Synthesis without Solvents: The Cluster (Nanoparticle) Beam Route to Catalysts and Sensors

CONSPECTUS: It is hard to predict the future of science. For example, when C 60 and its structure were identi ﬁ ed from the mass spectra of gas phase carbon clusters, few could have predicted the era of carbon nanotechnology which the discovery introduced. The solubilization and functionalization of C 60 , the identi ﬁ cation and then synthesis of carbon nanotubes, and the generation and physics of graphene have made a scale of impact on the international R&D (and to some extent industrial) landscape which could not have been foreseen. Technology emerged from a search for molecules of astrochemical interest in the interstellar gas. This little sketch provides the authors with the con ﬁ dence to present here a status report on progress toward another radical future  the synthesis of nanoparticles (typically metals) on an industrial scale without solvents and consequently e ﬄ uents, without salts and their sometimes accompanying toxicity, with minimal prospects for unwanted nanoparticle escape into the environment, with a high degree of precision in the control of the size, shape and composition of the nanoparticles produced and with applications from catalysts and sensors to photonics, electronics and theranostics. In fact, our story begins in exactly the same place as the origin of the nanocarbon era  the generation and mass selection of free atomic clusters in a vacuum chamber. The steps along the path so far include deposition of such beams of clusters onto surfaces in vacuum, elucidation of the key elements of the cluster − surface interaction, and demonstrations of the potential applications of deposited clusters. The principal present challenges, formidable but solvable, are the necessary scale-up of cluster beam deposition from the nanogram to the gram scale and beyond, and the processing and integration of the nanoclusters into appropriate functional architectures, such as powders for heterogeneous catalysis, i.e., the formulation engineering problem. The research which is addressing these challenges is illustrated in this Account by examples of cluster production (on the traditional nanogram scale), emphasizing self-selection of size, controlled generation of nonspherical shapes, and nonspherical binary nanoparticles; by the scale-up of cluster beam production by orders of magnitude with the magnetron sputtering, gas condensation cluster source, and especially the Matrix Assembly Cluster Source (MACS); and by promising demonstrations of deposited clusters in gas sensing and in heterogeneous catalysis (this on the gram scale) in relevant environments (both liquid and vapor phases). The impact on manufacturing engineering of the new paradigm described here is undoubtedly radical; the prospects for economic success are, as usual, full of uncertainties. Let the readers form their own judgements.


INTRODUCTION
By definition, nanoparticles (NPs) or nanomaterials are small objects in the range of 1−100 nm, 1 usually defined as an agglomeration of atoms or molecules which exhibit a wide range of size-dependent properties. 2 NPs with tailored morphologies and chemical composition have attracted great attention due to their unique properties. They can lead to highly advantageous characteristics in applications such as heterogeneous catalysis, 3,4 plasmonics, 5,6 or biomedicine. 7,8 The fabrication methods of nanomaterials are classified into two categories: (i) top-down methods and (ii) bottom-up methods. The latter is of interest in this Account and can be divided into two approaches, the chemical and physical methods. The chemical approaches are the most widely used, as they offer a high control over the size, shape and chemical composition. 9−11 However, the chemical approaches often involve chemical agents associated with environmental toxicity. Thus, it is highly desirable to find novel green synthesis approaches for NP production, either by using environmental-friendly solvents methods 12 or solvent-free synthesis such as gas-phase methods.

