Single atom catalysis for electrocatalytic ammonia synthesis

Abstract

Ammonia is a vital base molecule for modern agriculture and industry. As the commercially-mature approach for NH₃ production, the traditional Haber–Bosch process has achieved great success; however, it also suffers from drawbacks such as massive energy consumption and huge CO₂ emissions. The electrocatalytic nitrogen reduction reaction (eNRR) process, powered by intermittent renewable energy, can realize the carbon-neutral and on-site generation of NH₃ under mild conditions directly from air/N₂ and water, and has gradually emerged as a multidisciplinary research hotspot. Owing to their advantages such as maximum atom utilization as well as special coordination environment and electronic structure, single atom catalysts (SACs) exhibit great potential in enhancing the faradaic efficiency and ammonia yield rate of the eNRR process. This review analyzes recent theoretical and experimental advances in the development of SACs used for the eNRR, including their synthesis and in situ characterization, as well as the classification and activity of noble metal-, non-noble metal- and nonmetal-based SACs. In addition, we summarize the unique structure and electronic properties of SACs and their enhanced activity in electrochemical reactions, to guide the rational design of efficient catalysts. Finally, we discuss major challenges related to the synthesis of NH₃via SACs, and outline future prospects in their development.

---

1. Introduction

Ammonia (NH₃) is one of the most important and widely applied base chemicals. Its global production is 235.34 Mt and is expected to reach 290 Mt in 2030.¹ In addition, the global ammonia market is expected to reach 76.64 billion US dollars by 2025.² An 88% proportion of the synthesized NH₃ is used to produce agricultural fertilizers.^3,4 NH₃ production accounts for >40% of the overall industrial revenue and contributes to global population growth.⁵ In addition, NH₃ is used as a raw material for manufacturing medicines, synthetic fibers, explosives, plastics, rubber, etc.^6,7 Following the rapid development of renewable energy technologies, NH₃ plays a promising role as a H₂ energy carrier, owing to its advantages such as high energy density (17.6% hydrogen content) and relatively low-cost transportation.^8,9 As a carbon-free molecule, renewable NH₃ holds promise for replacing the current fossil fuel-based energy industry with a zero-emission approach.

Although the atmosphere is rich in nitrogen, the inert nature of N₂ makes the synthesis of NH₃ undisputedly challenging. The breakage of the N≡N bond requires an extremely high energy (∼941 kJ mol⁻¹).¹⁰ The first-generation artificial nitrogen fixation method was the Birkeland–Eyde process, which directly combines nitrogen and oxygen on an electric arc to produce NO. The development of this process was hindered by the low energy efficiency and immature electricity market in the early 20th century.¹¹ Similarly, the modern Frank–Caro process, based on the reaction between nitrogen and calcium carbide, was also limited by its poor productivity and energy efficiency.

Today, ammonia production heavily relies on the Haber–Bosch process (HBP), one of the most important reactions in the 20th century, invented by Fritz Haber and Carl Bosch over one century ago. In practice, the HBP uses Fe-based catalysts and is performed at a pressure of 250–350 bar and a temperature of 450–550 °C. The main H₂ source for the HBP is steam reforming of CH₄ from natural gas, which results in ∼400 Mt CO₂ emitted for NH₃ production every year (∼1.2% of the global emissions per year). The HBP also consumes 1–3% of the annual global energy production.^12–15 Highly centralized HBP infrastructures require huge capital costs, and in most developing countries and remote areas their creation is hampered by insufficient financial resources and poor transportation.¹⁶ Hence, developing decentralized, eco-friendly, and sustainable alternatives for NH₃ production using renewable resources is a task of great significance (Fig. 1).

Fig. 1 Development of the nitrogen fixation process since the 20th century.

The electrochemical nitrogen reduction reaction (eNRR) process produces NH₃ from using water and N₂ (eqn (1)). This process has attracted considerable interest for many years in light of its great potential for distributed NH₃ production when powered by renewable energy (solar, wind, etc.).¹⁷

N₂ + 3H₂O → NH₃ + 3/2O₂ (1)

The US Department of Energy (DOE) has set specific performance targets for profitable NH₃ production from the eNRR, including a NH₃ productivity of 10⁻⁴ mol h⁻¹ cm⁻² at a faradaic efficiency (FE) of 90%.¹⁸ A large variety of electrocatalysts have been investigated to date (Table 1). Unfortunately, their eNRR performance remains low:^9,17,19,20 in particular, the FE has been limited to below 50% at NH₃ productivities of 10⁻⁶–10⁻⁴ mol h⁻¹ cm⁻².²¹ In addition, to address the challenge of activating N₂, catalysts must be designed to suppress the competing hydrogen evolution reaction (HER), which is likely to become kinetically favored over the eNRR at appropriate reaction potentials.

Table 1 Comparison of different eNRR demonstrations under mild operating conditions

Electrocatalyst	Electrolyte	Reactant	Potential (V versus RHE)	Production rate (μg cm^−2 h^−1)	Production rate (μg mgcat.^−1 h^−1)	FE (%)	Ref.
Au1/C3N4	H2SO4	N2	−0.1	—	1.9	11.1	24
AuSAs-NDPC	HCl	N2	−0.2	2.3	—	12.3	25
Boron@graphene	H2SO4	N2	−0.5	9.8	—	10.8	26
CSA/NPC	Na2SO4	N2	−0.2	14.6	—	10.5	27
NC–Cu SA	KOH	N2	−0.35	—	53.3	13.8	28
	HCl	N2	−0.3	—	49.3	11.7
FeMo@NG	LiClO4	N2	−0.4	—	14.9	∼10	29
			−0.2	—	3.4	41.7
ISAS-Fe/NC	Phosphate buffer solution	N2	−0.4	62.9 ± 2.7	62.9 ± 2.7	18.6 ± 0.8	23
FeSA–N–C	KOH	N2	0	—	7.48	56.55	30
Fe–N/C–CNTs	KOH	N2	−0.2	—	34.8	9.28	31
Fe–N–C	NH4Cl	N2	−0.05	0.9	—	4.51	32
SA-Mo/NPC	KOH	N2	−0.3	—	34.0 ± 3.6	14.6 ± 1.6	33
P-CNTs	LiClO4	N2	−1.1	—	24.4	∼0.9	34
Ru SAs/N–C	H2SO4	N2	−0.2	—	120.9	29.6	35
Ru@ZrO2/NC	HCl	N2	−0.11	—	1	21	36
Ru/NC	HCl	N2	−0.21	—	10.995	∼8.87
Ru SAs/g-C3N4	NaOH	N2	0.05	—	23	8.3	37
Y1/NC	HCl	N2	−0.1	23.2	—	12.1	38
Sc1/NC	HCl	N2	−0.1	20.4	—	11.2
Au1Co1@GO	K2SO4	N2	−0.2	—	36.82	22.03	39
CNS	LiClO4	N2	−1.19	97.18	—	11.56	40
MoS2/CC	Na2SO4	N2	−0.5	4.94496	—	1.17	41
MoN NA/CC	HCl	N2	−0.3	18.4212	—	1.15	42
VN NPs	Nafion	N2	−0.1	20.196	—	6	43
o-Fe2O3–Ar/CNT	KOH	N2	−0.9 V vs. Ag/AgCl	0.46	—	6.04	44
Fe/Fe3O4	Phosphate buffer solution	N2	−0.3	0.19	—	8.29	45
Au nanoparticle@N-doped porous carbon	HCl	N2	−0.2	36	—	22	46
AuHNCs	LiClO4	N2	−0.5	3.98	—	14.8	47
			−0.4	2.35	—	30.22
NiCoS/C	LiSO4	N2	0	2.6	—	12.9	48
Ag/CPE	HCl	N2	−0.6	2.82744	—	4.8	49
Au1Cu1/GCE	H2SO4	N2	−0.2	—	154.91	54.96	50
Ru@MXene	KOH	N2	−0.4	39.1	—	13.13	51
Ru2P–rGO	HCl	N2	−0.05	—	32.8	13.04	52
Nb2O5 nanofiber	HCl	N2	−0.55	—	43.6	9.26	53
B4C/CPE	HCl	N2	−0.75	—	26.57	15.95	54
Bi NS	Na2SO4	N2	−0.8	2.54 ± 0.16	—	10.46 ± 1.45	55
BNS/CP	Na2SO4	N2	−0.8	—	13.22	4.04	56
MnO/TM	Na2SO4	N2	−0.39	6.72	—	8.02	57
FL-BP NSs/CF	HCl	N2	−0.7	—	31.37	3.09	58
			−0.6	—	20.87	5.07
O-CNT/CP	LiClO4	N2	−0.4	—	32.33	12.5	59
BP/CP	HCl	N2	−0.6	—	26.42	12.7	60
OVs-MoO2	HCl	N2	−0.15	—	12.2	8.2	61

---

Shi et al. observed that Au sub-nanoparticles (sub-NPs) embedded in a Ti substrate showed the best eNRR performance at a size scale of 0.5 nm. In contrast, the yield and FE corresponding to nanoparticle sizes of 37 and 4 nm were very low, which highlighted the advantages associated with sub-nano-scale sizes and high dispersion of nanoparticles.²² Reducing the size of metal materials can increase the number of unsaturated active sites, broaden the distribution of electronic energy levels, and enhance the interaction between the metal and substrate, thus improving the catalytic activity of the metal atoms for specific reactions. It is reasonable to speculate that a more substantial improvement in electrocatalytic performance can be achieved by further reducing the size of the material, even reaching the atomic dispersion level. Downsizing electrocatalysts typically provides a means to enhance their activity toward sluggish processes, and the simpler structure is also helpful to understand the corresponding reaction mechanisms. Hence, single-atom catalysts (SACs) have emerged as promising candidates for overcoming the inert nature of N₂. SACs provide uncoordinated active sites and unique electronic structures that may facilitate the breakage of N≡N bonds. A record NH₃ productivity of 0.062 mg_NH3 cm⁻² h⁻¹and a faradaic efficiency of 18% at −0.4 V_RHE were achieved using a Fe SAC.²³ Moreover, owing to their simple and tunable structures and active sites, SACs can also enable a better understanding of the eNRR mechanism, as shown in Fig. 2.

Fig. 2 Schematic illustration of the eNRR with single-atom catalysts.

In this paper, we reviewed recent progress on SACs for electrocatalytic NH₃ synthesis, focusing on the following aspects: (1) synthetic strategies for single-atom eNRR catalysts; (2) latest advances in eNRR performance using SACs; (3) reaction mechanism of the SAC-catalyzed eNRR to NH₃. Finally, we discuss key challenges and future perspectives for developing more active and selective SACs toward NH₃ production from the eNRR.

2. Synthesis methods and characterization of SACs

2.1 SACs synthesis methods

The most critical challenge in the synthesis of SACs is their tendency to aggregate. Owing to the high surface free energy of isolated single atoms, their migration and aggregation into nanoclusters or nanoparticles are thermodynamically favorable in the absence of stable chemical coordination.⁶² In addition, the scale-up synthesis of SACs is also an essential aspect for their practical applications. Many synthetic strategies have been developed to date, including atomic layer and electrochemical depositions, as well as wet chemical and pyrolysis methods.⁶³

2.1.1 Atomic layer deposition method. Atomic layer deposition (ALD) is a pulse vapor deposition technology that relies on chemical precursors. This approach can precisely control the size, morphology, loading, and active site distribution of SACs by systematically varying the number of deposition cycles.⁶⁴ Typically, the SACs processed by ALD have loadings of less than 2 wt%.⁶⁵

For example, Sun and co-workers used a Pt precursor and O₂ to synthesize a Pt single-atom catalyst using ALD (Fig. 3a).⁶⁶ The oxygen-based functional groups on the surface of graphene nanosheets reacted with the MeCpPtMe₃ ligand to form a Pt monolayer. Then oxygen was exposed to the Pt surface to form a new layer of surface O⁻ species that reacted with MeCpPtMe₃ in the next cycle. Therefore, Pt deposition could be controlled precisely by adjusting the number of ALD cycles.

