1 Introduction

Thermocatalysis plays a central role in the modern chemical industry, providing innovative solutions to improve the efficiency, sustainability, and product yields in various thermo-chemical processes. However, traditional supported metal catalysts (nanoparticle catalysts) typically suffer from insufficient atomic utilization and non-uniform size and shape distribution of metal particles, which can result in high costs of catalyst preparation (especially for noble metal catalysts) and compromise the selectivity of desirable products. Single-atom catalysts (SACs), with 100% atom utilization and uniform isolated active sites, have been receiving intensive attention over the past decade. SACs have found extensive applications in heterogeneous catalysis and exhibit exceptional performance in various thermocatalytic reactions, such as CO oxidation [1, 2], water–gas shift reaction [3,4,5], methane oxidation [6, 7], selective hydrogenation [8, 9]/dehydrogenation [10, 11] etc. SACs are generally stabilized by metal-support interactions. The typical examples include Pt1/FeOx [1], Pt1/CeO2 [12], Pt1/Al2O3 [13], Pd1/TiO2 [14], Au1/Co3O4 [15], and Pt1/N–C [8], etc.

However, the implementation of SACs in industrial applications is still in its infancy mainly because of their unsatisfied stability under continuous reaction processes. SACs are thermodynamically unstable due to the increased surface free energy of single metal atoms [16]. The change of single metal sites during the long-term catalytic process will deprive the advantage of SACs, leading to undesired catalytic behaviors. The unsatisfactory catalytic efficiency and resulting increased production cost greatly hinder SACs’ large-scale application.

In recent years, numerous reviews have been published on the synthesis, structure, and applications of SACs [17,18,19,20,21,22]. However, there has been a lack of discussion and summarization regarding the stability issues of SACs. In this mini-review, we will focus on the topic of SACs stability in thermocatalysis. Specifically, we discuss the deactivation mechanism, deactivation types, protective strategies, and regeneration methods of SACs. We also provide the challenges and opportunities faced by SACs in terms of their further development.

2 The deactivation of SACs

An accurate diagnosis and understanding of the deactivation mechanisms of SACs are essential for designing catalysts with excellent stability. Performance degradation of catalysts can occur due to various factors such as thermal, chemical, or mechanical effects on the metal, support materials, or reactor [23]. While the deactivation of SACs is fundamentally similar to conventional catalyst deactivation, there are also notable differences that require attention. In this section, as shown in Fig. 1, we aim to explain the deactivation behaviors of SACs caused by 1) sintering and agglomeration, 2) embedding, 3) poisoning, 4) coking, and 5) vapor–solid reaction (metal atoms loss).

Fig. 1
figure 1

Schematic illustration of representative deactivation types of SACs

Full size image

2.1 Sintering and agglomeration

The high temperature required for catalytic reactions and the accompanying heat exchange in endothermic or exothermic catalytic reactions are the primary factors that induce deactivation of the SACs by sintering [24, 25]. They can affect both the single metal atom and the support material, leading to a reduction in the amount of effective active sites. Endothermic or exothermic catalytic reactions undergo local heat exchange around the catalytic site. This can serve as a driving force for migrating the single metal atoms immobilized on the support material, leading to single metal site coalescence and particle growth of the SAC. For instance, highly active Pd SACs dispersed on the La-Al2O3 supports rapidly sintered into large particles even at temperatures as low as 100 °C [25]. Moreover, a reducing atmosphere also facilitates the agglomeration of single metal sites. The atomically dispersed Pt in Pt1/CHA (zeolite) can be easily converted into subnanometer clusters or 1 nm NPs in H2 treatment, depending on the reduction temperatures [26]. The sintering of a single metal site is usually due to the stronger metal–metal interactions than metal-support interactions after changing the external conditions. Therefore, great attention should be paid to the strategies to prevent the migration of single metal sites of SAC by support selection and engineering. Additionally, deactivation due to the sintering of support materials is also a fundamental factor that decreases the performance of the catalyst.