CLUSTER BEAM DEPOSITION (CBD) METHOD
Despite the technical demands of gas-phase synthetic methods, the CBD method has emerged as one of the most promising techniques for producing supported NP assemblies. Typically, gas-phase synthesis relies on homogeneous nucleation of a supersaturated vapor, for example metal atoms within a lowpressure rare gas environment, followed by particle growth via condensation and coalescence. 13 Among the variety of CBD methods available to date, the magnetron-sputtering, inert-gas condensation method 14 has emerged as one of the most versatile as it allows control of the growth process by varying the experimental parameters such as the magnetron power, the inert-gas pressure and the length of the aggregation zone ( Figure 1).
In recent years, a new generation of CBD sources has emerged 15,16 which enables not only the synthesis of singlecomponent NPs, but also multicomponent NPs via the use of two or more independent target materials ( Figure 1). The capability to use multiple independent targets opens new avenues for the design and preparation of multimetallic NPs with control of their morphology and composition, in a single synthetic protocol. For binary NPs, three main types of NP morphologies ( Figure 1) have been reported, which differ in the atomic arrangements of the two elements (A and B) within the same NP: (i) NP alloys (AB or BA), which can be either random 17 or ordered; 18,19 (ii) core@shell NPs (A@B), 20 which consist of one type of atoms (B) surrounding a core of another type of atoms (A); and (iii) the commonly named Janus NPs (A-B), which consist of two parts of different elements sharing a common interface. 21 Discussion of these morphologies has invoked thermodynamic considerations which determine the driving forces and stability of the different NPs. 22,23 However, a more in-depth understanding of the underlying mechanisms of NP growth has also highlighted kinetic effects, for example, to explain the formation mechanism of cuboid shapes in gas-phase synthesis. 24 Among these different morphologies, CBD methods also offer the ability to size-control the NPs by postgrowth separation phase (Figure 1), such as time-of-flight 25,26 (TOF) or quadrupole mass-filtration. 27 However, size separation challenges the utilization of gas-phase condensation techniques for large scale production of nanoparticles. 25 In this Account, we highlight methods that focus on accurate determination of nanoparticle sizes during the growth, instead of producing a wide size-dispersion first followed by mass-filtration afterward. In particular, we consider the potential of nanoparticle selfarrangement during gas-phase growth, and a new class of CBD source, the Matrix Assembly Cluster Source (MACS) that promises a breakthrough in large-scale production of clusters.

Size-Control during Formation in the Gas-Phase
The control of the particle size remains a primary aim in the preparation of metal NPs, whether by physical or chemical means. In CBD, this control is commonly ensured by sizeselection after NP growth, which limits the obtainable material yield considerably. Thus, it is highly desirable to find new routes to size-selection during the formation process. A novel methodology to ensure simultaneous size and shape control during cluster formation was recently reported; this work introduced the ferromagnetic (Fe) target thickness as a decisive experimental parameter. 28 Previous studies showed the limiting effects of screening by the high-permeability of a magnetic target, but this factor can also be exploited as a tool to control NP growth. On the grounds that nucleation happens within a short distance of the sputtering target, this effect can tune the NP size. The magnetic target thickness was adjusted from 0.3 to 1 mm, with the effects illustrated in Figure 2a. A switch from bimodal size distribution to a well-defined single peak in the mass spectrum is observed. The work revealed a close correlation between the width of the target erosion profile (Figure 2b) and the shape of the cluster sizedistribution, mainly its width (standard deviation). This is consistent with modification of the electron trap region close to the surface by the magnetic screening induced by the target. In addition, to validate the approach regarding the accurate control of the size during formation, similar experiments were performed with and without the quadrupole mass filter. Interestingly, there was no gain in standard deviation using mass filtration, but a significant decrease of the material yield compared to experiment without mass filtration was observed, as shown in Figure 2c.

Shape-Control of Iron Nanoparticles
Besides the size-control ability shown previously using the gasphase method, the morphology of NPs, in particular the shape, is of importance in some aspects of nanotechnology. 29,30 A striking example of NP shape control in CBD is the ability to switch readily the shape of Fe NPs from cubic ( Figure 3a) to quasi-spherical ( Figure 3b) by only adjusting the dc magnetron power of the sputtering source. 24 The magnetron power Nanoparticles fly toward the deposition chamber by differential pressure forming a cluster beam of monatomic species A and B, or binary nanoparticles such as mixed structure (AB), core−shell structure (A@B), and Janus structure (A-B). The nanoparticle beam is deposited onto planar supports or powders after passing through a focusing stage and mass filtration unit for size-selection.