Fig. 3 (a) Schematic illustrations of the Pt ALD mechanism on graphene nanosheets. This figure has been adapted from ref. 66 with permission from Springer Nature, copyright 2013. (b) Schematics of thermally stable Pd₁ catalysts synthesized using ALD. This figure has been adapted from ref. 67 with permission from Royal Society of Chemistry, copyright 2016. (c) The graphene oxide solution in the presence of H₂O₂ and metal precursors was hydrothermally treated to form a 3D graphene hydrogel. This figure has been adapted from ref. 68 with permission from Springer Nature, copyright 2018. (d) Protocols for the synthesis of Ni/graphdiyne and Fe/graphdiyne. This figure has been adapted from ref. 70 with permission from Springer Nature, copyright 2018. (e) Electrochemical deposition mechanism. a and b Schematic of cathodic and anodic deposition of Ir species. This figure has been adapted from ref. 71 with permission from Springer Nature, copyright 2020. (f) Schematic illustration of the formation of Fe-ISAs/CN. This figure has been adapted from ref. 73 with permission from Wiley-VCH, copyright 2017. (g) Scheme of the transformation of nanoparticles to single atoms and structural characterization of Pd single atoms. This figure has been adapted from ref. 74 with permission from Springer Nature, copyright 2018.

Similarly, Lei and co-workers demonstrated an approach to stabilize single-atom Pd catalysts using ultrathin metal oxide protective coatings prepared by ALD (Fig. 3b).⁶⁷ Palladium hexafluoroacetylacetonate (Pd(hfac)₂) was first chemisorbed on the surface of Al₂O₃ using ALD. Then, ultrathin TiO₂ coatings were deposited onto the substrate by ALD. The TiO₂ layer selectively grew on the surface without Pd(hfac)₂ to form nanocavities. Finally, the Pd SACs were formed in the TiO₂ nanocavities after removing the hfac ligands using formalin (HCHO). High-precision instruments are required to precisely achieve ALD.

2.1.2 Wet chemical method. Wet chemical methods involve simple procedures and exhibit promising potential for large-scale SACs synthesis. Typically, they use two successive steps to uniformly anchor single metal atoms to their support. First, hydrothermal synthesis is used to attach metal atoms to the support or reduce metal ions to single atoms in the liquid phase. Freeze-drying methods and gas-phase reduction are then employed to obtain SACs.¹⁵

For instance, Fei and co-workers hydrothermally treated a mixture of graphene oxide, metal precursor, and H₂O₂ to obtain metal-doped graphene hydrogels as the precursors for SACs (Fig. 3c).⁶⁸ After being freeze-dried, the gel was annealed at 900 °C in an NH₃ atmosphere. Then, metal atoms were uniformly anchored on the nitrogen-doped graphene surface in an MN₄C₄ coordination configuration.

Similarly, Choi and co-workers prepared an atomically dispersed Pt catalyst on zeolite-templated carbon (ZTC), with a high Pt loading of ∼5 wt%.⁶⁹ S-doped ZTCs were synthesized by acetylene/H₂S chemical vapor deposition in a NaX zeolite template at 823 K. Subsequently, the Pt catalysts were impregnated in the ZTCs using H₂PtCl₆. The results revealed a positive correlation between the Pt loading and the S content of the ZTCs.

In most cases, the number of anchoring sites and the agglomeration tendency of a metal atom limit the metal loadings of SACs processed via wet chemical approaches to less than 1 wt%.⁶⁵ Therefore, low-concentration metal precursors and high-surface area supports are desirable for the application of wet chemical methods.

2.1.3 Electrochemical deposition method. Electrodeposition approaches can tune the mass loading, coordination, and active site density of SACs by regulating the metal precursor concentration and deposition time. Typically, reducing the diffusion rate of metal ions leads to the growth of single-atom structures on the substrate.⁶⁵

For example, Xue and co-workers prepared Ni/graphdiyne and Fe/graphdiyne SACs through a two-step strategy (Fig. 3d). In the first step, a three-dimensional (3D) graphdiyne foam (GDF) was fabricated on carbon cloth (CC) surfaces via an acetylenic cross-coupling reaction using hexaethynylbenzene (HEB) as the precursor. Then, Ni/Fe atoms were anchored on the GDF by a facile electrochemical reduction method.⁷⁰

Moreover, Zhang and co-workers proposed a general electrochemical deposition method to synthesize more than 30 different kinds of single-atom catalysts (Fig. 3e). They varied the types of metal precursors and supports as well as the electrodeposition reactions that occurred at the anode or cathode to obtain different single-atom structures.⁷¹ For example, the authors mixed IrCl₄ with a KOH aqueous solution to obtain the metal precursor. Ir SACs could then be synthesized through both cathodic and anodic deposition. At the cathode, the Ir SACs were formed via the deposition of IrCl³⁺; at the anode, Ir(OH)₆²⁻ was the precursor of the Ir SACs formed on Co(OH)₂ nanosheets. In brief, the deposition rate of the metal precursor can be adjusted by regulating the electrochemical current.

2.1.4 Pyrolysis method. Pyrolysis is commonly used to treat metal–organic, covalent organic, or zeolitic imidazolate frameworks (MOFs, COFs, and ZIFs, respectively) and obtain single-atom catalysts supported on heteroatom-doped porous carbon materials.⁷² In a typical process, metal-containing precursors are first prepared using a spatial confinement strategy with the assistance of porous materials (such as zeolites, MOFs and COFs) or organic/inorganic protective layers (such as surfactants, amorphous carbon, and silica). The subsequent pyrolysis removes ligands and other stabilizers from the precursor to form isolated metal atoms stabilized by the support. Pyrolysis can generate atomically dispersed metal sites and doped heteroatoms (such as N and S) on the catalyst support, providing high structural stability and metal loadings.

For example, Chen and co-workers impregnated iron acetylacetone (Fe(acac)₃) in the zeolitic imidazolate framework ZIF-8 (Fig. 3f).⁷³ High-temperature pyrolysis transformed this composite into a Fe SAC supported by porous N-doped carbon (Fe-ISAs/NC). The Fe loading reached 2.16 wt%.

Similarly, Wei and co-workers reported the conversion of noble metal nanoparticles into thermally stable and highly active SACs based on Pd (Fig. 3g).⁷⁴ They grew ZIF-8 nanocrystals on Pd NPs by mixing the Pd NPs, a Zn(NO₃)₂ solution, and a 2-methylimidazole solution. The Pd-NPs@ZIF-8 nanocomposites were then annealed at 900 °C under an inert atmosphere for 3 h. The Pd SACs were formed on the nitrogen-doped carbon support after annealing, as confirmed by high-resolution high-angle annular dark field-scanning transmission electron microscopy (HAADF-STEM), extended X-ray absorption fine structure (EXAFS), and X-ray absorption near-edge structure (XANES) measurement.

In addition, taking advantage of the pore confinement effect of MOFs, Zhong et al. proposed a MOF precursor-based co-pyrolysis approach for the synthesis of SACs.^75,76 After mixing the precursors for fabricating Co-MOF, different amounts of prepared Co-MOF and urea were employed for co-pyrolysis. Finally, a series of metal atomically dispersed Co–C₃N₄ SACs with different metal contents were obtained. Benefiting from the confinement function of the MOF structure and the close interaction between the MOF and graphitic C₃N₄ (g-C₃N₄), Co single atoms could be uniformly dispersed on the substrate during g-C₃N₄ formation. Therefore, the introduction of porous framework materials as precursors promotes the subsequent atomic dispersion.

The general characteristics of the different SAC synthesis methods are summarized in Table 2. ALD can achieve precise control through accurate and expensive equipment. Wet chemical methods involve simple procedures and equipment, but usually require repeated operation. Electrochemical deposition can control the deposition rate by changing the current. Pyrolysis can achieve high structural stability and metal loadings at the expense of a higher-temperature atmosphere. Therefore, improving the different synthetic methods according to their respective characteristics is a task of great importance to avoid atom aggregation and increase the yield of SACs.

Table 2 Comparison of SACs synthesis methods

Synthesis methods	Advantages
Atomic layer deposition	Precise control of the SACs synthesis process
Wet chemical method	Simple operation; potential for large-scale SACs synthesis
Electrochemical deposition	The deposition rate can be controlled by the current; simple and flexible equipment
Pyrolysis	High structural stability and metal loading

---

2.2 Characterization techniques for SACs

Accurate and suitable characterization techniques are crucial for developing SACs, owing to their unique structure and specific coordination environment. Based on an efficient characterization, the geometric, electronic and coordination structures of SACs can be correlated with their catalytic performance, providing an in-depth understanding of the relevant electrochemical reactions.

Scanning electron microscopy (SEM) and transmission electron microscopy (TEM) are used to reveal the morphology of the catalyst, i.e., its geometric structure (Fig. 4a). HAADF-STEM with atomic resolution (Fig. 4a) can further qualitatively distinguish single-atom catalysts and nanoparticles.^35,77 The HAADF-STEM images and the corresponding energy-dispersive X-ray spectroscopy (EDS) elemental mappings can be used to examine the homogeneity of the element distribution in a selected area of a SAC. For instance, Geng Z. et al. used HAADF-STEM and EDS to reveal the uniform distribution of Ru on a carbon substrate.³⁵ Using the same techniques, both Wang et al. and Zhang et al. confirmed the presence of isolated Fe sites anchored on an N-doped carbon support for a single-atom Fe catalyst (Fig. 4a).^30,32

Fig. 4 (a) Typical SEM, TEM, and HAADF-STEM images, and corresponding EDS elemental mapping results. This figure has been adapted from ref. 35 with permission from Wiley-VCH, copyright 2018. This figure has been adapted from ref. 77 with permission from the Royal Society of Chemistry, copyright 2019. (b) XRD patterns of Ru SAs/N–C and Ru NPs/N–C. This figure has been adapted from ref. 35 with permission from Wiley-VCH, copyright 2018. (c) XPS spectra of the N 1s and Ru 3p_3/2 for Ru SAs/N–C and Ru NPs/N–C. This figure has been adapted from ref. 35 with permission from Wiley-VCH, copyright 2018. (d) Ru K-edge XANES spectra and EXAFS spectra of Ru SAs/N–C, Ru NPs/N–C, RuO₂, Ru(acac)₃, and Ru foil. This figure has been adapted from ref. 35 with permission from Wiley-VCH, copyright 2018. (e) In situ XRD intensity map, and (f) in situ Raman spectra of the transformation from COF/NC to Eex-COF/NC. This figure has been adapted from ref. 79 with permission from Springer Nature, copyright 2019.

