Learning from Metal Nanocrystal Heterogeneity: A Need for Information-Rich and High-Throughput Single-Nanocrystal Measurements

Abstract

Metal nanocrystals (NCs) show utility in a variety of applications due to their unique structure-dependent properties. Isolating these structure–property relationships is crucial for NC design, but heterogeneities present in NC ensembles as well as limitations in NC characterization strategies complicate this goal. Herein, we describe the various types of intraparticle and interparticle heterogeneities common to NC ensembles and then provide a detailed description and comparison of single-particle techniques that can be used to characterize these different heterogeneities. Case studies then showcase the use of multimodal characterization approaches, where multiple, primarily single-NC techniques are used in tandem to provide new insights into metal NC structure–property relationships. We conclude with a critique of single-NC techniques that motivates the development of new high-throughput and high-resolution single-NC characterization approaches as well as computational tools, with a proposed workflow outlined to accelerate NC design and discovery.

Metal nanocrystals (NCs) are of interest in research areas as far ranging as biosensing to space exploration. − The interest in and usefulness of metal NCs arise from their unique nanoscale features, such as large surface area-to-volume ratios, the close proximity of energetically distinct surface sites, and the rise of localized surface plasmon resonances (LSPRs). − Further, the properties of NCs can be tuned through their size, shape, and composition. − However, colloidally synthesized NCs are not identical within the ensemble, displaying both intraparticle and interparticle heterogeneity, defined as structural and compositional inconsistencies within a single particle or among a grouping of particles, respectively. This heterogeneity arises from small energetic differences between certain features during NC formation (e.g., twinned versus single-crystalline seeds, which can vary in their total energy as a function of size). Significantly, this heterogeneity may dilute the properties of the predominant NC structure within an ensemble, falsely enhance the ensemble’s properties in ensemble-averaged measurements, and mask NCs with exceptional properties. For example, Nie and Emory proposed that the enhanced Raman signal observed when using Ag nanoparticles originated from less than 1% of the nanoparticles within the ensemble. This groundbreaking identification led to decades of research, largely leveraging theory and single-NC approaches, to accurately elucidate structure–property relationships, where now surface-enhanced Raman spectroscopy (SERS) is a robust analytical tool.

This example highlights the importance of deconvoluting the structure–property relationships of NCs to efficiently design NC candidates for various applications and understanding the tolerances of various heterogeneities in an NC ensemble toward an application. However, this example also highlights a major challenge for the field of nanoscience. Most efforts directed at structure–property correlation rely on routine bulk measurements (e.g., absorption spectroscopy, cyclic voltammetry, etc.), which can lead to misidentification of these relationships on account of interparticle and intraparticle heterogeneity. To obtain reliable structure–property correlation, property measurements of individual NCs are required, which directly correlate with their composition, size, shape, and defects. Single-NC measurements are state-of-the-art in terms of structure–property correlation. − However, these methods often suffer from experimental complexity, low throughput, and insufficient correlated data. There is a need for accessible, high-throughput single-NC methods that are also information-rich. For example, conceptual breakthroughs such as those that made SERS robust could be accelerated by obtaining correlated structure–property measurements from thousands of individual NCs in seconds, but a major bottleneck is reliance on electron microscopy to obtain structural and compositional information at the level of single NCs. Additionally, some property measurements are not available at the level of single NCs currently (e.g., selectivity and active surface area of single-NC electrocatalysts, particle density, ligand placement, etc.).

Herein, NC heterogeneity across length scales is described by outlining common intraparticle defects, ranging in scale from atomic level (point) to linear (one-dimensional) to planar (two-dimensional) defects, before concluding with a description of ensemble-level (interparticle) heterogeneity. Then, analytical techniques for single-particle structural characterization of metal NCs are described, comparing and contrasting their advantages and disadvantages relative to those of other techniques. Case studies in both the intraparticle and interparticle size regimes are included, showcasing the current state of structure–property characterization and correlation across these length scales. We conclude this Perspective with a proposed characterization workflow for accelerating NC design through both high-throughput and high-resolution single-particle techniques, outlining the frontiers of single-NC studies where big data and computational and machine learning (ML) approaches are expected to play an increasingly important role.

Nanocrystal Heterogeneity across Length Scales

Figure summarizes the different types of intraparticle and interparticle heterogeneities common to metal NCs and their relationship to one another as a function of length scale. The smallest-scale sources of intraparticle heterogeneity possible within NCs are point defects. These atomic-scale defects include the replacement of an atom with one of a different element (substitutions) or the absence of an atom (vacancies) within a crystal lattice. Interstitial defects can also occur where an atom occupies a site outside of the points formed by the crystal lattice.

1.

Scheme depicting different types of intraparticle and interparticle NC heterogeneities across length scales, where blue atoms contribute to intraparticle examples, brown atoms contribute to interparticle examples, and black atoms represent a secondary composition in multielemental particles or a substituted atom.

Linear defects in NCs are restricted to a row of atoms in a crystal and are also known as dislocations. The most diverse defect types found in NCs are planar defects, which range from defects within the crystal to defects on the NC surface. We note that the NC surface is a planar defect itself when compared with the bulk crystal due to the presence of dangling bonds. Two primary types of planar defects within crystals are stacking faults and grain boundaries. Stacking faults occur when the stacking order of atomic planes within a crystal lattice is interrupted, while grain boundaries result from the interface of two crystal lattices with different orientations or two different types of crystal lattices. Twin planes are a specific type of grain boundary where the crystalline regions meeting are mirror images of each other. A type of planar defect isolated to the NC surface is surface truncation, where the vertices or edges of an NC truncate to decrease surface energy. Steps and kinks are further possible surface defects, where steps are variations in the stacking height of the surface atoms, and kinks are point defects within these steps. Lastly, compositional intraparticle heterogeneity can be defined as non-uniform elemental distribution across a multielemental NC and can arise during synthesis or through their use in applications such as catalysis. ,

Although intraparticle heterogeneity is inherent to colloidal NCs, interparticle heterogeneity is more commonly characterized, as these differences are often larger and more easily visualized. Interparticle heterogeneities include differences in NC size and shape within the ensemble and particle-to-particle variations in composition and truncation. Ligands on the surface of NCs, introduced from the NC synthesis or by post-synthesis NC functionalization, can be an additional source of both intra and interparticle heterogeneity due to differences in ligand density across a single NC or from particle-to-particle. We note that in some cases, intraparticle heterogeneities introduce interparticle heterogeneities; that is, when intraparticle defects are present in NCs used as seeds (i.e., sites for heterogeneous nucleation) for growth to larger NCs, propagation of such defects to a larger scale is possible. The propagated defects can lead to increased polydispersity, especially in relation to the NC shape in the resulting NC ensemble. For example, when synthesizing Ag nanocubes, often a subset of the ensemble consists of shape impurities such as Ag right bipyramids, Ag decahedra, and even Ag nanowires with pentagonal cross-sections; these shape impurities arise because the initial seeds formed during the synthesis contain twin planes while the seeds producing Ag nanocubes are single-crystalline. However, intraparticle defects in seeds can also be intentionally engineered to access monodisperse populations of specific shapes and symmetry-reduced structures. , Further, intraparticle defects in catalytic materials can be intentionally formed to alter the material’s electronic properties and surface structure toward more selective and active catalysis.

Analytical Techniques to Characterize Single-Particle Heterogeneity

This section describes and contrasts different analytical techniques that are capable of characterizing the structural features of single metal NCs at different length scales, whether they are already well-established tools for single-NC analysis (e.g., electron microscopy) or less common but promising tools for single metal NC characterization. Though techniques such as X-ray diffraction (XRD), X-ray fluorescence, and X-ray photoelectron spectroscopy (XPS) are common methods used in materials characterization and may indicate the presence of some interparticle heterogeneities (e.g., size polydispersity, sample compositional and phase purity, etc.), − they are limited to bulk-scale, not single NC, analysis and thus will not be discussed herein. Further, techniques that are primarily used to measure single-NC properties (as opposed to techniques that characterize structure only or can measure both structure and properties) were excluded from this section, with two single-NC property techniques, scanning electrochemical cell microscopy (SECCM) and electrochemical collision methods, being highlighted in the following case study sections.

