Atomic Layer Deposition: A Gas Phase Route to Bottom-up Precise Synthesis of Heterogeneous Catalyst

WANG Hengwei , LU Junling ,2,＊ Department of Chemical Physics, Hefei National Laboratory for Physical Sciences at the Microscale, iChEM, University of Science and Technology of China, Hefei 230026, P. R. China.

2 CAS Key Laboratory of Materials for Energy Conversion, University of Science and Technology of China, Hefei 230026, P. R.China.

Abstract: Heterogeneous catalysts are usually synthesized by the conventional wet-chemistry methods, including wet-impregnation,ion exchange, and deposition-precipitation. With the development of catalyst synthesis, great progress has been made in many industrially important catalytic processes. However, these catalytic materials often have very complex structures along with poor uniformity of active sites. Such heterogeneity of active site structures significantly decreases catalytic performance, especially in terms of selectivity, and hinders atomic-level understanding of structureactivity relationships. Moreover, loss of exposed active components by sintering or leaching under harsh reaction conditions causes considerable catalyst deactivation. It is desirable to develop a facile method to tune catalyst active site structures, as well as their local chemical environments on the atomic level,thereby facilitating reaction mechanisms understanding and rational design of catalysts with high stability.

Key Words: Atomic layer deposition; Supported metal catalyst; “Bottom-up” synthesis; Single-atom catalyst;Dimeric metal catalysts; Metal-oxide interfaces; Confinement effect

Prof. LU Junling received his PhD degree from Institute of Physics, Chinese Academy of Sciences under the supervision of Prof. Hongjun Gao in 2007. During his PhD studies, he visited Prof. Hans-Joachim Freund group at Chemical Physics Department, Fritz-Haber-Institute, Max Planck Society as an exchange student in 2004–2006. After graduation, he spent three years in Prof.Peter C Stair’s group at Northwestern University and then about two and a half years in Dr. Jeffrey W. Elam’s group at Argonne National Laboratory as a Postdoc. In March. 2013, he became a professor at University of Science and Technology of China. His current research interest is atomically-precise design of new catalytic materials using a combined wetchemistry and atomic layer deposition (ALD) approach for advanced catalysis.

1 Introduction

Efficient chemical transformations from fossil-based feeds(oil, gas and coal) and renewable resources (biomass, carbon dioxide and water) to commodity chemicals and fuels in a sustainable and environmentally benign manner are highly desirable to meet the global demands of energy, resources and the living environments1–4. Catalysis is the essential technology for accelerating and directing the transformations from feeds to valuable products5,6. Solid heterogeneous catalysts are the preferred form in industrial processes since they can be easily separated from the final products and reutilized3,5,7. In the last century, many approaches based on the wet-chemistry have been developed for heterogeneous catalysts synthesis, including impregnation, ion exchange, and precipitation8,9. Although remarkable progress has been achieved in many industrial catalytic processes5, these catalytic materials, in many cases,have poor homogeneity of the structures of the active centers,which might contribute to different reaction pathways and produce undesired side-products10. Moreover, the nonuniformity of the active sites also makes it extremely difficult to link the catalytic properties with the specific catalytic structure6.

Design highly selective catalysts like enzymes, which convert a molecule to a specific product, is the goal in 21th century but challenging3. Improving the homogeneity of the active site structure with atomic-level control is essential to improve the catalyst selectivity. Fundamentally, the improved active site homogeneity could also facilitate the understanding of the reaction mechanisms and the establishment of the “structureactivity” relationships. Great efforts have been devoted to advanced catalyst synthesis in the 20th century9. For instance,colloidal synthesis shows great opportunities to finely control over the size, composition, and morphology of metal nanoparticles (NPs)/nanocrystals with the additive of capping ligands11–13. In this case, the ligands (mostly organic compounds) might significantly influence the catalytic performance due to steric effect or electronic effect14,15.Moreover, grafting such colloidal NPs onto catalyst supports and removing the capping ligands sequentially for catalytic applications might again cause considerable metal aggregations9.It is an urgent need to develop a method which allows tuning the catalyst active site structures as well as their local chemical environments at the atomic level, and thus speeding up our understanding of reaction mechanisms and catalyst rational design.

One promising approach is atomic layer deposition (ALD)6,16.It is a gas-phase thin film growth technique, originally developed for producing high quality thin films in electroluminescent flat panel displays17. ALD is similar to chemical vapor deposition(CVD) except that the deposition is proceeded in cycles, each of which is based on two successive and alternative self-limiting surface reactions between precursor molecule vapors and a substrate16,18,19. After each step of surface reaction, an inert gas is introduced to the reaction chamber to purge out any unreacted precursors and volatile products, as illustrated in Fig. 1. Due to the unique feature of molecular-level “self-limiting” surface reactions, ALD shows a superior capability of conformal coating on complex substrates with high aspect ratio and/or high surface areas20–26. Moreover, the film thickness can be tuned precisely at the atomic level by varying the ALD cycles. So far, more than 1000 ALD processes have been developed for a number of elements in the periodic table in the forms of metals, oxides,nitrides, and sulfides etc.19.

Owing to the capability of precise control, ALD has attracted great attention in broad applications including catalysis,photovoltaics, batteries, fuel cells, polymers, and microdevices.Particularly, the success of conformal deposition on highsurface-area complex substrates with nearly atomic precision,sparks great interest to the catalysis field in recent years. ALD is expected to enable precise construction of the target catalytic materials on high-surface-area supports step-by-step as increasing ALD cycles, thus providing a “bottom-up” approach for synthesis of supported catalysts with high uniformity at the atomic level6,27,28. Consequently, catalyst precise synthesis offered by ALD brings great opportunities to facilitate atomiclevel understanding of reaction mechanisms and rational design of advanced catalysts with high performance.

In this review, we summarize the recent developments of ALD in nanoscale bottom-up constructing various catalytic structures of supported metal catalysts including monometallic, bimetallic catalysts, metal single-atom catalysts (SACs), dimeric metal catalysts (DMCs) and metal-oxide interfaces engineering for improved catalytic performance.

2 Precise synthesis of supported metal catalysts

Transition-metal NPs supported on oxide or carbon-based support are of great interest in heterogeneous catalysis duo to their unique catalytic properties7. They are widely used in many practical reactions, including petroleum refining29, steam reforming30,31, water-gas shift32,33, automobile exhaust treatment1,34,35, Fischer-Tropsch synthesis36,37, biomass conversions38,39, chemical upgrading40–42, and any other processes43–45.The catalytic performances of metal catalysts strongly depend on the size, composition, and structure of metal NPs14,46–53.Therefore, precisely tuning these factors in wide range is essentially important to optimize the catalyst performance and to understand the reaction mechanism. Great efforts have been devoted to achieving a better control over these factors54,55.Nevertheless, a facile and efficient method is still missing.

2.1 Synthesis of monometallic NPs

Metal ALD on various oxides and carbon-based materials has been widely investigated. Instead of metal films, metal NPs are often formed on oxide or carbon-based support due to the high mobility of metal atoms at elevated temperatures used for ALD growth. This motivates the application of ALD in catalysis. The size and loadings of metal NPs can be tuned precisely by varying the number of ALD cycles and deposition temperatures6,19.Interestingly, it has been found that the metal particle size distribution is often very narrow.