X-ray diffraction (XRD) allows analysis of the crystal structure of a catalyst. In a typical SAC, the diffraction peaks of the corresponding bulk metal, metal oxides, and other crystalline compositions are generally absent. Fig. 4b shows the (002) and (101) facets of graphitic carbon in N-doped-carbon-supported Ru SACs (Ru SAs/N–C) and Ru nanoparticles (Ru NPs/N–C).³⁵ The absence of the XRD peaks of Ru crystals' for Ru SAs/N–C indicates that the Ru atoms in the catalysts are likely to be single sites.³⁵

X-ray photoelectron spectroscopy (XPS) can reveal the chemical composition and valence states of SACs. For example, the XPS profiles of the N 1s and Ru 3p_3/2 levels for Ru SAs/N–C and Ru NPs/N–C confirmed that the derivative of ZIF-8 was transformed into N-doped carbon, and the oxidation state of Ru in Ru SAs/N–C was approximately +3 (Fig. 4c).³⁵

X-ray absorption spectroscopy (XAS) is used to examine structures whose X-ray absorption coefficients vary with the energy of X-ray. Upon absorbing X-ray energy, the core electrons of the samples are excited into vacant orbitals, providing information about the XANES. When core electrons are excited into a continuous bond and form superposition of waves with the surrounding atoms, the corresponding spectra are labeled EXAFS. XANES spectra can be used to analyze chemical valence states. Fourier-transform EXAFS spectroscopy is a powerful tool to inspect the coordination structures of atoms in a catalyst.⁷⁸Fig. 4d shows the Ru K-edge XANES profile of Ru SAs/N–C, whose energy absorption edge matched well with that of Ru(acac)₃.³⁵ The valence state of Ru in Ru SAs/N–C was approximately +3, whereas the Ru species in Ru NPs/N–C had an oxidation state of 0. The EXAFS spectrum of Ru SAs/N–C contained only one peak corresponding to Ru–N bonds. The absence of Ru–Ru metallic bonds in the spectrum confirmed that Ru atoms in the Ru SAs/N–C structure were fully isolated.

The above techniques have been widely used to characterize the morphology and structure of SACs. Under electrochemical reaction conditions, the catalysts may undergo a series of complex dynamic evolutions and reconstructions. Hence, in situ/operando techniques are of great importance for investigating dynamic changes and understanding mechanisms in relevant catalytic processes. Given that SACs have well-defined active site structures, they are ideal platforms for mechanistic studies, particularly of the eNRR, which involves a complex reaction pathway with multiple electrons and protons transferred. In situ/operando characterization techniques can provide insight into the active centers and critical intermediates under working conditions. This information is crucial for devising highly selective and active SACs. For instance, Liu and co-workers studied a N-doped-carbon-supported COF (COF/NC) catalyst for the eNRR using in situ XRD and Raman spectroscopy (Fig. 4e and f).⁷⁹ The in situ XRD results revealed the amorphization of the COF structure during the eNRR process (Fig. 4e). In contrast, under Ar-saturated conditions, the crystallinity of the COF/NC catalyst was almost unchanged. These results indicated that the transition of COF/NC from its crystalline to the amorphous state was induced by the eNRR process. The authors also performed in situ Raman tests to observe the structural changes taking place in the COF/NC catalyst under eNRR conditions (Fig. 4f). They observed the gradual appearance of the stretching vibrational band at approximately 800 cm⁻¹, while no such change occurred in an Ar atmosphere. These results indicated that the lattice distortion of COF/NC originated from the chemical adsorption of nitrogen or nitrogen-containing intermediates on the catalyst surface.

3. Classification of various SACs for the eNRR

SACs consist of isolated single atoms and have unique electronic structures, which improve their electrocatalytic activity toward various reactions. The use of SACs as eNRR catalysts could inhibit the HER, enhance the adsorption of nitrogen and intermediate reactants, and change the rate-limiting step, thus improving the selectivity and activity in the eNRR. The existing SACs for the eNRR can be classified into noble metal, non-noble metal and non-metal SACs.

3.1 Noble metal SACs

To date, Ru-, Au-, and Pt-based SACs have been utilized as electrocatalysts for the eNRR. Ru SACs are intriguing because of their high activity as Ru-based counterparts in the Haber–Bosch process.^80,81 Geng and co-workers synthesized Ru SAs/NC (Ru mass loading 0.18%) by pyrolyzing a Ru-containing derivative of ZIF-8 (Fig. 5a). The Ru SAs/N–C exhibited a FE of 29.6% for NH₃ production by the eNRR, and the yield rate reached 120.9 μg_NH3 h⁻¹ mg_cat⁻¹ at −0.2 V vs. reversible hydrogen electrode (RHE).³⁵ (Fig. 5b). The HAADF-STEM images, Ru K-edge XANES, and EXAFS spectra of Ru SAs/N–C of Ru SAs/N–C revealed that the Ru atoms in Ru SAs/N–C were completely isolated and coordinated by N atoms with a coordination number of approximately 3.4. Compared to Ru NPs/N–C, the Ru SAs/N–C catalyst exhibited a stronger N₂ chemisorption. The authors concluded that the enhanced N₂ adsorption on the surface of Ru SAs/N–C led to a lower Gibbs free energy (ΔG) for the rate-determining step (RDS) of the eNRR than that of Ru(101) (Fig. 5c). As a result, Ru SAs/N–C exhibited excellent electrocatalytic performance for the eNRR.

Fig. 5 (a) Scheme of the synthetic procedure for Ru SAs/N–C. This figure has been adapted from ref. 35 with permission from Wiley-VCH, copyright 2018. (b) Yield rate of NH₃ production at different applied potentials on Ru SAs/N–C and Ru NPs/N–C. (c) Free energy diagram of the N₂ electrochemical reduction with a distal pathway on Ru₁–N₃, Ru₁–N₄, and Ru (101). * represents an adsorption site. This figure has been adapted from ref. 35 with permission from Wiley-VCH, copyright 2018.

Suppressing the H₂ evolution to reduce the fraction of electrons participating in this competing reaction is regarded as an effective approach to enhance the selectivity toward NH₃. For this purpose, Sun and co-workers annealed Ru-incorporated UiO-66 to produce Ru@ZrO₂/NC, which achieved a remarkable NH₃ FE of 21% at an overpotential of 0.17 V (ref. 36) (Fig. 6a). The authors claimed that ZrO₂ can suppress the HER and enhance the eNRR selectivity toward NH₃. Density functional theory (DFT) calculations showed that Ru@Zr₃₂O₆₃ with an O vacancy and Ru/NC₂ had a lower energy barrier for the potential-determining step than the Ru (0001) metal substrate (Fig. 6b). The authors concluded that adding the unsaturated –NH₂ groups during pyrolysis avoided atomic aggregation and improved the catalytic activity of Ru and the NH₃ yield rate, and suggested that the presence of ZrO₂ reduced the activity for the HER.

Fig. 6 (a) The FEs of NH₃ over Ru@NC, Ru@C, Ru@ZrO₂/NC, and Ru@ZrO₂/C at various applied potentials. (b) ΔG_PDS for the eNRR on various reaction sites. These figures have been adapted from ref. 36 with permission from Elsevier, copyright 2018.

Au-Based SACs have also been widely studied because of their low theoretical affinity for H, which is expected to reduce the HER activity and increase the NH₃ faradaic efficiency.^22,46,82 Wang and co-workers prepared Au atoms on carbon nitride(Au₁/C₃N₄) and Au nanoparticles supported on C₃N₄ (Au NPs/C₃N₄) for the eNRR²⁴ (Fig. 7a). According to DFT calculations, the electrons of the Au atoms migrated to g-C₃N₄ and shifted its d-orbital position toward the Fermi level, leading to a lower ΔG for the rate-determining step (N₂ to *NNH) of Au₁/C₃N₄ than Au(211) (Fig. 7b). As a result, Au₁/C₃N₄ had an NH₃ FE of 11.1% and a NH₄⁺ yield rate of 1305 μg h⁻¹ mg_Au⁻¹ at −0.1 V vs. RHE; the yield was 22.5 times as high as that of Au nanoparticles (Fig. 7c). Thus, Au₁/C₃N₄ had a better eNRR electrocatalytic activity and selectivity than Au NPs/C₃N₄ and Au(211).

Fig. 7 (a) Schematic of an electrolyzer for the N₂ electroreduction test. (b) Free energy profile of the eNRR on Au₁/C₃N₄ and Au (211) at zero potential following the alternating mechanism. (c) NH₄⁺ yield rates normalized by Au mass. These figures have been adapted from ref. 24 with permission from Elsevier, copyright 2018.

Achieving the optimal coordination between the substrate and single atoms also has a great impact on the performance of SACs. Qin and co-workers reported an electrocatalyst consisting of Au single sites supported on N-doped carbons (AuSAs-NDPCs), with outstanding activity and stability for the eNRR.²⁵ Compared with AuSAs-NDPCs, the Au atoms showed obvious agglomeration in nitrogen-free salt-templated carbons (STCs), which resulted in a weaker eNRR performance (Fig. 8a). The AuSAs-NDPCs catalyst showed an excellent NH₃ yield of 2.32 μg h⁻¹ cm⁻², with a maximum FE of 12.3% at −0.2 V vs. RHE, which was attributed to the interaction of Au with N and C atoms that stabilized the single sites, and to the highly open porous structure of the NDPCs, which facilitated the access to active sites and mass transport (Fig. 8b).

Fig. 8 (a) Faradaic efficiency of NH₃ with different catalysts at −0.2 V versuss RHE. (b) Faradaic efficiency and yield rate of NH₃ at different potentials. These figures have been adapted from ref. 25 with permission from Wiley-VCH, copyright 2018.

Although Pt is a highly effective catalytic component in many reactions, it also shows a higher affinity for H than N in the electrolyte; hence, its surface tends to be covered by H, resulting in poor electrocatalytic activity for ammonia synthesis in an aqueous environment.⁸³ However, the first-principles calculation of Yin and co-workers revealed that Pt-embedded monolayer graphitic carbon nitride (Pt/g-C₃N₄) showed significant deviations from the scaling relationship between the energetics of the *N₂H and *NH₂ intermediates⁸⁴ (Fig. 9a). In particular, Pt/g-C₃N₄ exhibited stronger and weaker interactions with *N₂H and *NH₂ species, respectively, compared with TM₁/g-C₃N₄ materials (TM₁ = Sc ∼ Cu, Mo, Ru, Rh, Pd, Re, Ir, and Pt); this identified Pt/g-C₃N₄ as the most suitable electrocatalyst for the eNRR. The tight anchoring of Pt atoms on g-C₃N₄ prevented diffusion and aggregation, resulting in an outstanding stability. Compared with pristine g-C₃N₄, the Pt–N orbital hybridization of Pt/g-C₃N₄ reduced the band gap, thus enhancing the electronic conductivity (Fig. 9b). Moreover, the strong hydrophobicity of Pt/g-C₃N₄ balanced the fact that Pt tends to adsorb H instead of N, effectively suppressing the HER. Hence, Pt/g-C₃N₄ showed remarkable activity with a low limiting potential of −0.24 V (Fig. 9c). Therefore, Pt/g-C₃N₄ was selected as the most suitable electrocatalyst for ammonia synthesis in the TM₁/g-C₃N₄ series.

Fig. 9 (a) Adsorption energies of N₂H and NH₂ as a function of atomic N adsorption strength on TM/g-C₃N₄. (b) Density of states for Pt/g-C₃N₄. (c) Free energy profiles for the eNRR on Pt/g-C₃N₄ at 0 V (black) and UL (red) via the alternating mechanism. These figures have been adapted from ref. 84 with permission from the Royal Society of Chemistry, copyright 2019.

3.2 Non-noble metal SACs

The high cost and scarcity of noble metal-based catalysts limit their wide application. In addition to their cost-effectiveness, non-noble metal-based SACs also have great potential as efficient catalysts for the eNRR. Transition metal (TM)-based SACs exhibit outstanding ability as eNRR electrocatalysts, because the TM contributes electrons to the antibonding π* orbital of N₂, while N₂ transfers σ electrons to the vacant d orbital of the TM, enhancing the N₂ adsorption onto the TM active sites. Therefore, transition metals with both vacant and occupied d orbitals are promising electrocatalysts for the synthesis of ammonia. Mo and Fe are important components of biological nitrogen fixation reactions, and Fe also plays an important role in the Haber–Bosch process. Moreover, Co-based catalysts are very active in the electrocatalytic reduction of O₂ and CO₂. These ideas, combined with DFT calculations, highlight the promising potential of non-noble metal-based SACs as eNRR electrocatalysts.

