectivity. Moreover, they can substantially contribute to lower the cost and abundancy challenges associated with raw materials. Significant breakthroughs have been achieved over the past decade, for instance, in terms of synthesis methods for SACs, their catalytic activity, and the mechanistic understanding of their functionality. Still, great challenges lie ahead in order to render them viable for application in important fields such as electrochemical energy conversion of renewable electrical energy. We have identified three particular development fields for advanced SACs that we consider crucial, namely, the scale-up of the synthesis, the understanding of their performance in real devices such as fuel cells and electrolyzers, and the understanding and mitigation of their degradation. In this Perspective, we review recent activities of the community and provide our outlook with respect to the aspects required to bring SACs toward application.

1. Introduction

The majority of current environmental challenges can be traced back to our fossil fuel-based, consumeristic economy. Over the last decades, renewable electrical energy based predominantly on solar and wind has become increasingly available around the globe, initiating a change toward a more environmentally adapted energy culture. Moreover, research and development on coupling different energy sectors have been intensified so that in the future green electrical energy can also be utilized for mobility and industrial processes. Electrochemical energy conversion is thought to play a decisive role in this sector coupling scenario and thus to relinquish our dependency on fossil fuels while still satisfying humanity’s energy needs. In addition to battery technology, which is commercially available for industrial and even private mobility applications, the field has reached the proof-of-concept stage, and technologies are used for specialty applications. Continuous electrochemical reactors such as electrolyzers and fuel cells (FCs) rely on electrocatalysts to mediate the activation barrier of the reactions of interest. State-of-the-art commercial devices use platinum group metals (PGMs) as electrocatalysts. The price of PGMs has so far hindered the widespread commercialization of such technologies. To decrease their overall loading, an increase of PGM utilization, i.e., their surface-to-volume ratio, is necessary. Therefore, PGMs are usually synthesized in the form of highly dispersed nanostructures.¹⁻³ To further increase the utilization, single-atom catalysts (SACs) have emerged as somewhat of a holy grail of catalysts. Their maximized dispersion, diversity, and tunability offer numerous applications ranging from thermal heterogeneous to electrochemical catalysis.⁴⁻⁸ SACs have demonstrated extraordinary catalytic properties in diverse electro-,⁹⁻¹⁴ photo-,^15,16 and thermocatalytic reactions.¹⁷⁻¹⁹ By virtue of the new opportunities they offer, SACs have established a new frontier in the field of catalysis. For electrocatalysis, in which conductive supports are required, metal-nitrogen-doped carbon (MNC) SACs have played the most prominent role in replacing or reducing PGM loadings so far.^11,20−22

Due to its tunability with heteroatoms such as N (but also S, P) as electronic and ligating promoters, carbon has attracted much attention as a SAC support. It is well understood that the local environment of the oxygen reduction reaction (ORR) active iron-nitrogen-doped carbon (FeNC) sites governs their reactivity. Therefore, many efforts have been devoted to the development of advanced in situ and ex situ characterization techniques to probe and understand the nature of the active site.⁷ Electron microscopy offers the unique ability to visualize single-atom sites and show their distribution within the carbon network.²³ Mößbauer spectroscopy^24,25 and synchrotron-based extended X-ray absorption fine structure^26,27 are commonly used as powerful spectroscopy tools that probe the local coordination environment of the single-atom site. Further, X-ray photoelectron spectroscopy provides two essential parameters: an overall compositional analysis and specification of the dopants’ chemical state.^28,29 Finally, one of the most important macroscopic properties of a catalyst material is its mass transport characteristics, which are mostly reliant on the pore structure and can be characterized by nitrogen sorption in the most prominent MNC type catalysts.³⁰ Through these in-depth investigations, the past years generated great insight into the FeNC catalyst systems. The physical state of the active site has so far been linked to catalyst performance and selectivity in aqueous electrolytes.^20,31 Many of the great advances over the past decade on specific synthesis procedures, the activity of the catalysts, and the mechanistic understanding have been summarized in excellent reviews.^{7,21,29,32−34}

In a logical next step toward application, small-scale devices have been demonstrated to prove the concept of SAC as fuel cell cathode catalysts.^35,36 For practical applications, however, there remain a number of critical challenges, including a scale-up issue in synthesis, a knowledge gap between fundamental electrochemistry and device- or system-level application, as well as a stability problem.³⁷ Over the past years, a myriad of synthesis pathways have been proposed for anchoring single metal atoms on a support medium. Depending on the field of application, the support can be of metallic, oxidic, or nonmetallic nature.⁸ In this Perspective, we highlight many promising approaches of SAC synthesis and assess their scalability to an industrial level. In electrochemical energy conversion, especially, carbon has always played a crucial role as a porous, conductive support material. Stability aspects were recently revealed in more detail by online inductively coupled plasma mass spectrometry (ICP-MS).^20,38−41 Because the atomically dispersed metal species are highly unstable and prone to agglomeration due to their high surface energy,^42,43 their stabilization on supporting materials is indispensable. Under electrochemical operation conditions, dissolution or leaching of active metal species has been suggested as one of the critical degradation paths hampering the prolonged electrocatalysis of MNCs.⁴⁴ The formation of metal clusters or nanoparticles (NPs) via agglomeration has also been reported.^45,46 So far, these degradation mechanisms were only studied on a fundamental level, while their extent and rate in real devices are rarely addressed. To bridge this gap, we introduce gas diffusion electrode (GDE) half-cell investigations as a promising tool to investigate such fundamental processes under more realistic conditions in the future.

In contrast to the previously mentioned reviews with a focus on synthesis, activity, and mechanistic understanding, this Perspective aims to emphasize the important next steps in SAC research to bring this technology to application. We point out necessary steps in general synthetic procedures for SAC, particularly for their industrial use, mass production, or large-scale synthesis. Since MNCs comprise some of the most investigated SACs, we further focus on their applicability as a future game-changer in green-energy conversion. Therefore, the electrochemical methods needed for understanding SAC performance under operational conditions are elucidated as the example of FeNC in fuel cells. In addition, a special focus is set on durability and a comprehensive understanding of degradation processes, which is essential for the successful industrial implementation of SACs in various electrochemical energy conversion devices. Finally, we illustrate the interconnection between these fields and suggest paths forward to achieve a level of understanding that facilitates rational design of stable SACs ready for mass production and large-scale application.