2.2 Embedding

The pathway for single metal atom migration in SACs driven by thermal energy can occur not only on the surface of support materials but also internally. This phenomenon is known as the embedding of the SAC through solid-state diffusion. It contrasts with the exsolution phenomenon, where elements precipitate and emerge from the surface of the particles in a homogeneous solid-state solution [27]. Under thermodynamically favorable conditions such as temperature, atmosphere, pressure, and composition, the formation of a solid-state solution by embedding metal atoms is more favorable towards minimizing total surface energy than the presence of single metal atoms on the support surface. As a result, the isolated single metal atoms that were originally anchored on the support surface can diffuse into the internal structure of the support material. For example, single Pt or Fe metal sites were shown to diffuse into the Cu2O subsurface in CO oxidation reactions, leading to catalyst deactivation [28, 29]. However, interestingly, there are cases where SAC deactivation due to strong metal-support interaction can be confused with embedding. Han et al. studied the dynamic behavior of Pt1/TiO2 SAC and Pt NPs/TiO2 in the H2 environment [30]. They found that activity loss of Pt SACs is due to Pt coordination saturation by H and Ti3+, unlike the SMSI-induced embedding of Pt NP by a TiOx layer. In sum, embedding directly excludes the SACs from the surface/interface, restricting their participation in the reaction and causing a detrimental loss of catalytic sites.

2.3 Poisoning

Strongly chemisorbed species on catalytic sites act as a barrier that hinders the access of reactants and are difficult to remove. This type of deactivation caused by strong chemisorption is commonly referred to as poisoning [31]. In addition to physically obstructing adsorption sites, adsorbed poisons have the potential to induce changes in the electronic or geometric structure of the surface or trigger the formation of compounds. The species that can cause poisoning include not only reactants and products but also by-product impurities. Therefore, it is important to consider the possible reactions that can occur between all of these components and the SACs.

Interestingly, SACs can exhibit different poisoning behavior compared to conventional metal nanoparticle (NP) catalysts. For instance, carbon monoxide (CO), which is strongly adsorbed on metallic Pt or Pd NP, is a well-known poison in low-temperature fuel reforming reactions [32]. In other words, at temperatures below 200 °C, where the catalyst is unable to efficiently convert CO to CO2, and trace amounts of CO hinder the reactions of other reactants, leading to rapid catalyst deactivation. However, unlike Pt NP catalysts, Pt SACs with a cationic nature can induce weak CO adsorption, making them potential active sites for low-temperature CO conversion. This finding challenges the conventional belief that Pt and Pd SACs are relatively stable against CO poisoning compared to conventional metal Pt or Pd NP catalysts [33]. However, it should be noted that even with weaker CO adsorption, these SACs are not completely free from poisoning, as cases have been reported in various catalytic applications [34, 35]. For instance, Au1/S-1 SAC can be poisoned by high propylene coverage in the oxidation reaction of propylene [34]. In the liquid phase hydrogenation of p-chloronitrobenzene, the catalytic performance of Pt1/C SACs is drastically reduced due to poisoning of Pt active sites by solvent molecular acetonitrile [35].

2.4 Coking

Coking refers to the physical deposition of carbon species onto a catalyst surface, leading to decreased activity due to blocked active sites or pores of support materials. This can occur through various reactions such as hydrocarbon polymerization, dehydrogenation, carbonization, and condensation, resulting in the formation of high molecular weight species or carbon species [36,37,38,39]. These reactions can encapsulate/cover catalytic metal sites, blocking the access to reactants and thus leading to reduced reactivity, and block mesopores and micropores of supports, hindering the transport of reactants. In extreme cases, the accumulated carbon in the pores can stress the support materials, causing their destruction and leading to the disintegration of catalyst pellets and damage to the reactor. Therefore, an accurate diagnosis of how and where species that cause coking are formed and deposited is crucial. For example, nickel (Ni) is a representative catalyst used for C-H bond activation [40, 41]. Moreover, Ni is an effective catalyst for breaking C–C bonds and has suffered from coke formation due to its high carbon solubility within the lattice. Carbons formed by C–C bond breakage dissolve into the nickel lattice and subsequently precipitate onto the catalyst surface, covering it. Comparatively, Ni SACs are still effective catalysts for C-H bond activation and C–C bond dissociation [42, 43] but do not possess a nickel crystal lattice into which carbon can dissolve, allowing them to be free from carbon deposition caused by precipitation. However, carbon radical species formed by C–C/C-H bond dissociation can still induce coking through mechanisms other than precipitation, so this should be taken into consideration. For example, the carbon radicals formed by C-H bond activation during nonoxidative coupling of methane can be coupled to form soft coke (i.e., oligomers of small olefins), which is deposited on the Pt1/CeO2 SACs and thus causes rapid catalyst deactivation [44]. Notice that the solid carbon product directly produced from the side reactions (eg. methane pyrolysis) can also result in coking deactivation of SACs. In the reaction of dry reforming of methane, Ni1/CeO2 SAC underwent quick activity loss as a result of the carbon deposition caused by CH4 decomposition (CH4 → C + 2 H2) [45].