In most biological nitrogenase systems, the Mo-based structure is the active center; this provides a possible route for the design of single-atom catalysts for ammonia synthesis. Luo and co-workers manufactured a catalyst based on single Mo atoms anchored on N-doped porous carbon (SA-Mo/NPC).³³ XPS measurements showed that the valence state of Mo in SA-Mo/NPC consisted of 74% Mo⁶⁺ and 26% Mo⁴⁺. The catalyst showed a high NH₃ yield rate of 34.0 mg_NH3 h⁻¹mg_cat.⁻¹ and a large FE of 14.6%, because the 3D porous structure of NPC can facilitate access to active sites and improve mass transport (Fig. 10a). Inductively coupled plasma-atomic emission spectroscopy (ICP-AES) measurements showed that a Mo loading of 9.54 wt% in SA-Mo/NPC achieved the maximum ammonia production (Fig. 10b). A too low Mo content would lead to limited improvements, while an excessively high content would generate nanoclusters and reduce the number of Mo–N bonds, with a similar weakening effect on the ammonia production rate. Interestingly, Mo–N bonds were considered to be active for the eNRR process.

Fig. 10 eNRR electrochemical performances of SA-Mo/NPC in 0.1m KOH. (a) NH₃ yield rate (red) and FE (blue) at each given potential. (b) NH₃ yield rates of SA-Mo/NPC samples with different Mo loadings. These figures have been adapted from ref. 33 with permission from Wiley-VCH, copyright 2019.

Similarly, because Fe is the active center of the Haber process and biological nitrogenase, Fe-based electrocatalysts are likely to achieve efficient synthesis of ammonia. However, as the theoretical potential of the hydrogen evolution reaction is close to zero, controlling the reaction potential of nitrogen reduction in a reasonable range can effectively inhibit the HER and improve the FE. Wang and co-workers investigated a catalyst consisting of single iron atoms dispersed on N-doped carbon (Fe_SA–N–C).³⁰ Molecular dynamics simulations revealed that the energy barrier for N₂ to access the single iron atom was only 2.38 kJ mol⁻¹, and DFT calculations showed that the binding Gibbs free energy of *N₂ (−0.28 eV) was much lower than that of *H (2.91 eV), suggesting that the inhibition of the HER by Fe_SA–N–C was significantly stronger than that achieved by common transition and noble metals (Fig. 11a). The released energy could be supplied to the first hydrogenation step, successfully overcoming the eNRR reaction bottleneck. Thus, the Fe_SA–N–C catalyst achieved a substantial FE of 56.55% and a desirable yield rate (7.48 μg h⁻¹ mg⁻¹) at 0 V vs. RHE, with an ultralow onset potential of 0.193 V (Fig. 11b).

Fig. 11 (a) Calculated energy barriers of the adsorption of hydrogen and nitrogen. (b) NH₃ faradaic efficiencies and mass-normalized yield rates at each given potential of Fe_SA–N–C. Copyright 2019, Springer Nature. These figures have been adapted from ref. 30 with permission from Springer Nature, copyright 2019.

Generally, Co-based catalysts show good activity in the O₂ and CO₂ reduction reactions,^85,86 in which the Co–N coordination increases the magnetic moment of the Co atom. This structure plays an important role in activating the reactants and leads to Co-based catalysts having a higher catalytic activity than traditional noble metal catalysts. Sardroodi and co-workers performed first-principles DFT calculations to explore the stability and eNRR mechanism of single Co atoms supported on defective N-doped graphene (Co/N₃–Gr).⁸⁷ By analyzing the adsorption energy and the formation energy of Co/N₃–Gr, they concluded that the Co atom was strongly bound to the neighboring N atoms. N₂ molecules were most likely to attach to the Co sites in the end-on configuration, owing to its stronger adsorption energy and electron transfer (Fig. 12a). The Co/N₃–Gr system is expected to catalyze the eNRR by an enzymatic rather than a distal or alternating mechanism in the water phase, owing to the lower overpotential (Fig. 12b). Compared to the eNRR, Co/N₃–Gr required a higher overpotential for the HER (Fig. 12c). Therefore, owing to the high-spin polarization of the Co atom, the non-noble metal Co/N₃–Gr catalyst may be an excellent candidate for the eNRR.

Fig. 12 (a) Optimized structure (above) and PDOS plots (below) of N₂ adsorption on Co/N₃–Gr for end-on configurations. (b) Reaction energy (left) and Gibbs free energy (right) diagrams for the N₂ reduction reaction on Co/N₃–Gr through enzymatic mechanisms at zero and applied potential (limiting potential). (c) The energy diagram and relaxed geometries of species involved in the hydrogen evaluation reaction over Co/N₃–Gr. These figures have been adapted from ref. 87 with permission from Wiley-VCH, copyright 2019.

3.3 Non-metal SACs

Non-metal SACs exhibit weak adsorption of H⁺, because the non-metal atoms coordinate carbon atoms and show a high positive charge, forming Lewis acid sites, while the carbon atoms possess a negative charge density. Hence, the non-metal atoms tend to bind the weak Lewis base N₂ and promote the eNRR. Moreover, non-metal atoms and the surrounding coordinated species can act as electron acceptors and donors upon tuning the eNRR activity through structural distortion. For instance, Wen et al. achieved a considerable eNRR performance based on boron and sulphur co-doped carbon nanofibers.⁸⁸ Therefore, exploring the catalytic performance of non-metal SACs is an important task.

The boron-doped graphene skeleton can retain the original sp² hybridization and conjugated plane structure; the lower electronegativity of the B (2.04) than the C (2.55) atoms results in a change in the electron density of the carbon ring structure. Hence, the electron-deficient B site enhances the capture of N₂ and inhibits the adsorption of H⁺. Yu and co-workers experimentally found that the ammonia yield of B-doped graphene at −0.5 V vs. RHE was 9.8 μg h⁻¹ cm⁻², with a FE of 10.8%.²⁶ Moreover, DFT calculations showed that, among different B-doped carbon structures, BC₃ had the lowest energy barrier for electrocatalytic ammonia synthesis. Liu and co-workers constructed 21 models with a single boron atom decorated on eight popular two-dimensional (2D) materials, involving different boron states⁸⁹ (Fig. 13a). Unlike the d orbitals of transition metals, the boron atom could accept the lone-pair electrons of N₂ in the sp³ hybrid orbital, forming B-to-N π-back bonding. To identify promising eNRR catalysts, the authors calculated the binding energy of a single N in each state and the maximum energy required for the HER and eNRR. Single boron atoms supported on graphene and substituted into h-MoS₂ exhibited high selectivity and energy efficiency with relatively low energy barriers of 0.31 and 0.46 eV in the distal pathway (Fig. 13b).

Fig. 13 (a) Proposed 2D materials and potential B-sites. The numbers indicate different bonding environments described in the text. Black, blue, rose, yellow, purple, and cyan spheres represent C, N. B, S, P, and Mo, respectively. (b) Computational screening of 14 catalyst combinations: ΔG^HER_maxvs. ΔG^NRR_max. These figures have been adapted from ref. 89 with permission from American Chemical Society, copyright 2019.

Likewise, when a high electropositive P single atom is loaded on a carbon substrate, it coordinates electronegative C atoms. As a result, the P atom becomes a Lewis acid site and the active center for N₂ adsorption. Hu and co-workers developed an efficient metal-free electrocatalyst for the eNRR consisting of phosphorus-doped carbon nanotubes (P-CNTs).³⁴ The Raman spectrum showed that the P-CNTs had more defects on their surface than the pristine CNTs. Moreover, the pristine CNTs and a bare carbon cloth exhibited a negligible NH₃ yield compared to that of the P-CNTs, which was up to 24.4 μg h⁻¹ mg⁻¹_cat. (Fig. 14a). The P-dopant loading and the NH₃ yield showed an almost linear relationship, which revealed that the phosphorus centers were the real active sites for the eNRR. The Mulliken charge density distribution showed that the phosphorus atoms of P-CNTs formed Lewis acid sites to combine with N₂ (Fig. 14b). Compared to P-doped porous carbon nanosheets, the highly conjugated π network of P-CNTs also enhanced the activity for the eNRR. According to in situ FTIR spectroscopy measurements and DFT calculations, the eNRR in P-CNTs proceeded by the distal mechanism rather than direct N₂ dissociation, reflecting the reaction pathway of metal-free catalysts (Fig. 14c).

Fig. 14 (a) NH₃ yields of P-CNTs in N₂/Ar-saturated electrolyte at −1.1 V and in N₂-saturated electrolyte at open circuit potential (OCP), and NH₃ yields of P-CNTs, CNTs, and carbon cloth in N₂-saturated electrolyte at −1.1 V. (b) Mulliken charge distribution of the P-CNTs model. (c) Free energy profiles for distal and alternating pathways on the phosphorus site in P-CNTs. These figures have been adapted from ref. 34 with permission from Springer Nature, copyright 2020.

3.4 Effect of metal content of SACs

It is well known that the full exposure of active sites is crucial for the performance of catalysts. Compared with metal nanoparticles, SACs can maximize the exposure of the active atomic sites, which indicated the increase of active sites. Therefore, the metal content plays a very significant role in these systems. Jiao et al. obtained iron SAC-implanted N-doped porous carbon (Fe_SA–N–C) with 1.76 wt% metal content, which exhibited an excellent kinetic current density (J_k) at 0.85 V (23.27 mA cm⁻²), exceeding that of the commonly used Pt/C catalyst for the oxygen reduction reaction.⁹⁰ For further improvement, they introduced SiO₂ in the synthesis process of the MOF, synergistically inhibiting the agglomeration of metal species.⁹¹ As a result, Fe_SA–N–C achieved a high Fe loading of 3.46 wt%. The optimized SACs exhibited an even more higher J_k of 37.19 mA cm⁻², indicating an apparent performance promotion. Therefore, increasing the metal content can significantly improve the activity of SACs. However, Li and co-workers proposed a Co-based SAC on nitrogen-doped graphene (SACo@NG) with a high Co content of 4.1 wt%, which showed >90% conversion of benzyl alcohol (BzOH); this catalyst was much more efficient than Co NPs loaded on nitrogen-doped graphene (6.7 wt%).⁹² It can be inferred that, despite the increased metal loading, the number of atoms actually participating in the reaction may be lower, because the metal atoms in the nanoparticles are not fully exposed. Similarly, Gong et al. observed that the photocatalytic activity of CO₂ reduction exhibited a volcano-type trend.⁷⁵ When the amount of Co-MOF used for preparing Co SACs was 25 mg, the yield of CO reached the peak value of 394.4 μmol h⁻¹ g⁻¹, which was notably higher than that obtained with 10 and 35 mg Co-MOF; this might be ascribed to the aggregation of Co active sites. Consequently, increasing and controlling the metal loading of SACs within a reasonable range can significantly contribute to improving the activity of the catalysts. Although only a few studies have focused on the influence of the metal content in the eNRR, we envisage that these considerations will inspire further investigation of this subject.

4. Mechanisms of the eNRR

The mechanisms of the eNRR typically involve (i) N₂ adsorption, (ii) multiple electron and proton transfer steps to the intermediates, and (iii) NH₃ desorption. According to detailed mechanistic analyses, the eNRR mechanisms can be classified into two pathways, i.e., dissociative or associative (Fig. 15).⁹³ The Haber–Bosch process follows the dissociative pathway, in which the complete cleavage of the N≡N bond occurs before the hydrogenation of N and an extremely high energy input is required.¹⁵

Fig. 15 Reaction pathways for the eNRR by proton-coupled electron transfer, including the dissociative pathway, distal and alternative associative pathway, and enzymatic pathway. Copyright 2019, the Royal Society of Chemistry. This figure has been adapted from ref. 93 with permission from the Royal Society of Chemistry, copyright 2019.