2. Single-Atom Catalyst Synthesis

The amount of synthetic pathways to a successful SAC seems overwhelming when doing an initial literature survey. They range from simply mixing a metal oxide with a metal dopant precursor to calcine into single PGM atoms on a transition metal oxide, over pyrolyzing organic material that includes all important ingredients (C, N, S, P, and the active metal ion) to yield rather undefined MNC catalysts all the way to well-defined compositions and porous structures by templating approaches with organic precursor molecules. In the following section, we group various established methods and discuss their viability toward a scale-up. Finally, we summarize challenges that the field of SAC synthesis still has to overcome and highlight the paths forward with an already commercialized example of FeNC.

2.1. Established Synthesis Methods

Even under the most established methods, a wide range for SAC preparation is available, including traditional mixing-and-activation, conversion of metal–organic frameworks (MOFs), thermal transformation, deposition under ultrahigh vacuum, electrodeposition, and galvanic replacement.

The most widely exploited approach to SACs is the mixing-and-activation method, with more than 60% of related literature employing this method. This method consists of mixing a metal precursor with a support and subsequent activation. The mixing step is carried out by wet impregnation, incipient wetness impregnation, precipitation, and dry impregnation (mechanochemical) methods. The activation can be achieved by thermal treatment under inert or reactive gas, photochemical reduction, and microwave treatment. We note that although the mixing-and-activation routes based on wet impregnation, incipient wetness impregnation, and precipitation are generally applicable to a wide range of supports, they suffer from agglomeration of metal species into NPs with high metal loading, typically above 1 wt %. In this sense, it is noteworthy that the dry impregnation-based mixing-and-activation method could generate SACs in higher metal content (1.9 wt %). Moreover, a simple mechanical mixing by an agate mortar or ball milling in the solid-state can make this process simpler and appealing for large scale-up than other methods. The significance of dry impregnation for large-scale synthesis will be discussed later in detail.

The MOF conversion route has recently emerged as one of the most popular methods, particularly for carbon-supported SACs such as FeNC.⁴⁷ Zeolitic imidazolate frameworks, one family of MOFs, have been most widely used for this purpose.

They are constructed with three-dimensional (3D) or 2D networks of metal centers and nitrogen- and carbon-containing organic ligands, and their thermal conversion under appropriate conditions can yield carbon-supported single-site M–N species.

Thermal transformation strategy refers to the conversion of metal or metal oxides in the form of NPs or bulk particles into single metal atoms on support surfaces by high-temperature treatment.⁴⁸ This method is counterintuitive because metal atoms generally tend to agglomerate into NPs to reduce their surface energy, and thus, most catalyst preparation methods try to suppress the agglomeration. Interestingly, some metal and support combinations can reversely generate single atomic sites from large particles in high temperatures.

Deposition in vacuum (e.g., atomic layer deposition), electrodeposition, and galvanic replacement are also reported as important fabrication methods for SACs. However, the production yield of these methods is merely milligram scale, which is significantly lower than those of the other introduced methods. Even more, for vacuum deposition, the use of an expensive apparatus is indispensable. Therefore, in the current status, these methods appear to be more appropriate to prepare laboratory-scale catalysts for the purpose of fundamental mechanistic studies.

2.2. Advances toward Scale-Up

Toward turning a promising catalyst from the laboratory scale to the industrial scale with minimal effort, some factors should be recognized from the stage of laboratory-scale synthesis. We surveyed two hundred selected papers in the field of SACs and categorized the catalysts according to their preparation methods (see the Supporting Information). For the respective methods, average production scale and average metal content were summarized and their scalability was assessed (Table 1). The production scale was estimated by the amount of precursors used in the preparation steps. We used median values of the collected results to estimate an average index due to the considerable deviation of the production scale among the surveyed literature.

Table 1. Production scale of preparation methods for atomically dispersed catalysts^a.

preparation method	average production scale (g)	highest production scale (g)	metal content (wt%)	scalability	number of references
mixing-and-activation process
wet impregnation	0.30	1600	0.56	+	81
incipient wetness impregnation	0.08	0.5	0.47	–	19
precipitation	0.90	2	0.17	–	13
dry impregnation	2.25	1000	1.90	+	12
MOF conversion	0.20	2	1.21	+	18
thermal transformation	0.90	10	1.00	+	18
deposition in vacuum	n/a	0.5	0.35	–	11
electrodeposition	0.38 mg cm–2	2 mg cm–2	1.48	–	10
galvanic replacement	0.06	0.5	0.21	–	9

^aThe average production scale over all surveyed reports as well as the highest ones reported. The average metal content achieved by the individual methods as well as our verdicts of their scalability. (See the Supporting Information).

For mixing-and-activation processes, the production scale and average metal content are dependent on the mixing method. The most widely used wet impregnation-based catalysts have an average production scale of 0.3 g and average metal content of 0.56 wt %, some of which are, however, far from industrial-scale production. The highest reports on the other hand have production scales of up to 1.6 kg, which shows the potential scalability of this method. The incipient wetness impregnation-based method produces SACs on a lower production scale, and the precipitation method affords lower metal content than the wet impregnation. The mechanochemical method-based dry impregnation, in contrast, allows for a large, kilogram-scale synthesis of SACs in batch production, making this method the most appropriate route to scale-up. We note that, despite the promise of mechanochemical methods, its success so far has been limited to oxide-supported SACs. Given the tremendous rate of progress in the field of SACs, its implementation for carbon-supported SACs that are more relevant to electrocatalysis is expected soon to be realized.

As for MOF conversion, a survey of relevant literature indicates a relatively low production scale (0.2 g). The low yield may stem from the sacrificial nature of the MOFs during the pyrolysis evaporating a significant amount of their constituents. Further, the synthesis of MOF precursor itself requires additional processing steps. For the scale-up of this method, the development of simple synthesis and processing of MOFs and the treatment of emitted, sometimes toxic, gases should be considered.

On the other hand, the direct thermal transformation method is a proper way to produce single atomic site catalysts in a flow system by continuously supplying fresh supports. As far as a suitable combination of metal and support being found, this method can serve as an attractive pathway for large-scale synthesis of SACs, due mainly to its simplicity. Indeed, the average production scale and metal content of thermal transformation are comparatively high, with values of 0.9 g and 1.0 wt %, respectively. However, the thermal transformation may have limitations as a general route to SACs, as specific metal and support interactions need to be considered.

