Review Article: Catalysts design and synthesis via selective atomic layer deposition

Catalysts are widely utilized to accelerate chemical reactions by decreasing reaction barriers in various industrial syntheses, environment pollution control, energy conversion, and so on. Composite catalysts based on metal-oxide with designed structures perform an irreplaceable role for most applications.^1–4 The catalytic performance of composite catalysts strongly depends on their size, heterointerfaces, active sites, etc.^5–7 To design and obtain catalysts with the desired activity, selectivity, and stability, synthesis strategies to build precise configurations with direct modulation of reactive sites are of great importance. Among various synthesis methods, atomic layer deposition (ALD) has been recently developed as an effective method to synthesize composite catalysts.^8–14 ALD is based on successive and alternative surface reactions from gas phase to fabricate thin films and overlayers in the nanometer range. So far, significant number of elements and their oxides can be synthesized via ALD.^15–17 Taking the advantage of self-limiting surface adsorption nature of ALD, the target materials can be deposited with controllability and uniformity in atomic level. Thus, it enables direct modification of the surfaces and structures,¹⁸ as well as adjustment of the shape and size of materials deposited on complex substrates.

The use of ALD for synthesis of heterogeneous catalysts has been developed rapidly in the past few years.^19,20 For instance, ALD was utilized to fabricate highly dispersed, size controllable metal nanoparticles, such as Ru, Pd, Pt, Ir, and Ni, during the nucleation stage.^21–25 Recently, ALD has shown its potential in synthesizing single atom catalysts as well.^26–29 By adjusting different ALD processes, it is possible to fabricate core–shell structured or alloys nanoparticles with well-defined compositions ratio.³⁰ Besides metal ALD processes, ALD of oxides has demonstrated a great potential in preventing catalysts from sintering. The well-engineered oxide overcoating layers could encapsulate supported nanoparticles to enhance the catalytic performance in both thermal stability and activity.³⁰ In detail, the oxide coating layers could serve as physical barriers, or form strong metal-oxide interactions to anchor metal nanoparticles. The heterointerfaces between metal and oxides are also beneficial for activity enhancement. So far, the development of ALD for the synthesis of catalytic materials (metal and oxide) has aroused great interests in the design and controllable fabrication of unique catalytic structures.

In fabrication of composite catalysts, the selective approaches of ALD are of importance and necessary. Area-selective ALD has been developed to exert spatial control to fabricate 3D nanostructures. A patterned area of substrate is initially activated or passivated with assistance of electron/ion beam, self-assembled monolayers (SAMs), or polymer resists.^31,32 Then during the deposition, materials can be deposited only where needed. The area-selective ALD has been applied to fabricate defect-free 3D patterns and nanostructures for electronic applications.^32,33 In catalysis, the motivations of selective-ALD enable directionally and precisely tailoring of the structural parameters, interfaces, and active sites, that is of great significance for advanced catalysis. In this article, fabrication of composite catalysts via selective ALD methods is introduced. Figure 1 shows the classification of catalytic structures discussed in this review, including core shell structures, discontinuous coating structures, and embedded structures. The preparation of core shell nanoparticles with three strategies and their applications will be reviewed initially.^34–36 Consequently, oxide overcoating catalysts fabricated via selective ALD are discussed. The discontinuous coating structures range from random porous coating to more ordered structures, such as selective deposition on edges sites or facets are highlighted in this section.^37–42 Finally, by applying multistep approaches such as selective blockage, nanostructured templates, the nanotrap or nanoparticles embedded structures can be fabricated.^43–46 In conclusion, a comprehensive summary and outlook are elaborated.

Core shell nanostructures are important categories of catalytic structures. Their properties are facile to tune through core-shell interactions and choice of surface materials. Compared with one-component catalysts, the formation of core shell structures can enhance the activity, selectivity, and stability^47–49 through the lattice strain, bonding interactions, and electron transfer.^50–52 At the same time, replacing the precious metal with a nonprecious metal as cores is commercially favorable.⁵³ In this part, bimetallic core shell nanoparticles fabricated via selective ALD are introduced as examples.