Under ambient conditions, the eNRR is more likely to proceed via the associative pathway, which includes “distal”, “alternating”, and “enzymatic” mechanisms. Along this pathway, the breaking of the N≡N bond and the hydrogenation occur simultaneously, and the N≡N bond is not completely broken until the first NH₃ molecule is released. N₂ can bind to the catalyst surface in two possible configurations: end-on and side-on. In the alternating and distal pathways, the N₂ molecule is adsorbed on the surface in the end-on configuration, which is a linear structure with only one N atom bound to the surface. The main difference between the distal and alternative pathways concerns the NH₃ release mechanism. In the distal path, the N atom not bound to the surface is preferentially hydrogenated to NH₃. In contrast, both nitrogen atoms of N₂ have the same probability to undergo hydrogenation in the enzymatic pathway. In the enzymatic pathway, N₂ is adsorbed in side-on mode, namely, with the two sides of the molecules simultaneously binding the catalyst, and the hydrogenation of the two N atoms takes place in sequence.^94,95

For different SACs, the N₂ adsorption configurations, potential-limiting steps, and corresponding reaction and activation energies for the eNRR are typically determined by examining the active site structures. For example, Wang and co-workers explored the catalytic behavior of the single-atom catalyst M₁–N₁C₂ (M = Cu, Pd, Pt, and Mo) via first-principles calculations.⁹⁶ Compared to other metals, Mo single atoms coordinated with N₁C₂ possessed a stronger N₂ adsorption in both end-on (−1.18 eV) and side-on (−1.19 eV) modes. Mo₁–N₁C₂ tended to adopt the enzymatic pathway; the overpotential was as low as 0.24 V, and the potential-limiting step was the *N₂ reduction to *N₂H. Compared to the enzymatic mechanism, the potential-limiting steps of the alternating and distal pathways were found to be the *NH–NH₂ to *NH₂–NH₂ and *NH₂–NH₂ to *N₂ reduction, with corresponding the overpotentials of 0.62 and 0.57 V, respectively. These results suggested that the eNRR performance of Mo₁–N₁C₂ was superior to that of other M₁–N₁C₂ structures. Moreover, the ΔG for NH₃ desorption from Mo₁–N₁C₂ was 0.47 eV, indicating that in this case, NH₃ can be rapidly released from the surface, which is also beneficial to the overall eNRR.

5. Rational design of SACs for the eNRR based on the reaction mechanism

The N₂ activation on the catalyst surface is the most crucial step of the eNRR. To activate N₂, the catalyst must provide an empty d orbital to receive the lone-pair electrons from N₂. The unique structure of the SACs may create more empty d orbitals in the metal centers, owing to the strong electronegativity of the ligands and support in the catalysts. In addition, the densely-decorated structure of the substrate can substantially enhance the electrical performances. For example, Li et al. achieved a well-designed single crystal ReS₂ device with prominent electrical and photoelectrical performances.⁹⁷ Therefore, this provides significant guidance for rational design of catalysts.

For instance, the electronic structure of the Mo atom is 4d⁵5s¹, whose sp³d² hybridization leads to the absence of empty orbitals. Ou and co-workers established a model of single Mo atoms on the N-doped black phosphorus substrate Mo₁N_iP_3−i (i = 0, 1, 2, 3) by DFT calculations⁹⁸ (Fig. 16a). They found that a higher number of N atoms led to a lower overpotential for the eNRR; thus, Mo₁N₃ required the lowest overpotential (0.02 V). This is due to the strong electronegativity of the coordinated N atoms, whose role is to receive electrons from the Mo atomic orbitals and accept lone-pair electrons from N₂ in an empty d orbital. The authors also explored the charge transfer through the distal mechanism based on Bader charge analysis. In the first three steps, the electron transfer only occurred between Mo₁N₃ and the intermediate reactant N_xH_y, while black phosphorus actively participated in the electron transfer in the last four steps to promote the formation and desorption of the second NH₃ molecule (Fig. 16b).

Fig. 16 (a) Top and side views of BP without Mo₁N₃, the Mo₁N₃ center, and the adsorbed *N_xH_y species, of the *N–N intermediate. This figure has been adapted from ref. 98 with permission from the Royal Society of Chemistry, copyright 2019. (b) Charge variation of the three moieties and distal mechanisms on Mo₁N₃ at zero and applied potentials. This figure has been adapted from ref. 98 with permission from the Royal Society of Chemistry, copyright 2019. (c) Optimized structure of FeN₃-graphene and FeN₄-graphene, and their isosurface of the spin-resolved density pictures. This figure has been adapted from ref. 94 with permission from American Chemical Society, copyright 2016.

Li and co-workers further compared and explored the eNRR performance of FeN₄- and FeN₃-embedded graphene⁹⁴ (Fig. 16c). The dipole moment of the FeN₄ center was found to be split into upper and lower components by the interaction with graphene; the smaller dipole moment of this center makes it difficult for it to polarize N₂. In FeN₃-graphene, N doping in a single vacancy resulted in the Fe dipole moment increasing from 0 to 3.16 μ_B. The larger dipole moment resulted in an electron cloud distribution more concentrated near the Fe atom, which was exposed in an apical position on top of the graphene layer, explaining the superiority of FeN₃-graphene for N₂ adsorption. Moreover, after the N₂ adsorption, the electron cloud was uniformly distributed on N₂ and FeN₃, confirming the activation of N₂.

The Sabatier principle, first proposed in 1920, states that the interaction between the catalyst and the reactive species in catalytic reactions should be neither too strong nor too weak: if the interaction is too strong, the reaction product is difficult to desorb, whereas an overly weak interaction prevents the reactants from combining with the catalyst, as illustrated by the volcano plot in Fig. 17a.⁹⁹ On the left side of the plot, when the bond strength is too high, the catalytic reactivity is limited by the product desorption. As the bond strength slowly decreases, the product becomes prone to desorption, and the rate gradually increases. On the right side of the plot, when the bond strength is too weak, the activation of the reactants is difficult, and this process limits the catalytic reaction rate. As the bond strength slowly increases, the reactants become easier to activate, and the rate will gradually increase. Accordingly, the maximum of the volcano plot corresponds to the highest bond strength and optimum activity. However, the Sabatier principle does not clarify which descriptor should be used for the bond strength between intermediate reactants and catalysts, or the bond strength value associated with the optimum catalyst; hence, it cannot be used as a quantitative method to identify the optimum catalyst.⁹⁹

Fig. 17 (a) Schematic representation of the qualitative Sabatier principle. This figure has been adapted from ref. 99 with permission from Elsevier Inc, copyright 2015. (b) Combined volcano diagrams (lines) for the flat (black) and stepped (red) transition metal surfaces for reduction of nitrogen with a Heyrovsky type reaction, without (solid lines) and with (dotted lines) the H-bond effect. The data points are the DFT values of DG for a given reaction step. This figure has been adapted from ref. 95 with permission from the Royal Society of Chemistry, copyright 2012. (c) Comparison of HER and eNRR limiting-potential volcanoes. This figure has been adapted from ref. 103 with permission from Wiley VCH, copyright 2015.

In electrochemical nitrogen fixation, the adsorption energies of the hydrogen-containing species NH_x (x = 1, 2) and N₂H_x (x = 0, 1, 2), as well as of H atoms, exhibit a linear scaling relationship with the chemisorption energy of N on selected transition metals.^100,101 In addition, the free energy of each elementary step, which is a combined function of the applied bias U and the adsorption energies, scales as the adatom binding energy.¹⁰² Thus, based on this scaling relationship, the reaction free energy can be considered as the descriptor of the electrocatalytic activity, and a volcano plot can be used to predict the most efficient electrocatalyst for ammonia synthesis. Skúlason et al. assessed the scaling of the activation energy toward the eNRR with the binding energy of N-adatoms on flat and stepped surfaces of transition metals. The volcano plot in Fig. 17b shows the scaling relationship between the onset potential and binding energy of N species.⁹⁵ It is clear that the transition metals Mo, Fe, Rh, and Ru at the top of the volcano plot, which preferentially attach to *H rather than *N species, have the highest catalytic activity. Moreover, the metals on the right side of the volcano plot such as Rh, Ru, Ir, Co, Ni, and Pt are consistently covered by H atoms, leading to the HER taking place instead of the eNRR. Only some of the flat surfaces of early transition metals such as Sc, Y, Ti, or Zr would bind N atoms more strongly than H atoms, and required a bias of only 1 to 1.5 V vs. standard hydrogen electrode (SHE) for the reduction of N₂ to NH₃ through dissociative mechanisms.

Montoya and co-workers introduced a new perspective on the role of nitrogen scaling relationships, suggesting that the electrochemical synthesis of NH₃ is limited by the linear scaling relationship between the energetics of two intermediates, *N₂H and *NH₂, as shown in Fig. 17c.¹⁰³ Most of the metal surface facets suffered from the reduction of *N₂ to form *N₂H, but the (111) terraces on the left side of the volcano plot were limited by the protonation of *NH to form *NH₂, while the reductive desorption of *NH₂ affected the (211) steps on reactive surfaces. Fig. 17c shows that the limiting potentials required by the eNRR were larger than those corresponding to the HER, with the minimum difference being as high as 0.5 V for Re(111), suggesting that the selectivity of the considered transition metals was biased toward the HER. The authors considered it challenging to design a catalyst that can selectively stabilize *N₂H or destabilize *NH₂.

Although metal alloys can significantly improve the performances of gas-phase and electrochemical catalysts,¹⁰⁴ the binding configurations of the key *N₂H, *NH, and *NH₂ adsorbates, in which adsorption takes place through a single N atom, are too similar to be distinguished; hence, alloys are unlikely to break the scaling relationships for the eNRR. To achieve an efficient heterogeneous catalyst for the electrochemical synthesis of ammonia, the use of bifunctional surfaces that simultaneously bind *N₂H and *NH₂ onto different active sites or multiple sites would make it feasible to break the scaling relationships.

6. Optimization of the catalytic system

The electrochemical nitrogen reduction reaction is commonly performed using water as a protic solvent. Many effective metal catalysts preferentially adsorb protons rather than N₂; thus, their active sites are covered by H⁺ species.^105,106 Unfortunately, because the limiting potential for the eNRR is negative enough to preferentially drive the HER, the latter usually dominates on the catalyst surface.¹⁰ Therefore, the aqueous eNRR suffers from low selectivity relative to the HER.¹² Because the rate-limiting step of the eNRR exhibits a zeroth-order dependence on the proton density, while that of the HER shows a first-order correlation,¹⁰ limiting the accessibility of protons from the electrolyte to the electrocatalyst surface can be considered an effective approach to suppress the HER; this would favor the preferential adsorption of nitrogen molecules on the electrocatalyst, facilitating the subsequent hydrogenation.

The conventional aqueous electrolyte can also be replaced with an ionic liquid (IL) or another water-free donor, only adding a small amount of water molecules to provide protons, and greatly reducing the probability of the hydrogen evolution reaction. MacFarlane and co-workers proposed the application of an aprotic, highly fluorinated IL to suppress the availability of H⁺, achieving an excellent FE for the eNRR of 60%.¹⁰⁷ The IL also had excellent N₂ solubility, but its viscosity led to poor mobility, which limited N₂ transport, resulting in an NH₃ yield rate of only 10⁻¹² mol s⁻¹ cm⁻². Given the liquidity, the authors chose highly fluorinated 1H,1H,5H-octafluoropentyl 1,1,2,2-tetrafluoroethyl ether (FPEE) as a solvent, with the highly soluble 1-butyl-1-methypyrrolidinium tris(pentafluoroethyl) trifluorophosphate ([C4mpyr][eFAP]) as an IL.¹⁰⁸ At a moisture concentration of 114.3 ppm, the highest FE of 32% for the eNRR and an NH₃ yield rate of 2.35 × 10⁻¹¹ mol s⁻¹ cm⁻² were obtained when the mole fraction of IL(X_IL) was 0.23; however, the eNRR performance apparently decreased when X_IL was 0.46, reflecting the significant importance of FPEE for mass transport, as shown in Fig. 18a. The rational design of aprotic solvents to regulate H⁺ adsorption greatly improves the selectivity and FE parameters. Nørskov and co-workers compared the effect of the non-aqueous proton donor 2,6-lutidinium (LutH⁺) and hydronium on the selectivity, as shown in Fig. 18b.¹⁰⁹ Hindered by the metal surface puckering, LutH⁺ required more energy than H₃O⁺ to complete the Volmer reaction (LutH⁺ + e⁻ ⇌ Lut + *H, H₃O⁺ + e⁻ ⇌ H₂O + *H), which further increased the difficulty for H species to reach the catalyst surface. Thus a non-aqueous proton donor is appropriate to suppress the HER and enhance the eNRR performance.