Considering a minimum industrial or commercial production scale of a catalyst is ∼100 kg,⁴⁹ currently reported synthetic scales are overall far from the industrial-scale production. In recent several years, though, the attention to large-scale synthesis of SACs has been increased. We note that most methods use solvent during the SAC preparation. For plant-scale preparation, the usage of solvent may invoke several problems, including the uneven local concentration of reactants during mixing and the difficulty and environmental impact in the removal of huge amounts of solvent. Moreover, flammable organic solvent can react vigorously with metal catalysts, leading to safety problems.^49,50,52,53 Among the methods surveyed above, the dry impregnation-based mechanochemical method allows for solvent-free processes, engendering of process simplicity, and cost effectiveness.

In a recent notable example, Ma and co-workers have demonstrated the effectiveness of mechanochemical synthesis for the scale-up preparation of SACs.⁵⁰ After diluting Pd(acac)₂ on Zn(acac)₂ (acac, acetylacetonate) with the weight ratio of 1:400, the precursors were thoroughly ground through a ball milling process. Subsequently, the mixture was calcined at 400 °C for 2 h in air. The resulting catalyst was composed of single Pd atoms supported on ZnO (Pd₁/ZnO), and the production can be increased into the kilogram scale (Figure 1a). This approach could be further extended to Rh and Ru metal centers and CuO and Cu supports, demonstrating its broad utility. Ji and co-workers also showed the large-scale synthesis through ball-milling.⁵² Similar to the work of the Ma group, Pt(acac)₂ and Co(acac)₂ were ground with a weight ratio of 1:500 by ball milling, followed by sequential calcination and reduction. As a result, the PtCo single atom alloy (alternatively: dual metal dimers) can be easily obtained on the kilogram scale. This method also has wide applicability to other types of metals such as Pd, Rh, Ir, and Ru.

Figure 1.

Exemplary scaled up SAC synthesis methods. (a) Simple illustration of mechanochemical synthesis of atomically dispersed catalyst and high-angle annular dark-field scanning transmission electron microscopy (HAADF-STEM) image of resulting Pd₁/ZnO catalyst with a photograph showing kilogram-scale production. Reprinted with permission from ref (50). Copyright 2020 Elsevier Inc. (b) Schematic illustration of direct thermal transformation of Cu₂O into atomically dispersed Cu and HAADF-STEM images of resulting catalyst with photographs showing gram-scale production. Reprinted with permission from ref (51). Copyright 2019 Springer Nature.

The other preparation method that is amenable to scale-up is the direct transformation of bulk metals or metal oxides into single atomic sites.^54,51 This method is a kind of thermal transformation, but with the important difference of the absence of a mixing process of metal and support sources compared to the conventional approach. Bulk metals or metal oxides are directly converted or sublimated as mobile vapor at high temperatures, trapped onto support surfaces as forms of single atomic sites. Since the metal–support mixing and activation steps take place simultaneously, the preparation step can be simplified compared to the conventional thermal transformation method, which may be beneficial to scale-up.

Recently, Li and co-workers reported the direct transformation of bulk Cu into single atomic Cu sites.⁵⁴ Under ammonia flow, Cu is coordinated with NH₃ to form volatile Cu(NH₃)_x species, which are then trapped on N-doped carbon support. Through this method, single-site Cu catalysts can be obtained in a gram-scale (∼2 g). Likewise, using Cu₂O powder as a metal precursor, Wu and co-workers also demonstrated the direct transformation from bulk to single-atom with gram-scale production (Figure 1b).⁵¹ Under an inert condition (N₂) at 1000 °C, the Cu₂O was evaporated and transported to N-doped carbon support. The Cu₂O vapor in the vicinity of the support surface was trapped and reduced to form Cu SACs.

Beyond literature reports, the most noteworthy example in industrial SAC preparation is the commercialized FeNC produced by Pajarito-powder LLC. Here nitrogen-rich-organic and metal precursors are mixed in solution with a sacrificial silica support. After pyrolyzing the organic and metals to FeNC, the silica support is dissolved, leaving behind a porous carbon network with active FeN₄ sites. Initial reports of this modified sacrificial support method from 2014 state a production volume of 200 g,^22,55 which has been further optimized ever since. This catalyst can be found in a broad selection of research literature from different groups over the past years, and large fuel-cell manufacturers show interest in the product to replace expensive Pt.

2.3. Challenges and Paths Forward

Large-scale production is the final and critical process for the industrialization of catalysts but an undervalued technology in the research community. Recognition of the scale-up issue in the early state of catalyst preparation thus can accelerate time to market of new catalyst materials.

Generally, there is a huge gap between research catalysts prepared in a laboratory, or bench scale, and technical catalysts where the research catalysts are reformed or reproduced for the purpose of industrial-scale operation.⁵⁶Research catalysts are fabricated to screen a series of materials in the early stages of new catalyst development for improving catalytic activity, selectivity, and stability. Thus, catalysts are typically prepared on a small scale using high-purity analytical reagents through elaborate synthetic procedures.⁴⁹ On the other hand, technical catalysts are manufactured to evaluate the possibilities of their practical applications on the industrial-scale. Hence, technical-grade reagents are employed due to economic consideration, and multiple additives are included to enhance physical (mass or heat transfer), chemical (functionality), and mechanical (strength) properties of a catalyst for a long-term operation in harsh conditions.⁵⁶ For SACs, given their relatively low metal contents compared to NP-based catalysts, the conversion per unit volume of the reactor should also be considered. To bridge the gap between research catalysts and technical catalysts, it is necessary to understand the importance of the scale-up technology and learn the requirements of catalysts in the aspects of industrial applications. The starting point is a synergistic collaboration between academia and industry. By sharing the respective experiences and needs, design criteria toward the scale-up can be taken into account from the early stage of catalyst development, which may in turn facilitate the industrialization of the catalysts. For SACs, severe difficulties are further expected in the product homogeneity of the scale-up process due to their sensitive catalytic structure. Therefore, in this beginning period of the development of SACs, an intimate connection between academia and industry is imperative toward the advancement into commercial catalysts.

Using the example of commercial FeNC, some considerations remain in improving their activity toward the ORR in fuel cells as well as their stability. The following sections highlight these two major challenges. Therefore, the path forward can be seen as a feedback loop where research catalysts act as a method to understand these processes, which can then be applied as an iteration in an optimized, ideally large-scale synthesis approach.