2.5 Vapor–solid reaction (Metal atoms loss)

Reactions between specific species and the SACs can lead to the formation of volatile compounds, such as metal carbonyls, oxides, sulfides, and halides, which can induce vapor transport of single metal sites even at temperatures much lower than the vaporization temperature of the metal [46, 47]. As a result, the vaporization of the SAC can directly lead to the loss of metal atoms, despite the reaction temperature being significantly lower than the vaporization temperature of metal. Amsler et al. studied the resistance of Rh SACs in vapor–solid reactions by CO using DFT calculations [48]. By calculating how the adsorption energy between the support and Rh-CO compounds changes with the number of CO molecules coordinated to Rh metal, they were able to identify suitable supports for stable immobilization of Rh SACs. When the lattice planes of the support can readily facilitate a high degree of coordination with SACs, it could make the coordination of CO with SACs more challenging, ultimately hindering the metal loss by vapor–solid reaction. However, SACs that were too strongly coordinated to the support suffered losses in catalytic activity because their interaction with the reactant was reduced. Metal active site loss of SACs through vapor–solid reaction is an irreversible deactivation route that can happen during catalysis. Such deactivated catalysts are practically impossible to regenerate. Restoring metal sites on a support material that has experienced metal loss is closer to a re-synthesis. Therefore, this type of deactivation should be carefully considered during catalyst synthesis and catalysis stages, emphasizing the critical need for establishing appropriate protective strategies.

3 Protective strategies of SACs

SACs are susceptible to deactivation through the structure change of single metal sites (sintering, embedding, and metal atom loss), chemisorption of poisoned species (poisoning), and physical deposition of carbon species (coking). To enhance the stability of SACs, it is important to stabilize the single metal atom against migration or protect the active sites from poison species adsorption or carbon deposition. In the past decade, various protective strategies have been developed to extend the lifespan of SACs while maintaining or improving their activity. In this section, we will discuss recent advances in protecting SACs from two perspectives: 1) tuning metal-support interaction and 2) confining metal atoms onto the support.

3.1 Tuning metal-support interaction

The metal-support interaction (MSI) should be regulated properly to immobilize the individual metal atoms while maintaining accessible active sites. SACs are thermodynamically unstable under reaction conditions such as high temperatures, especially on supports with weak MSI [49]. By contrast, a strong metal support interaction can give rise to a coordination saturation of isolated metal sites, leading to activity loss [30]. Therefore, selecting suitable supports and fine-tuning their interaction with metals provide a feasible solution to stabilize SACs. Reducible supports like CeO2 [50] and Fe2O3 [51] can trap mobile metal atoms to stabilize SACs through forming optimal MSI. For example, Abhaya et al. found that Pt atoms on the Pt/La-Al2O3 catalyst would migrate at high temperatures and be trapped by CeO2 to form sinter-resistant SAC [50]. For tuning MSI, increasing the number of defects on the support surface represents one promising way to enhance MSI, as the defects could serve as the anchoring sites of the single metal atom. For example, Murayama et al. reported defective NiO as a stabilizer for Au single-atom catalysts (Au1/NiO), which exhibited high catalytic stability for CO oxidation under 120 °C for 120 h (Fig. 2a) [52]. DFT calculations showed that the formation of Au single-atom sites was more favorable on a NiO surface with Ni vacancy sites than on a clean NiO surface, indicating higher structural stability (Fig. 2b-2c). In addition, crystal facets also have a great impact on the stability of SACs. Li et al. demonstrated that Pd species primarily existed as single atoms on CeO2 (100) facets, whereas they tended to aggregate into clusters on CeO2 (111) facets [53]. This is because a larger amount of oxygen vacancies were generated on the (100) facet of CeO2 in a reducing atmosphere, and Pd atoms could be trapped and thus exhibit atomic dispersion. Such a difference in microscopic structure between Pd1/CeO2 (100) and Pd cluster/CeO2 (111) in turn led to vastly different catalytic performances in the N-alkylation reaction.