Accounts of Chemical Research
Article dictates the NP shape, by providing a way to control the number of metal atoms ejected from the target materials (Fe atom density) as well as the thermal environment by increasing the probability of metal−metal atom collisions instead of metal−Ar collisions.
For comparison, in chemical synthesis methods, the wellknown ability of surfactants to bind preferentially to specific facets allows different facet growth rates, thus providing a means of tuning the NP shapes. 31 But what can explain the control of the cuboid NP growth using physical methods? One way to scrutinize subtle kinetics effects on the growth, such as the competition between surface diffusion rates on specific facets with the rates of atomic deposition, is to employ a combination of molecular dynamics and kinetic Monte Carlo simulations. As shown in Figure 3c, the competition between a layer-by-layer growth mode (Frank-van-der-Merwe-like (FM)) on the {100} facet and an island growth mode (Volmer− Weber-like (VW)) on the {110} facet is the driving force for growth of cubic rather than near-spherical shapes. This study has revealed the possibility to stabilize cubic shape which is energetically unfavorable at thermodynamic equilibrium using gas-phase methods.

Bimetallic Iron-Based Nanocubes
While improved control of size and shape of monometallic NPs is a significant development, it is insufficient to fulfill the demand for optimized multifunctional nanomaterials. Our deposition approach has been recently extended to bimetallic NPs using a co-sputtering process with nonmagnetic dopants, 28,32 while keeping uniformity in size, shape, and crystallinity.
As a demonstration of binary nanoparticle control, Figure 4 shows a study of the Fe−Pd system using three independent targets simultaneously. Well-defined FePd nanocubes are obtained with a narrow size distribution, as shown in Figure  4a. A core−shell morphology with a highly crystalline core is revealed after air exposure, as shown in the inset to Figure 4a, in high-angle annular dark field image (HAADF) in scanning TEM (STEM) mode. Furthermore, spatially localized electron energy loss-spectra (EELS) from the STEM clearly distinguish the presence of both Fe and Pd within the core while only Fe oxide exists in the shell (Figure 4b). Multiple characterization techniques indicate that the FePd core has a random alloy arrangement with an estimated chemical composition of 10− 15% of Pd. All these results show the highly controllable features of these techniques, such as the capacity of noble metal (Pd, Au) doping to form bimetallic nanoparticles while keeping the highly uniform size and shape seen previously.

SCALE-UP OF THE CLUSTER PRODUCTION RATE
The successful control of size, shape and composition of the clusters formed by synthesis in the gas-phase has made the cluster beam technique a powerful tool to investigate the

Accounts of Chemical Research
Article fundamental properties of clusters on surfaces and has also stimulated studies of size-dependent catalytic activity on the 1 cm 2 surface area scale, i.e., in the domain of surface science. However, its application to making catalysts at the standard chemistry laboratory research scale, let alone industrial levels, is hampered by the low production rates. For example, a conventional cluster beam source can typically produce a sizeselected cluster beam current of about ∼0.1−1 nA (equivalent to 0.05−0.5 μg Au 100 clusters/h). Thus, a prohibitive time is needed even to produce 1 g of catalyst loaded with 10 mg of clusters (1 wt % loading) for standard catalytic R&D measurements. Therefore, very significant scale-up of the cluster production rate is a crucial challenge if the cluster beam technique is to contribute at this level. Thus, reported developments in high flux cluster beam instruments are of particular interest. However, we should also be aware that the current studies have not yet been able to achieve the size/ shape/composition control and high cluster deposition rate at the same time. Hopefully, in the future, both features could be obtained together with the help of more advanced cluster beam techniques. Here we highlight two innovations of high flux cluster beam instruments.

Magnetron-Sputtering Gas Condensation Cluster Source
A scale-up of 2 orders of magnitude has been achieved with a traditional magnetron sputtering, gas condensation cluster source, 33,34 simply by extraction of the cluster beam before the mass spectrometer, which is only used for monitoring the cluster size distribution; the scheme is shown in Figure 5a. The instrument consists of four sections: magnetron sputtering, gas condensation chamber; ion optics chamber; mass filter chamber; and cluster deposition chamber. The sputtering chamber includes two magnetrons mounted in parallel, which allow for tuning of the composition of binary clusters. An electrostatic "octosphere" deflector allows the ionized cluster beam to fly directly into the mass filter to measure the size distribution or deflects it down by 90°toward the deposition chamber. Here a future innovation is introduced. Powder supports loaded into a stainless-steel cup are mechanically agitated close to the beam deposition to maximize exposure of powder to the incoming cluster beam. The typical cluster beam deposition current is around 20−50 nA, which enables the deposition of 0.1 wt % Au 300 clusters onto 1 g of powder in 10 h. An example of early catalytic studies with the clusterdecorated powders produced is discussed in section 5.2.