3. Electrocatalytic Performance

With suitable synthesis approaches at hand, the electrocatalytic performance toward the reaction of interest is the key toward understanding and optimizing the catalyst further. Here we separate the workflow of performance evaluation from a fundamental understanding of the catalysts’ current response to an applied potential to a full cell where many other processes come into consideration. Especially the knowledge gap between these two performance metrics will be addressed. Note that we focus specifically on the example of FeNC as an ORR catalyst in hydrogen FCs because of the unparalleled understanding that has been already achieved in recent years for this specific SAC and reaction. This particular knowledge and the outlook can, however, be directly transferred and will also be essential for other upcoming types of SACs.

3.1. Three Electrode Characterization

Fundamental three-electrode testing has been the workhorse of the SAC research community to compare the physical state of the active site with a descriptor of the catalysts’ activity. It uncovered many structure/performance relationships imperative for our advanced mechanistic understanding. Through this technique, SACs have shown their enormous potential for a variety of electrochemical reactions (including nitrogen, oxygen, and carbon dioxide reduction, methanol oxidation, hydrogen oxidation and evolution, and most intensively oxygen reduction).^7,31,57−60

Using a pipet, an aliquot of a catalyst ink is drop cast manually on the electrode and left to dry. Depending on the composition of volatile ink components, the temperature and airflow can significantly influence the film quality after drying. The diligence required for good quality layers is high and should not be neglected in assessing the performance of the catalyst layer.⁶¹⁻⁶³ When done correctly, rotating disk electrode (RDE) measurements can provide significant insight into the activity and selectivity of the catalyst toward many reactions under controlled hydrodynamic conditions. The major advantages of this fundamental method are the easily accessible equipment, the relatively short workflow between drop-casting and result, and the fast screening capabilities.

In the following, we pay closer attention to the example of the ORR at fuel cell cathodes. The activity of a catalyst toward this reaction depends on many factors such as the support porosity and site accessibility,⁶⁴ the coordination environment of the metal and therefore its redox potential,⁶⁵ and the nature of electrolyte.^20,66 These findings come from extensive in situ and ex situ characterization of MNCs over the years, and they have allowed the development of new and improved structures. When assessing the activity of a new catalyst, the surface area and porosity should be addressed first to have a reliably normalized activity descriptor. As mentioned previously, N₂ sorption analysis is a common tool but also electrochemical cyclic voltammograms can give valuable insights. For example, the Fe²⁺/Fe³⁺ redox peak or the capacitive currents can be an initial quantifiable descriptor. Even advanced electrochemical nitrite stripping⁶⁷ and CO cryo-chemisorption⁶⁴ methods have been developed to quantify the active site density. To then assess the activity of the catalyst, a linear sweep voltammogram in oxygen-purged electrolyte can be used to extract the onset potential, exchange current density, and limiting currents of the ORR (Figure 2a). With the help of Koutecky–Levich analysis, the kinetic current can be extracted from RDE measurements as the most common activity descriptor, that excludes mass transport limitations. Given the controlled hydrodynamic conditions, the diffusion of O₂ to the surface is dictated by the rotation rate, kinematic viscosity, and diffusion coefficients in the given electrolyte. As an ideal pathway of the ORR, the limiting current, where the reaction rate is in a steady state with incoming O₂, can provide insights on the selectivity and the reaction mechanism. The rotating ring disk electrode (RRDE) technique can help with the selectivity determination even further by measuring the oxidative current of H₂O₂ or HO₂^– species that evolve from the disk.

Figure 2.

Electrocatalytic performance testing on various application scales. (a) RRDE data on four differently prepared FeNC catalysts reproduced from ref (24). (b) Single-cell PEMFC tests from Seeberger et al.,⁷⁰ Serov et al.,²² and Zitolo et al.²⁴ as well as recent AEMFC results of Santori et al.³⁰ and Firouzjaie et al.,⁷¹ operated with pure oxygen. (c) Recent GDE experiments from Ehelebe et al. to evaluate FeNC catalysts in a half cell with aqueous electrolyte at high current densities. Reproduced with permission from ref (72). Copyright 2020 Elsevier.

These easily accessible descriptors, together with accelerated stress tests (ASTs), can further unlock information about the durability in classical R(R)DE measurements, especially with regards to side reactions and H₂O₂, which will be discussed further in a later chapter.^20,68,69

3.2. Fuel Cell Testing

Nowadays, there is reasonable doubt concerning the validity of catalyst screening in a liquid electrolyte at room temperature, especially at the typically low current densities imposed by the low mass concentration and mass transport. Since the ORR at FC cathodes occurs at the triple-phase boundary of the gas phase, hydrated ionomer, and catalyst, the actual measured current densities are orders of magnitude larger due to improved mass transfer. Therefore, to move forward with the implementation of SACs into real energy conversion devices, it takes more than the established RDE descriptors to understand their performance in real devices.

As a commendable example, FeNC research on the ORR has employed more time-consuming single-cell FC measurements since the early 2000s to support the structure/activity relationships drawn from physical and electrochemical testing.^35,36,73 Only in this simplified version of a market-ready fuel cell stack, the complex interplay between the processes in flow fields, gas diffusion layer (GDL), catalyst layer, and membrane can be tested under operating conditions. Such studies often show reasonable correlations between the trends in RDE and FC measurements, thus mending the concerns about fundamental three-electrode measurements. Jaouen et al. as well as Zitolo et al. have shown that, for three differently prepared FeNC catalysts, the activity in RDE (20 °C) and proton exchange membrane (PEM) FC (80 °C) experiments can differ up to a factor of 5, but at least the trend between the different catalysts remained the same.^24,74 Furthermore, they show that this 5-fold discrepancy can be attributed to the poising by liquid electrolyte anions such as sulfates.⁷⁴ Workman et al. have correlated RDE and PEMFC performance of seven differently prepared catalysts and revealed the same dependency on physical characteristics.⁷⁵

Still, many performance-related questions remain, which can only be addressed under operating conditions. For example, large issues with water management and flooding in anion exchange membrane (AEM) FCs have been reported in the gas diffusion layer due to the high rate of water formation at the anode.⁷⁶ Furthermore, the stability of some next generation membrane materials needs to be addressed under operating conditions.^77,78 Finally, the impact of mass transport, selectivity, and the possible formation of harmful H₂O₂ at high current densities remain largely unexplored.