Scanning electron microscopy (SEM) and transmission electron microscopy (TEM) have long been foundational techniques in single-NC studies, being the gold standard of verifying that a single NC is being analyzed, although familiarity with signal intensities (e.g., electrical current, light scattering) in other single-NC techniques can provide an internal reference. In SEM, a focused electron beam is scanned across the sample, generating backscattered and secondary electrons that are detected on the same side of the sample as the electron source to build an image with three-dimensional (3-D) topographical information. Alternatively, TEM uses an electron beam that penetrates and is transmitted through the sample, generating elastically and inelastically scattered electrons that are collected on the opposite side of the sample as the electron source to construct a two-dimensional projection of the sample. Due to the reliance on the transmission of electrons through the sample, TEM is limited to a sample thickness of a few hundred nanometers, although focused ion beam milling can thin specific regions of a material for TEM analysis and be used to reveal the internal structure of nanomaterials. SEM is limited only by the maximum distance the instrument allows between the sample stage and the electron beam/detector, which is on the scale of centimeters. Further, SEM requires conductive samples to limit the charging of the sample by the electron beam, but TEM can image both conductive and non-conductive specimens. Overall, SEM offers easier sample preparation and analysis, faster experiments, and less sample damage at the cost of reduced spatial resolution compared to TEM, which is on the order of nanometers in SEM studies as opposed to angstroms in TEM studies. SEM is also limited in the number of compatible auxiliary techniques when compared to TEM, although an energy-dispersive X-ray spectroscopy (EDS) detector can be inserted into both types of instruments to obtain compositional information from element-specific X-rays generated as the electron beam interacts with the sample.

High-resolution transmission electron microscopy (HRTEM) and scanning transmission electron microscopy (STEM) are two popular modes of TEM capable of reaching subangstrom resolution. In HRTEM, a phase contrast image is formed by the interference of the direct electron beam with a diffracted electron beam, whereas in STEM, a single highly focused electron beam is raster scanned across the sample, increasing the time required to collect an image but also increasing the resulting spatial resolution. , The detector used in STEM imaging can also be varied to further optimize the resulting signal. For example, annular bright field (ABF)-STEM collects low-angle scattered electrons and is optimal for imaging light elements, whereas high-angle annular dark-field (HAADF)-STEM collects electrons scattered at high angles and is optimal for imaging heavy elements. HAADF-STEM has superior Z-contrast (Z = atomic number) compared to HRTEM, making atomic features often easier to visualize and differentiate. Aberration correction in some HRTEM and STEM instruments can further increase the spatial resolution of these techniques, as well as Z-contrast in HAADF-STEM, through real-time correction of electron beam distortions while imaging.

In addition to static imaging, auxiliary methods make it possible to capture 3-D morphology, obtain compositional information, and monitor dynamic processes in state-of-the-art TEM studies. Electron tomography is one such auxiliary technique, in which the 3-D structure of a single NC is reconstructed from a series of two-dimensional TEM or STEM images collected at various tilt angles, providing snapshots of NC structure in processes such as early-stage nucleation and atomic-level restructuring. , Electron tomography can also be used to qualitatively evaluate complex NC structures and morphology, such as helicity measurements to report on single-NC chirality. Another auxiliary technique is 4D-STEM, where an electron diffraction pattern is collected at each probe position as the electron beam is scanned across the sample, forming an information-rich four-dimensional data set with a high spatial resolution (∼1 nm). 4D-STEM has been used to obtain high-resolution information about strain, lattice orientation, and dielectric field across single NCs, − though the large amount of data collected in this technique makes data processing time-intensive.

Compositional mapping of single NCs can be obtained by coupling STEM with either electron energy loss spectroscopy (EELS) or EDS. EELS measures the loss in energy of electrons by means of inelastic scattering as the electron beam interacts with the sample. EELS is superior compared to EDS in mapping light elements (Z < 11), making it a powerful tool for visualizing the distribution of organic capping ligands at the NC surface. EELS also has the ability to reveal information about bonding and valence states within materials and is commonly used to spatially map “hot spots,” or areas of electromagnetic field enhancement, in both single plasmonic NCs and at the interface formed between particles in close proximity. , In contrast, STEM-EDS measures element-specific X-rays and is much better at mapping heavy elements than STEM-EELS, being used to provide compositional information about complex alloy NCs and multimetallic site-specific growth. − Depending on the resolution of the STEM instrument and EDS/EELS detectors being used, both STEM-EDS and STEM-EELS can achieve atomic-level compositional resolution. ,

Advances in in situ TEM and STEM have made the visualization and study of various dynamic processes occurring at the single-NC level possible. Specifically, advances in TEM sample holders and use of graphene liquid cells provide gas- and solvent-tight chambers, − while specialized e-chips and holder ports allow imaging and tracking in a variety of environments, including thermal, electrochemical, pressurized/gas flow, , and liquid flow. Further, advancements in cameras used in this technology allow dynamic processes occurring at the millisecond time scale to be captured. EDS, EELS, and 4D-STEM can also be used in tandem with in situ (S)­TEM to study changes in composition and the crystal lattice as well as changes in electronic properties of single NCs in real time. ,, A significant question that continually plagues in situ electron microscopy experiments is, to what extent does the electron beam affect, even at low doses, the dynamic processes being recorded? To minimize these effects, careful protocols and best practices have been implemented by the community, with control experiments and benchmarking with ex situ data being important components in validating in situ results.

In comparison to electron microscopy, optical microscopy techniques can be inexpensive, high-throughput alternatives to identify interparticle heterogeneity, especially in plasmonic NC samples. However, since optical microscopy techniques probe samples with much lower energy than electron microscopy, optical images of single NCs are diffraction-limited, meaning that small-scale intraparticle defects cannot be directly visualized. This limitation can be overcome using super-resolution imaging, where single NCs are tagged with a series of probes that can be localized at the NC surface with a spatial resolution of ∼10 nm, revealing NC structure in optical reconstructions. Such studies can also be used to reveal the effect of both interparticle and intraparticle heterogeneity on single-NC catalytic activity, , though these super-resolution techniques suffer in part from lower throughput and challenging sample preparation compared to other optical methods as well as possible alteration of the NC properties resulting from probe functionalization.

Without additional probes, wide-field optical microscopy techniques can be used to capture interparticle heterogeneity in NC samples, where many single particles are imaged simultaneously. As the optical response of a single NC is highly sensitive to its physical features, NCs of different sizes, shapes, and compositions can be compared and distinguished through differences in their optical properties. Further, the short integration times and in situ capabilities of optical microscopy make it an ideal candidate to track dynamic changes in single-NC structures under various conditions and applications. , Of course, the “structural” information obtained from diffraction-limited optical data is qualitative and must be inferred, where correlated electron microscopy is often used to confirm such interpretations and determine structure–property relationships.