Fig. 1 Schematic illustration of ALD process 6.

A pioneer work of ALD in catalysis is to deposit Ni NPs on alumina using nickel (II) acetylacetonate (Ni(acac)2) and air56,57.The activity of the obtained materials in toluene hydrogenation showed a “volcano-shape” trend as a function of Ni ALD cycles,similar to the behaviors of Ni/Al2O3catalysts synthesized by traditional method58. Later, Medlin and co-workers59synthesized ALD Ni/Al2O3catalysts using nickelocene(Ni(Cp)2) and H2as precursors at 300 °C. It was shown that the Ni particle size was tuned form 2.4 to 3.3 nm (Fig. 2) and the Ni loadings was adjusted from 4.7% to 16.7% (w, mass fraction) by varying the Ni ALD cycle from 1 to 15. On the contrary, A Ni/Al2O3catalyst synthesized by the incipient wetness method showed a large particle size of 15 nm at a Ni loading of 8.2%(w). They found that the ALD Ni catalyst showed much higher hydrogenation and hydrogenolysis activity than such conventional Ni catalysts, due to the much more abundant defect sites (steps and kinks) on ALD Ni NPs, as shown in Table 1.

Forming smaller Ni NPs using ALD than the traditional wetchemistry methods was also found on the silica support. Liang and co-workers60reported that the Ni particle size was only 2.4 nm after 1 cycle of Ni ALD and slightly increased to 2.6 nm after 15 cycles. They further showed that the ALD Ni/SiO2catalyst was highly active and selective for in the transfer hydrogenation of aryl nitro compounds into the corresponding amines due to its high dispersion and uniform particle size distribution61.Independently, Kim and co-workers62also reported that Ni ALD on silica support using Ni(Cp)2and water as precursor under 250 °C formed highly dispersed Ni NPs. In dry reforming of methane, the ALD Ni/SiO2catalysts exhibited higher activity and better stability against sintering and coking than unsupported NiO materials, even under 800 °C working condition.

Fig. 2 Transmission electron microscopy and high-resolution transmission electron microscopy (HRTEM) images of Ni/Al2O3 catalysts prepared by 1 (a), 5 (b), and 15 (c) ALD cycles 59.

Table 1 Propylene hydrogenation and hydrogenolysis turnover frequencies (TOF) at 225 °C 59.

Pd ALD have been studied extensively in the recent several years16,18,63–67. Typically, Pd ALD was performed using Pd(II)bishexafluoroacetylacetonate (Pd(hfac)2) and formalin as precursors at 200 °C18,66. The growth rate is about 0.022 nm per cycle. Elam et al.65demonstrated that Pd ALD on ZnO-coated silica gel and alumina-coated silica gel both formed highly dispersed and highly uniform Pd NPs (Fig. 3). In the methanol decomposition reaction, the obtained ALD Pd/Al2O3catalysts exhibited ～2 times higher activity than the Pd catalysts synthesized by other methods due to the ultra-uniform dispersion and smaller particle size68,69. Later they further showed that decreasing the ALD growth temperature, the Pd particle size can be further reduced. They found that ultra-small Pd clusters around 0.8 nm with a much narrow size distribution (SD = 0.2 nm) were form on alumina by lowering the ALD temperature from 200 to 100 °C followed by applying protective ALD alumina overcoat67. These subnanometer Pd clusters showed ～2 times higher turn-over frequency (TOF) than that of 2 nm Pd NPs. These well-controlled catalyst synthesis illustrates the advances of ALD in improving the catalytic performance by precisely tuning the particle size.

Fig. 3 Scanning transmission electron microscopy image of Pd NPs supported on Al2O3 prepared by 1 cycle Pd ALD and the correlated Pd particle size distribution 65.

Pt ALD is another ALD process investigated extensively in recent years18,70–73. Pt ALD is usually performed using(methylcyclopentadienyl)trimethylplatinum (MeCpPtMe3) and oxygen as precursors at relatively high temperatures between 200 and 300 °C. Here the oxygen is used to combust off the ligands during each cycle, and the Pt growth rate is about 0.03 and 0.05 nm per cycle at 200 and 300 °C, respectively74. Pt ALD has been widely performed on various catalyst supports ranging from carbon70,75, silica76, alumina77,78, and titania73. It was found that the Pt particle size generally depends mainly on the deposition temperature and the number of ALD cycles, but not strongly influenced by the substrates6. At 300 °C, 1-cycle Pt ALD on oxide supports often results in the formation of highly dispersed Pt NPs with ～1–2 nm size, decreasing the temperature would form subnanometer Pt clusters27,79. Marshall and coworkers80synthesized highly dispersed (～32%) uniform Pt NPs with high Pt loadings (up to 10.4% (w)) on sphere Al2O3at 300 °C (Fig. 4). In the water-gas-shift reaction, they pointed out that the ALD Pt/Al2O3catalysts exhibited excellent catalytic activity and longer catalyst life compared to those of catalysts prepared by impregnation81.

Fig. 4 TEM images of one to three cycles of Pt ALD at 300 °C on sphere-Al2O3 with O2 treatment 80.

Fig. 5 TEM images of the Pt/CNTs catalysts prepared by ALD 84.

Uniform Pt deposition on carbon-based supports was also demonstrated27,70,82,83. Qin and co-workers84reported the formation of highly dispersed Pt sub-nanometer clusters on carbon nanotube (CNTs) with a Pt loading as high as 2.3% (w)after 10 cycles of Pt ALD at 300 °C (Fig. 5). By precise control of the Pt particle size, they demonstrated that the subnanometer Pt clusters with a particle size around 0.5–0.7 nm exhibited the highest activity in styrene hydrogenation reaction84. Bent et al.70reported that Pt NPs deposited on carbon aerogels by ALD exhibited high catalytic activity for oxidation of CO even at an ultralow Pt loading of ～0.05 mg·cm-2with a conversion efficiency of nearly 100% in the 150–250 °C temperature range.This example illustrates improvement of the efficiency of Pt utilization can be achieved by carefully controlling the Pt particle size and loadings.

Besides the particle size control, it is also possible to tune the morphology of metal NPs using ALD by controlling the support structure. Marks and co-workers showed that when Pt ALD was performed on a crystalline SrTiO3support, the Pt NPs have a cube-cube epitaxy with the so-called “Winterbottom construction” controlled by the thermaldynamics between nanoparticle and the facets of the support (Fig. 6)85,86. In hydrogenation of acrolein, Pt supported on SrTiO3was more selective for the allyl alcohol product, because of their higher proportion of (111) facets6. Nonetheless, it is worth noting that tailoring the morphology of metal NPs by ALD has been much less explored.

In addition to the metal catalysts reviewed above, fabrication of other metal catalysts, such as Ru72,87, Ir88, Cu89,90, Co91,92and Ag93, has been also achieved by ALD. In general, due to the self-limiting reaction character, ALD synthesis showed great advantages in precise control of metal particle size with high uniformity, thus providing ideal model catalyst systems for elucidating reaction mechanisms.