Unfortunately, the various components and their assembly in a FC also lead to a variety of factors, other than the catalyst, influencing the performance of the cell. Figure 2b highlights some polarization curves in PEM and AEMFCs that, despite operating with similar or identical catalysts, show drastically varying performances. Here, some aspects, including contact resistance, catalyst layer utilization, membrane conductivity, mass transport, and water management, are just a few of the reasons why there is a large knowledge gap between the fundamental electrochemistry of SACs and their performance in real devices.⁷²

3.3. Half Cells at Realistic Conditions

While single-cell FC performance is the ideal way to assess the SAC under realistic conditions on a laboratory scale, the vast amount of interactions between processes makes it challenging to separate contributions to the overall performance in the full device. Therefore, a relatively new approach to bridge the gap between fundamental RDE and single-cell performance under real conditions is the use of GDE half cells.⁷⁹⁻⁸¹ These half cells have the advantage of employing flow field, GDE, and even membranes together with a counter and a reference electrode. This three-electrode configuration allows relatively fast and detailed kinetic studies at the triple-phase boundary without mass transport limitations. Therefore, such GDE approaches can be an ideal tool to validate new SACs or their interaction with various membranes and electrolytes under realistic conditions.

Only recently, a GDE setup was employed to investigate the kinetics of FeNC catalysts for AEMFC in alkaline electrolyte. Here, commercial FeNC catalyst (Pajarito Powder) ink using an alkaline binder (Aemion) was coated on a carbon GDE and measured in 1 M KOH at room temperature. The simple measurement method was used to deduct ideal preconditioning procedures for alkaline FeNC electrodes in a short period of time⁷² but, on the other hand, still revealed a large gap to state-of-the-art AEMFC operation;⁷¹ see Figure 2c. Next to the influence of elevated temperature on this system, especially the local pH at the catalyst remains an enigma when operating with pure humidified gases and an AEM instead of liquid electrolyte. With many issues to address in the future, GDE half-cell investigations hold promise to improve the understanding of processes at the catalyst layer, GDL, and membrane under more realistic conditions than in RDE, yet having easier applicable and more controlled conditions than those in full cell studies.

3.4. Challenges and Paths Forward

With this example of FeNC as a FC cathode catalyst, we point out the necessity for SAC research to provide performance data in real devices and also for other proposed reactions.

The understanding of the ORR reaction on FeNC has led to new synthesis approaches yielding remarkable improvements of their intrinsic activity. However, the knowledge gap between three-electrode measurements and the combined processes in a full cell remains large. Therefore, we consider the technique of half-cell GDEs to be extremely helpful for future studies. The community needs to increase the awareness of the interplay of parameters that lead from the intrinsic catalyst activity to the apparent performance in a real device. The overall goal is to understand the performance at high currents, elevated temperatures, and ideally realistic local pH values of a membrane interface. All of these conditions can lead to a change in apparent performance. Thus, understanding interconnected parameters could lead to a similar improvement of apparent performance in a full cell by knowledge-based optimization. The summarized paths forward for the example of FeNC as an ORR catalyst are especially true for the less investigated but highly interesting topics of research that were mentioned at the beginning of this section. Here, the proof-of-concept provided by three-electrode measurements of CO₂ reduction and further reactions needs to be implemented in full-sized energy conversion devices to prove their relevance.

Along the way to application, many more challenges such as operational stability can occur, which might bring the technology back to the drawing board. Therefore, the next section on SAC stability can be a game-changer in terms of moving forward with SAC for electrocatalytic renewable energy conversion.

4. Stability of SACs

Over the past few years, many works focusing on the synthesis of SACs and their electrochemical applications have been reported, successfully leading to remarkable advances in their initial activity and thereby showing potential feasibility for industrial implementation. Contrary to the recent achievements in activity improvement, however, stability issues for the SACs have been much less investigated. In many cases, stabilization of single-atom metal species (or ions) has been achieved by their covalent integration with p-block elements (e.g., nitrogen, sulfur, phosphorus, etc.) doped in the supporting substrates.^24,82 Therefore, the stability of the supporting materials and the chemical nature of active metal coordination have also been considered as additional key factors determining the SACs’ stabiltiy.^45,83,84 Here, the largest body of recent literature addressing stability is also related to our example of FeNC, with which we aim to raise the awareness of what will also become a necessity for any other type of novel SAC.

4.1. Demetalation of Single-Atom Metal Sites

Isolated metal species play a key role as a main active site and determine the electrocatalytic properties of SACs. Therefore, dissolution or leaching of such metal species directly links to the loss of the catalytic site.^38,85,86 Besides the catalytic deactivation induced by the demetalation, the dissolved metal cations can also be detrimental to the prolonged operation of the device since they can deteriorate the ionic conductivity of the membrane and ionomer in electrochemical devices.⁴⁴ However, there have been only a few reports which deal with the demetalation process of SACs carefully to date.

Using operando spectroscopic techniques (i.e., online ICP-MS coupled to a modified scanning flow cell system; SFC/ICP-MS), Choi et al. observed Fe leaching from FeNC catalyst at a relatively low potential region (below 0.7 V_RHE) in a deaerated 0.1 M HClO₄ electrolyte at room temperature (Figure 3a).³⁸ Surprisingly, no significant loss of ORR-active Fe–N_x sites was found on the polarized catalyst, as evidenced by its negligible activity drop after the demetalation. They reported that a trace amount of Fe NPs, which are not active toward ORR, is responsible for the Fe dissolution. Once potential increases to higher than 1.0 V_RHE, however, carbon corrosion to CO₂ derives considerable demetalation of Fe–N_x sites with significant activity drop, suggesting that the demetalation process is a function of the potential applied. On the other hand, Santori et al. revealed that NH₃ treatment of the FeNC catalyst leads not only to a significant increase in initial ORR activity but also to an almost 10 times enhanced Fe dissolution compared to nontreated FeNC catalyst at potential below 0.75 V_RHE.²⁰ A significant ORR activity drop of ca. −30% after the AST, which was not found for the nontreated sample in alkaline electrolyte, suggests possible demetalation of active Fe–N_x sites in acidic electrolyte. Even for the nontreated FeNC catalyst, the non-negligible dissolution of Fe–N_x sites and consecutive ORR activity drop were also verified at a potential below 1.0 V_RHE at a high temperature of 80 °C in the presence of oxygen.⁸⁷ In practice, the specific demetalation of Fe–N_x active sites was disclosed by the Dodelet group during real PEMFC operations.⁸⁵ They thoroughly explored the instability behavior of the FeNC catalyst through FC testing at various voltages. Using ⁵⁷Fe Mößbauer spectroscopy, completely overlapping trends between remaining amounts of Fe–N_x species and fuel cell performance were identified (Figure 3b). They concluded that the severe activity loss of FeNC catalysts at the beginning-of-life stage of fuel cell operation mainly results from the demetalation of ferrous moieties. A series of previous works with identical or similar FeNC catalysts have consequently identified that the Fe demetalation rate is highly affected by experimental conditions such as potential, temperature, electrolyte pH, and chemical environment.