The wide variety of optical techniques available allows for many different NC systems and applications to be studied. Scattering-based techniques such as dark-field optical microscopy are perhaps the simplest methods to monitor many single NCs simultaneously but are limited to observing large (>40 nm), plasmonic nanoparticles. , Absorption-based techniques such as photothermal microscopy allow for smaller plasmonic nanoparticles to be imaged and are believed to provide greater information about NC morphology, but they require more complex experimental setups and suffer from low throughput. , Similarly, interference-based approaches allow for small, non-plasmonic NCs to be imaged, such as those desirable in catalysis, but again, they require more complex experimental setups. , Strategies such as point-spread function engineering and hyperspectral imaging have also been developed to extract higher-resolution information about NC characteristics. , One such technique employing point-spread function engineering, calcite-assisted localization and kinetics (CLocK) microscopy, will be discussed in detail in the Case Studies: Interparticle Heterogeneity section. Additionally, ML methods have been implemented with various optical microscopy and spectroscopy techniques, showing promise for obtaining NC size, shape, and aggregation state all-optically. − The current bottleneck in such a workflow is the large amount of correlated electron microscopy data required to train and validate an ML model; however, this bottleneck may be overcome by using high-quality simulations in place of correlated electron microscopy or automated electron microscopy data collection methods. −

Scanning probe techniques, where a sharp tip is scanned across a region to collect information locally about the sample surface, are powerful tools for single-NC characterization due to their high lateral and vertical spatial resolution. This high spatial resolution (sub-nanometer) arises due to the sharpness of the probe tip, which can be as small as a single atom, as well as the ability to minimize the distance between the tip and the sample surface to sub-nanometer distances through piezoelectric control. Due to the sensitivity of probe techniques for controlling the tip–surface distance, these techniques are also sensitive to atomic height variation on the sample surface, something that is challenging to visualize by electron microscopy due to the two-dimensional projection generated in (S)­TEM and the insufficient spatial resolution of SEM. One of the most mature scanning probe techniques is scanning tunneling microscopy (STM), in which electrons tunnel from a conductive probe tip to a conductive sample surface as a result of an applied bias voltage. The resulting tunneling current is highly dependent on the distance between the probe tip and the sample surface, affording the high vertical spatial resolution (∼1 Å) and atomic lateral spatial resolution of this technique. The high spatial resolution of STM allows for small defects, such as grain boundaries and even vacancies, as well as molecules adsorbed onto metal surfaces to be visualized. , To achieve atomic resolution, flat metal substrates are imaged under ultrahigh vacuum conditions, with a limited number of examples having achieved a high enough resolution to visualize surface ligands and defects for single NCs due to their 3-D structure/curvature and drift. , While single-crystalline and polycrystalline metal substrates are regularly used as model systems in atomic-resolution STM studies to estimate NC surface structure and adsorbate interactions, it remains unclear how well these models translate to NCs, given their 3-D structure and the complex surface energies arising from features such as vertices and edges. Beyond these challenges with atomically resolved NC structural characterization, scanning tunneling techniques have been shown to be a powerful means of analyzing single-NC electronic properties, giving insight into single-particle conductivity, density of states, and LSPR, which can be correlated to NC size and structure within the same instrument. −

In atomic force microscopy (AFM), both conductive and non-conductive samples can be analyzed as the measurement is based on the strength of atomic forces across the sample surface instead of tunneling current, like in STM. To form a topographical image in contact mode AFM, the height of a cantilever in contact with the sample surface is adjusted as it scans across the sample to maintain a specific force, which is affected by both attractive and repulsive intermolecular forces between the tip and sample. Because of weak NC–substrate interactions, NCs are often imaged using non-contact or tapping AFM modes, where minimal contact between the particle and cantilever can prevent movement of the particle on the substrate during imaging. For both bulk materials and single NCs, the spatial resolution (nanometers) is lower than that of STM, but AFM is superior in its ability to maintain more of its spatial resolution in non-ultra-high-vacuum environments, including in solution and at many temperatures and pressures. Besides its ability to characterize NC size, structure, and surface topography, , other contact, non-contact, and dynamic modes for AFM have been used to measure single-NC properties such as adhesion forces, Young’s modulus, NC–adsorbate binding interactions, surface energy and charge, , and hydrophobicity. Further, technique adaptations such as modifying the cantilever surface or applying a bias voltage between the cantilever and sample can be used to measure single-NC magnetic properties or simultaneous topography/surface electrochemical variations across the particle, respectively. −

Compared to scanning electron microscopy techniques (SEM and STEM), both STM and AFM have lower temporal resolution (s^–1 to ms^–1 scan rate as opposed to μs^–1 scan rate for SEM/STEM), limiting the event types that can be captured in situ. However, STM and AFM eliminate the concern of electron beam effects in in situ studies that afflict electron microscopy studies. Electron microscopy also benefits from the ability to obtain compositional information on single NCs through commercially available coupled techniques, though STM and AFM can measure particle height with high accuracy and obtain electrical, magnetic, and mechanical information on single NCs. A significant benefit to STM and AFM is that they can characterize the size, structure, height, and surface topography of thin two-dimensional NCs, whereas (S)­TEM can usually only capture the lateral dimensions and shape of these materials, and SEM often struggles to visualize these weakly electron-scattering NCs at all.

Mass spectrometry (MS) generates ions from organic or inorganic compounds using an ion source, then separates these ions according to their mass-to-charge ratio (m/z), and detects them quantitatively and qualitatively in a mass analyzer. Traditionally, MS has been used in nanoscience to provide elemental identity and abundance, and when coupled with particle size information, particle concentrations can be calculated for colloidal samples. The amount of sample needed to generate ions in enough abundance is extremely low, sometimes nanograms of the sample; however, the sample consumed by the analysis cannot be recovered, which differs from other analytical techniques such as nuclear magnetic resonance (NMR) spectroscopy or infrared (IR) spectroscopy, as well as many of the techniques described above. Numerous mass spectrometers proclaim a sensitivity at the attomole level; however, this sensitivity comes at a cost since an attomole of a sample may not reflect the whole sample composition but instead only a fraction of the ions in a sample. As the field of MS has advanced, various MS techniques have proven to be powerful tools in both high-throughput and high-resolution single-particle characterization of nanostructures. While we confine the discussion of MS techniques hereafter to those focused on analyzing the inorganic NC core, some MS techniques, such as matrix-assisted laser desorption/ionization (MALDI) MS, have been used to analyze the ligand shell environment at the NC surface. These techniques are limited in the types and sizes of ligands that can be studied and are not typically considered to be single-particle techniques. −

Single-particle inductively coupled plasma MS (SP-ICP-MS) has arisen as a high-throughput method to obtain size and compositional information on single NCs, where hundreds of individual NCs can be analyzed in seconds. In the traditional use of ICP-MS for NC analysis, particles in a suspension are first digested using acid to break the NCs into metal ions, then the ionic solution is introduced to the ICP instrument to obtain bulk elemental concentrations. Conversely, in SP-ICP-MS, single particles are directly introduced to the plasma torch for atomization and ionization, negating the need for bulk acid digestion and retaining single-particle information within segregated ion plumes that reach the mass analyzer in discrete intervals. This separation is achieved by diluting the aqueous NC suspensions being injected into the nebulizer by approximately 1000-fold, which simultaneously minimizes matrix effects during analysis resulting from the NC synthesis (such as capping ligands). As many NCs are formed in non-aqueous conditions, the limitation of incompatibility of this technique with most organic solvents can be bypassed through ligand exchange protocols. After measurement, NC diameter or edge length can be calculated using shape-dependent density equations, where density refers to the density of the bulk metallic material and particle mass in the equations is obtained from individual ion plume intensities from SP-ICP-MS measurements. , A more detailed explanation of the shape-dependent density equations and data processing has been provided by Koolen et al. SP-ICP-MS has been used to make size histograms from thousands of single-particle measurements for monometallic NCs with diameters as small as sub-5 nm and shapes beyond spheres, including cubic, octahedral, and tetrahedral NCs. − The predominant particle size estimated by SP-ICP-MS agrees well with electron microscopy data, though the large sampling number of SP-ICP-MS has allowed for subpopulations within the size distributions to be visualized, which can be missed by the small number of particles often measured in time-consuming electron microscopy measurements. SP-ICP-MS has also been used to quickly obtain the composition of individual multimetallic NCs within large NC populations, , though extracting accurate information about multimetallic NC size from mass data is challenging due to complicated particle densities and atom size differences in multimetallic NCs. Currently, SP-ICP-MS suffers from the inability to distinguish NC shape, requiring the user to often confirm the predominate NC shape within the sample by other techniques such as TEM or SEM. The reliance of SP-ICP-MS data processing on shape-dependent density equations can lead to inaccuracies in the resulting mass-to-size calculations if more than a single shape is present within the sample, though various separation techniques have been paired with SP-ICP-MS in an effort to increase the types of NC heterogeneity that can be distinguished. ,