2.2 Synthesis of bimetallic NPs

Supported bimetallic NPs often show superior catalytic performance compared to their monometallic counterparts due to the synergistic effect, which may originate from the ensemble effect and/or the electronic effect40,94–96. Precise control of the size, structure and composition of bimetallic NPs is prerequisite to optimize the synergistic effect. A number of wet-chemistry methods have been developed for bimetallic catalysts synthesis,such as successive impregnation97, co-impregnation98,99,controlled simultaneous reduction and electroless deposition100,101.Nevertheless, these methods are often lack of precise control over the structures of bimetallic NPs and yield a mixture of NPs with different compositions and structures102,103. A facile method for bimetallic NPs synthesis with atomically-precise control is still missing.

Fig. 6 Low resolution (left) and high resolution (right) TEM images of Pt nanoparticles on SrTiO3 nanocubes 85.

Adding a second metal component to a supported monometallic NPs by ALD provides the possibility for bimetallic NP synthesis27. The major challenge in synthesis of supported bimetallic NPs is to make sure the secondary metal only depositing on the primary metal NP surface, while avoiding their nucleation on the support to form monometallic NPs27,104,105.

Recently, we demonstrated a novel strategy of selective ALD on metals to synthesis atomically precisely controlled bimetallic catalysts by using a lower deposition temperature and a proper reactant, which was called low-temperature selective ALD79,104.We found that the metal substrate can facilitate the nucleation of metal precursor on metals at lower temperatures due to their catalytic nature, thus enabling the metal ALD process on metals at low temperatures. On the contrary, oxides are usually inert,thus unable to react with metal precursors at low temperatures to initiate the nucleation79. Consequently, at a properly low deposition temperature, the nucleation of metal precursor on oxide supports will be inhibited while only allowing the growth of metal on metal NP surfaces. In situ quartz crystal microbalance (QCM) measurements demonstrated such selective growth, where at 150 °C, the ALD of Pd, Pt and Ru essentially stopped on oxide surfaces (TiO2, Al2O3and ZrO2),but on metal surfaces these metal ALD proceeded smoothly, by showing a linear growth as increasing the ALD cycles (Fig. 7)79.Along with decrease of the deposition temperature, the choice of co-reactant can also facilitate selective metal ALD. For example,usage of H2as the co-reactant in Pd ALD can suppress the Pd growth on oxides more efficiently than HCHO at 150 °C79,105.

To demonstrate the atomic-level precise control of the composition and structures of bimetallic NPs using ALD, we further developed a method of in-situ Fourier transform infrared spectroscopy (FTIR) of CO chemisorption to track the gradual changes in surface composition of bimetallic NPs during ALD synthesis. For example, when selective low-temperature Pt ALD was executed on a Pd/Al2O3catalyst to grow Pd@Pt core-shell NPs using the sequence of MeCpPtMe3and O2at 150 °C, in situ FTIR CO chemisorption was performed after each Pt ALD cycle.FTIR showed that the bridge-bonded CO peak on Pd at 1917 cm-1decreased gradually along with a simultaneous increase of the linear CO on Pt at 2056–2075 cm-1, providing direct evidence of continuous deposition of Pt on Pd NPs with increasing Pt ALD cycles (Fig. 8).The bridge-bonded CO peak on Pd disappeared after ～10 Pt ALD cycles, indication of the formation of a complete Pt shell. According to in situ QCM (Fig.7b) and high-angle annular dark-field STEM (HAADF-STEM)measurements (Fig. 9a), the Pt shell thickness on Pd NPs can be precisely tuned by an increment of ～0.15 ML (monolayer) per cycle.

Changing the order of ALD sequences can further finely tune the structure of bimetallic NPs (eg, alloys vs core-shell). For instance, PtPd alloy NPs can be synthesized by combining Pt and Pd ALD together at 150 °C with the alloy ALD sequence of MeCpPtMe3-O2-H2-Pd(hfac)2-H2-O2. Here adding one H2reduction step after each Pt ALD cycle, and one O2oxidation step after each Pd cycle was used to complement these two different surface chemistries. HAADF-STEM along with energy-dispersive X-ray spectroscopy (EDS) mapping confirmed the PdPt alloy formation (Fig. 9b). Moreover, Pd ALD on Pt/Al2O3at 150 °C using the sequence of Pd(hfac)2-H2can also form Pt core-Pd shell bimetallic NPs (Fig. 9c). The method of low-temperature selective ALD is general and has been extended to other bimetallic systems, such as Ru@Pt79,Cu@Pt106, Cu@Pd107, Au@Pd, etc. for broad applications.

Fig. 7 In situ QCM measurements of ALD metal on metal and oxide surface 79.

Fig. 8 In situ FTIR CO chemisorption measurements during ALD of Pt shell on an ALD Pd/Al2O3 sample at 150 oC 79.

Fig. 9 Structures of ALD PtPd bimetallic NPs 79.

In addition to the strategy of low-temperature selective ALD developed in our group, another two ALD approaches for synthesis of core-shell bimetallic NPs have been also suggested:i) Reducing the oxygen partial pressure to reduce the secondary metal nucleation on support108,109; ii) using self-assembled monolayers (SAMs) to modify the substrate to achieve areaselective ALD (AS-ALD)110,111. In the latter case, Chen et al.111reported that using octadecyltrichlorosilane (ODTS) selfassembled monolayers (SAMs) to block the active sites on substrate, while only allowing metal ALD growing in the nanoscale pinholes to obtain the selective deposition (Fig. 10).With this strategy, Pd@Pt core shell NPs was successfully prepared.

Fig.10 Schematic illustration for fabricating the core-shell NPs through area-selective ALD on ODTS modified substrate 111.

The above manipulation of PdPt bimetallic structures demonstrate the capability of atomically-precise control offered by ALD. The total number of ALD cycles determines the bimetallic particle size, the relative number of ALD cycles used for each component dictates the composition and the order in the individual cycles controls the structure of the bimetallic NPs.Such cycle-by-cycle “bottom up” construction of bimetallic (or multimetallic) NPs on supports shows obvious advantages over conventional methods such as successive impregnation. It is worthy to note that selective metal ALD on metal surfaces but not on supports to exclude any monometallic NP formation,ensuring the high uniformity of ALD bimetallic catalyst. Such finely controlled synthesis of supported bimetallic catalysts facilitates understanding of the correlation between the bimetallic NP structures and their catalytic performance.

AuPd bimetallic catalysts showed superior catalytic performance in many reactions including CO oxidation112,113,direct synthesis of hydrogen peroxide114,115, direct synthesis of vinylacetate116, formic acid decomposition117, and alcohol oxidation40,118,119due to the synergistic effect95,96. The lack of precise control of the structure and composition of AuPd bimetallic NPs at atomic scale makes it difficult to optimize the catalyst performance and establish the structure-activity relations. For example, Hutchings and co-workers showed that in solvent-free selective oxidation of benzyl alcohol, an Au rich core-Pd rich shell bimetallic catalyst synthesized by the coimpregnation method exhibited superior benzaldehyde selectivity to almost 95% at ～60% conversion, much higher than pure Au (85%) and pure Pd (40%), while the activity of Au@Pd bimetallic catalyst was significantly lower than the pure Pd NPs118.However, Yang et al.119observed an opposite result, by showing that Au rich core-Pd rich shell bimetallic catalyst was slightly active than pure Pd.