Figure 3.

Demetalation and agglomeration of single-atom metal sites. (a) Online SFC/ICP-MS Fe dissolution (blue) of a FeNC catalyst under varying potential (black) over time. Reproduced with permission from ref (38). Copyright 2015 Wiley-VCH. (b) Relative change in the current density and the number of Fe–N_x sites in the cathode of a membrane–electrode assembly (MEA) versus time. The current density was provided from fuel cell measurements, and the number of Fe–N_x sites was estimated from Mößbauer spectra and iron contents at the cathode of MEAs. Reproduced with permission from ref (85). Copyright 2018 Royal Society of Chemistry. (c) Fe–N_x stability versus dissolution diagram as a function of pH and potential. Reproduced with permission from ref (86). Copyright 2020 American Chemical Society. (d) Mößbauer spectra of the FeNC cathode after H₂/O₂ PEMFC operation at 0.5 V and 80 °C for 50 h. The inset shows the Tafel plot trace as a function of time (from right to left). Reproduced with permission from ref (88). Copyright 2021 Springer Nature.

Despite many experimental parameters that make experimental approaches rather complex, computational approaches seem to greatly corroborate a deeper understanding of the demetalation process of SACs. Based on density functional theory (DFT) calculations, Holby et al. proposed a model and descriptor for leaching of single-atom metal species.⁸⁶ By applying various experimental and reaction variables into their calculations (e.g., reaction environments, intermediates, etc.), they drew a stability Pourbaix diagram of FeNC catalysts (Figure 3c) that can explain the previous experimental findings of the instability of FeNC catalysts.

4.2. Agglomeration of Single-Atom Metal Sites

Aside from the demetalation of active metal centers, their agglomeration is also of major concern for SAC types of catalysts. Diffusion and consecutive formation of clusters or NPs are thermodynamically viable to reduce the high surface energy of the single-atom metal species.⁶⁹

During PEMFC operation, the in situ Mößbauer spectra of FeNC catalysts exhibited a complete switch of a doublet signal (D1) to a sextet, clearly indicating the agglomeration of ORR-active Fe–N_x moieties (D1) into ORR-inactive (or less active) Fe₂O₃ clusters (sextet; Figure 3d).⁸⁸ The agglomeration of single-atom Fe was also observed for NH₃-treated FeNC catalysts even after mild load cycling in a rotating disk electrode system.⁶⁹ Moreover, the formation of Fe cluster was highly temperature-dependent; namely, agglomeration of Fe–N_x moieties occurred after the cycling at 80 °C but were not observed at 60 °C. The results implied that temperature might play a key role in determining the agglomeration rate of single-atom Fe. A reason for surface mobility and cluster formation can be the extent of interaction between the metal ion and the coordinating heteroatom. Very recently, Speck et al. revealed transformations of single-atom Pt stabilized in S-doped carbon into NPs after potential cycling up to 1.5 V_RHE while they are stable at a potential below 1.0 V_RHE.⁴¹ This work suggested that the transformation process of single-atom Pt to clusters or particles is closely related to the oxidation of coordinating ligands and the corrosion of the support.

4.3. Corrosion and Chemical Modification of the Support

Since supporting substrates chemically and/or physically host the single-atom metal species, their chemical phase transition or physical destruction can introduce significant performance drops of SACs. As exemplified in sections 4.1 and 4.2, corrosion of supporting substrates derives a loss of the single-atom metal species.^38,41 Moreover, lowered electron conductivity due to corrosion also affects the overall electrocatalytic performance once the surface of the supporting substrates is passivated during the operations. Hence, activity loss of SACs induced by the modification of supporting substrates is inevitably and strongly dependent on the (electro)chemical stability of the supporting substrates.

For the FeNC catalysts, it has been suggested that carbon corrosion plays a critical role in the integrity of active Fe–N_x sites, as evidenced by postmortem ⁵⁷Fe Mößbauer spectroscopy.^89,90 Potential-resolved differential electrochemical mass spectroscopy (DEMS) study further revealed an increase in CO₂ and CO productions from the carbon corrosion of the FeNC catalyst as the potential exceeds ca. 1.0 V_RHE (Figure 4a), ultimately leading to a significant ORR activity decay due to the destruction of active sites.³⁸ Despite relatively low thermodynamic potentials of carbon corrosion to CO₂ (0.207 V_RHE) and CO (0.518 V_RHE), its slow kinetics widen the stable potential window (up to nearly 1.0 V_RHE) as confirmed by previous DEMS investigations with various carbon substrates.⁹¹ Therefore, stability tests in a half-cell system at room temperature and potential below 1.0 V_RHE (e.g., chronoamperometry analysis) have sometimes shown insignificant activity drop. However, these results may not promise their high stability in practical applications since operating conditions of the real devices are not identical with those of the half-cell measurements. For instance, carbon corrosion is an issue in real devices due to due to higher operating temperatures (Figure 4b)⁹² and, more importantly, due to potential excursions beyond 1.0 V_RHE that occur during practical operations of electrochemical devices (e.g., start–stop and fuel starvation conditions of PEMFC).⁹³ One possible strategy to mitigate the undesirable corrosion would be the employment of carbon substrates having excellent crystallinity (e.g., multiwalled carbon nanotubes (MWCNTs)) and strong corrosion-resistance, besides controlling the operation to remain within the typical potential range. For instance, Lim et al. successfully stabilized isolated Pt ions on N-doped MWCNT and verified their stable electrocatalysis toward chlorine evolution reaction above 1.4 V_RHE, at which carbon corrosion is extremely severe for typical carbon-based materials.⁹⁴

Figure 4.