Matrix Assembly Cluster Source (MACS)
A new type of nanoparticle beam source, MACS, 35 has recently been invented and demonstrated in an attempt to reach cluster

Accounts of Chemical Research
Article beam intensities sufficient to 1 g of catalyst in 1 h. The principle of this new system is different from the traditional method of condensing hot atoms in a cold gas. Instead, the MACS is the impact of an ion beam (e.g., Ar + ) upon a condensed matrix of rare gas atoms impregnated with atoms of the desired material. The impact of an ion initiates a cascade of collisions inside the matrix, causing clusters to nucleate and/or ripen. A cluster can be ejected from the matrix by the same (or subsequent) ion beam impact. Figure 5b shows a schematic diagram of a prototype MACS. 36 The solid matrix is prepared on a cryogenically cooled (∼10 K) matrix support by condensation of rare gas atoms while the cluster material is evaporated at the same time. Once the matrix is prepared, it is then moved to the second chamber in this system for sputtering. A carousel holding up to 21 glass slides is designed to collect the clusters sputtered out of the matrix. The cluster flux is linearly proportional to the incident ion beam current (∼1 cluster is emitted per 100 Ar + impacts), 33 so further scaleup can be achieved by employing a more powerful ion source (here 100 μA). Indeed, a scale up of 5 orders of magnitude has been achieved, from 100 nanograms of clusters per hour to 10 mg per hour with an Ar ion beam of ∼1 mA current. In addition, compared with a traditional gas condensation source in which much of the material (>90%) is deposited on the walls of the condensation chamber, the MACS offers in principle 100% utilization, since all the material loaded into the matrix can be sputtered out and collected or recovered from the matrix support plate after use.
In the emerging understanding of how clusters are formed in the MACS matrix, molecular dynamics simulations have played a valuable role in illustrating the ion bombardment of three Ar matrices with Ag content of 5%, 10%, and 20%. 37 It was found that clusters in the matrix ripen via multiple ion impacts. Figure 6a shows the average size of Ag clusters formed in the matrix as a function of the number of impact Ar ions. For all three cases, the Ag clusters grow with successive collision cascades, especially for the matrix with 20% Ag loading. But how do such clusters escape from the matrix? The emission process was simulated for a model system containing an individual Ag cluster embedded in an Ar matrix around 1 nm below the surface (Figure 6b). The simulations showed that a direct collision between the incident Ar atom and the embedded Ag cluster. The cluster temperature rise was sufficient to boil the Ar layer around the cluster surface. Moreover, the momentum transferred from incident Ar atom to cluster accelerated the cluster toward the bulk of the matrix. The recoil due to the compression of the matrix was sufficient to drive the cluster out of the matrix.

Gas Sensing: NO 2 Detection
Chemoresistive metal oxide gas sensors 38,39 have potential applications in environmental monitoring and biomedical diagnostic. For examples, nitrogen oxides (NO x , mainly consisting of NO and NO 2 ) are industrially and biologically important gas species, which can be potential markers of pathologies such as asthma. 40,41 In the field of breath diagnosis, the need to fabricate highly reproducible, stable, and highly sensitive NO 2 sensors with an accuracy of 1 ppb in a range of 1−500 ppb remains a significant challenge. The optimization of semiconducting metal oxide materials in terms of selectivity 42 and low operation temperature 43 using catalytic NPs has been extensively investigated. However, the use of expensive metals (Pd, Pt, Au, etc.) remains a limitation, as well as the lack of information on the interplay between their atomic-scale structure and their influence on the sensing properties. The development of scalable nanomaterials preparation methods may play a role in addressing this problem. CBD is well suited for CMOS-compatible sensor production because it allows single-step NP deposition at room temperature under high vacuum conditions without the presence of precursors and surfactants. A sensor fabricated with the Fe NP production technology described in section 3.1 is shown in Figure 7. The clusters were deposited onto an electrode array (Figure 7a) and produced a percolating film of Fe nanocubes (Figure 7b). The circularshaped metal electrodes were fabricated on a micro hot plate 44 support by lithographic techniques, which allows local heating during sensor operation. Gas sensing experiments were performed in a commercial probe station, measuring the electrical resistance changes during exposure. Figure 7c shows the resistance change of the Fe nanocube film in dry synthetic air during exposure to pulses of NO 2 (concentration range 3− 100 ppb) at temperature of 200°C. A sensitivity down to the 3-ppb level was established, which meets the specification for asthma detection. This approach shows better performance than conventional Fe-based sensors, 45 as well as competing favorably with other nanomaterial such as 1D metal oxides 46 or 2D materials. 47 Results with other toxic gases are shown in the