Fig. 11 Schematic illustration of synthesis of Au@Pd core-shell bimetallic NPs using low-temperature selective Pd ALD 104.

To address the above issue, low-temperature selective ALD was further carried out to synthesize Au@Pd core-shell bimetallic catalyst on a silica support104, as illustrated in Fig. 11.Therein, we firstly synthesized an Au/SiO2catalyst using the deposition-precipitation method. Then Pd ALD was carried out on the Au/SiO2catalyst at 150 °C using the sequence of Pd(hfac)2-H2. Inductively coupled plasma atomic emission spectrometer (ICP-AES) analysis confirmed the selective deposition. HAADF-STEM showed a linear growth of the Pd shell at ～0.08 nm per cycle, as indicated by the increases in both mean particle size and the Pd shell thickness (Fig. 12). Diffuse reflectance infrared Fourier transform spectroscopy (DRIFTS)CO chemisorption measurements revealed that the structure of the Pd shell changes from isolated Pd adatoms/tiny aggregates to larger ensembles, and further to continuous islands or films as increasing Pd ALD cycles. X-ray photoemission spectroscopy(XPS) showed the thinner the Pd shell, the more negative shift in Pd 3d binding energy, indicating that the electronic modulation of Pd by Au depends on the Pd shell thickness tightly98. In solventfree oxidation of benzyl alcohol, we observed the TOFs of Au@Pd core-shell bimetallic catalysts exhibited a “volcanoshape” trend as a function of the Pd shell thickness, wherein the Au@Pd core-shell catalyst with a Pd shell thickness of ～0.8 nm showed a maximum TOF of 27600 h-1(Fig. 13). More importantly, by precisely tuning the shell thickness, the activity of optimized Au@Pd catalyst was ～9 times and ～90 times higher than that of pure Pd/SiO2and pure Au/SiO2catalyst,respectively, which is the largest activity enhancement for Au@Pd catalyst ever reported in literature120–125. The benzaldehyde selectivity in all these samples were about 90%.This work demonstrated both ensemble and electronic effects played the crucial role in benzyl alcohol oxidation. In brief,precise synthesis of bimetallic catalysts systems at the atomic level using selective ALD would greatly facilitates understanding of the correlation between the bimetallic NP structures and their catalytic performance and optimization of catalyst performance.

Fig. 12 Aberration-corrected HAADF-STEM images of different bimetallic catalysts at low and high magnifications 104.

Fig. 13 Initial activities of Au, Au@Pd and Pd catalysts in benzyl alcohol oxidation 104.

2.3 Synthesis of metal SACs

Atomically dispersed metal catalysts have attracted rapidly increasing attention due to their unique catalytic properties and maximal atom efficiency126–137. They are considered as a bridge between homogeneous and heterogeneous catalysis126,138. A number of wet-chemistry-based methods such as coprecipitation126,130,139, impregnation140, pyrolysis synthesis141,surface grafting142, and atomic trapping137have been developed for SACs fabrication. Significant reduction of metal loadings,sometimes even to 0.01% (w) was often applied to minimize metal aggregations126,138,139. However, such ultra-low metal loadings might largely restrain their practical applications. On the other hand, metal single atoms are prone to sintering under reaction conditions due to the high surface free energy.

Synthesis of SACs using ALD from gas phase can be also challenging, since ALD is often performed at elevated temperatures at which metal aggregations are accelerated remarkably. Nonetheless, given the self-limiting feature,fabrication of SACs using ALD is still possible. Sun and coworkers did the pioneer work and reported that Pt ALD on reduced graphene at 250 °C resulted in the formation of a mixture of atoms, clusters and NPs134,143. This material showed significantly improved catalytic activity in the methanol oxidation reaction up to 10 times higher than of the commercial Pt/C catalyst. Botton and co-workers also reported the formation of a mixture of Pt atoms and clusters on a N-doped graphene support using Pt ALD144. Later, Sun et al.134further showed that Pt1single atoms can be formed on N-doped graphene by 50 cycles of Pt ALD at 250 °C (Fig. 14). The first principle calculations suggested that the Pt1atoms bond to N-sites strongly, leading to a remarkable enhanced activity (～37 times)and highly stability in comparison to the commercial Pt/C catalyst for the hydrogen evolution reaction (HER)134.

Very recently, by a careful control over the type and amount of oxygen-containing functional groups on graphene, we successfully achieved atomically dispersed Pd1after applying one cycle of Pd ALD at 150 °C on graphene using the sequence of Pd(hfac)2-HCHO (Fig. 15)145, therein no any Pd clusters or NPs was found. XPS measurements revealed that removal of the weakly bonded oxygen species, while leaving isolated phenol groups only on graphene through high-temperature annealing is the key to form Pd1single atoms (Fig. 15a). X-ray adsorption fine structure spectroscopy (XAFS) further confirmed the atomic dispersion. In selective hydrogenation of 1,3-butadiene, the single-atom Pd1/graphene catalyst demonstrated superior catalytic performance. 100% butenes selectivity, particularly the highest ever 1-butene selectivity of ～70% at 95% conversion was achieved at a mild reaction condition of about 50 °C (Fig.16a, b). In the presence of an excess of propene, which was used to simulate the industrial working conditions, the propene stream was effectively protected with a conversion limited to only 0.1%at a 1,3-butadiene conversion of 98% (Fig. 16c, d), implying the great potential for industrial applications. More importantly, the single-atom Pd1/graphene catalyst showed excellent durability against deactivation during a total 100 h of reaction time on stream without any visible activity decline or selectivity change.We speculate that butadiene adsorbs on the isolated Pd atoms very likely via the mono-π-adsorption mode, rather than the di-π-adsorption (Fig. 16e), since the later one usually requires a large ensemble of Pd surface. The mono-π-adsorption mode encourages 1,2 hydrogen addition to form 1-butene, explaining the extraordinarily high 1-butene selectivity. These results suggest that metal SACs might open a new dimension in selectivity hydrogenation reactions with excellent selectivity,stability and comparable activity.

Fig. 14 ADF STEM images of ALD Pt/NGNs samples with 50 and ALD cycles and schematic illustration of the Pt ALD mechanism on NGNs 134.

Fig. 15 (a) Schematic illustration of single-atom Pd1/graphene catalyst synthesis via a process of anchor sites creation and selection and Pd ALD on pristine graphene. Representive HAADF-STEM images of Pd1/graphene at low (b, c) and high (d) magnifications.Atomically dispersed Pd atoms in image (d) are highlighted by the white circles 145.