A challenge for synthesis of core shell structured nanoparticles is the control of second metal to deposit exclusively on the primary metal cores but not on the substrate.⁵⁴ In general, metal ALD processes share very similar surface chemistry. The second metal is likely to deposit on the first metal as well as form new nuclei on the substrate, which leads to the formation of mixture of monometallic with core shell structured nanoparticles. Several approaches to achieve selective deposition for core shell nanoparticles have been developed, including ALD processes adjustment from precursors' partial pressure reduction, deposition temperature reduction, and substrate modification assisted with SAMs.

During the ALD reaction, the growth rate of metal nanoparticles is strongly depended on the chemisorption of precursors with the substrate.¹⁰ For example, the growth of Pt with the precursors of (methylcyclopentadienyl)-trimethylplatinum (MeCpPtMe₃) and O₂ is affected by the dissociation of chemisorbed O₂. O₂ dissociation usually prefers to take place on a metal surface, leading to the formation of active oxygen atoms to decompose and activate MeCpPtMe₃ molecules for the following reactions.^55–57 Kessels and coworkers reported that when the O₂ partial pressure decreased to 7.5 mTorr, the growth of Pt on Al₂O₃ was inhibited.³⁶ No Pt growth was observed on Al₂O₃ surface even after 600 cycles. While under such condition, Pt could still initiate its growth on Pd without delay and form core shell structure exclusively. The dissociation of chemisorbed O₂ on the catalytic Pt group metal was a key step in the selective ALD mechanism.^55,57 The Pt ALD process at 7.5 mTorr could be considered as selective condition to Pd versus Al₂O₃ substrate. Figure 2 shows the high-angle annular dark-field scanning transmission electron microscopy (HAADF-STEM) image and electron energy loss spectroscopy (EELS) line scan to confirm the Pd@Pt core shell structure fabricated with this selective ALD method. The density of nanoparticles before and after the Pt ALD was similar, indicating that Pt mainly deposited on the Pd surface but not on the substrate to form new monometallic Pt nuclei. In a following study, they demonstrated this selective ALD method enabled to prepare core shell nanoparticles on high surface area support.⁵⁶ The substrate was high density array of Al₂O₃ covered GaP nanowires that were interest for fuel cells' or sensors' applications. Along with reducing the reactant partial pressure, the choice of coreactant could also facilitate selective ALD for core shell nanoparticles fabrication. In this situation, the dissociation of precursors played a key role. For instance, using H₂ instead of HCHO as the coreactant for Pd ALD effectively suppressed Pd ALD on Al₂O₃, TiO₂, or ZrO₂ surfaces at 150 °C, as H₂ was hard to dissociate and activate these oxides substrates, while maintaining Pd growth on Pt surfaces to form Pt@Pd core shell structure at the same deposition condition.³⁵ 

Controlling the growth temperature of shell materials is another way to form core shell nanoparticles, because the nucleation rate of metallic ALD is strongly depended on deposition temperature. When lower the deposition temperature, the growth of metal on oxide substrates usually exhibits a longer nucleation delay or even no apparent growth.⁵⁸ However, the nucleation on another metal surface usually has lower energy barrier and could be initiated without delay. Thus by adjusting a proper deposition temperature, selective deposition of metal on another metal surface but not on oxide can be achieved.

Lu and coworkers performed Pd ALD on a high surface area silica support using Pd(hfac)₂–H₂ at 150 °C and measured no Pd content after 8 ALD cycles.⁶⁶ In contrast, the Pd loadings increased considerably on an Au/SiO₂ catalyst with the number of Pd ALD cycles at the same deposition condition. Evidently, the Pd grew selectively on Au nanoparticles but not on the SiO₂ support and exclusively formed Au@Pd core shell structure. The Pt@Pd core shell nanoparticles could be synthesized by a similar process.⁵⁹ As compared to the physical mixture of monometallic ALD Pt and Pd catalysts, the Pt@Pd core shell nanoparticles showed higher selectivity and yield to propylene in propane oxidative dehydrogenation.