Accounts of Chemical Research
Article inset of Figure 7c. The nonzero response found for H 2 S gas compared with CO is well described by an n-type metal oxide semiconductor model. H 2 S can be easily distinguished from NO 2 by the sign of the resistance change. The reproducibility of the sensors across two different batches is shown in Figure  7d. The analysis of the sensing mechanism is something of a puzzle in the literature. Indeed, with elevated temperatures and oxidizing/reducing gas atmospheres, a severe impact on the nanoscale morphology of the sensing elements can be anticipated. In this case, environmental-TEM revealed the formation of voids at the centers of the NPs under operational condition, 28,32 which gives a possibility to understand the sensing performance and stability after operation. The excellent sensing properties due to specific NP morphologies combined with the inherent advantages of NP gas-phase synthesis make the CBD approach a highly promising candidate for future large-scale production of miniaturized, high-performance gas sensor devices integrated with standard microelectronic components.

Catalysis Applications
Physically produced metal clusters deposited onto powder supports with tunable interaction represents a new paradigm in making heterogeneous catalysts for different applications, from fine chemistry to petrochemistry, from electrocatalysis to photocatalysis. Scale-up of the rate of cluster beam deposition, especially at the gram scale, opens the way to explore catalytic behaviors under realistic reaction conditions. Here we review two recent examples of cluster catalysis, one in the liquid phase and one in the vapor phase, at the test tube or beaker rather than surface science scale.

4-Nitrophenol Reduction in Solution.
The reduction of nitrophenol by sodium borohydride (NaBH 4 ) is an attractive model reaction to evaluate cluster catalytic activities in solution, since the reagents and products can be measured precisely by real-time spectroscopy. In this study, Au/Cu binary clusters (spherical shape; diameter in the range of 3−5 nm) with three different compositions (termed Au-rich, Au/Cu-balanced and Cu-rich), were deposited onto MgO powder supports by the dual magnetron sputtering gas condensation cluster beam source discussed in Figure 5a. 17 Cluster composition measurements indicated a strong relationship between the Au/Cu atomic ratio and the power applied to the two sputtering targets. Figure 8a and b shows a HAADF-STEM image and corresponding diameter distribution for the Au/Cu-balanced cluster catalyst. The measured average size on the MgO support (4.6 nm) was larger than the value estimated from the mass spectra (2.4 nm for a spherical shape), indicating that a degree of surface aggregation happens when the clusters landed on the powders. The Au/Cu catalysts produced were directly used to catalyze the reduction of 4nitrophenol by NaBH 4 in solution. Figure 8c shows the evolution of the UV−vis absorbance spectra of the reaction solution after adding 30 mg Au/Cu cluster catalysts. For the Au/Cu-balanced cluster catalyst, the intensity of the peak at

Accounts of Chemical Research
Article 400 nm, which indicates the concentration of 4-nitrophenol in the solution, decreases quickly with the reaction time. For the Au-rich and Cu-rich cluster catalysts, the peak intensity drops much more slowly, suggesting the Au/Cu-balanced cluster catalyst is more active than the Au-rich and Cu-rich cluster catalysts. Since the cluster loading for the different compositions was not identical, their activities are best compared through the normalized reaction rate constant, K nor (min −1 ·mol metal −1 ). For the Au/Cu-balanced cluster, the K nor is 8.9 and 6.6 times that of Au-rich and Cu-rich clusters, respectively, which indicates that the Au/Cu-balanced cluster is much more active. DFT calculations on the binding energies of model cluster-reagent and cluster-product systems suggested that binding of the reagent molecules to an adjacent pair of Au and Cu atoms at the cluster surface optimized the relative energies for regent adsorption and product desorption, which could account for the enhanced catalytic activity. 17 This work shows an insight into design of bimetallic clusters at the atomic scale for highly active catalysts.