Graphitic carbon nitride (g-C3N4) has received much attention for photocatalysis. The unique six-fold N-coordinated cavities have strong interaction with metal atoms, enabling the design of g-C3N4-based SACs. Very recently, we reported that single-atom Pd1/C3N4catalyst can be achieved by applying 1 cycle of Pd ALD on g-C3N4at 150 °C (Fig. 17) with a Pd loading of ～0.7%(w)146. HAADF-STEM confirmed the dominant presence of isolated Pd1atoms without any Pd NPs formation. In selective hydrogenation of acetylene in excess ethylene, the single-atom Pd1/C3N4catalyst exhibited superior catalytic performance of high ethylene selectivity as well as high coking-resistance. In addition to Pd, we also showed that 1 cycle of Co ALD on g-C3N4at 150 °C using bis(cyclopentadienyl)cobalt (Co(Cp)2) and O3as precursors resulted in atomically dispersed Co1atoms on g-C3N4147. EXAFS characterization revealed a formation of Co-N4structure. In photocatalytic H2evolution, the Co1/C3N4exhibited a robust H2production activity up to 10.8 μmol·h-1, 11 times higher than that of pristine counterpart.

Fig. 16 Catalytic performances of Pd1/graphene, Pd-NPs/graphene,Pd-NPs/graphene-500C, and Pd/carbon samples in selective hydrogenation of 1,3-butadiene 145.

Although SACs on carbon-based supports using ALD have been achieved in a number of studies as mentioned previously,synthesis of metal single atoms on oxides using ALD has been rarely reported. Recently, we showed that Pt1single atoms can be synthesized on a CeO2support by exposing the support to MeCpPtMe3at 150 °C followed by calcination at 200 °C in 10%O2in Ar to remove the ligands148. The Pt loading was ～0.2%(w) by applying 1 cycle of Pt ALD on CeO2. STEM images at different locations revealed that isolated Pt1atoms were mainly located on the Ce rows of CeO2(110) facet and on the CeO2(100)facet (Fig. 18). However, it was surprising that no obvious Pt atoms were found on CeO2(111) facet. DRIFTS CO chemisorption revealed a CO peak at 2089 cm-1on Pt1/CeO2(Fig. 18e). XPS measurements showed the Pt 4f7/2binding energy was 72.9 eV on Pt1/CeO2(Fig. 18f), indicating that the oxidation state of the Pt single atoms is between Pt2+and Pt4+.Theoretical calculations on CeO2(110) facet showed that Pt1single atom substitution in the surface Ce vacancy is most stable where the positively charged Pt atom is stabilized by the lattice oxygen via six Pt―O bonds (inset of Fig. 18d).

Fig. 17 Synthesis of single-atom Pd1/C3N4 catalyst using Pd ALD 146.

2.4 Synthesis of dimeric catalyst

Supported metal clusters with only a few atoms are of great interest in catalysis. Therein, one atom variation might drastically alter their electronic structures and catalytic properties. Such “atom-dependent” catalytic behaviors have been unambiguously demonstrated on the model catalysts of mass-selected metal clusters, which were fabricated by physical vapor deposition combined with the mass selection and softlanding techniques under ultra-high vacuum conditions149–152.Unfortunately, such synthesis approach is only limited for model catalysis studies and is hardly applicable to high-surface-area supports for practical applications. Synthesis of atomically precise ultrafine metal clusters, such as dimers on high-surfacearea supports remains grand challenge.

Fig. 18 Single-atom Pt1/CeO2 catalyst synthesized by Pt ALD 148.

Very recently, we reported that Pt2dimers can be “bottom-up”fabricated on a graphene support through proper nucleation sites creation, Pt1single atoms deposition and assembling secondary Pt atom selectively on the preliminary one using ALD (Fig. 19)153.Firstly, we prepared a single-atom Pt1/graphene catalyst at 250 °C on the reduced graphene with isolated phenol groups or phenol-carbonyl pairs as the nucleation sites. The formed isolated Pt1atoms were further utilized as nucleation sites for anchoring the second MeCpPtMe3molecule in the following cycle, which was performed at a lower temperature of 150 °C.O3was used in the second ALD cycle to remove the ligand efficiently. The steric effect and self-limiting surface reactions allow chemisorbing one MeCpPtMe3molecule only on each Pt1atom, ensuring the Pt2dimers formation. HAADF-STEM characterization provided solid evidence of the formation of both Pt1single atoms and Pt2dimers (Fig. 20a–f). Density functional theory (DFT) calculations and XAFS simulations confirmed that Pt1single atoms and Pt2dimers are in the oxidized form of Pt1O4and Pt2O6, respectively (inset of Fig. 20a, d). In hydrolytic dehydrogenation of ammonia borane (AB) for hydrogen generation, we showed that Pt2/graphene exhibited an unprecedentedly high activity of 2800 mol·mol-1·min-1at room temperature, which was about 17- and 45-folds higher than Pt1single atoms and Pt NPs supported on graphene, respectively.More importantly, the Pt2/graphene catalyst was stable under the current reaction conditions and also in the inert environment at below 300 °C. This work demonstrates the great potential of ALD in synthesis of supported metal clusters catalysts with atomic precision for practical applications.

3 Nanoscale engineering of metal catalysts via oxide overcoating

Fig. 19 Schematic illustration of bottom-up synthesis of dimeric Pt2/graphene catalysts 153.

Fig. 20 Morphology of the single-atom Pt1/graphene and dimeric Pt2/graphene catalysts 153.

For supported metal catalysts, the support is not only a support, but also significantly modulates the electronic property and morphology of metal NPs via electron transfer and metaloxide interaction, respectively. Moreover, the support, especially a reducible support, such as TiO2, FeOx, CeO2etc. can also participate in catalytic reactions at the metal-oxide interface.Consequently, the metal catalyst activity depends tightly on the support, which is so-called support effect5. When an oxide overcoating layer was applied onto metal NPs, the electronic properties, morphology of metal NPs can be further altered,along with the formation of new metal-oxide interfaces. Thus,oxide overcoat on metal NPs provides additional opportunities to further promote the catalytic performance7. Moreover, the physical oxide overcoat also improves the stability of metal NPs against sintering and leaching under harsh reaction conditions.The atomic-level precise control over oxide film thickness with high uniformity on high-surface-area materials makes ALD to be an ideal tool to engineer metal catalysts in the nanoscale for improving catalytic performance, without generating masstransfer issues. In this section, we summarize the strategies of tailoring the catalytic performances of metal catalysts in terms of activity, selectivity and stability using ALD oxide overcoating.

3.1 Boosting activity

Metal-oxide interfaces play very crucial roles in a wide range of catalytic reactions, such as CO oxidation154,155, water gas shift reaction156, benzyl alcohol oxidation157, methanol synthesis158,159,etc. While for supported metal catalysts, the interfacial sites are limited to the perimeter sites along the metal-support contact, as illustrated in Fig. 21. Oxide deposition on metal NPs allows creating abundant new metal-oxide interfaces, thus enhancing the catalytic activity remarkably. In one early example, Kim and co-workers62reported that TiO2ALD on Ni powders using titanium tetraisopropoxide (TTIP) and H2O as precursors 180 °C,created TiO2nano-islands on the Ni surface. They found that these TiO2nano-islands increased significantly the catalytic activity in CO2reforming. Moreover, the carbon deposits were also suppressed considerably to extend the catalyst life.

