Single atoms that act as catalysts are widely found in nature and play important roles in vital molecules, e.g., Mo in nitrogenase[1], Fe in heme [2], and Mg in chlorophylls [3]. In vivo, single atoms that act as catalysts are coordinated with biomacromolecules and work cooperatively. Similarly, synthetic chemical ligands can provide single atoms in biomimetic functions and other applications. For example, Wilkinson's catalyst (RhCl[P(C₆H₆)₃]₃), the first successful homogeneous catalyst, is commonly used for hydrogenation, hydroformylation, hydrosilylation, and hydroboration of alkenes and alkynes. Such homogeneous catalysts usually have excellent catalytic turnover numbers because they take full advantage of each metal atom, but the instability of these organic molecular ligands and the difficulty of catalyst recovery are major drawbacks. Inorganic solid materials can also serve as specific ligands for single metal atoms [4], and the concept of single-site [5, 6] or single-atom catalysts (SACs) [7, 8] is being increasingly developed [8–12]. Among heterogeneous supports, traditional bulk metal oxides, e.g., MgO, Al₂O₃, SiO₂, TiO₂, FeO_x, ZnO, and CeO₂, zeolites, and metal-organic frameworks are generally used to stabilize various single metal atoms [9]. The excellent catalytic performance of a SAC arises from both the coordinatively unsaturated state of the active center and its stable coordination with the support. Two-dimensional (2D) materials with special physical and chemical properties could therefore provide a new class of promising and ideal supports for single atoms, but this has not been widely studied.

The successful exfoliation of graphene [13] has led to an increase in the use of 2D materials, and the use of analogous materials such as h-NB, C₃N₄, and MoS₂ is expanding [14–16]. These 2D materials are not only excellent catalysts in themselves [17] but also have may specific advantages in constructing SACs. First, 2D materials with unique geometric and electronic structures [17] can modulate the catalytic behavior of single atoms in unusual ways. Secondly, 2D materials always have large specific surface areas, which can create more active sites by anchoring single atoms on the surface. Thirdly, strictly single-atom-layer 2D materials can facilitate the adsorption and diffusion of reactive molecules on confined single atoms from two sides. Fourthly, such composites provide model catalysts for which uniformly active sites can be well identified and catalytic performances can be predicted using theoretical chemical methods [18]. Finally, the confined single atoms can promote or activate the intrinsic catalytic activities of 2D materials.

In this perspective, we summarize recent progress in research on catalysts based on the confinement of single atoms in 2D materials, mainly graphene, C₃N₄, and MoS₂. We focus on how 2D materials and confined single atoms work cooperatively in catalysis, the identity of the real active sites, and the potential practical applications of such novel catalysts.

Graphene consists of single-atom-layer sp² carbon, and can be easily doped with non-metal heteroatoms such as N, B, P, and S. Heteroatom-doped graphene materials are widely used in electrocatalysis [19] and photocatalysis [20]. The N-doped graphenes have been widely researched and various methods for their preparation by direct synthesis or post-treatment synthesis have been developed [21]. Doping of N into the graphene lattice polarizes the carbon network, which changes the electronic properties of graphene. This N-induced electronic perturbation can be directly observed using scanning tunneling microscopy (STM) [22, 23]. Deng et al. reported that the STM images of N-doped graphenes show brighter sections that are distinct from the main graphene network (Fig. 1(a) and (b)); this has also been observed in other studies (Fig. 1(c)). These observations indicate substitution of N atoms in graphene. Density functional theory (DFT) calculations and STM simulations have confirmed that these bright spots are not images of N atoms but of C atoms neighboring N atoms, and are the result of increased electron densities of states caused by doping with N atoms [22, 23]. Such doped graphitic N single atoms (Fig. 1) not only facilitate the dissociation of O₂ molecules on adjacent C atoms but also promote the formation of strong C–O chemical bonds through electronic interactions. The catalytic activity of N-doped graphene is therefore better than those of pure graphene and commercial XC-72 in oxygen reduction reactions (ORRs) [22].