The reactivity of the precursors especially of the organic ligands within precursors also influences the initial selective growth behavior. As an example, the selective deposition of Cu on Pd and Pt metal surfaces using Cu(thd)₂–H₂ was reported.⁶⁰ It was found that the temperature window for selective Cu growth ranged from 190 to 260 °C. At higher temperatures, the selectivity was lost without distinguishing substrate materials. Selective growth was explained by the different reaction mechanisms of Cu(thd)₂ on hydroxylated oxide substrate versus metal surfaces. On the hydroxyl-terminated oxide substrate, Cu(thd)₂ reacted with the hydroxyl group and one of the thd ligands was removed as a volatile H(thd) molecule. The second ligand could not be removed at lower temperatures due to poor reactivity; thus, the deposition process on oxide substrate was inhibited. However, Cu(thd)₂ could dissociate on metal surfaces, as a consequence, selective deposition of Cu was achieved on Pd and Pt metal surfaces but not on the oxides supports.

Through an ABC type of ALD approach, it is possible to fabricate well-mixed bimetallic alloy nanoparticles at the same deposition temperature. The alloys can be accomplished by alternating between the ALD sequences for individual metals. For example, Pd/Pt alloys could be fabricated by alternately performing Pd and Pt ALD cycle. The sequence was performed with MeCpPtMe₃-O₂-H₂-Pd(hfac)₂-H₂-O₂ at temperature of 150 °C. By varying the deposition cycles of two metal precursors, it was facile to regulate the composition of bimetallic nanoparticles.^35,61 Figure 3 is high-angle annular dark-field TEM images and corresponding energy dispersive x-ray spectroscopy line scans for Pd@Pt, Pt@Pd. Ru@Pt core shell nanoparticles and Pd/Pt alloys.

The previous two approaches are based on the adjustment of ALD processes, such as deposition temperature, precursor partial pressure reduction. Modifying substrate assisted with SAMs is another way to fabricate core shell nanoparticles and avoid changing ALD process parameters. We reported using octadecyltrichlorosilane (ODTS) SAMs to selectively passivate the active sites on substrate to achieve core shell nanoparticles.³⁴ Figure 4 is a schematic illustration of the synthesis strategy of the Pd@Pt core shell nanoparticles exploiting the ODTS SAMs modified substrate. The SAMs were formed with nanoscale pinholes which exposed the hydroxyl (–OH) group of the substrate.^62–64 Other parts were covered with closely packed ODTS molecules with inert methyl (–CH₃) end groups.⁶⁵ During the metal cores formation, the nucleation of Pd took place in the pinhole sites. During the second metal deposition, the Pt precursors could selectively bind to Pd cores, but not on the closely packed SAMs layer, resulting in the formation of Pd@Pt core shell nanoparticles. In comparison, the nucleation of Pt on the substrates could hardly be avoided on a bare oxide substrate without introducing adjustment of ALD processes, leading to a mixed phase of Pt, Pd, and Pd@Pt nanoparticles. For fabricating the Pd@Pt nanoparticles, it was found that the Pt shell growth rate on Pd is liner at 82 ng cm⁻²cycle⁻¹, which corresponded to approximately 0.15 Pt atomic monolayer (ML) per cycle. The liner control of shell thickness was also confirmed with particles size distribution measured from TEM.¹³ 

The above mentioned ALD methods enable the precise control of nanoparticles' size, composition, and shell thickness by adjusting the ALD temperature and number of cycles. For example, with fabricating of Pd@Pt core shell nanoparticles, the average diameter of the nanoparticles (NPs) increased linearly with the number of Pt ALD cycles, implying that the Pt shell thickness grows at a constant rate. Weber et al. demonstrated that when fabricating Pd@Pt core shell nanoparticles, the Pt shell thickness growth rate was 0.03 ± 0.01 nm/cycle corresponding to ∼0.14 Pt atomic monolayer per cycle.⁵⁶ The results were similar to our findings.¹³ In addition, the ALD methods are versatile since they allow for independent tuning of the core and shell diameter. The size distribution can be controlled to be unimodal and stay narrow, which is of particular interest for many catalytic applications, since a precise definition of nanoparticles' size and shell thickness allows for optimizing the catalytic activity and/or selectivity.