Fig. 21 Oxide overcoating on supported metal catalyst to create new metal-oxide interfaces.

Elam et al.160investigated the ZnO-promotion of Pt catalysts in aqueous phase reforming of 1-proanol using ZnO ALD. They prepared two types of ZnO-promoted catalysts: i) Depositing ZnO on Al2O3as a supported layer before Pt ALD to obtain a Pt/ZnO/Al2O3catalyst; ii) Depositing ZnO after Pt ALD to form a ZnO/Pt/Al2O3catalyst, as shown in Fig. 22. They found that Pt/ZnO/Al2O3had a better H2selectivity than Pt/Al2O3, while ZnO/Pt/Al2O3improved both the activity and selectivity considerably, by yielding a H2formation rate of 0.44 mol·mol-1·min-1, higher than both Pt/Al2O3and Pt/ZnO/Al2O3(Fig. 22f). The ZnO/Pt/Al2O3catalyst also showed a good stability (Fig. 22g). They suggested that the improved reactivity and selectivity were due to having both Pt-ZnO and Pt-Al2O3interfaces. These results suggest that construction of multiple metal-oxide interfaces by ALD is important for optimizing catalyst performance.

Qin et al. reported a strategy of multiple confinements for catalyst activity enhancement, therein carbon nanocoils was used as sacrificial templates, and Ni NPs was grown was in Al2O3nanotubes (ANTs) by ALD to maximize the metal-oxide interfaces161. In selective hydrogenation of cinnamaldehyde(CA), they found that the Ni-in-ANTs catalyst exhibited a ～5 times higher activity than Ni-out-ANTs (Fig. 23a), along with a higher reusability (Fig. 23b). TEM and ICP-AES analysis revealed that most of Ni NPs are still confined in the ANTs for Ni-in-ANTs after the fourth run (Fig. 23d). However, Ni NPs in Ni-out-ANTs were significantly leached after the fourth run, and caused the catalyst deactivation (Fig. 23f). Temperature program reduction and XAFS revealed that the increased Ni-Al2O3interfacial sites are responsible for the greatly enhanced catalytic activity. Such enhancement of hydrogenation activity by metaloxide interfaces were also found in ALD FeOx-coated Pd/Al2O3catalysts162.

Recently, Chen et al.163developed another strategy of construction Co3O4nanotraps to create Pt-Co3O4interfaces using AS-ALD. Therein, 1-octadecanethiol (ODT) was used as the block agent to protect Pt NPs to be covered by the subsequent Co3O4deposition. Thereafter, Co3O4(50 cycles) was deposited onto the Al2O3supports using AS-ALD. Finally, the Co3O4nanotraps are formed after removing the ODT blocking agents via calcination in air, re-exposing the Pt sites and the newly formed Pt-Co3O4interfaces (Fig. 24a). In CO oxidation, they showed that the T50(the temperature at which CO conversion is 50%) of Co3O4/Pt/Al2O3is lowered by about 49 °C compared with that of Pt/Al2O3catalyst (Fig. 24b). In contrast, the Co3O4@Pt/Al2O3catalyst without using the ODT tapping agent exhibited a much worse activity due to the blocking of Pt NP active sites by the oxide overlayer. Their work suggests formation of the nanotraps is the key to keep Pt NPs exposed and accessible for catalytic function. In addition, they showed that this material had excellent sintering resistance up to 600 °C calcination.

Owing to the extraordinarily high activity in the CO oxidation reaction, Au/TiO2catalysts have been extensively investigated over decades164. Nonetheless, the nature of active sites has been controversial, both the Au low-coordination sites (LCSs) and Au-TiO2interfaces have been suggested to be the active sites.One major reason is that changing the Au particle size to vary the fraction of Au LCSs inevitably accompanied by the change of the Au-TiO2perimeter length makes it extremely difficult to identify their individual roles165.

Fig. 22 ZnO-promoted Pt catalysts 160.

Fig. 23 Catalytic performance of multiple confined Ni catalyst in selective hydrogenation of cinnamaldehyde 161.

Fig. 24 Co3O4 nanotrap-anchored Pt NPs on Al2O3 supports based on AS-ALD 163.

Recently we reported a new strategy to isolate them by applying TiO2overcoat to Au/Al2O3and Au/SiO2catalysts via ALD where the new Au-TiO2interfacial length was precisely tuned at a wide range while preserving the particle size. TEM,atomic force microscopy (AFM) and DRFITS CO chemisorption all confirmed that the TiO2overcoat preferentially decorated the low-coordinated sites of Au NPs and generated Au-TiO2interfaces. In CO oxidation, we demonstrated a remarkable improvement of the catalytic activities of Au/Al2O3and Au/SiO2catalysts by the ALD TiO2overcoat165,166. Where T50was significantly decreased from 225 °C to room temperature (Fig.25), although the LCSs of Au NPs were blocked by the TiO2overcoat. Therefore, this work provides solid evidence that the Au-TiO2interface is the active site, rather than the lowcoordination Au atoms. TiO2ALD was further applied to Au/TiO2catalysts with different Au particle sizes of (2.9 ± 0.6),(5.0 ± 0.8) and (10.2 ± 1.6) nm. In CO oxidation, we found that the CO conversion on Au/TiO2-2.9 nm with 50 cycles of TiO2overcoat was 47% at 298 K, still impressively larger than the uncoated Au/TiO2-5.0 nm (22%) and Au/TiO2-10.2 nm (13%),even though the population of the exposed low-coordination Au sites was about two orders of magnitude less than the latter two(Fig. 26). This result clearly demonstrates that the Au particle size effect in CO oxidation is not due to the size-related changes in the number of Au LCSs either. The perimeter sites at the Au-TiO2interface could certainly play a dominant role in the Au particle size effect. Nonetheless, the contributions from the quantum-size effect and strain effect cannot be excluded out,based on our results.

These results demonstrate flexibility of constructing nanostructured metal catalysts at the atomic level using ALD.Applying oxide overcoat on metal NPs provide an efficient approach to optimize the metal-oxide interfacial effect in catalysis. The above methods can be general, and have been extended to other systems (Pt, Pd, Ru, Fe, Co, Cu, etc.)167,168.Specially, selective blocking the LCSs of metal NPs with oxide ALD provides a very useful tool to discriminate the role of lowcoordination metal sites in catalytic reactions.

3.2 Enhancing selectivity

Fig. 25 Plot of the reaction temperature for 50% CO conversion as a function of the number of TiO2 ALD cycles 165.

Fig. 26 CO conversion as a function of the number of exposed low-coordination Au sites on Au/TiO2-2.9 nm, xc-Au/TiO2-5.0 nm and xc-Au/TiO2-10.2 nm catalysts with different ALD TiO2 overcoating at 298 K 166.