Fig. 2
figure 2

a, Durability test of CO conversion over 0.93 wt % Au1/NiO at 80, 100, and 120 °C. Reaction conditions: 150 mg catalyst, 1 vol % CO in air (50 mL min–1), SV of 20,000 mL h–1 gcat.–1. Deposited structures of Au single atoms on clean and defective b, NiO (100) and c, NiO (110) facets. Gray spheres: Ni atoms, red spheres: O atoms, yellow spheres: Au atoms. d, Recycled testing of synthesized Rh1/CeTiOx catalyst for reaction of 5 h at five runs. Reaction conditions: catalyst (30 mg), H2O (20 mL), 250 °C, initial pressure (3.0 MPa, H2/CO2 = 3:1), 5 h, 400 rpm. The optimized structures of Rh atoms dispersed on e, CeO2, f, TiO2, and g, CeTiOx surface. Adapted with permission from [52, 54]

Full size image

Doping of the support is another feasible way to enhance SACs’ stability because dopants can regulate the electronic structure of supports and thus provide a suitable local environment to immobilize single metal atoms. For example, Liu et al. improved the stability of Rh1/CeO2 by doping Ti ions onto CeO2 support to form CeTiOx [54]. Compared to Rh1/CeO2 and Rh1/TiO2 (Fig. 2d), Rh1/CeTiOx SAC exhibited excellent cycle stability for CO2 hydrogenation to ethanol. DFT calculations suggested that the doping of Ti into CeO2 led to the structural deformation of CeO2, which facilitated the formation of shorter Rh-O bands and gave a higher energy barrier for the aggregation of Rh in Rh1/CeTiOx (Fig. 2e-2g). Similarly, Lu et al. reported a stable Na-doped Rh1/ZrO2 SAC, which demonstrated excellent stability for CO2 hydrogenation to CO within 200 h at 300 °C [55]. They found that Na+ reinforced the electronic Rh1-support interactions, endowing the Rh1 atoms more electron deficient, which improved the stability against sintering and inhibited deep hydrogenation of CO to CH4. Moreover, Zheng et al. reported a Na+ dopant-stabilized Ru1/Al2O3 SAC through an indirect protection pathway, where the stability improvement was attributed to the strong interaction between dopant Na+ and H* intermediate species, suppressing the agglomeration of Ru induced by the reduction of Ru1Ox sites with H* [56]. Especially, one can stabilize SACs by directly doping the single metal atom into the support through electronic metal-support interactions (EMSI). For instance, Li et al. developed a stable Ir1–WO3 catalyst with a stronger EMSI than Ir1–CeO2 and Ir1–TiO2 by substituting W atom with Ir atom in the WO3 lattice [57]. The desirable EMSI effect afforded the Ir1–WO3 catalyst excellent activity and stability for the CO2 cycloaddition of styrene oxide to styrene carbonate.

Tuning metal-support interaction is proven an efficient strategy to enhance the stability of SACs. However, the possible trade-off between activity and stability of SAC resulting from regulated metal-support interactions also requires further attention. For example, Yang et al. reported that Pdδ+(0 < δ < 2) single site at surface non-defect sites of ceria exhibited higher activity than Pd2+ and Pd4+ that anchored on the defect sites for methane combustion [58]. The under-coordinated Pdδ+ site could better accept and donate lattice O for easier methane activation. Differently, Zhang et al. applied a simple water treatment to subtly regulate the strong covalent metal-support interaction of Pt1/CoFe2O4 SACs [59]. The weakened interaction between Pt and CoFe2O4 spinel increased catalytic activity without changing the dispersion and stability of Pt atoms. Moreover, Ma et al. found that the anti-sintering ability of Pd1 species on nanodiamond@graphene (Pd1/NDG) is attributed to the redistribution of the Pd single atoms induced by generated coke during propane direct dehydrogenation [60]. The in-situ surface regeneration and the preservation of highly dispersed Pd1 atoms improved the stability of Pd1/NDG SAC. Therefore, it’s necessary to explore the proper metal-support interaction of SACs based on specific catalysts and reactions toward desirable catalytic performance.