The advantages to control the core shell structural parameters precisely open up opportunities to study the fundamental insights with catalytic performance directly. Lu and coworkers reported that for the Au@Pd core shell catalysts, the catalytic activities toward solvent-free aerobic oxidation of benzyl alcohol showed a volcanolike trend as a function of Pd shell thickness.⁶⁶ Small coverage of Pd adatoms on Au@Pd had a lower activity than continuous Pd shells. The turnover frequency (TOF) and specific activity gradually increased with Pd shell thickness, and reached maximums at 27 600 and 9800 h⁻¹ on the Au@Pd core shell catalysts with 8 cycles Pd (shell thickness ∼0.8 nm). Then, the TOF and specific activity both decreased as further increasing Pd shell thickness. The results indicated that aerobic oxidation was a structure-sensitive reaction. The enhanced activities could be attributed to synergistic effect from electronic modification that Pd 3d level drew electrons from Au in the bimetallic system. For preferential oxidation of CO under excess H₂, we reported that the Pd@Pt core shell catalysts showed enhanced activities compared with monometallic Pd, Pt catalysts and the well-mixed Pd/Pt alloys. The Pd@Pt core shell catalysts with shell thickness around ∼1 atomic ML of Pt showed optimal activity and selectivity for CO oxidation [Figs. 5(a) and 5(c)]. From density function theory (DFT) calculation and Arrhenius slops [Fig. 5(b)], the barrier/activation energies of CO oxidation significantly decreased upon Pt coating, and a monolayer decoration showed the lowest barrier energy value.¹³ 

The core shell structure is to form continuous and uniform coating on the target materials,³⁰ discontinuous coating of oxides on catalysts is another type of structure to realize the sintering resistance for dispersed metal nanoparticles catalysts. The early demonstrations of oxide overcoating techniques originated from sol gel, CVD methods, etc.^67,68 For the overcoating catalysts, the thickness and configuration of the coating layer are of great significance. The catalytic activity often suffers as the nanoparticles are fully covered with thick oxide layers which prevent the reactants from contacting with active sites. Discontinuous coating structure is desired to maintain the access of reactants to metal nanoparticles.⁶⁸ Generally, there are several ways to tune the oxide coating configurations with selective ALD approaches. The structures can be divided into random porous coating structures and more ordered coating configurations, such as selective passivation on edge/low coordinated sites or facets.

The porous catalytic structures fabricated via ALD mainly come from two strategies. The first way is to calcinate the oxide coated catalysts. The dehydrogenation or oxidation of residual carbon contaminations creates nanotunnels in the oxide layers. In calcination, pores can also be formed through the relaxation of stress induced by densification of the ALD films during phase transitions/crystallization.⁷⁰ As an example, Stair and coworkers showed that with 8-nm mesoporous coatings on Pd nanoparticles could enhance and sustain catalytic activity and selectivity toward dehydrogenation reaction. The initial formed 8-nm Al₂O₃ overcoating was continuous, thermal treatment to the coated catalysts under oxidative environment was used to re-expose active sites.^69,70 The ∼2 nm pores were formed by dehydrogenation from the carbon residues in the Al₂O₃ coating layers. The coated catalysts showed high selectivity of ethylene yield (23%) in the oxidative dehydrogenation of ethane reaction at 675 °C, the yields of undesired CO, CO₂, and CH₄ were suppressed at 5.1%, 3.9%, and 0.9%. The coke formation was significantly reduced by 94%. At the same time, Al₂O₃ coating layer provided physical barrier that inhibit sintering of Pd particles to improve long term stability. While for uncoated catalysts, substantial deactivation of Pd was due to heavy coking by filamentous carbon and particle sintering.