Another high-throughput MS technique, charge-detection MS (CD-MS), which is often used to analyze large biomolecules, has also been used to obtain single-particle information on nanoscale particles. In CD-MS, the charge of an ion (a single, charged particle in the case of NCs) is measured through the charge (z) induced by the ion on the walls of a conducting tube, while the m/z is simultaneously obtained from the time-of-flight of the ion through the tube. Because both the z and m/z are measured for each charged particle, the mass of thousands of single particles can be calculated and converted into size histograms using shape-dependent density equations like in SP-ICP-MS data processing. While CD-MS cannot provide precise compositional data like SP-ICP-MS because the particles are not atomized, CD-MS has been shown to be sensitive to the particle surface, where particles with similar masses but different surface properties (e.g., capping ligands, etc.) can be distinguished due to differences in z after ionization. However, CD-MS instruments are still much less common than quadrupole ICP-MS instruments, which are able to analyze mono- and bimetallic NCs in SP-ICP-MS (whereas the more rare and expensive time-of-flight ICP-MS is required for compositions >2). , Currently, most nanoparticle CD-MS studies have focused on polymeric or oxide nanoparticles, and particle masses between 1 MDa and a few hundred MDa have been found to be sufficient for analysis. , For reference, a 10 nm Au nanoparticle (assumed to be perfectly spherical with a density equal to bulk Au, 19.3 g/cm³) is expected to have a mass around 6 MDa, which theoretically falls within the typical reported analysis range for non-metallic nanoparticles by CD-MS, though detection limits for metal NC compositions have not yet been thoroughly investigated for CD-MS as they have for techniques like SP-ICP-MS. This data will be of growing interest and necessity for relevant metal compositions as CD-MS likely continues its expansion into the high-throughput metal and metal oxide NC characterization space.

Recently, single-particle mass and volume measurements of large populations of silica nanoparticles, acquired by CD-MS and resistive-pulse sensing (RPS), respectively, allowed for the observance of size-dependent silica particle density changes for particles ranging from 19 to 72 nm in diameter. While not an MS technique, RPS is a high-throughput microfluidic technique capable of estimating certain single-particle structural features through signal processing and geometric modeling, similar to SP-ICP-MS and CD-MS. In RPS, ion displacement resulting from the movement of a particle through an electrically biased nanopore results in a current pulse that has an amplitude proportional to the particle volume (Figure a). Further, the particle dwell time in the pore (i.e., current pulse width) may give insight into particle shape and length, and the pore-to-pore time (i.e., the time between current pulses) can provide information about particle ζ-potential (Figure b). While being unable to obtain NC composition like in SP-ICP-MS, the biggest advantage of RPS over both SP-ICP-MS and CD-MS is the ability to estimate the size of single particles with unknown densities (i.e., multimetallic NCs) by solving volume equations directly, although these equations are still shape-dependent. Analysis of metal NCs by RPS can be complicated compared to other analytes due to particle conductivity and surface charges affecting ion flux through the pore and therefore the characteristics of the resulting current pulse. , The impact of NC conductivity can be minimized through adjustment of pore size, electrolyte salt concentration, and applied bias voltage as well as the use of particle calibration standards to decrease the error in particle volume calculated from current pulse amplitude. Still, there are currently few examples of analysis of metal NCs by resistive-pulse techniques, − though the non-destructive continuous flow analysis afforded by this technique makes it uniquely capable of direct coupling to flow synthesis reactors and other single-particle characterization methods compared to the majority of other techniques discussed herein.

2.

(a) SEM image of an example detecting region used in resistive-pulse sensing measurements, where particles in solution enter the channel through a nanofilter to limit aggregates entering and clogging the nanopores, and multiple pores in series are used to increase the measurement precision. (b) Current vs time trace resulting from resistive-pulse measurements of a single 50 nm silica particle passing through the three-nanopore series, where the current pulse amplitude (Δi), dwell time (t _d), and pore-to-pore time (t _pp) are labeled. (c) Schematic showing one preparation method for APT specimens containing NCs, where Au@Ag core@shell nanoparticles were dried onto a Si substrate before being sputter coated with a Cr layer and then covered with a Pt layer by chemical vapor deposition. A focused ion beam instrument is then used to carve and lift out a probe tip-shaped portion of the layered specimen (black dashed line), with the nanoparticles suspended in the middle of the formed tip. (d) HAADF-STEM image of an APT specimen fabricated using a similar method as shown in part (c) except using Pd@Au core@shell nanoparticles (visualized as white spots in the inset magnified image) embedded in a Ni matrix. (e) Three-dimensional APT reconstruction (top) showing heterogeneity on the surface of an isolated Au@Ag nanoparticle from the sample described in part (c) as well as a 2 nm thick cross-sectional slice through the particle showing the core@shell morphology when Au and Ag are mapped together (middle) and showing a small internal cluster when just Ag is mapped (bottom). Parts (a–b) adapted with permission from ref . Copyright 2024 American Chemical Society. Parts (c, e) adapted with permission from ref . Copyright 2014 Wiley-VCH Publishing. Part (d) adapted with permission from ref . Copyright 2020 Elsevier B.V.

For high-resolution MS techniques, there are two types of resolutions that should be considered: mass resolution and spatial resolution. Mass resolution is determined by the mass analyzer and is the smallest difference in m/z that can be separated, whereas spatial resolution refers to the minimum sampling spot size where an isolated mass spectrum is obtained in MS mapping techniques. One such high-spatial-resolution MS technique capable of analyzing single NCs is atom probe tomography (APT), which allows for the 3-D compositional mapping of a sample with a sub-nanometer spatial resolution and parts per million elemental sensitivity. This sensitivity is accomplished by applying a high voltage to a sample, resulting in the ionization and desorption of individual atoms that are then characterized based on their m/z. Each data point is then used to build a 3-D reconstruction of the sample volume containing spatially resolved compositional information, which also gives information about particle size and shape. An important distinction of APT is that the sample must be formed into the shape of a probe tip with a tip diameter of around 100 nm, through processes such as focused ion beam milling for bulk materials, in order to achieve the electrostatic field magnitudes (∼10 V nm^–1) required to remove the single atoms from the material surface. Many methods have been designed for preparing NC samples for APT analysis such as coating a Si probe tip with NCs by electrophoresis, or encasing one or multiple NCs within a matrix material through electrodeposition before sharpening to a tip using a focused ion beam (Figure c,d). APT has been used to visualize individual NC phenomena such as trace Na and Cl impurities in 7 nm carbon-supported Pt nanoparticles as well as to visualize phase-segregated clusters hidden within individual core@shell nanoparticles (Figure e). − The small phase-segregated clusters found inside of the nanoparticles by APT would likely not be possible to visualize by traditional STEM-EDS/EELS due to each point in a 2-D STEM-EDS/EELS map corresponding to the compositional average of that point along the height of the particle (i.e., the z -axis). APT has also been used to visualize thin shells (down to a single atomic layer) in core@shell nanoparticles, where the thin shell could not be visualized by STEM-EDS due to detection limits of the EDS detector and the core and shell metals not having a large enough Z-contrast to be visually distinguished. Although three-dimensional compositional mapping can be performed using EDS tomography within the STEM instrument, it does not match the compositional spatial resolution of APT, and issues can arise due to detector shadowing and sample damage from long electron beam exposure time. Despite the unmatched intraparticle compositional detail that APT can provide for individual NCs, it is still not widely adopted in single-NC studies given the limited number of instruments available across institutions, its high cost, as well as challenging sample preparation.