Research on graphene-confined SACs has increased recently because of their broad range of applications. Sun et al. used atomic layer deposition (ALD) to anchor Pt single atoms on graphene [24] or on N-doped graphene [25, 26], and these Pt SACs were used in the electrocatalytic methanol oxidation reaction (MOR) [24], ORR [25], and hydrogen evolution reaction (HER) [26]. The activities of Pt SACs are 10 times and 37 times higher than those of their Pt nanoparticle (NP) counterparts (commercial Pt/C) in the MOR and HER, respectively. Based on DFT calculations and X-ray absorption fine structure (XAFS) analysis, the authors proposed that the low-coordination and partly unoccupied densities of states of the 5d orbitals of the Pt single atoms are responsible for the observed excellent catalytic performances [24, 26]. Pt SACs also showed much higher CO tolerance during the MOR [24] and long-term stability (1000 cycles) during the HER [26]. This is attributed to the high binding energy between Pt single atoms and C (or O) atoms at the graphene edges [27], which can be enhanced by doping with N [25].

Similarly, Yan et al. [28] synthesized a Pd₁/graphene SAC using atomic layer deposition. In the selective hydrogenation of 1, 3-butadiene, Pd₁/graphene showed about 100% selectivity for butenes (70% selectivity for 1-butene) at 95% conversion at 50 ℃. Under the same conditions, graphene-supported Pd NPs (5.5 nm) gave only 70% selectivity for butenes and 15% selectivity for 1-butene. The authors suggested that changes in the 1, 3-butadiene adsorption mode and enhanced steric effects at Pd single atoms are possible reasons for these results (Fig. 2(a)). Similar to the situation for graphene-supported Pt SACs [24–26], XAFS analysis indicated that Pd single atoms coordinate with defective C and phenolic O atoms (Fig. 2(a)) and therefore show +2 valence state behavior [28]. Such strong bonding interactions endow Pd₁/graphene with excellent durability against metal sintering and carbon deposition for a total 100 h of reaction time on stream [28].

Qiu et al. [29] used chemical vapor deposition to grow graphene on nanoporous Ni substrates. Three-dimensional (3D) graphene-confined Ni SACs can be acquired by subsequent acid etching, and can therefore be used for HERs in acidic solutions (0.5 mol/L of H₂SO₄). The overpotential (50 mV) and Tafel slope (45 mV/dec) for Ni SACs were lower than those for conventional Ni-based catalysts. It is proposed that the unusual catalytic performance of Ni₁/graphene is the result of sp-d orbital charge transfer between the Ni single atoms and the surrounding carbon atoms. Unlike Pt₁/graphene [24–26] and Pd₁/graphene [28], in which the single metal atoms are mainly anchored on graphene vacancies, a single Ni atom substitutes for one C atom in the graphene lattice and chemically bonds with three C atoms (Fig. 2(b)). DFT calculations show that graphene with substituted Ni atoms should be more efficient in the HER (Fig. 2(b)) [29], and the formed C–Ni bonds ensure excellent cycling stability (1000 cycles).

Co SACs confined in N-doped graphene for HERs have also been reported [31]. An overpotential of 30 mV was reached, and the durability was high (1000 cycles) under both acidic and basic conditions. The catalyst was prepared by simply heat-treating (750 ℃) graphene oxide and cobalt salts in an Ar/NH₃ atmosphere. XAFS analysis indicated that Co single atoms strongly coordinate with N atoms, which are the catalytically active sites. However, the authors did not report the exact configuration. Zhang et al. [30] synthesized an analogous Co SAC using a similar pyrolysis method. They found that the active sites were Co–N–C structures with pyridinic N as coordinating atoms (Fig. 2(c)) in aerobic oxidative C–C coupling reactions, and the Co SAC showed the highest turnover frequency by far for these reactions. Furthermore, pyridinic N-stabilized single atoms of Pt-group metals [32, 33] or base metals [34–36] were also reported to be the catalytically active sites in the decomposition of formic acid [32, 33] or the ORR [34–36], respectively.

Although significant progress has been made in single-atom dispersed MN_x catalysts, their accurate synthesis and characterization are still a challenge [37]. Recently, Deng et al. [38] developed a facile ball-milling method for the large-scale synthesis of high-density and highly dispersed FeN₄ species confined in a graphene matrix. The atomic structure of FeN₄ centers in graphene were shown for the first time [38, 39]. High-angle annular dark-field scanning transmission electron microscopy (HAADF-STEM) images (Fig. 3(a) and (b)) show homogeneously scattered Fe single atoms on graphene as small bright dots. The electron energy loss (EEL) atomic spectrum (Fig. 3(c)) of the bright dot in Fig. 3(b) clearly shows that the Fe and N atoms are adjacent to each other, suggesting the formation of Fe–N bonds. Low-temperature STM shows the atomic structure of the FeN₄ center embedded in the graphene matrix (Fig. 3(d)), which can also be well reconstructed using simulated STM (Fig. 3(e)). The Fe single atom is resolved as a bright spot, and the neighboring N and C atoms are brighter than the other C atoms in the graphene lattice, suggesting that the Fe single atom significantly modifies the densities of states of the adjacent atoms and vice versa. STM images across the FeN₄ center (Fig. 3(f)) show a sharp resonance state at −0.63 eV below the Fermi level, indicating that the Fe single atom strongly interacts with the graphene lattice and thus introduces a new electronic state near the Fermi level.