The second way to create porous structures utilizes the initial island growth stage of oxide on metal. It was found that ALD of Al₂O₃ at the island growth stage in the first few cycles formed discontinuous coating films and expose part of surface sites of metal nanoparticles to the reactants.^39,71 Figure 6 demonstrates the TEM images of Al₂O₃ films grown on Pd nanoparticles with increasing cycles, a schematic diagram besides shows the porous coating structures. It was found that the thickness of Al₂O₃ coating layers grew linearly with the number of ALD cycles (0.16 nm/cycle). The porosity in the Al₂O₃ overcoats was confirmed with in situ quadrupole mass spectrometry, carbon monoxide-diffuse reflectance infrared Fourier transform spectroscopy (CO-DRIFT) characterizations. From Brunauer-Emmertt-Teller measurements, the pore size was indicated to be around 1 nm for Al₂O₃ layer fabricated with 20 cycles. When the thickness of coating layer was thick, the porous structures disappeared and coating films became continuous. So far, the porous films fabricated via ALD have also been used to protect base metal catalysts like Cu for liquid-phase hydrogenation of furfural alcohol^73–75 (especially The Cu/γ-Al₂O₃ system^76,77), Co for solid oxide fuel cell cathode⁷⁸ and aqueous-phase hydrogenation reactions,⁷⁹ Au for CO oxidation,⁸⁰ Ni for dry reforming of methane,^81,82 Ag for plasmonic photocatalysis,⁸³ and so on.

Tailoring the coating thickness and pores size can greatly improve catalytic performance of coated catalysts. Elam and coworkers studied the coating thickness effect of Al₂O₃ on Pd nanoparticles toward methanol decomposition,⁸⁴ and a single ALD Al₂O₃ cycle was sufficient to suppress sintering of Pd nanoparticles. The activities of coated catalysts were slightly higher than the uncoated Pd sample, which was attributed to the selective blockage of the low-coordinated sites of Pd nanoparticles, since these sites contributed to the dehydrogenation reaction pathway. Beyond 16 cycles ALD of Al₂O₃, the activity decreased due to the blockage of most surface reactive sites, indicating the importance of precise control over the coating thickness. Micropores size is another important factor to tune catalytic activity and selectivity. In selective hydrogenation of 1,3-butadiene reactions, it was found that at high conversions, the butenes increased with ALD alumina overcoat thickness.⁷¹ For 30 cycles ALD of Al₂O₃ coated Pd catalysts, the selectivity researched 99% at a conversion of 95%. On the contrary, the selectivity was 46% at the same conversion on uncoated Pd catalysts (Fig. 7). The remarkable promotion of butenes selectivity was partly due to the confinement effect of the microspores with dominated size ∼1 nm. The Pd surfaces exposed by the micropores would consist of small Pd ensembles on which the adsorbed 1,3-butadine might favor selective hydrogenation due to steric effect.

Besides the above methods to form nanoporous structures, Detavernier et al. elaborated examples of applying TiO₂ ALD to form porous structures and tailor pores' size with zeolites nanotemplates.⁷² Weimer and coworkers reported the molecular layer deposition with trimethylaluminium (TMA) and ethylene glycol to form highly porous alumina films on supported Pt nanoparticles (∼2 nm). The pores were formed from thermal oxidizing of aluminum alkoxide hybrid films that re-exposed the metal surface sites.⁸⁵ However, they found that the porous coating decreased the catalytic activity which might due to the small size of the pores.

The porous oxide coating on metal nanoparticles increases the sintering resistance, while catalytic activity will degrade if the pores size is less controlled. As reported, the sintering effect may be caused by the migration or gasify of high-energy, edge/low-coordinated sites. For some catalytic reactions such as oxidative dehydrogenation of ethane, methanol decomposition, etc., the low-coordinated sites favor C-C bond scission to produce carbon fragments that leads to coke.⁴² Selective passivation of these sites leaves most of the reactive facets/sites accessible for reactants and enhance the stability simultaneously.⁸⁴ In addition, the edge sites are highly reactive for undesired products in some catalytic applications, selective blockage of these sites are also beneficial to enhance selectivity of targeted products.