3.2 Confining metal atoms onto the support

Strongly chemisorbed species and physical deposition of carbon species on single metal sites result in the poisoning and coking of SACs, thus leading to catalysts’ deactivation. Therefore, it’s crucial to explore the rational design of SACs with lowered adsorption affinity of poisoning species [61] or physically protective shells of single metal sites [62]. Supports that could spatially confine the metal atoms provide a feasible solution to enhance SACs’ stability.

Porous materials like zeolites [62], metal − organic frameworks (MOFs) [63], and covalent − organic frameworks (COFs) [64] are receiving sustainable attention as their well-defined channels and structures can not only confine metals to single sites and protect them against aggregation but also adjust product distribution and inhibit the generation of by-products, thus preventing catalyst deactivation from poisoning [65]. For example, Xing et al. reported an in situ template-guiding strategy to selectively encapsulate single Pt atoms within the six-membered rings of sodalite cages inside Y zeolite (Fig. 3a) [65]. They demonstrated that the Pt@Y-SOD catalyst exhibited remarkable resistance to poisoning species such as CO and thiophene in hydrogenation reactions, with good catalytic selectivity and stability (Fig. 3b-3c). Besides, 2D materials like hexagonal boron nitride (h-BN) [66], graphitic carbon nitride (g-C3N4) [67] and layered double hydroxides (LDHs) [68] are also widely investigated as the engineered defects in these materials can spatially confine the single metal atoms in plane and their 2D structure can promote the diffusion of the products to suppress the metal-product interactions. For example, Lu et al. developed a Pd1/g-C3N4 catalyst (Fig. 3d) that displayed a higher selectivity and coking-resistance than the conventional Pd/Al2O3 and Pd/SiO2 nanoparticle (NP) catalysts in acetylene hydrogenation reaction. For the Pd1/g-C3N4 catalyst, no green oil (carbonaceous species) was observed after a stability test of 100 h. By contrast, considerable green oil was formed for the Pd/C3N4-NP catalyst (Fig. 3e-3f). The improved coking-resistance on Pd single site catalyst was probably because coke species could not be formed at 2D g-C3N4 supported single Pd site. However, the agglomeration of the Pd single atoms still occurred during the reaction [36]. This is a typical example that SACs exhibit excellent coke-resistance ability benefiting from isolated active metal sites. Notice that poisoning and coking-induced catalyst deactivation are due to the covering of formed species on catalyst surface, and therefore removing these species to reuse the spent catalysts can be possible (named SAC regeneration). More details will be given in Sect. 4.4.

Fig. 3
figure 3

a, Schematic illustration of the synthesis strategy of Pt@Y-SOD by selectively encaging single Pt atoms into the six-membered rings of SOD cages within Y zeolite. b, In situ DRIFT spectra of CO adsorption on Pt@Y-SOD and Pt/Y catalysts. c, H2-D2 exchange experiments over pristine Pt@Y-SOD and Pt/Y catalysts and the corresponding thiophene-treated catalysts. d, Schematic illustration of the synthesis of Pd1/C3N4 catalyst by atomic layer deposition on pristine g-C3N4. Durability test on e, Pd/C3N4-NP(WI) and f, Pd1/C3N4 in the selective hydrogenation of acetylene in excess ethylene for 100 h at about 55 °C. The insets are photographs of the reactor outlet after the long-term stability test. Adapted with permission from [36, 65]

Full size image

4 Regeneration of SACs

The regeneration of SACs from their deactivation is a promising way to reuse the spent catalysts in terms of economic and environmental benefits. However, the employed regeneration strategy greatly depends on the specific deactivation behavior and mechanism of spent catalysts. SAC deactivation involves the agglomeration of single metal sites, the embedding of metal sites into supports, poisoning, and coking. In this section, we’ll discuss how to reactivate spent SACs in each case.