Such coordinatively unsaturated Fe sites (FeN₄/GN) are highly active in the catalytic oxidation of benzene to phenol at room temperature (Fig. 4(a)) and even at 0 ℃ because of the easy formation of Fe=O intermediates [38]. DFT calculations (Fig. 4(b)) showed that a H₂O₂ molecule can be easily dissociated on the FeN₄ center to form Fe=O and O=Fe=O intermediates successively; this is consistent with the XAFS analysis. The adsorption and activation of benzene on the O=Fe=O sites should be the rate determining step, with an energy barrier of only 0.59 eV, which is moderate and enables low-temperature reactions. Furthermore, the FeN₄ center can directly dissociate an O₂ molecule [40], and the dissociated O on the single Fe atom is much more strongly adsorbed than those on the C atoms (Fig. 4(d)) because of the notably higher distribution of the density of states of Fe near the Fermi level than of those of the C atoms (Fig. 4(e)). The ORR activity of FeN₄ SAC is therefore significantly high and comparable to that of commercial 40% Pt/C (Fig. 4(c)). FeN₄ SAC also has a higher stability and tolerances to SO_x, NO_x, and methanol than does Pt/C, probably because of its preferential adsorption of an O₂ molecule [40].

The high-energy ball-milling method [38] was used to synthesize various graphene-confined base-metal SACs with MN₄ (M = Mn, Fe, Co, Ni, or Cu) centers; the materials were used as counter electrodes in dye-sensitized solar cells (DSSCs) [41]. Among all these MN₄/GN catalysts, CoN₄/GN was the most active and stable in interconversion of the redox couple I⁻/I₃⁻, and its power conversion efficiency (PCE) surpassed that of a sputtered Pt electrode (Fig. 5(a)). DFT calculations showed that CoN₄/GN has superior properties because the adsorption energy of iodine on the confined Co sites is appropriate, leading to a good balance between adsorption and desorption processes (Fig. 5(c)). In detail (Fig. 5(b)), the optimum binding energy of iodine on MN₄, E_{ad, I}, is predicted to be about −0.80 eV, which exactly corresponds to that on CoN₄/GN. Furthermore, the confined CoN₄ structure is the most stable among these MN₄/GN catalysts because of the higher energy of Co–N bonds (inset in Fig. 5(b)). Both examples [40, 41] show the potential of graphene-confined MN₄ SACs for replacing precious metals as catalysts.

The results described above show that graphene has various sites for confining single atoms, such as defective C or O, doped N, and even S [42], and these functional groups can modulate the catalytic performance of the SAC. A graphene-confined SAC usually shows high stabilities in many catalytic reactions, highlighting the potential for practical applications. The development of more efficient and facile methods for the large-scale production of graphene-confined SACs with high loadings is therefore an important issue.

Graphitic carbon nitride (g-C₃N₄) is one of the oldest synthetic polymers [43] and has a broad range of applications in photocatalysis [44, 45], organocatalysis [46], and other areas [47]. Unlike graphene or heteroatom-doped graphene, which are typical zero-bandgap materials, 2D g-C₃N₄ is a typical polymeric semiconductor with an sp² π-conjugated system [48]. g-C₃N₄-confined SACs therefore have specific catalytic properties and applications [49–51]. Mesoporous g-C₃N₄ has the same microstructure as 2D g-C₃N₄, and is therefore also covered in this perspective [49, 50].