Fig. 18 (a) X_IL dependence of the NH₃ yield and FE (%) at −0.65 V vs. NHE. This figure has been adapted from ref. 108 with permission from American Chemical Society, copyright 2018. (b) 2,6-Lutidinium adsorption configurations on Pt(111). This figure has been adapted from ref. 109 with permission from the Royal Society of Chemistry, copyright 2018. (c) Schematic illustration of the fabrication procedure of Au/TS and Au-PTFE/TS. This figure has been adapted from ref. 110 with permission from Elsevier Inc, copyright 2018. (d) Scheme depicting the strategy to transform water into an efficient eNRR proton source by regulating access of water to the electrocatalytic surface. This figure has been adapted from ref. 111 with permission from American Chemical Society, copyright 2020. (e) Effective rate of ammonia production and effective faradaic efficiency when using the M@ZIF-OAm working electrode as the amount of water in the electrolyte is regulated. This figure has been adapted from ref. 111 with permission from American Chemical Society, copyright 2020.

Modification of the surface structure of the catalyst, such as coating with a hydrophobic layer, increases the kinetic barrier for water molecules to reach the catalyst, and also effectively improves the selectivity toward NH₃. Zheng and co-workers deposited a porous poly(tetrafluoroethylene) (PTFE) layer on a p-type Si wafer with magnetron sputtered Au nanoparticles, as shown in Fig. 18c.¹¹⁰ The presence of the hydrophobic PTFE player resulted in an excellent FE of 37.8% and an ammonia yield rate of 18.9 μg cm⁻² h⁻¹, whereas the photoelectrochemical performance in the ammonia synthesis was significantly reduced without PTFE, which highlighted the role of PTFE as a HER inhibitor. However, the corresponding current density was only 0.1 mA cm⁻², far lower than that of the pure Si photoanodes. Ling and co-workers modified the surface of a zeolitic-imidazolate framework-71 (ZIF-71) layer with hydrophobic oleylamine (OAm) and filled the pores of ZIF-71 with butanol, which was more preferentially adsorbed in ZIF-71 over water modules, as shown in Fig. 18d.¹¹¹ The use of both OAm and ZIF-71 prevents water from smoothly entering the pores of ZIF-71. The decorated ZIF-71 coating encapsulated a bimetallic electrocatalyst (metal@ZIF-OAm) to concentrate the reactants on the surface. For metal@ZIF-OAm, the maximum FE in the presence of 0.1% v/v water was 18%, nearly four times as high as that in the absence of water. This suggested that the increased water content effectively promoted the eNRR performance. For metal@ZIF without the oleylamine layer, the FE and ammonia yield with 0.075% v/v water were close to zero, which highlighted the effect of hydrophobicity in thermodynamically excluding moisture and suppressing the HER, as shown in Fig. 18e. This effect transforms water from an unavoidable hindrance to a superior proton source for the eNRR, possibly combined with the direct use of atmospheric air as a raw N₂ and proton source for a sustainable eNRR.

Therefore, the optimization of the electrolytic system, especially in terms of limiting the competing hydrogen evolution reaction, involves two aspects, namely, the rational design of (1) electrolyte composition and (2) surface structure of the electrodes. Tuning the mass transfer and gas diffusion can hinder the direct contact of protons with the surface of the catalyst. In addition, the conductivity of the catalysts and the synergy between the electrolyte and catalytic active sites can also be optimized to improve the electrocatalytic selectivity toward the synthesis of ammonia.

7. Ammonia detection

The ubiquitous presence of ammonia in the laboratory environment may interfere with the experimental results, because the NH₃ yield rate in the eNRR is usually very low. The analysis must verify that the ammonia detected in the eNRR experiment is electrochemically reduced by nitrogen rather than other exogenous sources. At present, the main colorimetric methods for ammonia detection are ion chromatography and spectrophotometry. Ion chromatography is highly selective, reproducible and efficient, but requires complex and expensive equipment. Spectrophotometry measurements are relatively cheap, and include the indophenol blue and Nessler's reagent methods. The principle of the former spectrophotometric approach is that ammonia absorbed in dilute sulfuric acid, in the presence of sodium nitroprusside and sodium hypochlorite, reacts with salicylic acid to form a blue-green indophenol blue dye; the final quantification of the NH₃ concentration is based on the depth of the reagent color. In the Nessler's reagent method, the iodide and mercury ions in Nessler's reagent react with ammonia under alkaline conditions to produce a reddish-brown mixture. The absorbance of the above complex is directly proportional to the ammonia nitrogen content of the solution. Therefore, the ammonia content can be determined by measuring the absorption intensity of the reacted solution. However, considering that the reagent is highly toxic and the detection is prone to interferences, sample pretreatment is needed. The three methods mentioned above are accurate when the ammonia concentration is lower than 500 μg L⁻¹, whereas the indophenol blue method is less accurate at high ammonia concentrations (acidic medium). In terms of accuracy of the results, the ion chromatography method is suitable for acidic media, whereas the indophenol blue approach is suitable for alkaline media, and the Nessler's reagent method is suitable for both.¹¹² Apart from colorimetric methods, the main electrode methods for NH₃ detection are based on ammonia gas-sensing and ammonium ion-selective electrodes. Generally, these methods do not require pretreatment of samples. The principle of the ammonia gas-sensing electrode is the addition of a concentrated alkali solution to the target solution to convert ammonium ions to free ammonia. The latter enters the internal buffer of the electrode through a hydrophobic membrane, which alters its pH value. The instrument can measure the ammonia concentration in the solution by monitoring the pH change, which is directly proportional to the NH₃ concentration. The ammonia gas-sensing electrode has the advantages of simple operation, short test time, large measurement range and high accuracy. However, care should be taken to avoid contamination on the surface of the membrane electrode, which would otherwise block the pores, resulting in the need for expensive electrode replacement.¹¹³ On the other hand, the selective membrane of the ammonium ion-selective electrode is in contact with the ammonium ion solution, creating a potential difference between the interior and exterior of the membrane, which depends on the activity of free ammonium ions in the solution. The ammonium ion concentration can thus be calculated from the potential difference. This method has the advantages of simple operation, wide linear range, small relative error and high accuracy.¹¹⁴ However, its anti-interference performance needs to be improved.

Before the experiment, the ammonia content in the feed gas should be quantified. Similarly, it is necessary to measure the ammonia content of each part of the experimental system, improving the gas purity in the experimental procedure. Moreover, the repeatability of the experiment should be confirmed to improve the reliability of the results, as shown in Fig. 19.¹¹⁵ In summary, before ammonia detection, the interferences from the external environment should be eliminated to ensure the purity of the ammonia source; multiple detection methods can be used at the same time to improve the accuracy and precision of the results.

Fig. 19 Updated and simplified eNRR experimental protocol. This figure has been adapted from ref. 115 with permission from Springer Nature, copyright 2020.

Conclusions

Ammonia is a crucial chemical in modern society. The development of new sustainable and economical methods for ammonia synthesis has become an urgent task to support the growing global demand associated with the increasing population and carbon emissions. The electrocatalytic synthesis of ammonia at ambient temperature is one of the most promising approaches, with potentially zero carbon emission, reduced energy consumption, and easy implementation in many underdeveloped regions. Furthermore, SACs have abundant active sites, unsaturated coordination environment, and unique electronic structure superior to normal catalysts, which have the potential to achieve sustainable ammonia green production. In addition, because of their simple structure, these catalysts can be used as a model for fundamental studies aimed to further understand the mechanism of the eNRR process.¹¹⁶ However, the low affinity for N₂ of the catalysts and the competition of the hydrogen evolution reaction for active sites, electrons, and protons, along with the large overpotential and high energy barrier of the ammonia synthesis reaction, result in low faradaic efficiency, poor ammonia yield, and limited selectivity of most catalysts. Therefore, several key challenges remain to be resolved for designing and producing suitable single-atom catalysts for electrocatalytic ammonia synthesis:

1) Even though each atom of SACs shows efficiency, the practical application of SACs is limited by the low atomic loading, and an excessive increase of metal content will lead to the formation of nanoparticles. Thus, it is vital to explore more rational synthesis methods based on atomic-level designs to increase anchor sites and active sites, improving electronic structure and adapting to industrial application, such as the defect engineering strategy, spatial confinement strategy and coordination design strategy. Besides, exploring coupled SACs would allow taking full advantage of the synergetic effect between adjacent single atoms, removing the constraints of linear relationships, efficiently adsorbing reactants while rapidly releasing products, and improving the performance of electrocatalysts for ammonia synthesis. For the industrial development of the eNRR, further efforts should be made to strengthen economic analysis and evaluate the compatibility with renewable energy. Further work should be expanded to modulate porous substrates and metal catalysts.

2) In situ characterization techniques need to be further developed and applied at the atomic level, to reveal the detailed structure of the catalyst, and offer a variety of visualizations of the real-time and real-space progress of the electrochemical process. Although computational DFT can limitedly verify the reaction mechanism, experimental verification of which still has challenges. Consequently, it is essential to visualize the coordination environment of isolated single atoms based on advanced in situ characterization, and then determine the structure–performance relationships in SACs as well as the evolution of the actual processes taking place at catalytically active sites, providing the guidance for the rational design of eNRR catalysts.

3) Considering that transition metal surfaces tend to adsorb H rather than N₂, the competitive reaction HER occupies the active sites and inhibits the further development of the eNRR. In order to enhance selectivity, on the one hand, it's necessary to reduce the accessibility of protons to avoid H⁺ ions from occupying the active sites and favoring the hydrogen evolution reaction, on the other hand, providing H⁺ for the subsequent reaction of the eNRR is also critical. This requires regulating the ion type, concentration, pH and temperature of the electrolytes. For example, using ionic liquids instead of water is a valuable option to raise selectivity, which can also improve the solubility of N₂ and increase the concentration of gaseous reactants. More research is needed to explore similar electrolytes.

4) More intuitive approaches for the quantitative determination of ammonia/ammonium ion yields also require efficient and accurate detection methods. However, owing to the low ammonia yield rate, the decomposition of ammonia, and the interference of trace ammonia gas in the environment, the accuracy and reliability of the results need to be assessed. Reliable data that can be compared to the actual catalytic performance can only be obtained by strictly following the experimental protocols, excluding all sources of ammonia contamination, and minimizing the possibility of false positive results. Further work should be implemented for more advanced and accurate real-time ammonia detection and more rigorous detection processes.

In conclusion, many challenges still need to be addressed to achieve the large-scale and simple synthesis of SACs, including the study of the catalytic mechanism of the eNRR, the development of in situ characterization techniques, and the accurate determination of NH₃. Fundamental progress in these areas will support future studies, aimed to develop green catalytic processes, and eventually achieve the sustainable production of NH₃.

Conflicts of interest

There are no conflicts to declare.

Acknowledgements

This work was supported by the National Natural Science Foundation of China (51976191, 51806193) and Academy of Ecological Civilization, Zhejiang University.