4.1 Regeneration of agglomerated SACs

Agglomeration of single metal sites into metal nanoparticles (NP) during the reaction may be the most common deactivation phenomenon of SACs. This is usually due to the stronger metal–metal interactions than metal-support interactions after changing the external conditions. Therefore, the regeneration of agglomerated SACs will be a reverse transformation from NP to single metal sites. Corma et al. reported that H2 reduction treatment would convert single Pt sites in Pt1/CHA into Pt NPs, while O2 treatment could fragment the agglomerated Pt NPs into Pt single sites (Fig. 4a and 4b) [26]. Their result indicated that the interconversion between Pt1 sites and Pt NPs can be achieved by controlling the redox properties of the atmosphere. Datye et al. demonstrated that aggregated Pt–Sn clusters/CeO2 catalysts can be easily regenerated to atomically dispersed Pt1-Sn/CeO2 SAC under oxidizing conditions through atom trapping [69]. Ding et al. applied CO/CH3I thermal treatment to redisperse Ir NPs to Ir single atoms on activated carbon (AC) (Fig. 4c-4e) [70]. But they found NPs with a size larger than about 5 nm are difficult to disperse due to strong Ir–Ir bonds together with the kinetic stability (Fig. 4f-4h). In sum, this NP-to-single-atom transformation regeneration can be realized when the interactions between the support and metal single atoms are stronger than the original metal − metal bonding at optimized conditions. Besides, further studies are needed to demonstrate the applicability of this (oxidizing) approach for the regeneration of non-Pt SACs.

Fig. 4
figure 4

a, The interconversion between Pt single atom and Pt NP by changing reducing/oxidizing atmosphere; b, Fourier-transformed extended X-ray absorption fine structure (FT-EXAFS) spectra of Pt-CHA sample after various thermal treatments. c, TEM image and corresponding particle size distribution of Ir/AC reduced by CO/H2 at 513 K, d, HAADF-STEM image of Ir/AC after regeneration, and e, EXAFS analyses of Ir/AC after regeneration; f, HAADF-STEM image of Ir/AC reduced by H2 at 573 K and the corresponding particle size distribution, g, HAADF-STEM and h, TEM image of Ir/AC after regeneration. Adapted with permission from [26, 70]

Full size image

4.2 Regeneration of embedded SACs

SACs deactivation by metal sites embedded into the supports generally results from the surface reconstruction of SACs, which buries single metal sites underneath the support. The metal sites of SACs are not accessible to reactants and thus deactivated. Therefore, regenerating embedded SACs should be re-exposing embedded metal sites to the environment with unchanged geometric and electronic structures. However, the relevant reports are rare to date probably due to the lack of a feasible strategy. More attention should be paid to reactivating buried SACs or preventing SACs embedding.

4.3 Regeneration of poisoned SACs

SAC poisoning is due to the coverage of single metal atomic sites by strong chemical adsorption/bonding of reactants/intermediates/products/impurities in the feed. Common SAC poisoning includes sulfur poisoning, carbon monoxide poisoning [1, 71, 72], etc. The key to regenerating the poisoned SACs is to remove the covered species. However, many studies prove that the SACs are not easily regenerated once poisoned, and they will be deactivated rapidly even if SACs are regenerated. This is because the active metal atoms have a high affinity for poisoned species. The promising way to avoid SAC poisoning is to develop poisoning-resistant SACs by lowering the affinity between active metal atoms and poisoned species. For example, Gates et al. found that the iridium pair-site catalysts are active for ethylene hydrogenation even after poisoning by CO. This is because formed CO ligands are gradually replaced by reactants as catalysis proceeds [73]. Similarly, Ding et al. developed a single-atom Rh1/porous organic polymer (POPs) catalyst with self-recovered ethylene hydroformylation activity after the removal of H2S from the feed. They attributed it to the easy elimination of sulfur-poisoned species by the reactant (CO/H2) [74]. Besides, Ding et al. also developed a single-atom Rh1/porous ionic polymer(PIPs) catalyst with sulfur-promoted hydrocarboxylation of olefins. They found that the addition of sulfur species forms a particular sulfur ligand, which could lower the energy barrier of the rate-determining step of migration and insertion of CO [75]. This is an excellent example of utilizing sulfur species as a promoter for sulfur-resistant catalysis.