Recently, Qin and coworkers reported that FeO_x oxides (ferrocene and O₃ as precursors) preferentially initiated its growth on the low coordinated sites of Pt.⁴² The structure was confirmed with both DFT calculations and CO-DRIFTS results. The selective passivation of low coordinated sites (which favor C=C bond) generated Pt-FeO_x interfacial perimeter sites (which favor C=O bond) and the selectivity for the hydrogenation of cinnamaldehyde to cinnamyl alcohol was improved. In this work, they also observed that the catalytic selectivity was sensitive to the coating thickness, for the first few cycles, FeO_x decorated the low coordinated sites and increased the selectivity to cinnamyl alcohol. With increasing ALD cycles, continuous Fe₂O₃ coating films formed that decreased the Pt-FeO_x interfacial perimeter sites, leading to decreased selectivity. Compared with conventional impregnation method, all of the Pt active sites were simultaneously blocked and the catalytic selectivity (62%) was much lower than that of catalyst prepared by ALD (84%).

Similar selective ALD growth mode was found in the deposition of TiO₂ growth on Au. It was observed that TiO₂ films grown with titanium tetraisopropoxide (TTIP) and H₂O as precursors preferentially nucleated on the low-coordinated sites of Au surfaces.⁸⁶ The CO chemisorption DRIFTS characterization showed that the intensity of CO linear chemisorption peak on Au's low coordinated sites decreased dramatically as ALD cycles of TiO₂ increased. This might be due to the low coordinated sites were more active to react with TTIP precursors than terrace sites and facets. In fact, the oxide growth behavior is also affected with the choice of precursors, Biener and coworkers reported that using TiCl₄ and H₂O to grow TiO₂ films with 10 cycles on Au tended to form continuous and smooth coatings.⁸⁷ The continuous coating completely encapsulated the Au surfaces resulted in the rapid disappearing of activity in CO oxidation. The differences between two kinds of growth modes could be attributed from the initial nucleation differences of two Ti precursors.

The oxide growth behavior on metals is closely related with nucleation and chemisorption of precursors in the first few cycles. With DFT calculations, the binding energies of precursors on metal surface can be obtained and are helpful to illustrate the selective growth sequence. As an example, Fig. 8 shows the calculated barrier energies of TMA on Pd (111), Pd (211) and Pt (111), Pt (211). The DFT results preformed on Pd surfaces demonstrated TMA adsorbed strongly on the threefold (hcp) site and quickly dissociated to Al(CH₃)* and CH₃*. The edge sites were thermodynamically more favorable for TMA dissociative adsorption than the Pd (111) terraces with 0.81 eV energy difference. However, the difference in the free energy change for TMA on Pt (211) edges and Pt (111) facets was 0.35 eV, which was much lower than that on Pd surfaces.³⁹ The differences explained the preferential precursors' chemisorption on step edges of Pd but not on Pt observed with previous studies using STM.

For Al₂O₃ thin films grown with the precursors of TMA and H₂O, it is porous in the first few cycles. When its thickness increases to ∼1 nm, Al₂O₃ will rapidly form a continuous film and cover all the surface of substrate. For oxides such as Fe₃O₄ on Pt, TiO₂ on Au, there is a strong tendency to first nucleate on the egde/low coordinated sites as previously discussed. The reactivity of precursors is a key factor to influence the selective growth sequences and behaviors on metals. We recently reported a facet selective ALD method,⁴¹ by which cerium oxide {tetrakis(2,2,6,6-tetramethyl-3,5-heptanedionato) cerium(IV) [Ce(thd)₄] and O₃ as precursors} had been utilized to form nanofences around Pt NPs (shown in Fig. 9). The facet selectivity was realized through the intrinsic differences in binding energies of Ce precursor fragments chemisorbed on Pt surfaces, and a nanofence like structure was directly formed around Pt nanoparticles. During the initial growth stage, CeO_x highly preferred to nucleate on the Pt (111) facet while leaving the Pt (100) surface intact. The selectivity could be maintained for a number of cycles until CeO_x on Pt (111) was thick enough to trigger horizontal epitaxy. The nanofence structure exposed active facets for CO oxidation, and formed perimeter sites at intimate ceria–metal interfaces that were beneficial for activity enhancement. The nanofence structure around Pt nanoparticles also provided physical blockage that suppressed particle migration and coalescence. Huang and coworkers reported deposition of Ga₂O₃ [Ga(CH₃)₃ and O₂ as precursors] on Pd particles to tune the exposed facets of Pd.⁸⁸ Ga₂O₃ adlayers were preferentially deposited at edges and facets other than the {111} facets of Pd particles, which transformed the low-coordinated sites of the {111} facets into catalytically {111} terrace like sites. The structure significantly suppressed the formation of poisoning polymeric carbon deposited on the sites and facets other than {111}, and greatly improved activity and stability in the selective hydrogenation reaction of acetylene to ethylene.