Notes and references

1. Ammonia annual production capacity globally 2030, https://www.statista.com/statistics/1065865/ammonia-production-capacity-globally/, (accessed July 6, 2021).
2. Ammonia Market Size Worth $76.64 Billion By 2025 | CAGR 5.3%, https://www.grandviewresearch.com/press-release/global-ammonia-market, (accessed July 6, 2021).
3. S. Giddey, S. P. S. Badwal and A. Kulkarni, Int. J. Hydrogen Energy, 2013, 38, 14576–14594.
4. L. Lassaletta, G. Billen, B. Grizzetti, J. Anglade and J. Garnier, Environ. Res. Lett., 2014, 9, 105011.
5. Ammonia Market Size Worth $76.64 Billion By 2025 | CAGR 5.3%, https://www.grandviewresearch.com/press-release/global-ammonia-market, (accessed July 7, 2021).
6. Q. Hao, C. Liu, G. Jia, Y. Wang, H. Arandiyan, W. Wei and B.-J. Ni, Mater. Horiz., 2020, 7, 1014–1029.
7. R. Schlögl, Angew. Chem., Int. Ed., 2003, 42, 2004–2008.
8. J. H. Park, H. C. Yoon, J.-N. Kim, C.-H. Jeong, E.-Y. Jeong, D. S. Yun, H. Yoon, S. H. Park, M.-H. Han and C.-Y. Yoo, Korean J. Chem. Eng., 2018, 35, 1620–1625.
9. G.-F. Chen, S. Ren, L. Zhang, H. Cheng, Y. Luo, K. Zhu, L.-X. Ding and H. Wang, Small Methods, 2019, 3, 1800337.
10. A. R. Singh, B. A. Rohr, J. A. Schwalbe, M. Cargnello, K. Chan, T. F. Jaramillo, I. Chorkendorff and J. K. Nørskov, ACS Catal., 2017, 7, 706–709.
11. J. A. Bakken, Pure Appl. Chem., 1994, 66, 1239–1246.
12. S. Wang, F. Ichihara, H. Pang, H. Chen and J. Ye, Adv. Funct. Mater., 2018, 28, 1803309.
13. J. Guo and P. Chen, Chem, 2017, 3, 709–712.
14. H. Xu, K. Ithisuphalap, Y. Li, S. Mukherjee, J. Lattimer, G. Soloveichik and G. Wu, Nano Energy, 2020, 69, 104469.
15. X. Yan, D. Liu, H. Cao, F. Hou, J. Liang and S. X. Dou, Small Methods, 2019, 3, 1800501.
16. J. N. Galloway and E. B. Cowling, Ambio, 2002, 31, 64–71.
17. M. Li, H. Huang, J. Low, C. Gao, R. Long and Y. Xiong, Small Methods, 2019, 3, 1800388.
18. Y. Yao, J. Wang, U. B. Shahid, M. Gu, H. Wang, H. Li and M. Shao, Electrochem. Energy Rev., 2020, 3, 239–270.
19. Y. Hao, X. Dong, S. Zhai, H. Ma, X. Wang and X. Zhang, Chem. – Eur. J., 2016, 22, 18722–18728.
20. C. Tang and S.-Z. Qiao, Chem. Soc. Rev., 2019, 48, 3166–3180.
21. N. Lazouski, M. Chung, K. Williams, M. L. Gala and K. Manthiram, Nat. Catal., 2020, 3, 463–469.
22. M.-M. Shi, D. Bao, B.-R. Wulan, Y.-H. Li, Y.-F. Zhang, J.-M. Yan and Q. Jiang, Adv. Mater., 2017, 29, 1606550.
23. F. Lü, S. Zhao, R. Guo, J. He, X. Peng, H. Bao, J. Fu, L. Han, G. Qi, J. Luo, X. Tang and X. Liu, Nano Energy, 2019, 61, 420–427.
24. X. Wang, W. Wang, M. Qiao, G. Wu, W. Chen, T. Yuan, Q. Xu, M. Chen, Y. Zhang, X. Wang, J. Wang, J. Ge, X. Hong, Y. Li, Y. Wu and Y. Li, Sci. Bull., 2018, 63, 1246–1253.
25. Q. Qin, T. Heil, M. Antonietti and M. Oschatz, Small Methods, 2018, 2, 1800202.
26. X. Yu, P. Han, Z. Wei, L. Huang, Z. Gu, S. Peng, J. Ma and G. Zheng, Joule, 2018, 2, 1610–1622.
27. Y. Liu, Q. Xu, X. Fan, X. Quan, Y. Su, S. Chen, H. Yu and Z. Cai, J. Mater. Chem. A, 2019, 7, 26358–26363.
28. W. Zang, T. Yang, H. Zou, S. Xi, H. Zhang, X. Liu, Z. Kou, Y. Du, Y. P. Feng, L. Shen, L. Duan, J. Wang and S. J. Pennycook, ACS Catal., 2019, 9, 10166–10173.
29. Y. Li, Q. Zhang, C. Li, H.-N. Fan, W.-B. Luo, H.-K. Liu and S.-X. Dou, J. Mater. Chem. A, 2019, 7, 22242–22247.
30. M. Wang, S. Liu, T. Qian, J. Liu, J. Zhou, H. Ji, J. Xiong, J. Zhong and C. Yan, Nat. Commun., 2019, 10, 341.
31. Y. Wang, X. Cui, J. Zhao, G. Jia, L. Gu, Q. Zhang, L. Meng, Z. Shi, L. Zheng, C. Wang, Z. Zhang and W. Zheng, ACS Catal., 2019, 9, 336–344.
32. R. Zhang, L. Jiao, W. Yang, G. Wan and H.-L. Jiang, J. Mater. Chem. A, 2019, 7, 26371–26377.
33. L. Han, X. Liu, J. Chen, R. Lin, H. Liu, F. Lü, S. Bak, Z. Liang, S. Zhao, E. Stavitski, J. Luo, R. R. Adzic and H. L. Xin, Angew. Chem., Int. Ed., 2019, 58, 2321–2325.
34. L.-P. Yuan, Z.-Y. Wu, W.-J. Jiang, T. Tang, S. Niu and J.-S. Hu, Nano Res., 2020, 13, 1376–1382.
35. Z. Geng, Y. Liu, X. Kong, P. Li, K. Li, Z. Liu, J. Du, M. Shu, R. Si and J. Zeng, Adv. Mater., 2018, 30, 1803498.
36. H. Tao, C. Choi, L.-X. Ding, Z. Jiang, Z. Han, M. Jia, Q. Fan, Y. Gao, H. Wang, A. W. Robertson, S. Hong, Y. Jung, S. Liu and Z. Sun, Chem, 2019, 5, 204–214.
37. B. Yu, H. Li, J. White, S. Donne, J. Yi, S. Xi, Y. Fu, G. Henkelman, H. Yu, Z. Chen and T. Ma, Adv. Funct. Mater., 2020, 30, 1905665.
38. J. Liu, X. Kong, L. Zheng, X. Guo, X. Liu and J. Shui, ACS Nano, 2020, 14, 1093–1101.
39. Q. Wang, G. Zheng, S. Hao, X. Liu, J. Zheng, Y. Wang, Z. Su, N. Xu, Y. He, L. Lei and X. Zhang, ACS Sustainable Chem. Eng., 2020, 8, 44–49.
40. Y. Song, D. Johnson, R. Peng, D. K. Hensley, P. V. Bonnesen, L. Liang, J. Huang, F. Yang, F. Zhang, R. Qiao, A. P. Baddorf, T. J. Tschaplinski, N. L. Engle, M. C. Hatzell, Z. Wu, D. A. Cullen, H. M. Meyer, B. G. Sumpter and A. J. Rondinone, Sci. Adv., 2018, 4, e1700336.
41. L. Zhang, X. Ji, X. Ren, Y. Ma, X. Shi, Z. Tian, A. M. Asiri, L. Chen, B. Tang and X. Sun, Adv. Mater., 2018, 30, 1800191.
42. L. Zhang, X. Ji, X. Ren, Y. Luo, X. Shi, A. M. Asiri, B. Zheng and X. Sun, ACS Sustainable Chem. Eng., 2019, 7, 1807–1807.
43. X. Yang, J. Nash, J. Anibal, M. Dunwell, S. Kattel, E. Stavitski, K. Attenkofer, J. G. Chen, Y. Yan and B. Xu, J. Am. Chem. Soc., 2018, 140, 13387–13391.
44. X. Cui, C. Tang, X.-M. Liu, C. Wang, W. Ma and Q. Zhang, Chem. – Eur. J., 2018, 24, 18494–18501.
45. L. Hu, A. Khaniya, J. Wang, G. Chen, W. E. Kaden and X. Feng, ACS Catal., 2018, 8, 9312–9319.
46. H. Wang, L. Wang, Q. Wang, S. Ye, W. Sun, Y. Shao, Z. Jiang, Q. Qiao, Y. Zhu, P. Song, D. Li, L. He, X. Zhang, J. Yuan, T. Wu and G. A. Ozin, Angew. Chem., Int. Ed., 2018, 57, 12360–12364.
47. M. Nazemi, S. R. Panikkanvalappil and M. A. El-Sayed, Nano Energy, 2018, 49, 316–323.
48. X. Wu, Z. Wang, Y. Han, D. Zhang, M. Wang, H. Li, H. Zhao, Y. Pan, J. Lai and L. Wang, J. Mater. Chem. A, 2020, 8, 543–547.
49. H. Huang, L. Xia, X. Shi, A. M. Asiri and X. Sun, Chem. Commun., 2018, 54, 11427–11430.
50. Y. Liu, L. Huang, X. Zhu, Y. Fang and S. Dong, Nanoscale, 2020, 12, 1811–1816.
51. A. Liu, M. Gao, X. Ren, F. Meng, Y. Yang, Q. Yang, W. Guan, L. Gao, X. Liang and T. Ma, Nanoscale, 2020, 12, 10933–10938.
52. R. Zhao, C. Liu, X. Zhang, X. Zhu, P. Wei, L. Ji, Y. Guo, S. Gao, Y. Luo, Z. Wang and X. Sun, J. Mater. Chem. A, 2020, 8, 77–81.
53. J. Han, Z. Liu, Y. Ma, G. Cui, F. Xie, F. Wang, Y. Wu, S. Gao, Y. Xu and X. Sun, Nano Energy, 2018, 52, 264–270.
54. W. Qiu, X.-Y. Xie, J. Qiu, W.-H. Fang, R. Liang, X. Ren, X. Ji, G. Cui, A. M. Asiri, G. Cui, B. Tang and X. Sun, Nat. Commun., 2018, 9, 3485.
55. L. Li, C. Tang, B. Xia, H. Jin, Y. Zheng and S.-Z. Qiao, ACS Catal., 2019, 9, 2902–2908.
56. X. Zhang, T. Wu, H. Wang, R. Zhao, H. Chen, T. Wang, P. Wei, Y. Luo, Y. Zhang and X. Sun, ACS Catal., 2019, 9, 4609–4615.
57. Z. Wang, F. Gong, L. Zhang, R. Wang, L. Ji, Q. Liu, Y. Luo, H. Guo, Y. Li, P. Gao, X. Shi, B. Li, B. Tang and X. Sun, Adv. Sci., 2019, 6, 1801182.
58. L. Zhang, L.-X. Ding, G.-F. Chen, X. Yang and H. Wang, Angew. Chem., Int. Ed., 2019, 58, 2612–2616.
59. J. Zhao, B. Wang, Q. Zhou, H. Wang, X. Li, H. Chen, Q. Wei, D. Wu, Y. Luo, J. You, F. Gong and X. Sun, Chem. Commun., 2019, 55, 4997–5000.