4.4 Regeneration of coked SACs

The coking of SACs in the catalytic reaction usually originates from the coverage of formed carbonaceous or carbon products of desired reactions or side reactions. The regeneration of coked SACs is easier than agglomerated or embedded SACs. Spent SACs can be reactivated by removing formed coking species. The commonly employed strategy is the oxidative treatment at optimized temperatures. For example, coked Co/N–C [76], Pt1Ga-Pb/SiO2 [77], Ru1/N–C [78] catalysts can be quickly regenerated by O2 oxidative treatment. As shown in Fig. 5a and 5b, once the coke species formed on Co1/N–C and Ru1/N–C catalyst surfaces are removed by oxidative treatment, their catalytic activity could be rapidly recovered. Furthermore, water vapor treatment is also reported to regenerate the deactivated SACs by facilitating the removal of coke or other poisoning species. For example, water co-feeding can contribute to the HCl desorption and coke species removal on Pt1-Co/HZSM-5 [79,80,81] during the catalytic oxidation of dichloromethane, thus enhancing the stability of catalysts (Fig. 5c).

Fig. 5
figure 5

a, The catalytic behavior and regeneration ability of the Co1/N–C catalyst for ethylbenzene dehydrogenation; b, The catalytic performance and regeneration ability of Ru1/NC catalyst for propane dehydrogenation; c, Stability and water regeneration ability of 0.01 Pt/HZSM-5, 20Co/HZSM-5 and 0.01 Pt-20Co/HZSM-5 for catalytic oxidation of dichloromethane. d, Thermal stability examination of SACs in regeneration conditions. Adapted with permission from [76, 78, 79]

Full size image

It is noted that the employed regeneration strategies should not change the original structure of SACs, since the coordination environment and electronic properties of SACs are key to their catalytic performance. However, limited studies have been carried out to systematically investigate the regeneration of spent SACs. Most reports are focused on the performance recovery of spent SACs after the regeneration process. It lacks examination of local geometric and electronic structure of active metal atoms in regenerated SACs (Fig. 5d). Furthermore, a fundamental understanding of the thermal stability of SACs in regeneration conditions is needed, which is helpful for the regeneration strategy design and optimization.

5 The conclusions and outlooks for future directions

By reducing the size of metal nanoparticles in catalysts to the single-atom level, it has been possible to achieve excellent atom utilization and exceptional catalytic performance. However, concerns have been raised regarding the stability of these catalysts due to the increased surface free energy of single metal atoms. In this mini-review paper, we summarized the deactivation mechanism of SACs, protective strategies of preventing SACs deactivation, and regeneration methods of spent SACs. We found that the activity, selectivity and stability of SACs largely depend on the metal-support interactions, the tuning of which could serve as effective means to regulate the structure and catalysis behaviors of SACs. Furthermore, reaction conditions such as temperatures, atmosphere (eg. H2O, O2 and H2), and reactants/products also affect the stability of SACs. Therefore, to design next-generation SACs, the tolerance of SACs’ structure in changed reaction conditions for a specific reaction should be considered.

Optimizing the stability of SACs necessitates finding the optimal material and structure for a reaction. This inevitably involves numerous attempts, however trying them out through conventional trial and error methodology is challenging. Data science and machine learning can help us to transform the data into actionable insights. To start with the selection of a target problem (eg. SACs stability), machine learning is expected to understand and quantify which catalyst properties are relevant for the SACs to attain desirable performance toward specific reactions. The screening of the materials in terms of composition and structure, based on critical activity descriptors can be conducted to target the best catalysts.

From an experimental perspective, the introduction of a new support [82], protecting layer, or secondary metal element [83, 84] can be a promising strategy to synthesize stable SACs. Especially, co-localizing single atom/few atoms around the mononuclear active site not only stabilizes/confines the single metal atom of SACs, but also tunes the electronic structure of active sites to efficiently facilitate elementary steps in the catalytic cycles. This method recently has shown great potential for improving SACs’ stability, activity, and selectivity [83, 84]. However, further exploration is required to expand the material family of stable SACs. By employing the high-throughput screening system using automated equipment, it is possible to perform the necessary experiments more efficiently, accelerating the exploration/discovery of desirable SACs.

Limited by the development of characterization techniques, our understanding of the electronic/geometric structure change of SACs and their deactivation dynamics in practical working states is still inadequate. Extremely high-resolution microscopy and spectroscopy techniques under operando conditions are needed. Nevertheless, the understanding of deactivation behaviors and the anchoring sites of single atoms, based on well-defined chemistry, have paved the way for the development of strategies for the protection and regeneration of SACs. The SACs with enhanced stability and outstanding performance hold great promise as breakthrough materials in industrial applications for economic benefits.