Secondary-ion MS (SIMS) is another high-spatial-resolution MS technique where an ion beam is focused and scanned across a sample surface (substrate-supported particles in the case of NC analysis), ejecting secondary ions from the sample that are then guided to a mass analyzer and characterized based on their m/z. Currently, SIMS is limited to a lateral spatial resolution of nearly 100 nm, but the compositional mapping resolution can be improved to sub-20 nm through correlative SIMS-microscopy imaging and data processing or by inserting SIMS instrumentation and an ion source directly into different types of microscopes, including helium ion, , SEM, and (S)­TEM instruments. The compositional data obtained by SIMS is highly surface-sensitive, as only secondary ions from the first few atomic layers are ejected from the sample surface and detected. This sensitivity significantly differs from compositional analysis by (S)­TEM-EDS/EELS, where a very high compositional mapping resolution is typical (∼1 nm), but the composition at each spot is an average across the entire sample height and only has surface-sensitivity at the sample edges. Advantages of SIMS over EDS/EELS include its much lower detection limit (ppm compared to >0.1 wt %) and its ability to detect both light and heavy elements and distinguish between isotopes, though coupled electron microscopy-EDS/EELS instrumentation is widely commercially available and protocols are well-established. Although not widely used as such now, SIMS may be used to assess quantitative differences in surface composition between single NCs in an ensemble, even for surface elements at very low concentrations. Future technique improvements may also allow for spatial resolutions favorable for intraparticle mapping of NCs, giving a unique insight into compositional variations across NC surfaces.

Table summarizes the opportunities of the various analytical techniques discussed, comparing their ability to characterize different structural heterogeneities within and between NCs and their in situ capabilities and general limitations. Overall, though many of these structural and compositional characterization techniques have the spatial resolution required to visualize interparticle and intraparticle NC characteristics, logistics such as finding and placing the probe at a single NC remain a time-consuming challenge. Further, as each of these techniques has its own limitations, the most robust analysis of NCs often requires multimodal approaches where structural techniques are coupled with other such techniques (often analyzing different length scales such as single particles versus the ensemble) as well as coupled with techniques that measure properties. The following case studies demonstrate the ability of some of these analytical techniques to be used together to correlate NC structure and properties at the nanoscale.

1. Features of Analytical Techniques Used to Characterize NC Heterogeneity on Different Length Scales.

technique	size	structure	composition	surface	in situ capabilities	limitations
electron microscopy
SEM (low throughput)	yes; spatial resolution: ∼1–10 nm	yes	EDS: spatial resolution: 10 nm–1 μm depending on beam energy; primarily qualitative	organic ligands can decompose during imaging, forming a “glue” like substance visualized around NC	yes: heating, stretching, compression, environmental, electrochemical	conductive sample required; particle damage by beam
					temporal resolution: limited by scan rate (μs–1 scan rate typical)
TEM (low throughput)	yes; spatial resolution: sub-nm	yes, but limited by 2-D projection	EDS: spatial resolution: 1–10 nm; elemental sensitivity: >0.1 wt %	yes; can visualize surface defects and sometimes ligand shells with HRTEM	yes: heating, gas (up to 1 atm), liquid, electrochemical	thin sample (<200 nm) required; particle damage by beam
	HRTEM: spatial resolution: <1 Å	electron tomography provides 3-D			temporal resolution: limited by camera used (acquisition times ranging from ms to s)
STEM (low throughput)	yes; ABF HAADF spatial resolution: <1 Å and aberration-corrected HAADF spatial resolution: ≪1 Å	yes, but limited by 2-D projection	EDS: spatial resolution: ∼1 nm; better for heavy elements (Z > 11)	yes; can visualize surface defects	yes: heating, gas (up to 1 atm), liquid, electrochemical	thin sample (<200 nm) required; atomic-resolution imaging requires clean particle surface (beam shower + plasma cleaning often necessary); particle damage by beam and/or cleaning
			EELS: spatial resolution: ∼1 nm; good for light elements; can differentiate oxidation states	EDS: can detect halide capping ligands	temporal resolution: limited by scan rate (μs–1 scan rate typical)
				EELS: can visualize organic capping ligands
optical microscopy
scattering-based techniques (high or low throughput depending on experimental geometry)	diffraction-limited; best signal comes from large (>40 nm), plasmonic NCs	point-spread function engineering, polarization-resolved imaging, hyperspectral imaging can be used to infer information about NC structure	indirectly: composition changes the refractive index, thus changing scattering color/LSPR spectra	indirectly: surface coverage changes the refractive index, changing scattering color/LSPR spectra	yes: heating, gas, liquid, electrochemical	diffraction-limited; NC size limited to those larger than what is often relevant in catalytic applications
					temporal resolution: acquisition times typically on the time scale of ms
absorption-based techniques (low throughput)	diffraction-limited; can image small (2–10 nm) plasmonic NCs	shape information can be inferred from absorption spectra	indirectly: NCs of different composition will have different thermal properties	indirectly: surface coverage will change the refractive index, changing LSPR spectra	yes: heating, gas, liquid, electrochemical	diffraction-limited; often limited to single-wavelength measurements; complex experimental setup using confocal geometry
					temporal resolution: acquisition times typically on the time scale of ms to s
interference-based techniques (high or low throughput depending on experimental geometry)	diffraction-limited; can image small (∼5 nm) plasmonic and non-plasmonic NCs	shape information can be inferred from spectra	indirectly: NCs of different composition will have different refractive indices and damping, affecting phase relationships	indirectly: NCs with different surface coverage will have different refractive indices and damping, affecting phase relationships	yes: heating, gas, liquid, electrochemical, probe techniques such as SECCM	diffraction-limited; complex experimental setup
					temporal resolution: acquisition times typically on the time scale of ms
probe techniques
STM (low throughput)	yes; lateral spatial resolution: subnm in ideal conditions and samples vertical resolution: ∼1 Å	yes	no	yes; can visualize adsorbed species and intraparticle surface defects in ideal conditions and samples	yes: liquid, heating, gas, electrochemical	conductive and flat sample required; lower temporal resolution compared to EM
					temporal resolution: limited by probe feedback loop and scan rate (s–1 to ms–1 scan rate typical)
AFM (low throughput)	yes; lateral spatial resolution: ∼1–10 nm vertical resolution: sub-nm	yes	no	yes; surface topography indirectly: ligand-NC interaction strength inferred from particle height measured	yes: heating, stretching, compression, liquid, gas, electrochemical	lower temporal resolution compared to EM; cantilever damage in contact mode
					temporal resolution: limited by probe feedback loop and scan rate (s–1 to ms–1 scan rate typical)
mass spectrometry
SP-ICP-MS (high throughput)	indirectly: particle mass-to-size conversion when monometallic NC shape assumed; monometallic minimum particle size: <5 nm	possibly with technique development and particle atomization (plume) engineering/analysis	yes; elemental sensitivity: down to parts per trillion (ppt)	no	no	destructive technique
CD-MS (high throughput)	indirectly: particle mass-to-size conversion when monometallic NC shape assumed; minimum particle size: not yet well-established for metal NCs	no	no	indirectly: gives information about particle charge, which can be affected by capping ligands	no	not widely accessible; particle damage
APT (low throughput)	yes; spatial resolution: subnm	yes	yes; elemental sensitivity: ppm typical	can detect impurities and thin shells on the NC surface that cannot be visualized by STEM-EDS/EELS	no	difficult sample preparation; destructive technique; expensive; not widely accessible
SIMS (low throughput)	not accurately; spatial resolution: down to sub-20 nm	can differentiate large shape differences	yes; elemental sensitivity: ppm typical	yes; only composition of first few atomic layers obtained (<2 nm depth)	yes: heating	coupled SIMS-microscopy instruments not widely accessible; particle damage by beam
other
RPS (high throughput)	indirectly: particle volume-to-size conversion when NC shape assumed; minimum particle size: >10–20 nm	sometimes: NC structure may influence current pulse width/particle dwell time in pore	no	volume displacement and particle dwell time affected by particle capping ligands and charge	yes: reagent addition/mixing, optical imaging techniques	NC aggregation in required ionic solution; particle volume calculation more complex for conducting (metal) particles

Case Studies: Intraparticle Heterogeneity

Despite their sometimes atomic-scale size, intraparticle heterogeneities can influence the electronic structure, catalytic activity, optical response, mechanical strength, and more of NCs by creating localized regions with distinct behavior. For example, compositional and crystallographic gradients may give rise to distinct catalytic activity and selectivity within an NC. This strong dependence of properties on structure makes precise and correlative characterization of these heterogeneities critical for various applications, with examples to the fields of catalysis and plasmonics described herein.