Vile et al. [49] prepared a g-C₃N₄-confined Pd SAC for three-phase flow hydrogenation of alkynes and nitroarenes. Single Pd atoms were stabilized in the N-coordinating cavities through electrostatic interactions (Fig. 6(a)), which led to much higher activity, selectivity, and stability. In the selective hydrogenation of 1-hexyne, for example, the activity of Pd₁/g-C₃N₄ is an order of magnitude higher than that of the Lindlar catalyst at 90% selectivity to alkene, and its stability is good (Fig. 6(b)). DFT calculations showed that the high activity and selectivity can be attributed to facile hydrogen activation and alkyne adsorption on single Pd atoms. g-C₃N₄ is regarded as a special ligand for stabilizing single Pd atoms, inhibiting the adsorption of potential poisons (CO), and promoting H₂ activation.

Chen et al. [50] used a bottom-up strategy to synthesize a g-C₃N₄-confined Ag SAC via copolymerization of cyanamide with silver tricyanomethanide. The results showed single Ag atoms of valence state +1 in cages (Fig. 6(c)). In the semihydrogenation of 1-hexyne, the activity of Ag₁/g-C₃N₄ is about twice as high as those of Ag NP counterparts, and its selectivity is constant at 100% under a wide range of reaction conditions. It should be noted that such a bottom-up method can achieve much higher loadings of single metal atoms (10 wt% for Ag [50] and 8.5 wt% for Cu [35]) compared with those achieved using other common methods, by limiting the amount of metal to within 1 wt% to achieve single-atom dispersion.

Recently, Li et al. [51] reported that highly stable and dispersed isolated single Pt atoms anchored on 2D g-C₃N₄ can significantly enhance photocatalytic H₂ evolution activity. Its specific activity is 8.6 and 50 times higher than those of Pt NP-decorated g-C₃N₄ and bare g-C₃N₄, respectively. The N/C-coordinated framework formed by the tri-s-triazine structure facilitates the binding of single Pt atoms in the g-C₃N₄ matrix (Fig. 6(d)). It is proposed that these single Pt atoms change the surface trap states of g-C₃N₄, resulting in longer lifetimes of photogenerated electrons, leading to higher H₂ evolution activity. Unlike the above-mentioned single metal atoms whose properties are usually modified by supports, single Pt atoms used as a cocatalyst mainly modulate the electronic structure of g-C₃N₄, and therefore improve the photocatalytic performance [51].

Research on g-C₃N₄-confined SACs is still in its infancy, especially in the case of 2D g-C₃N₄ [48]. Compared with N-doped graphene, a 2D g-C₃N₄ support has more specific N species and larger N-coordinating cavities for anchoring single metal atoms [52]. Most importantly, it is also a catalyst in itself, on which single metal atoms could stimulate its intrinsic nature in unusual ways. Such g-C₃N₄-confined SACs are expected to be excellent catalysts for various catalytic processes such as ORRs [53], oxygen evolution reactions [54], and photocatalytic reduction of CO₂ [55]. It is therefore an important challenge to achieve these energy-related aims and further expand the practical applications of these SACs.

Two-dimensional MoS₂ crystals, which have unusual physical and chemical properties, have been attracting increasing attention innumerous research fields during the past few years [14–16]. At present, there is a consensus that the active sites in MoS₂ correspond to coordinatively unsaturated S or Mo atoms along its edges [56–58], whereas the perfect in-plane domain of MoS₂ is always inert. Deng et al. [59] successfully activated the inert MoS₂ surface for HER catalysis by replacing Mo atoms with other single metal atoms. Among various transition metals, Pt is predicted to be the most effective for modulating the adsorption behavior of H atoms on neighboring S sites and consequently the HER activity (Fig. 7(a)). A one-pot solvothermal method was developed to achieve in-plane doping of single Pt atoms in 2D MoS₂. HAADF-STEM images (Fig. 7(c) and (d)) showed that the single Pt atoms (brighter white dots) were uniformly dispersed in the 2D MoS₂ plane and exactly occupied the positions of Mo atoms. Compared with pure few-layer MoS₂ (FL-MoS₂), Pt-doped 2D MoS₂ (Pt-MoS₂) showed significantly higher HER activity with a much lower overpotential (Fig. 7(b)). Moreover, its catalytic stability and poisoning resistance are superior to those of Pt supported on the 2D MoS₂ surface because of protection by the MoS₂ sandwich structure. This means that single Pt atoms do not directly catalyze the reduction of H₃O⁺ but trigger HER activity of neighboring S atoms [59].