Corrosion and chemical modification of supporting substrates. (a) Online SFC/DEMS results of FeNC catalyst. DEMS signals were collected in a deaerated 0.1 M HClO₄ electrolyte at 50 °C during a stepwise chronoamperometry experiment between 0 and 1.5 V_RHE. Reproduced with permission from ref (38). Copyright 2015 Wiley-VCH. (b) Influence of temperature on carbon corrosion rate for fully humidified reactant gases at ambient pressure. Carbon loss rate was determined by integrating the peaks in the cathode exhaust gas CO₂ content resulting from single potential cycles between 0.6 and 1.3 V. Reproduced with permission from ref (92). Copyright 2009 The Electrochemical Society. (c) O₂ binding energy of Fe–N₄ and electron work function calculated by first-principle DFT analysis with a Fe–N₄ moiety embedded in a basal graphene plane. The surface is either oxygen-free or oxidized with one or two oxygen-functionalities, at positions as indicated in the insets. Reproduced with permission from ref (96). Copyright 2018 Royal Society of Chemistry. (d) Dissolution of the respective metal and BMC when brought in contact with 0.1 M HClO₄ electrolyte (red) and during potential cycling relevant to the hydrogen evolution reaction (black). Reproduced with permission from ref (103). Copyright 2017 Wiley-VCH.

Not only the structural deformation but also the chemical modification of the support surface can influence the durable operation of SACs. Unlike corrosion, which is an electrochemical process and dependent on the applied potential, surface modifications of SACs are possibly carried out by chemical processes. During ORR electrocatalysis of MNC catalysts (FeNC in particular), the formation of reactive oxygen species (ROS) via the Fenton(-like) reactions between H₂O₂ byproduct and metal species has been indicated as one of the main degradation pathways in PEMFCs.⁹⁵⁻⁹⁷ By treating MNC catalysts with H₂O₂ ex situ, Goellner et al. found a direct correlation between activity decrease and amount of H₂O₂ produced.⁹⁵ The spectroscopic analyses revealed minor changes in the physicochemical characteristics of the center metal, and most of the M–N_x moieties (i.e., 80–90%) remained after the H₂O₂ treatment. The severe degradation compared to the minor loss in M–N_x active sites implied that H₂O₂ dramatically reduces the turnover frequency (TOF) of active M–N_x species. Later, it was further disclosed that the H₂O₂ treatment oxidizes the carbon surface of FeNC catalysts, forming oxygen functional groups on carbon and decreasing TOF without a significant collapse in Fe–N_x sites (Figure 4c).⁹⁶ Moreover, the extent of chemical oxidation is greatly enhanced as temperature increases by virtue of fast kinetics of the Fenton(-like) reaction, resulting in much higher deactivation of Fe–N–C catalysts.⁴⁰ The autocatalytic degradation model proposed by Yin and Zelenay also supports the formation of deactivation species during the ORR (i.e., H₂O₂) and its detrimental effect on long-term stability.⁹⁷ Based on these findings, many efforts have recently been made to secure the durable operation of fuel cells with SACs. Therefore, in order to mitigate the activity loss resulting from the chemical attack of ROS, several synthetic/system-level strategies have been examined, such as introducing secondary catalytic sites beneficial for fast H₂O₂ reduction or decomposition,⁹⁸ and altering the environment to higher pH values to form less active ROS.⁷¹ Especially, there has been recent great progress in improving catalyst stability by replacing the Fe center metal to other d-block (e.g., Co, Mn, etc.) or p-block metals (e.g., Sn) that have milder Fenton(-like) reactivity than Fe.⁹⁹⁻¹⁰²

Although most of the previous works regarding the instability of SACs have been conducted with FeNC or similar MNC catalysts, similar degradation events would possibly occur for other SACs with metal-based supporting substrates, e.g., carbides, nitrides, and phosphides. Using an online ICP-MS, Ledendecker et al. observed considerable dissolution of the binary metallic ceramics (BMCs) once they were brought in contact with the electrolyte (Figure 4d).¹⁰³

4.4. Challenges and Paths Forward

Previous works have pointed out chemical and physical changes in either active metal centers (demetalation and agglomeration) or supporting substrates (corrosions and chemical modifications) as a main degradation pathway of SACs during electrocatalysis. In some cases, the degradation of SACs seems to be intricately intertwined and to not originate from only a single parameter. Hence, based on the previously drawn deactivation pathways of SACs, many recent efforts have been devoted to improve their stability such as minimizing dissolution of inactive metal particles by increasing the atomic dispersion of single-atom sites,¹⁰⁴ mitigating carbon corrosion with highly graphitized carbon substrate,¹⁰⁵ or preventing chemical modification of supports by introducing a ROS scavenger.^106,107 Beyond those synthetic-level strategies, we also would like to emphasize an accurate understanding of the origin of deactivations, which is expected to be feasible with a separation of apparent activity drop of SACs into the changes in their TOF and active site density (SD) values, but is still hardly achieved owing to the lack of general method to evaluate SD of SACs under electrochemical conditions.¹⁰⁸