In the case of catalysis, the intersite distance (d _site) of single atoms on the surface of NCs (referred to as single-atom alloys and can be viewed as point defects) has been shown to alter the activity of NCs. Wang et al. studied the effect of Au d _site in Au–Ag single-atom alloys on the oxygen reduction reaction (ORR) using a single-particle collision electrochemistry method. Particle collision electrochemistry, also referred to as stochastic electrochemistry, allows for many single-particle events to be captured one-by-one at a micro- or nanoelectrode in a short amount of time by relying on the random collision of nanoparticles dispersed in the electrolyte with the electrode. In contrast, more traditional NC electrochemical methods rely on a single ensemble-averaged measurement involving hundreds of stationary NCs anchored to the electrode surface, which may misrepresent or oversimplify NC structure–property relationships. The high-throughput single-particle measurements afforded by collision electrochemistry methods sacrifice the ability to perform correlated experiments with structural techniques such as electron microscopy, as the NCs measured are not anchored to the electrode. Structure–property correlation using this technique is thus often reliant on statistical analysis and inference, though collision methods can be more relevant to a wider range of materials (e.g., non-catalytic materials through blocking methods) than traditional methods. In the collision method employed by Wang et al., single Ag nanoparticles that collide with the surface of an ultramicroelectrode first trigger deposition of single Au atoms on the Ag nanoparticle surface from the reduction of NaAuCl₄ in the surrounding electrolyte. This collision process is also coupled with an instantaneous measurement of the ORR activity of the newly formed single-atom alloy particle. The d _site in the formed single-atom alloys was controlled by varying the concentration of NaAuCl₄ in the electrolyte solution, and its value was estimated theoretically by calculating the number of Au atoms deposited during the collision event and assuming that the Au atoms were evenly distributed across the Ag particle surface. Single-atom alloy nanoparticles formed by collision with the ultramicroelectrode were collected for imaging by aberration-corrected HAADF-STEM by injecting Ag nanoparticles one-by-one toward the surface of the ultramicroelectrode using an injection nanopipette in close proximity to the electrode. This specialized preparation method greatly decreases the throughput of the collision method but limits the number of pure Ag nanoparticles in the resulting STEM sample. HAADF-STEM imaging of the Au single atoms on the Ag host particle for the largest NaAuCl₄ concentration studied (Figure a,b) revealed an Au d _site value (1.2 nm) in good agreement with the value predicted theoretically (1.3 nm).

3.

(a) Aberration-corrected HAADF-STEM image of Au single atoms on an individual Ag NC and (b) corresponding distance energy spectrum between two neighboring Au atoms (d _site). (c) TOF per Au atom for the ORR as a function of d _site, with the error bars representing the standard deviation between 3 independent experiments. (d) HRTEM images of the tip of a defect-free (“normal”) Au nanostar and the tip of a defective Au nanostar with twin boundaries and superlattice defects. (e) Finite-element method-based electromagnetic field distribution simulations of a defect-free Au nanosphere (left), defect-free Au nanostar (middle), and defective Au nanostar (right) under various excitation wavelengths. (f) SERS spectra of 4-nitrobenzenethiol (4-NTP) collected on an Au nanosphere film substrate (red) and a defective Au nanostar film substrate (black) under an excitation wavelength of 785 nm. Parts (a–c) adapted with permission from ref . Copyright 2024 American Chemical Society. Parts (d–f) adapted with permission from ref . Copyright 2022 American Chemical Society.

To assess the effect of d _site, the turnover frequency (TOF) per Au atom for the ORR was calculated and mapped for each Au–Ag single-atom alloy studied (Figure c). When Au d _site decreased from 4.0 to 1.9 nm, the TOF increased by 1.8 times; however, the TOF declined when d _site was further reduced to 1.3 nm. Notably, Au–Ag single-atom alloys with a d _site of 1.9 nm exhibited the highest TOF, making them the most efficient ORR electrocatalyst. The optimized Au d _site is believed to enhance O₂ adsorption and weaken the binding strength of ORR intermediates, thereby improving activity. These findings demonstrate that even the smallest-scale intraparticle defect, specifically low concentrations of surface substitutions, may have a significant impact on the resulting properties, such as on the catalytic activity, of a particle. Further, the use of a collision electrochemical method for these intraparticle catalytic studies allowed for the high-throughput synthesis and near-instantaneous characterization of single particles, which would be a much lower throughput and tedious task if other methods were employed.

Intraparticle defects also have the potential to influence the plasmonic properties of the metal NCs. For example, while it has been well-established that large-scale features such as particle size, shape, and composition affect the optical properties of an NC, , Guo et al. recently showed that twin boundaries in the tips of Au nanostars can significantly impact the electromagnetic field enhancement and corresponding SERS intensity in what are usually highly active samples. Observed using HRTEM, the twin boundaries and superlattice defects are believed to shorten the mean free path of electron scattering, increasing electron oscillation damping and in turn weakening the electromagnetic field enhancement and SERS activity of the nanoparticle (Figure d). In finite-element method-based electromagnetic field distribution simulations, the effect of such defects can be modeled using a bulk damping term (Γ_b). Figure e shows that the simulated electromagnetic field enhancement for a defective Au nanostar (25Γ_b, right) is significantly lower than the electromagnetic field enhancement of a normal Au nanostar (1Γ_b, center) and even a tip-less Au nanosphere (1Γ_b, left). The simulated trends agree with the experimental observation that Au nanospheres have a greater enhancement factor than defective Au nanostars, with a nearly 40 times greater enhancement factor being observed experimentally at an excitation wavelength of 633 nm. This enhancement leads to the SERS intensity of 4-nitrobenzenethiol (4-NTP) collected on a film of Au nanospheres being greater than the intensity of 4-NTP collected on defective Au nanostars, contradicting the notion that sharp NC features will result in greater electromagnetic field enhancement and therefore greater SERS intensity compared to rounded edges (Figure f). Such a result demonstrates not only the need for analytical techniques capable of capturing atomic-level features but also the need for greater examination of the role that small intraparticle defects play in resulting NC properties compared to more “obvious” features such as NC size, shape, and composition. This study also demonstrates the coupling of single-particle (HRTEM) and ensemble (SERS) techniques, bridged by computational insight, to build a complete understanding of this specific structure–property relationship. Further, using single-particle insights to better understand how defects propagate effects to the ensemble-level is especially important as some NC applications, such as biosensing and batteries, require NC ensembles.

Case Studies: Interparticle Heterogeneity

Interparticle heterogeneity can lead to broad distributions in the physical and chemical properties across an NC ensemble. This heterogeneity can impact collective behaviors like optical absorption (e.g., LSPR bandwidth), catalytic activity, and even electronic conductivity by introducing different local environments and reaction sites. − This heterogeneity can lead to an inaccurate understanding of structure–property relationships while also limiting the reproducibility of ensemble NC responses. Thus, characterizing interparticle heterogeneities and correlating them to properties is essential, as highlighted with examples from the fields of catalysis and plasmonics herein.