Selective
Hydrogenation of 1-Pentyne. The selective hydrogenation of 1-pentyne to 1-pentene in the vapor phase is an important reaction, as it produces a high purity alkene steam which is widely used in the petrochemical industry. 48 Pd and Pd/Au clusters (spherical shape; diameter in the range of 1−2 nm) deposited with the MACS onto carbon tapes were used to catalyze this reaction and compared with a Pd reference sample prepared by impregnation. A dicing technique 49 was adopted to transform carbon tapes decorated by deposited clusters into catalyst powders, as shown in Figure  9a. The carbon tape supports were first mechanically diced to generate grooves on the surface. Then the diced supports were loaded onto the carousel and presputtered by an Ar ion beam; the point defects formed thus provide anchor sites for clusters on the surface to reduce aggregation. After cluster deposition, the topmost layer of the supports was scratched off to obtain the catalyst powders. Figure 9b and c shows a HAADF-STEM image and corresponding diameter distribution for the Pd cluster catalyst. Dense Pd clusters were observed on carbon flakes with an average diameter of 1.21 ± 0.36 nm. The cluster catalysts produced were directly used to catalyze the 1-pentyne hydrogenation reaction. The plots in Figure 9d compare the catalytic activity of the Pd cluster catalyst with the Pd reference sample in terms of activity per metal atom. The Pd cluster catalyst was found to be more active than the reference sample (average particle diameter 2.38 nm), consistent with the smaller particle size obtained with the MACS; thus, the number of active surface sites per unit weight of metals is increased. The Pd (1.21 ± 0.36 nm) and Pd/Au clusters (1.31 ± 0.41 nm) had the similar diameters, so the effect of composition on the catalytic activity could be explored. The product yields per surface atoms are shown in Figure 9e. It can be seen that the surface atoms on the binary nanoalloys exhibit a similar catalytic activity to those of the pure Pd clusters. One reasonable explanation is that oxidation drives the Pd atoms in the Pd/Au clusters to the surface when they are stored in air. Although a synergetic effect is not observed in the Pd/Au binary clusters, this study suggests a route to decrease the catalyst cost by design of core−shell bimetallic systems with a base metal at the core.

CONCLUSION
Cluster beam deposition provides a means to synthesize and deposit atomic clusters in vacuum with precise control of the size, shape, and composition without need for solvents, effluents, or salts. The deposition of clusters onto chosen supports, whether planar or powder, allows for the fabrication of functional systems including catalysts and sensors. In this Account, recent progress in the morphology control of clusters produced with a magnetron-sputtering, gas condensation cluster source has been illustrated through the generation of quasi-spherical and cubic Fe clusters as well as Fe/Pd nanoparticles; the former exhibit promising performance in NO 2 detection. Current efforts to address the bottleneck of the cluster beam deposition rate have been highlighted. In particular, the Matrix Assembly Cluster Source (MACS), based on the ion beam sputtering of a metal loaded, cryogenically condensed solid matrix, can increase the deposition rate by 5 orders of magnitude compared with a standard size-selected cluster beam source. This scale-up makes it possible to explore powder-supported cluster catalysts, created at the gram scale, in liquid-phase and vapor-phase catalytic reactions. Sensors and catalysts are but

Accounts of Chemical Research
Article two of the potential applications which come into view via nanoparticle structure control and scale-up. Green manufacturing by cluster beam deposition may thus emerge as a new paradigm for the industrial synthesis of functional nanostructured materials. Only time will tell.

Author Contributions
The authors contributed equally to this work.

Notes
The authors declare no competing financial interest.