Fig. 7(a) also shows that Co is one of the best potential base metals for modulating the intrinsic activity of the MoS₂ surface, but the HER performance of Co-doped MoS₂ is greatly inferior to that of Pt-doped MoS₂ [59]. To improve the activities of such non-precious metal catalysts, Deng et al. [60] developed a multi-scale control strategy for constructing a 3D MoS₂-confined single Co atoms composite. Fig. 8(a) shows that the composite has uniform mesopores at the macroscale, which could facilitate the transport of H₃O⁺ and H₂, and increase the accessibility of MoS₂ surface. At the nanoscale, oriented vertical growth of MoS₂ nanosheets around the mesopores increases the number of edges, which are active sites. At the atomic scale, chemical doping with single Co atoms of mesoporous MoS₂ foam (mPF-MoS₂) can further enhance the intrinsic HER activity by triggering the activity of neighboring S atoms. The authors also found that the optimum Co content in the Co doped mesoporous MoS₂ foam (mPF-Co-MoS₂) for high-performance HER activity was 16.7% (Fig. 8(b)); this material showed long-term durability over more than 5000 cycles and an overpotential of only 156 mV at a current density of 10 mV/cm², comparable to those of Pt-MoS₂ (Fig. 7(b)) and many other excellent MoS₂-based electrocatalysts. DFT calculations confirmed that moderate Co doping could modulate H adsorption on MoS₂ to a suitable degree and simultaneously maintain the structural stability, enabling the HER activity to reach the optimum value (Fig. 8(c)).

The above studies [59, 60] explored the latent capacity of single metal atoms to trigger the intrinsic catalytic activity of the MoS₂ surface. This could equally be applied to other 2D transition-metal dichalcogenides (e.g., WS₂, MoSe₂, WSe₂, and MoTe₂). The potential influences of such inductive effects in 2D MoS₂-confined SACs on other catalytic reactions such as hydrodesulfurization [61, 62] and syngas conversion to higher alcohols [63, 64] are worth studying. Alternatively, substituting S atoms in MoS₂ with single non-metal atoms such as B, N, and P is another promising method for tuning electronic structures and consequently catalytic behaviors.

Here, we have summarized recent progress in the synthesis and catalytic applications of single atoms confined in 2D materials and discussed their possible catalytic mechanisms. Three 2D materials, namely graphene, g-C₃N₄, and MoS₂, were discussed. Of these, graphene is the most studied and usually only serves as a support on which suitable types and numbers of main-group elements (e.g., C, N, O, and S) are essential for stabilizing single atoms and simultaneously maintaining their coordinatively unsaturated states for catalysis. g-C₃N₄ has abundant N species and large N-coordinating cavities for anchoring single metal atoms; its 2D heterostructures are still under development. Recently, a 2D C₂N material was successfully synthesized and characterized [65, 66]. Its N content falls between those of regular N-doped graphene and g-C₃N₄, which should endow the supported single metal atoms with some unexpected catalytic activities.

Rather than the usual use of single atoms as active sites [8, 9], this perspective emphasized the special role of single atoms in promoting the intrinsic catalytic properties of 2D materials, if the 2D materials themselves have catalytic abilities, e.g., g-C₃N₄ for photocatalysis [51] and MoS₂ for electrocatalysis [59, 60]. Similarly to 2D transition-metal dichalcogenides [59, 60], 2D metal oxides (e.g., layered double hydroxides, TiO₂, and perovskite-type materials) [15] are efficient catalysts in various catalytic processes and have seldom been studied for the confinement of single atoms [67, 68]. This is therefore a large and promising field of research on catalysis for energy and environmental science.

The mutual benefits between 2D materials and single atoms make their composites excellent models for developing real catalysts. Efficient methods for the synthesis of SACs confined in 2D materials, rather than just the preparation of such 2D materials [69], is therefore important. Precise control of the structure can facilitate recognition of the real active sites in situ and can help to elucidate the catalytic mechanism objectively [70, 71]. Furthermore, the mass production of 2D materials with confined high-density and highly dispersed SACs is necessary for practical applications. How to combine both aspects is an ongoing issue in this pioneering field and will undoubtedly have a far-reaching impact on fundamental research and industrial applications.

Acknowledgments

We gratefully acknowledge the financial support from the Ministry of Science and Technology of China (2016YFA0204100, 2016YFA0200200), the National Natural Science Foundation of China (21573220, 21621063), the Key Research Program of Frontier Sciences of the Chinese Academy of Sciences (QYZDB-SSW-JSC020), and the Strategic Priority Research Program of the Chinese Academy of Sciences (XDA09030100).