For plasmonic NCs, the correlation of single-particle optical properties such as the LSPR spectrum to the size, shape, and/or composition of an NC is critical, as changes to NC structure can impact performance in applications such as refractive index sensing. Traditionally, this correlation requires linked optical and electron microscopy measurements, which makes the evaluation of the structure–function relationship of many single particles extremely inefficient. Recently, O’Dell et al. demonstrated the ability to obtain Au nanorod structure and its corresponding optical properties in the same image using CLocK microscopy. In CLocK microscopy, a rotating birefringent calcite crystal is placed in the emission path of an optical microscope, producing a polarization-resolved extraordinary ray (e_⊥) ring around a traditional polarization-averaged diffraction-limited spot (Figure a). The polarization-resolved nature of the e_⊥ ring allows for NC structure, orientation, and information about its polarization-resolved scattering spectrum to be captured in a single image in less than 1 s. As can be seen in Figure b, the orientation of the red/orange lobes of the e_⊥ ring in the CLocK image indicates the orientation of the long axis of an Au nanorod on a substrate, as confirmed by the correlated SEM image. The relative intensities of the red, green, and blue (RGB) color channels at various e_⊥ ring orientations in the CLocK image report on the positions and magnitudes of the longitudinal and transverse plasmon modes traditionally measured by using polarization-resolved single-particle dark-field scattering spectroscopy. The RGB polar plot obtained from the polarization-resolved spectrum agrees well with the RGB polar plot obtained by fitting the CLocK image, supporting the idea that color CLocK images accurately capture single-particle scattering spectral information. Thus, with shorter integration times and wide-field capabilities, CLocK microscopy can be used to more efficiently infer information about Au nanorod length and width, allowing for Au nanorods of various dimensions and orientations to be identified and differentiated all-optically without further post-processing or structural characterization by electron microscopy (Figure c). Using ML approaches, specifically a convolutional neural network, Au nanorod length, width, and aspect ratio were predicted directly from CLocK images within ∼10% of the true value measured using electron microscopy. This wide-field, all-optical approach of obtaining structure–property relationships is extremely valuable in quickly identifying interparticle heterogeneity in NC samples and can potentially be used to identify over- or underperforming NCs in various applications when coupled with in situ techniques and other property measurements.

4.

(a) Schematic of the CLocK microscope. (b) Correlated SEM image, CLocK image, and polarization-resolved dark-field scattering spectra for a single Au nanorod. The RGB polar plots are provided to compare the optical properties obtained from CLocK (top) with the optical properties measured by using a spectrometer (bottom). SEM scale bar = 100 nm. CLocK scale bar = 1 μm. (c) Wide-field color CLocK image with the corresponding numerically labeled and color-correlated SEM images. SEM scale bars = 100 nm. CLocK scale bar = 5 μm. Parts (a–c) adapted with permission from ref . Copyright 2024 American Chemical Society.

Single-particle property measurements have also made it possible to better understand the implications of interparticle heterogeneity for common NC applications such as catalysis. For example, SECCM, whereby an electrolyte-filled micro- or nanopipette forms a droplet at its tip that acts as an electrochemical cell (Figure a), has allowed for electrochemical properties of single NCs dispersed onto an electrode to be individually measured. Besides the use of SECCM to form a small electrochemical cell that completely envelopes a single NC for study, SECCM has also been used to map the electrochemical activity and topography across the surface of particles to assess intraparticle differences, , as well as to map the heterogeneity in various electrode materials, such as indium tin oxide (ITO). Though they are both single-particle electrochemical techniques, SECCM differs from collision electrochemical methods as the NCs of interest are anchored onto the electrode surface, and the electrochemical cell formed by the nanopipette droplet is the dynamic component as opposed to the NCs comprising the dynamic component in collision methods. The stationary state of the NCs in SECCM allows for correlated techniques, such as electron microscopy, to be performed after electrochemical measurement, as well as the length of the experiment to be controlled, though the throughput of the measurements is greatly reduced compared to collision methods. A significant challenge in SECCM is the placement of the pipette onto the electrode since scanning the cell in a grid pattern across the electrode surface is time-intensive, with the risk of only partially enveloping an NC of interest. Because electron microscopy can damage the NCs of interest and deposit contaminants such as carbon onto the NCs while imaging, optical techniques have been employed before SECCM measurement to better target NC locations on the electrode.

5.

(a) Schematic depicting the SECCM setup. (b) SECCM map and correlated SEM images for the CO₂RR (−1.3 V vs RHE). (c) SECCM-derived and (d) macroscale electrochemically active surface area (ECSA)-derived TOF_CO vs potential plots comparing the CO₂RR activity of {111}-terminated octahedra (OD), {110}-terminated rhombic dodecahedra (RD), and {310}-terminated truncated ditetragonal prisms (TDP). (e) SEM images of a single Au nanocube (left) and Au octahedron (right) and (f) the averaged electrochemical response with error bars for HER at individual Au nanocubes (Au NCs) and Au octahedra (Au ODs) obtained by SECCM. Parts (a–d) adapted with permission from ref . Copyright 2022 American Chemical Society. Parts (e–f) adapted with permission from ref . Copyright 2020 American Chemical Society.

Recently, SECCM has been used to isolate the catalytic activity of different NC shapes and facet types toward the CO₂ reduction reaction (CO₂RR) and the hydrogen evolution reaction (HER). − Specifically, in one study, single-crystalline Au NCs predominantly exhibiting specific crystallographic planes, including {111}-terminated octahedra, {110}-terminated rhombic dodecahedra, {310}-terminated truncated ditetragonal prisms, and {100}-terminated cubes, were synthesized for SECCM studies. For CO₂RR, the SECCM pipette was scanned across a glassy carbon substrate containing dispersed particles, creating a map of electrochemical activity that was correlated with SEM images (Figure b). The turnover frequency of CO (TOF_CO), a key metric reflecting the intrinsic activity, was calculated to compare the performance of different Au facets. SECCM-derived TOF_CO showed that the CO₂RR activity follows the order of {110}-terminated rhombic dodecahedra > {310}-terminated truncated ditetragonal prisms > {111}-terminated octahedra across almost the entire potential range studied (Figure c). This same activity trend was observed from macroscale (ensemble) data at low overpotentials (potentials more positive than −0.7 V vs RHE). However, at higher overpotentials (potentials more negative than −0.7 V vs RHE), mass transfer limitations in macroscale measurements altered the trend to rhombic dodecahedra ≈ truncated ditetragonal prisms > octahedra (Figure d), obscuring the actual differences in activity. By combining SECCM and macroscale results, {110}-terminated Au rhombic dodecahedra were found to offer a superior activity for converting CO₂ to CO compared to both {111}-terminated octahedra and high-index {310}-terminated truncated ditetragonal prisms. For HER, single-particle SECCM measurements revealed that Au nanocubes exhibit higher HER electrocatalytic activity than Au octahedra (Figure e,f). These findings are significant to the understanding of the implications and tolerances of various applications toward interparticle heterogeneity, as all NC ensembles contain interparticle heterogeneities such as shape differences, which may yield different catalytic activities and selectivities depending on the shapes present. These case studies are also significant, as they demonstrate a rigorous protocol for correlating SECCM and NC structural characterization, which can be used to investigate other types of NC heterogeneity, such as composition or ligand density.

Conclusions and Future Outlook

These case studies represent the state-of-the-art in single-particle characterization, linking directly NC features to critical property measurements. However, certain property measurements are still not possible at the level of single NCs. For example, SECCM provides electrochemical readouts of single-NC catalysts, but to obtain product quantification from reactions generating multiple products, additional tools would need to be coupled with SECCM. Moreover, many single-NC techniques have low throughput, especially when moving between multiple analytical techniques to acquire correlative data. These limitations, coupled with the notion that single-NC techniques are experimentally complicated, have limited their adoption. However, we are inspired by the ability of single-NC studies to reveal fundamental NC features that give rise to functional properties and envision a future in which single-NC studies are high-throughput, information-rich, and experimentally accessible to a diverse community of researchers.

Conceptually, the screening challenge presented by NC heterogeneity may be viewed similarly to that of drug discovery and design, where there is a large, multidimensional experimental space from which a lead target is to be identified. Notably, the process of drug discovery has been transformed over the past few decades by the creation of high-throughput assays to identify lead compounds that then undergo high-resolution characterization and computer-aided design until clinical candidates are identified. We propose that a similar workflow be adopted to accelerate single-NC screening and design, as outlined in Figure .

6.

Schematic of proposed workflow to accelerate single-NC screening of structure and properties.