Improvement of catalyst selectivity to desired products is highly preferable in the 21st century, which not only reduces the cost for the later products separation but also is friendly to the environments by reducing the industrial wastes3,28. For supported metal catalysts, metal NPs often contains LCSs and high coordinated terrace sites (HCSs), these different local geometries change considerably chemical bond breaking and making in catalysis, leading to different reaction pathways169–174.In structure sensitive reactions, selective blocking LCSs or HCSs while only leaving the other type of metal sites participating in reactions would improve the catalyst selectivity27,28,105,175.

Elam and co-workers176reported the first evidence of siteselective blocking on the ALD Al2O3overcoated Pd/Al2O3catalyst, where Al2O3ALD was performed at 200 °C using trimethylaluminum (TMA) and water as precursors. DRIFTS CO chemisorption found that after applying ALD Al2O3overcoating on Pd NPs, the bridged-bonded CO on LCSs of Pd NPs at 1973 cm-1disappeared much more quickly than the one on HCSs at 1933 cm-1175,176. Later, scanning tunneling microscopy (STM) and DFT calculations177further confirmed the preferential decoration of ALD Al2O3on Pd LCSs. In oxidative dehydrogenation of ethane (ODHE) at 675 °C, Stair et al.175reported that bare Pd/Al2O3catalyst suffered from heavy coke formation and severe metal sintering, which led to a very short catalyst life within 30 min and a low yield of ethylene, in agreement with Huff and Schmidt’s report178. After coating 8 nm Al2O3(45 cycle Al2O3ALD), a long-term yielding of ethylene more than 1,800 minutes can be achieved (Fig. 27)175.LCSs were recognized to be responsible for coke formation179and C1products (CO, CO2, CH4) through a C―C bond session of ethane. By selectively blocking these energetic Pd LCSs, both coking and C―C bond session pathways was effectively inhibited, thus improving the ethylene yield by more than 10 times. Moreover, the 8 nm Al2O3overcoat also improved the catalysts thermal stability against sintering under the extreme reaction conditions at 675 °C. To the best of our knowledge, it was the first example achieving both coking- and sinteringresistance metal catalysts.

Fig. 27 Products yield on the Pd/Al2O3 samples with and without ALD Al2O3 overcoat during ODHE reaction as a function of reaction time under identical reaction conditions 175.

By adopting the same strategy, Marshall and co-workers investigated ALD alumina overcoated Pd/Al2O3catalysts in the selective hydrogenation of furfural180. They found that the selectivity to furan increased with increasing the HCSs to LCSs ratio. DFT calculations revealed that the furfural hydrogenation to furfuryl alcohol on step sites is more thermodynamically favorable than on terrace sites, whereas the furfural hydrogenation to furan on step sites is less favorable than on terrace sites.

Besides on Pd, Al2O3ALD was also found to preferentially grow on the LCSs of Cu and Ir metals177. In addition to Al2O3overcoating, several new strategies have been reported to achieve selective blockage181–184. For example, based on DRIFTS CO chemisorption and DFT calculations, Qin and coworkers reported that Fe2O3ALD preferred nucleating on Pt LCSs184. In selective hydrogenation of CA, the selectivity of cinnamyl alcohol (COL) was significantly improved from 45%to 84% after applying 30 cycles of Fe2O3overcoating (Fig. 28).While further increasing the Fe2O3cycles decreased the COL selectivity. The decrease of selectivity at high Fe2O3coverages was attributed to the decreased Pt-Fe2O3interfacial sites, since continuous Fe2O3coating films were formed at high coverages.

Besides site-selective blocking, the micropores formed within the ALD oxide overcoating layer induces the confinement effect,thus having a considerable impact on selectivity6,27,185–187. For example, on a Pd/Al2O3catalyst with 20 cycles of ALD Al2O3overcoat, we observed a pore size of ～1 nm in diameter188. The pore size increased to 2–4 nm after calcined at 350 °C for 1h.Similarly, ～2 nm pores were also formed within an 8 nm thick ALD Al2O3overcoat after calcination at 700 °C for 2 h175. The confinement effect induced by the micropores can be utilized for tailoring the catalytic performance of Pd catalysts. In selective hydrogenation of 1,3-butadiene, we demonstrated that at high conversions, the selectivity to all butenes increased remarkably with increasing the ALD alumina overcoat thickness. After performing 30 cycles of ALD Al2O3on a Pd/Al2O3catalyst(30Al/Pd/Al2O3, an overcoat thickness of 3.8 nm, inset of Fig.29a), the selectivity to all butenes was almost 99% at a conversion of 95% (Fig. 29a). Such Al2O3overcoated catalyst exhibited comparable selectivity to Au catalysts, but still preserved the high activity of Pd189,190. Compared to trans-2-butene and cis-2-butene, the improvement of 1-butene selectivity with increasing overcoat thickness was the most pronounced (Fig. 29b). High temperature pretreatment of ALD alumina coated Pd samples formed larger pores (greater than 2 nm in diameter), resulting in a large decrease of selectivity to butenes. This confirms that the confinement effect plays the major role in the selectivity enhancement.

Fig. 28 Selectivity to cinnamyl alcohol over various Pt catalysts with different cycles of Fe2O3 overcoating with reaction time in selective hydrogenation of cinnamaldehyde 184.

Fig. 29 Catalytic performance of the Pd/Al2O3 samples with and without ALD alumina overcoat in the absence of propene 188.

In addition, Canlas and co-workers also reported a shapeselective catalyst by using AS-ALD191. As shown in Fig. 30, a template molecule was grafted onto the surface to block the TiO2active sites underneath. A thin Al2O3coating was then precisely constructed at the surroundings of template molecule using ALD. Finally, ozone was used to remove the organic template molecule. A sieving layer of Al2O3(thickness, 0.4–0.7 nm) with“nanocavities” (＜ 2 nm in diameter) on a TiO2photocatalyst was then created. The resulting material had a higher selectivity for the photooxidation of benzyl alcohol versus 2,4,6-trimethylbenzyl alcohol due to the shape-selective effect.

3.3 Improving stability

Sintering and leaching of supported metal catalysts often lead to catalyst deactivation, especially under severe reaction conditions. These are one major issues of supported metal catalysts in applied catalysis192–196. Improving the stability of metal NPs against sintering and leaching under reaction conditions has drawn tremendous attentions. Encapsulation of metal NPs within porous oxide layers is the most commonly used methods197–199. In this case, the thickness of the protective shell can be very critical for optimizing catalyst stability without decreasing the activity largely owing to the mass-transfer issue.

Fig. 30 Schematic illustration of the synthesis of shape-selective catalysts by oxide ALD 191.

Fig. 31 Proposed schematic model of Co/TiO2 catalysts with and without ALD TiO2 decoration on edge 204.