60. X. Zhu, T. Wu, L. Ji, C. Li, T. Wang, S. Wen, S. Gao, X. Shi, Y. Luo, Q. Peng and X. Sun, J. Mater. Chem. A, 2019, 7, 16117–16121.
61. G. Zhang, Q. Ji, K. Zhang, Y. Chen, Z. Li, H. Liu, J. Li and J. Qu, Nano Energy, 2019, 59, 10–16.
62. R. Qin, P. Liu, G. Fu and N. Zheng, Small Methods, 2018, 2, 1700286.
63. X. Li, X. Yang, Y. Huang, T. Zhang and B. Liu, Adv. Mater., 2019, 31, 1902031.
64. B. J. O'Neill, D. H. K. Jackson, J. Lee, C. Canlas, P. C. Stair, C. L. Marshall, J. W. Elam, T. F. Kuech, J. A. Dumesic and G. W. Huber, ACS Catal., 2015, 5, 1804–1825.
65. B. Mohanty, B. K. Jena and S. Basu, ACS Omega, 2020, 5, 1287–1295.
66. S. Sun, G. Zhang, N. Gauquelin, N. Chen, J. Zhou, S. Yang, W. Chen, X. Meng, D. Geng, M. N. Banis, R. Li, S. Ye, S. Knights, G. A. Botton, T.-K. Sham and X. Sun, Sci. Rep., 2013, 3, 1775.
67. M. Piernavieja-Hermida, Z. Lu, A. White, K.-B. Low, T. Wu, J. W. Elam, Z. Wu and Y. Lei, Nanoscale, 2016, 8, 15348–15356.
68. H. Fei, J. Dong, Y. Feng, C. S. Allen, C. Wan, B. Volosskiy, M. Li, Z. Zhao, Y. Wang, H. Sun, P. An, W. Chen, Z. Guo, C. Lee, D. Chen, I. Shakir, M. Liu, T. Hu, Y. Li, A. I. Kirkland, X. Duan and Y. Huang, Nat. Catal., 2018, 1, 63–72.
69. C. H. Choi, M. Kim, H. C. Kwon, S. J. Cho, S. Yun, H.-T. Kim, K. J. J. Mayrhofer, H. Kim and M. Choi, Nat. Commun., 2016, 7, 10922.
70. Y. Xue, B. Huang, Y. Yi, Y. Guo, Z. Zuo, Y. Li, Z. Jia, H. Liu and Y. Li, Nat. Commun., 2018, 9, 1460.
71. Z. Zhang, C. Feng, C. Liu, M. Zuo, L. Qin, X. Yan, Y. Xing, H. Li, R. Si, S. Zhou and J. Zeng, Nat. Commun., 2020, 11, 1215.
72. L. Zhao, Y. Zhang, L.-B. Huang, X.-Z. Liu, Q.-H. Zhang, C. He, Z.-Y. Wu, L.-J. Zhang, J. Wu, W. Yang, L. Gu, J.-S. Hu and L.-J. Wan, Nat. Commun., 2019, 10, 1278.
73. Y. Chen, S. Ji, Y. Wang, J. Dong, W. Chen, Z. Li, R. Shen, L. Zheng, Z. Zhuang, D. Wang and Y. Li, Angew. Chem., Int. Ed., 2017, 56, 6937–6941.
74. S. Wei, A. Li, J.-C. Liu, Z. Li, W. Chen, Y. Gong, Q. Zhang, W.-C. Cheong, Y. Wang, L. Zheng, H. Xiao, C. Chen, D. Wang, Q. Peng, L. Gu, X. Han, J. Li and Y. Li, Nat. Nanotechnol., 2018, 13, 856–861.
75. Y.-N. Gong, B.-Z. Shao, J.-H. Mei, W. Yang, D.-C. Zhong and T.-B. Lu, Nano Res., 2022, 15, 551–556.
76. J.-H. Zhang, W. Yang, M. Zhang, H.-J. Wang, R. Si, D.-C. Zhong and T.-B. Lu, Nano Energy, 2021, 80, 105542.
77. J. Liu, Chin. J. Catal., 2017, 38, 1460–1472.
78. M. Wang, L. Árnadóttir, Z. J. Xu and Z. Feng, Nano-Micro Lett., 2019, 11, 47.
79. S. Liu, M. Wang, T. Qian, H. Ji, J. Liu and C. Yan, Nat. Commun., 2019, 10, 3898.
80. S. Back and Y. Jung, Phys. Chem. Chem. Phys., 2016, 18, 9161–9166.
81. V. Kordali, G. Kyriacou and Ch. Lambrou, Chem. Commun., 2000, 1673–1674.
82. D. Bao, Q. Zhang, F.-L. Meng, H.-X. Zhong, M.-M. Shi, Y. Zhang, J.-M. Yan, Q. Jiang and X.-B. Zhang, Adv. Mater., 2017, 29, 1604799.
83. Y. Yao, S. Zhu, H. Wang, H. Li and M. Shao, J. Am. Chem. Soc., 2018, 140, 1496–1501.
84. H. Yin, S.-L. Li, L.-Y. Gan and P. Wang, J. Mater. Chem. A, 2019, 7, 11908–11914.
85. S. Gao, Y. Lin, X. Jiao, Y. Sun, Q. Luo, W. Zhang, D. Li, J. Yang and Y. Xie, Nature, 2016, 529, 68–71.
86. X. Sun, K. Li, C. Yin, Y. Wang, F. He, H. Tang and Z. Wu, Phys. Chem. Chem. Phys., 2017, 19, 17670–17676.
87. N. Saeidi, M. D. Esrafili and J. J. Sardroodi, ChemistrySelect, 2019, 4, 12216–12226.
88. Y. Wen, H. Zhu, J. Hao, S. Lu, W. Zong, F. Lai, P. Ma, W. Dong, T. Liu and M. Du, Appl. Catal., B, 2021, 292, 120144.
89. C. Liu, Q. Li, C. Wu, J. Zhang, Y. Jin, D. R. MacFarlane and C. Sun, J. Am. Chem. Soc., 2019, 141, 2884–2888.
90. L. Jiao, G. Wan, R. Zhang, H. Zhou, S.-H. Yu and H.-L. Jiang, Angew. Chem., Int. Ed., 2018, 57, 8525–8529.
91. L. Jiao, R. Zhang, G. Wan, W. Yang, X. Wan, H. Zhou, J. Shui, S.-H. Yu and H.-L. Jiang, Nat. Commun., 2020, 11, 2831.
92. J. Li, S. Zhao, L. Zhang, S. P. Jiang, S.-Z. Yang, S. Wang, H. Sun, B. Johannessen and S. Liu, Small, 2021, 17, 2004579.
93. W. Guo, K. Zhang, Z. Liang, R. Zou and Q. Xu, Chem. Soc. Rev., 2019, 48, 5658–5716.
94. X.-F. Li, Q.-K. Li, J. Cheng, L. Liu, Q. Yan, Y. Wu, X.-H. Zhang, Z.-Y. Wang, Q. Qiu and Y. Luo, J. Am. Chem. Soc., 2016, 138, 8706–8709.
95. E. Skúlason, T. Bligaard, S. Gudmundsdóttir, F. Studt, J. Rossmeisl, F. Abild-Pedersen, T. Vegge, H. Jónsson and J. K. Nørskov, Phys. Chem. Chem. Phys., 2012, 14, 1235–1245.
96. C. Ling, X. Bai, Y. Ouyang, A. Du and J. Wang, J. Phys. Chem. C, 2018, 122, 16842–16847.
97. X. Li, X. Dai, D. Tang, X. Wang, J. Hong, C. Chen, Y. Yang, J. Lu, J. Zhu, Z. Lei, K. Suenaga, F. Ding and H. Xu, Adv. Funct. Mater., 2021, 31, 2102138.
98. P. Ou, X. Zhou, F. Meng, C. Chen, Y. Chen and J. Song, Nanoscale, 2019, 11, 13600–13611.
99. A. J. Medford, A. Vojvodic, J. S. Hummelshøj, J. Voss, F. Abild-Pedersen, F. Studt, T. Bligaard, A. Nilsson and J. K. Nørskov, J. Catal., 2015, 328, 36–42.
100. F. Abild-Pedersen, J. Greeley, F. Studt, J. Rossmeisl, T. R. Munter, P. G. Moses, E. Skúlason, T. Bligaard and J. K. Nørskov, Phys. Rev. Lett., 2007, 99, 016105.
101. F. Calle-Vallejo, D. Loffreda, M. T. M. Koper and P. Sautet, Nat. Chem., 2015, 7, 403–410.
102. J. G. Howalt, T. Bligaard, J. Rossmeisl and T. Vegge, Phys. Chem. Chem. Phys., 2013, 15, 7785.
103. J. H. Montoya, C. Tsai, A. Vojvodic and J. K. Nørskov, ChemSusChem, 2015, 8, 2180–2186.
104. V. R. Stamenkovic, B. S. Mun, M. Arenz, K. J. J. Mayrhofer, C. A. Lucas, G. Wang, P. N. Ross and N. M. Markovic, Nat. Mater., 2007, 6, 241–247.
105. C. J. M. van der Ham, M. T. M. Koper and D. G. H. Hetterscheid, Chem. Soc. Rev., 2014, 43, 5183–5191.
106. Y. Nishibayashi, Inorg. Chem., 2015, 54, 9234–9247.
107. F. Zhou, L. M. Azofra, M. Ali, M. Kar, A. N. Simonov, C. McDonnell-Worth, C. Sun, X. Zhang and D. R. MacFarlane, Energy Environ. Sci., 2017, 10, 2516–2520.
108. B. H. R. Suryanto, C. S. M. Kang, D. Wang, C. Xiao, F. Zhou, L. M. Azofra, L. Cavallo, X. Zhang and D. R. MacFarlane, ACS Energy Lett., 2018, 3, 1219–1224.
109. L. Zhang, S. Mallikarjun Sharada, A. R. Singh, B. A. Rohr, Y. Su, L. Qiao and J. K. Nørskov, Phys. Chem. Chem. Phys., 2018, 20, 4982–4989.
110. J. Zheng, Y. Lyu, M. Qiao, R. Wang, Y. Zhou, H. Li, C. Chen, Y. Li, H. Zhou, S. P. Jiang and S. Wang, Chem, 2019, 5, 617–633.
111. C. S. L. Koh, H. K. Lee, H. Y. Fan Sim, X. Han, G. C. Phan-Quang and X. Y. Ling, Chem. Mater., 2020, 32, 1674–1683.
112. Y. Zhao, R. Shi, X. Bian, C. Zhou, Y. Zhao, S. Zhang, F. Wu, G. I. N. Waterhouse, L.-Z. Wu, C.-H. Tung and T. Zhang, Adv. Sci., 2019, 6, 1802109.
113. G. Zhao, J. Xuan, Q. Gong, L. Wang, J. Ren, M. Sun, F. Jia, G. Yin and B. Liu, ACS Appl. Mater. Interfaces, 2020, 12, 8573–8582.
114. C. F. B. Coutinho, A. A. Muxel, C. G. Rocha, D. A. de Jesus, R. V. S. Alfaya, F. A. S. Almeida, Y. Gushikem and A. A. S. Alfaya, J. Braz. Chem. Soc., 2007, 18, 6.
115. J. Choi, B. H. R. Suryanto, D. Wang, H.-L. Du, R. Y. Hodgetts, F. M. Ferrero Vallana, D. R. MacFarlane and A. N. Simonov, Nat. Commun., 2020, 11, 5546.
116. A. Wu, J. Yang, B. Xu, X.-Y. Wu, Y. Wang, X. Lv, Y. Ma, A. Xu, J. Zheng, Q. Tan, Y. Peng, Z. Qi, H. Qi, J. Li, Y. Wang, J. Harding, X. Tu, A. Wang, J. Yan and X. Li, Appl. Catal., B, 2021, 299, 120667.

---