While not the focus of this Perspective, the workflow would begin with model NC collection generation, which could include small-batch colloidally prepared NCs such as those highlighted in the case studies in the previous sections. However, NC collections could also be created by using high-throughput methods that refine NC ensembles to meet certain ensemble metrics. For example, automated batch and flow microreactor platforms have been used to synthesize nanoparticles, which can provide large experimental data sets that link process parameters to nanoparticle properties (e.g., fluorescence, absorbance, circular dichroism, etc.); the data sets from such experiments can then be used to train models for predicting synthetic outcomes using ML. − Additionally, combinatorial and megalibrary synthetic approaches offer spatially resolved nanoparticles on substrates compatible with a host of analytical techniques for accelerated materials discovery. −

Next, the model NC collections would be analyzed by high-throughput primary assays that provide rapid evaluation of ensemble heterogeneity and, ideally, of an NC feature or property of interest at the level of single NCs. Here, automation and artificial intelligence may offer opportunities to improve the efficiency of traditionally low-throughput techniques such as electron microscopy. At the same time, intrinsically high-throughput methods are being advanced to provide more correlative sample information. For example, optical methods offer high-throughput screening of ensemble heterogeneity, and the NC features that account for this heterogeneity are now starting to be obtained through all-optical methods such as CLocK microscopy. , These advances provide a tremendous opportunity to accelerate the identification of interesting NCs through the creation of hybrid tools (e.g., tools that combine optical with probe-based techniques) and by coupling other optical readouts as proxies for a property of interest (e.g., use of fluorescent probes in catalysis). , Additionally, SP-ICP-MS provides high-throughput screening of ensemble structural and compositional heterogeneity. Statistical analysis may link this heterogeneity to other property measurements as well. With this insight, single NCs identified as potentially interesting through these primary assays can then enter a high-resolution characterization loop.

As discussed, excellent tools for high-resolution characterization of NCs exist (e.g., analytical TEM), where now the most promising single NCs would be the focus after being identified through a high-throughput workflow rather than the whole NC ensemble. Central to the success of this workflow is the ability to analyze the same single NCs in both the high-throughput primary assays and high-resolution techniques providing property measurement and structural characterization. This logistical challenge is non-trivial, given the sample requirements of different analytical techniques; however, research is addressing this challenge. For example, optically transparent carbon electrodes have recently been reported and enable correlative single-NC measurements by optical microscopy techniques, electrochemistry, and TEM. , Notably, by first screening NC ensembles with high-throughput single-NC methods, scientists now could strategically select single NCs with different properties for high-resolution characterization rather than randomly characterizing as many single NCs as possible with the hope that the different particle types are captured in the measurements. In this manner, single-NC studies are accelerated while preserving the ability to obtain accurate structure–property relationships.

At the same time, we need to continue to push the frontiers of what types of high-resolution single-NC property measurements are possible, where, again, the creation of new hybrid tools for analytical nanoscience represents a key opportunity. With these advances, computational methods play a crucial role, assisting experimentalists in validating techniques and providing a better understanding of structure–property relationships down to the atomic scale. Traditionally, first-principles calculations based on density functional theory (DFT) and molecular dynamics (MD) have been used to investigate the electronic and structural properties of NCs. , These calculated properties can aid in predicting the thermodynamic and kinetic behavior of species interacting with NC surfaces, defects, and interfaces, including surface reactivity. For instance, DFT can predict the hydrogen binding energy at different facets, which is an important descriptor for reactivity in the HER. This capability, when coupled with single-NC measurements of HER by SECCM and SEM imaging, provides the atomic-level detail essential for active site identification, which is a challenge for NC ensembles with high levels of heterogeneity.

At this point, NC optimization and design become possible with reliable structural and compositional inputs. Modern supercomputers have facilitated high-throughput screening and the data-driven expedition of novel nanomaterials. Structure generation methodologies such as Monte Carlo sampling, Cluster Expansion (CE), and Genetic Algorithm (GA) facilitate computational sampling for material screening. The rise of artificial intelligence has sparked significant interest in its application to NC design by introducing machine-learned potentials (MLPs), which define the potential energy surface (PES) of materials. These MLPs provide a promising alternative to traditional methods for modeling complex NC systems with a far-reduced computational expense as compared with DFT, nearing comparable accuracy. Many open-source and pretrained models such as CHGNET and MACE have been developed to accelerate simulations of NCs. , Furthermore, generated computational and experimental data can be used to train and refine MLPs for better accuracy and efficiency in future modeling efforts.

With such design principles computationally identified, the workflow can leverage automated synthetic methods for NC collection generation to target and optimize an ensemble in a high-throughput manner. In this NC optimization step, a key consideration is the synthetic feasibility. That is, once an NC feature is identified as functionally important through the workflow outlined in Figure , this feature will ideally be amplified in the NC ensemble. For some features, synthetic tools likely already exist (e.g., expression of low index surface facets). However, other features may be difficult to replicate at scale, given the current synthetic toolkit. In this regard, the workflow may identify the frontiers of nanomaterial synthesis. Yet, at the same time, this possibility also highlights that there may be benefits to keeping NC heterogeneity within a manageable level at the initial stage of model NC collection generation. Imposing such limits may conceptually be considered akin to diversity-oriented synthesis in the field of drug discovery and design, where small-molecule collections are created by starting from a common, promising compound and modifying the compound through limited synthetic steps.

The proposed workflow outlined in Figure and the high-throughput and high-resolution techniques required for its implementation are anticipated to greatly accelerate NC design by accurately revealing fundamental structure–property relationships. Yet, we also note two additional sources of heterogeneity that are not tied to the lattice of the NCs but are critical to evaluate to ensure accurate structure–property measurements. First, the ligand shells that passivate the surfaces of colloidally derived NCs can be a source of heterogeneity. For structural measurements at the single-NC level, capping ligands on the NC surface may impact the perceived particle volume and particle surface charge, resulting in inaccuracies in the calculated particle sizes. Further, capping ligands can also impact NC properties, including the LSPR of plasmonic NCs, through changes in the local dielectric environment and the activity and selectivity of NC catalysts through blocking of specific surface sites. − Consequently, protocols are commonly adopted to remove the capping ligands before property measurement, but as Baker et al. showed, some methods do not fully remove these ligands and can introduce spatial variation. These protocols also likely depend on the composition and faceting of the NCs, but rarely are these protocols validated across material systems. Second, many of the property measurements of NCs are performed with the NCs supported on a substrate. For example, for optical studies of NCs, ITO is a widely used transparent support electrode because of its favorable conductivity and transparency in the UV–visible region. However, ITO has recently been shown to have nanoscale heterogeneities in its electrochemical properties, , which could lead to misinterpretation of results. As such, consideration of this heterogeneity is critical, along with the development of alternative substrate materials.

Given the strong correlation between the NC structure and its properties, a trend in nanomaterial synthesis has been to minimize heterogeneity in NC ensembles and move toward atomically precise nanostructures. However, NC heterogeneity can also be viewed as an asset, as an opportunity for NC discovery, and for screening diverse NC populations. By the development of new high-throughput and high-resolution tools to evaluate single NCs in an integrated workflow, the inherent heterogeneity of NCs can be leveraged to drive new science and innovation. In accurately identifying NC structure–property relationships, conceptual breakthroughs for NC design can be achieved. This design should lead to NCs with property enhancements but also a better understanding of what level and types of heterogeneity are allowable, and even beneficial, in an ensemble for a given application. Ultimately, this understanding is anticipated to address reproducibility concerns and may lead to more sustainable nanochemistry.

M.K. wrote the abstract, introduction, heterogeneity across length scales, scanning probe, and portions of the MS technique sections; she also integrated Perspective components and references and assisted in making figures. Z.J.O. wrote the electron and optical microscopy sections, along with two case study sections with figures. M.E.E. researched the probe-based and MS techniques sections and wrote portions of the MS section while assisting with figures. C.H. wrote two of the case study sections with figures. M.N. wrote the discussion of computational and ML methods. O.J.W. aided in manuscript planning and literature research of analytical techniques. L.A.B., G.H., X.Ye, X.Yan, and K.A.W. edited the Perspective and gave scientific directions for topics. S.E.S. oversaw this project and wrote the Conclusions and Future Outlook section with feedback from the entire team.

The authors declare no competing financial interest.