Unfortunately, a full picture of FeNC long-term degradation under operating conditions is still lacking, thus underlining the necessity of unified accelerated stress tests depending on the particular mode of operation. Currently, the experimental conditions for ASTs employed in the literature differ drastically. For Pt-based FC catalysts, the potential range for cathode load and start/stop cycles was originally defined to be 0.6–1.0 V_RHE and 1.0–1.5 V_RHE., respectively,¹⁰⁹ but various protocols are applied.¹¹⁰ Recently, a more fundamental understanding of the degradation mechanisms in play led to improvements for a combined covering the transition from load to start/stop conditions (0.6–1.5 V_RHE).^109,111,112 A major question remains whether this is appropriate for MNC based ORR catalyst systems. First, the degradation of the carbon-support can be assumed to be similar to state-of-the-art Pt-based systems. Second, the degradation of the active site can be traced to agglomeration from support corrosion and is linked to the first. Third, demetalation can be linked to various factors, including carbon corrosion, ligand oxidation or loss, as well as selectivity of the ORR and the possible production of ROS.^30,40,113 Therefore, the potential range of the AST should be similar to that of Pt-based catalysts. The only significant difference of MNC to the degradation of Pt-based catalysts is the absence of transient-dissolution due to redox process of the metal surface.¹¹⁴ In addition to these modes of degradation, the physical conditions, such as electrolyte temperature and gas saturation, are of key importance when it comes to non-PGM catalysts. For example, ASTs are often still carried out at ambient temperature,^20,22,72,115 even though the influence of temperature on degradation mechanisms is well reported for ORR catalysts.^38,116,117 A second parameter that often varies is the gas saturation of the electrochemical cells during the AST. On the one hand, it has been widely accepted to conduct AST under inert gas conditions with the reasoning that the applied potential governs the degradation. On the other hand, it has been proven that the degradation during the electrochemical reaction itself is not to be neglected. Especially in the case of ORR, experimental data suggests increased degradation during loading when the 4e^– selectivity is not sufficient.¹¹⁸ We, therefore, urge researchers in the field to conduct their ASTs at as relevant conditions as possible, including high temperature and in the presence of oxygen, to account for all modes of degradation. Furthermore, as explained in section 3.4, harmful side products of the ORR can increase exponentially when high current densities are achieved in real devices. Therefore, a better understanding of the degradation of SAC in realistic electrochemical devices without limitations in mass transport and operation at higher temperatures is required. Ideally, ASTs for single cells and stacks need to be developed to understand the extent to which the above-summarized degradation mechanisms occur in the final device and what lifetime can be expected. Moreover, depending on electrochemical systems other than the ORR, SACs can experience fully different potential excursions, electrolyte pHs, and temperatures. They also interact with different molecules or ions and facilitate chemical bond breaking and formation, during which single-atom metal sites and their support can be chemically altered and induce structural and electronic changes.^86,119,120 In all cases, investigation of SAC degradation under more relevant electrochemical conditions and eventual development of structure/stability relationship will be of central importance for a technological breakthrough.¹¹⁴ A last remaining challenge in the understanding and prevention of SAC degradation is the shelf-life conundrum. It has been addressed by the community at various conferences and has also been observed in our laboratories, but to the best of our knowledge it has not been investigated by a dedicated study.¹²¹ It entails that, depending on the age of the raw catalyst powder, the catalyst performance can degrade, possibly due to the thermodynamically favored agglomeration and/or oxidation. Overall, a number of pivotal questions still remain: how and why do SACs degrade during electrochemical operation? The insufficient level of fundamental understanding toward their instability thus renders the rational design of SACs highly vague, and the underrepresented stability issue needs to be highlighted.

5. Conclusion

While there have been tremendous breakthroughs on all levels of fundamental understanding, the next generation of SACs needs to account for the challenges of synthesis scalability, fast and reliable in operando characterization, and long-term stability to bring this technology toward application. In this Perspective, we highlighted the current understanding of SACs in these regards, derived existing challenges and provided suggestions for further research paths. As we have demonstrated particularly by the role model of FeNC as an ORR catalyst, newly developed SACs have to overcome some major issues in each of these individual aspects that go beyond activity optimization and mechanistic understanding. Moreover, the mentioned challenges are also interconnected and therefore have to be advanced in a concerted approach. So, for instance, the stability of SACs needs to be assessed, not only in fundamental studies but also in more relevant electrocatalytic conditions in order to fully understand their influence on performance degradation and the lifetime of real devices. In return, only an increased understanding of the gap between simulative fundamental research and realistic full-scale energy conversion devices can help in the design of proper stability assessment strategies. This, in turn, can provide design criteria for scalable synthesis approaches to overcome inherent stability issues with state-of-the-art SACs. In addition, the effect of technical synthesis conditions needs to be readily screened by the appropriate testing so that factors of scale and cost can be considered from the beginning.

6. Outlook

SACs, and particularly FeNC at the forefront as a PGM-free catalyst for fuel cell cathode catalyst layers, have achieved a technology readiness level (TRL) of ca. 4–5 with single-cell fuel cell tests becoming a standard testing procedure (TRL4: Technology validated in laboratories; TRL5: Technology validated in relevant environments).^22,35,73,123 In comparison, PEMFCs employing Pt-based catalysts can be classified as TRL9 (highest rating: proven in operational environments) with commercially available products in use. As an overview of our proposed paths forward, Figure 5 shows the TRL scale of SACs with respect to the methods employed to reach them, along with the complexity of the system. Here, the blue progress bar marks our current assessment of the FeNC technology based on the efforts conducted in the literature. Everything past that bar is our perspective of challenges that need to be addressed more intensively toward TRL9.

Figure 5.

Overview of the technology readiness level scale¹²² and the progress of FeNC as a fuel cell catalyst indicated by the blue bar, as our assessment of the technology to this date.

To this day, electrode manufacturing for FC tests on TRL4 often relies on manual hand spray or brush deposition onto the GDL or membrane.^{22,30,73,76,124} While some automated approaches have been demonstrated using doctor blading⁷² or ultrasonic spray coating,^55,70,75,125 a drastic scale-up of these procedures, along with efficient performance testing and quality control, will be required to reach higher TRLs. Since it is established that catalyst layer porosity and homogeneity are important factors for mass transport and flooding issues, catalyst layer deposition optimization is imperative. Here, various tomography approaches can help one to understand catalyst layer structures and porosity but are inherently elaborate and time-consuming.¹²⁶ As a time-saving method, GDE testing, integrated into the development process for optimized catalyst layers, could be used for an efficient feedback loop between deposition and performance. Furthermore, GDE in conjunction with ICP-MS should be used to monitor demetalation at relevant currents and temperatures, similar to what has recently been demonstrated for Pt-based catalysts.¹²⁷ To reach TRL6 and 7 of SACs on a FC-stack level, the community requires a drastic scale-up in synthesis and optimized coating capabilities to build uniform MEAs for stack units. Long-term operational stack performance with the help of meaningful ASTs needs to be proven for developing balance of plant (BOP) components and, finally, a commercial product such as a fuel cell stack employing SACs.

Supporting Information Available

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/jacsau.1c00121.

• Detailed literature overview which was used to create Table 1 (XLSX)

Author Contributions

The manuscript was written through contributions of all authors. All authors have given approval to the final version of the manuscript.

C.H.C. acknowledges financial support from the National Research Foundation of Korea (NRF) grant funded by the Ministry of Science and ICT (NRF-2019M3D1A1079309 and NRF-2020R1A2C4002233). F.D.S. and S.C. acknowledge financial support of the CREATE-project from the European Union’s Horizon 2020 research and innovation program under Grant Agreement No. 721065. S.H.J. was supported by the NRF of Korea funded by the Ministry of Science and ICT (NRF-2021R1A2C2007495).

The authors declare no competing financial interest.