ALD enables atomically-precise control over the thickness of protective layer, thus it has been demonstrated to an ideal approach for applying oxide overcoating, as we mentioned previously27,28,167,175,188,200,201. Dumesic and co-workers195reported that 45 cycles of ALD Al2O3overcoat on a Cu/Al2O3catalyst could effectively prevent sintering and leaching of the underlying Cu NPs in the liquid phase furfural hydrogenation.They showed that the ALD Al2O3preferentially covered the Cu LCSs, which mainly contribute to sintering and leaching. In their later work, they decreased the number of ALD cycles from 45 to 5 and achieved a higher activity while still maintaining stability202.In addition to Cu catalysts, Au catalysts always suffers from severe deactivation through sintering and leaching due to the relative weaker metal-support interaction. We showed that ultrathin Al2O3overcoat by ALD can remarkable enhance stability of Au NPs up to 600 °C203. These stabilized Au NP catalyst might be useful for applications under harsh conditions, such as automobile exhaust treatment and water gas shift reaction33,34.

Besides ALD Al2O3overcoat, ALD TiO2, ZrO2, SiO2have been also demonstrated for stabilizing NPs against sintering and leaching. Huber and co-workers204recently showed that TiO2overcoat can effectively stabilize Co/TiO2catalysts in liquidphase hydrogenation of biomass feedstock under harsh conditions (Fig. 31). After high temperature calcination, they found that the Co nanoparticle size can be preserved and the oxide overlayer cracked to form ～16 nm large pore. Due to the TiO2overcoat protection, the Co catalysts exhibited excellent stability for against leaching and coking under liquid-phase reaction, much more superior than traditional non-ALD Co catalysts, such as Co/Al2O3and Co/TiO2, as shown in Fig. 32.Moreover, the controllable pore size of the TiO2overlayer allowed to optimize exposure of Co active sites, leading to a high activity for reactions. Gorte and co-workers also reported a ～1 nm ZrO2overcoat (50 cycles of ZrO2ALD) surprisingly stabilized the Pd/Al2O3catalyst under harsh thermal treatment in methane oxidation reaction205. After 800 °C calcination, the ZrO2overcoat cracked to pores leaving Pd sites exposed, which exhibited a higher and stable methane conversion while the uncoated Pd/Al2O3showed obviously deactivation and sintering.

3.4 Bifunctional catalysts design

Bifunctional catalysts have attracted great attentions in many practical reactions206–209. Dumesic and co-workers210demonstrated that applying acidic niobium oxide onto ALD alumina coated copper catalysts, significantly improved the activity of furfural hydrogenation. Recently, we also reported acidic ALD Al2O3overcoating on Pt NPs strongly enhanced the metal-acid proximity thus boosting the synergistic effect. In glycerol hydrogenolysis, we showed that the increased metalacid proximity induced by alumina overcoat improved the activity by about 3 times higher than the uncoated Pt/Al2O3sample, even though the total acidity were not changed after overcoating.

Qin and co-workers211reported another type of bifunctional catalyst with multiple interfaces (Ni/Al2O3and Pt/TiO2) using template-assisted ALD for tandem catalysis. First, an Al2O3layer and Ni NPs were deposited on a carbon nanocoil sacrificial template. Second, a sacrificial polyimide layer was coated.Finally, Pt NPs and a TiO2layer were deposited. After calcination and reduction treatments, the confined tandem catalyst (Al/Ni-Pt/Ti) was obtained, therein Ni NPs were highly dispersed on the outer surface of the Al2O3nanotubes (Ni/Al2O3interface), and Pt NPs were confined in the TiO2nanotubes(Pt/TiO2interface). This Al/Ni-Pt/Ti bifunctional catalyst exhibited extremely higher activity in hydrogenation of nitrobenzene using N2H4·H2O as a hydrogen source compared to the individual Al/Ni and Pt/Ti catalysts as well as their physically mixtures (Fig. 33). They found that N2H4·H2O catalytically decomposed at the Ni/Al interface and produced active H*,which quickly transferred to the Pt/Ti surface to process highly active hydrogenation of nitrobenzene. The superior catalytic performance was attributed to both the synergistic effect between the two bifunctionalities and the proximal space between two catalytic interfaces for the quickly H* transfer.

These results demonstrate constructing two types of active sites proximally with a controllable structure at an atomic-level is crucial for bifunctional catalyst design. The ability to add bifunctionality with atomic precision can be an important and promising future area in heterogeneous catalysis.

Fig. 32 Percent of cobalt leached as a function of time on stream for aqueous-phase hydrogenation of furfural alcohol 204.

Fig. 33 The catalytic performance of different catalysts 211.

4 Conclusions and Outlook

Catalyst precise synthesis is essential for unravelling catalytic mechanism and ultimately for rational design of advanced catalysts for high performance. The unique surface self-limiting character of ALD allows conformal deposition of catalytic materials on high-surface-area support to construct well-defined architecture catalysts in a step-by-step “bottom-up” fashion at an atomic level. In this review, we demonstrated ALD is facile method to synthesize supported metal catalysts on high-surfacearea supports along with precise control over the particle size,composition and structures. By optimizing the ALD conditions,and surface nucleation sites, it has been proven to be possible to synthesize metal SACs using ALD at elevated temperatures.Importantly, the selective-deposition strategies developed, have greatly facilitated the application of ALD in catalyst synthesis,by providing additional flexibility to construct distinct nanostructures on a support at the atomic level. Among them,deposition of secondary metal atoms selectively on primary metal particle surface but not on the support, allows synthesis of well-defined bimetallic catalysts without having any mixture with monometallic NPs. Moreover, the successful bottom-up synthesis of Pt2dimers on high-surface area graphene support further sheds lights on the great potential of ALD for synthesis of subnanometer metal clusters with atomic precision. These ALD catalysts with high uniformity render them often exhibiting better or comparable catalytic performances compared to the corresponding ones prepared by the traditional methods.Fundamentally, these well-defined ALD materials could be also used as model catalysts to facilitate our understanding structureactivity relationships.

Precisely depositing oxide overcoat on metal catalysts using ALD provides another strategy to further improve their catalytic performances in terms of activity, selectivity and stability.Construction of enriched metal-oxide interfaces and addition of bifunctionality by functional oxide overcoat have been demonstrated to be the effective way to enhance the catalyst activity. On the other hand, site-selective blocking of metal NPs and confinement effect can be very useful for improving the catalyst selectivity. Finally, improving the stability of metal NPs by oxide overcoat is quite straight forward, while the thickness of oxide overcoat might be critical.

So far, all ALD catalyst synthesis has been carried out in the lab scale, although numbers of commercially-used ALD reactors and more than 1000 ALD processes have been developed. For the commercial application of ALD in catalyst synthesis is still very challenging. The major challenges are to reduce the cost for catalyst synthesis and to solve the technical issues for scalingup. Developing cheap ALD metal precursors is one important way to reduce the cost. In this case, transitional metal and/or metal oxide ALD such as Fe, Co, and Ni, might have much higher possibilities for scaling up, since their metal precursors are much cheaper. A combination of ALD with the conventional wet-chemistry methods can be very desirable for synthesis of nanostructured catalysts with low cost. Moreover, performing ALD at atmosphere pressure using a fluidized-bed reactor with a batch type process, or spatial ALD with continuous operation can be further beneficial to decrease the cost, and to simplify the requirements for the reactor design, thus facilitate the development of scaling-up process.