The composite structures with metal nanoparticles embedded in oxide support have also been developed to make a certain trade-off between the activity and thermal stability. The designed structures can form stable metal oxide interfaces with strong synergistic effect, which can prevent the sintering of metal nanoparticles. Moreover, the structures can make sure the formed interfaces being exposed to reactants to promote the catalytic activity. Therefore, the choice of the support oxides and the oxides surrounding metal nanoparticles are important. Generally, two types of methods have been developed for the synthesis of the embedded structures.

In synthesizing nanotrap structure, protecting organic groups are utilized to chemisorbed on the metal nanoparticles initially to form strong binding to the metal nanoparticles.⁶⁷ Stair and coworkers have systematically studied the blocking ability of different agents for Al₂O₃ ALD.⁴³ They found that ethylenediamine and decanethiol were more effective than acetonitrile and hexafluoroacetylacetonate. In many cases, the blocking groups can also be the original ligands of the ALD precursors. Generally, the process can be divided into four steps: metal nanoparticles fabricated with ALD, blocking agent growth, metal oxide ALD, and blocking agent removal.⁴³ The blocking materials used in this case are crucial to determine the final catalytic performance. It should not contain hydroxyl groups which can be active nucleation sites for metal oxide ALD. Finally, the blocking agents should be removable and have no influence on the catalytic activity.

The order of four steps mentioned earlier can also change.⁸⁹ For example, protecting the active sites by bulky organic moiety can prevent the ALD process. Then, ALD was performed to deposit a film, but the resulting layer grew only on the support surface and not on the blocking group. Finally, the blocking group was removed, and the nanotrap structure was formed. The size and depth of the traps can be determined by the size of the blocking group and number of ALD cycles. Notestein and coworkers demonstrated tuning of pore size and depth of ALD Al₂O₃ films on TiO₂ substrate by choosing the steric diameter of the blocking SAMs molecules.⁸⁹ In this case, calixarene was used to block parts of the TiO₂ surface to achieve the selective deposition of Al₂O₃. The formed Al₂O₃ nanobowl structure (<2 nm in diameter) resulted in the enhanced selectivity (up to 9:1) toward less hindered reactants in competitive photocatalytic oxidations and transfer hydrogenations.

Figure 10 shows a schematic for Al₂O₃ deposited on Pd nanoparticles with and without blocking agents.⁴³ Without the blocking agent, the nanoparticles would be encapsulated completely as the DRIFTS measurements showed no linear-bonded CO. While CO was able to adsorb on Al₂O₃−ODT−Pd/Al₂O₃ catalyst verified that Pd was exposed. The supported metal nanoparticles could be dispersed at the bottom of nanotraps with enhanced sintering resistance. This method is also applicable to form active oxide coated metal nanoparticles with motivations of simultaneously enhancing the catalytic activity and thermal stability.⁴⁶ We demonstrated that selective blockage of noble metal nanoparticles with SAMs to inhibit oxide ALD on metal nanoparticles to form nanotrap structure (Fig. 11).⁴⁴ In this way, Co₃O₄ could be deposited around Pt nanoparticles to greatly enhance the sintering resistance. The highly active oxide–metal interfaces also achieved CO oxidation conversion at room temperature.

The embedded catalytic structures can be fabricated utilizing nanotemplates with multistep ALD approaches. Qin and coworkers exhibited that confined Ni nanoparticles embedded in Al₂O₃ nanotubes (ANTs) were synthesized utilizing carbon nanocoils templates.⁴⁵ The Ni and Al₂O₃ were deposited on carbon nanocoils successively. After calcination to remove carbon nanocoils, the Ni nanoparticles embedded in the ANTs were achieved (Ni-in ANTs, Fig. 12). Such structure created massive Ni–Al₂O₃ interfaces, and Ni nanoparticles were confined in the cavities of Al₂O₃ shell, which were beneficial to the enhancement of both activity and stability toward hydrogenation reactions. At the same time, Ni nanoparticles on ANTs (Ni-out-ANTs) could be also obtained by exchanging the ALD sequence of Ni and Al₂O₃. The TEM images in Fig. 12 demonstrate the Ni-in ANTs and Ni-out-ANTs catalysts fabricated with this method.

In this review, the recent developments of selective ALD methods in the design and synthesis of catalytic nanostructures are summarized, including core shell structures, discontinuous coating structures, and embedded structures. The references for catalytic structures fabricated with selective ALD methods are briefly summarized in Table I. There are mainly three strategies reported for the synthesis of core shell structures, which are deposition temperature reduction, precursor partial pressure adjustment, and surface modification assisted with SAMs. Compared with single component catalysts, the formation of core shell structures can enhance the activity, selectivity, and stability. The enhanced properties may originate from the lattice strain, bonding interactions, and electron transfer due to the formation of core shell interfaces. For supported nanoparticles catalysts, discontinuous oxide overcoating structures have been reviewed. The oxide layer can be grown controllably on different sites of the metal nanoparticles, such as porous coating, selective deposition on low coordinated site, as well as passivation on specific facets. In porous coating structures, tailoring the coating thickness and pores size can greatly affect catalytic performance of coated catalysts. For example, in many catalytic applications, the activities show a volcanolike trend as increasing coating layer thickness. With few ALD cycles, the interfaces are created and activity is increased through synergistic effect. Optimizing coating thickness can maximize the interfaces between metal and oxides and avoid blocking too many active sites. The pores size is another factor that influences catalytic selectivity through steric effect, which only allows a certain size of reactants accessible to metal reactive surface sites. With more selective deposition or passivation on specific sites on catalysts (e.g., low coordinated sites, facets), the ordered coating structures provide more controllability on tuning the catalytic activity and selectivity. At the same time, the oxide overcoating layers can also provide physical barriers that are especially helpful to enhance the stability during long term usage. With the assistance of selective blockage or nanostructure templates, the fabrications of the nanotrap or metal nanoparticles embedded composite catalysts have been discussed.

These strategies of selective ALD have demonstrated unique advantages to design and fabricate the catalysts in atomic scale and provided insights to understand the structure–activity relationship in a more direct way. Challenges still exist; however, it needs to be aware of that current lab scale proofs-of-concept examples demonstrate superior catalytic properties of ALD catalysts. The practical industrial applications are needed to continue motivating research on scale-up. The main problem may originate from fabricating catalysts with huge specific surface area and maintaining uniformity and efficiency during synthesizing. The deposition and decoration of catalysts on powder substrates consume long period of time for precursors' diffusion and reaction, and agglomeration of nano/micron powders substrates also destroy the deposition uniformity. The solutions may rely on the development of ALD processes and study of equipment to accelerate deposition period and utilization efficiency of precursors. Furthermore, the reported catalytic structures are still limited, which mainly come from the experimental studies. The theoretical studies about the selective growth mechanism and prediction of new structures are lacking. It is essential to focus on the fundamental understanding of surface reaction mechanisms such as interactions and reactions of precursors with different substrates, the reaction energetic routes of the selective ALD processes. Besides, the choice of precursors for selective ALD approaches is quite limited at this moment, which is usually important for determining the growth behavior. It is necessary to develop the precursors and study them on theoretical level for selective ALD. The combination studies of both in situ and ex situ experiments and characterizations with atomic accuracy are essential to deeply understand the structural evolution and reaction processes.

The work was supported by the National Basic Research Program of China (No. 2013CB934800) and the National Natural Science Foundation of China (Nos. 51702106, 51575217). Rong Chen acknowledges the Thousand Young Talents Plan, the Recruitment Program of Global Experts, the Hubei Province Funds for Distinguished Young Scientists (No. 2015CFA034).