Surface plasmons (SPs), described as electromagnetic waves coupled to collective excitations of conduction electrons in metals^[1,2], have been the subject of considerable attention for decades. Somerfeld and Zenneck were the first to observe and numerically describe the surface wave at the end of the 19th century^[3]. In 1902, Wood observed anomalous diffraction phenomena when he illuminated a metal grating with a polarized light source, which is regarded as a seminal experimental discovery of SPs^[4]. The concept of SPs was first proposed by Fano in 1941 to explain the aforementioned phenomenon^[5]. In 1957, Ritchie observed characteristic energy losses of fast electrons passing through thin metal films and predicted the existence of self-sustained collective excitations at metal surfaces^[6]. Two years later, Powell and Swan provided experimental confirmation of Ritchie’s theory^[7]. Subsequently, there has been a notable advancement in both theoretical and experimental investigations of surface plasmons, uncovering diverse fundamental properties of solids in the domain of condensed matter and surface physics. In the 1990s, the advent of nanotechnology created novel experimental conditions that facilitated breakthroughs in plasmonic research, offering new avenues for investigation and discovery. Tsai and Shalaev used photon scanning tunneling microscopy to characterize localized surface plasmons in metal colloid clusters^[8]. Ebbesen discovered the abnormal transmission of metal films with subwavelength hole arrays^[9]. Their pioneering research contributed to the emergence of plasmonics and stimulated a period of intense investigation. In comparison with free-space propagating waves, SPs demonstrate a considerably diminished mode volume and have been shown to surpass the optical diffraction limits, facilitating the integration of a greater number of functionalities into a single device. This has led to the development of applications in optoelectronic devices^[10,11] and super-resolution imaging^[12,13]. As a light field confined at the metal-dielectric interface^[14], SPs offer a unique two-dimensional (2D) optical platform for the engineering of light-matter interactions^[15,16]. In recent years, the rapid development of plasmonics has not only promoted theoretical mechanisms and manipulation methods of SPs but also provided new opportunities for many other interdisciplinary studies, including electronic information^[17,18], chemistry^[19,20], biology, and medicine^[21–24].

Two particularly versatile entities for SPs are surface plasmon polaritons (SPPs) and localized surface plasmons (LSPs)^[25]. SPPs are propagating electromagnetic modes that exist at the interface between a metal and a dielectric. The most attractive feature of SPPs is their capacity to manipulate the electromagnetic field at the nanoscale within a 2D platform, serving as a foundation for on-chip information carriers. Over the past few decades, the regulation of SPPs through the utilization of subwavelength artificially structured devices has facilitated numerous promising applications, including subdiffractional lithography^[26,27], optical holography^[28,29], and high-density data storage^[30]. Meanwhile, line-guided plasmonic devices, such as stripes, gaps, and grooves, have exhibited considerable promise in the development of highly compact plasmonic devices for signal routing and processing, with applications in optical information technology and photonic circuits^[31,32]. In contrast to the propagating nature of SPPs, LSPs represent non-propagating excitations confined within metal particles, which are typically much smaller than the wavelength of the incident light. They are capable of tailoring collective resonance and amplifying local electromagnetic fields^[33]. The highly localized and intense fields render LSPs exquisitely sensitive to minute changes in the local refractive index. Consequently, relevant research propels advancements in various applications, including electro-optic modulation^[34,35], nonlinear enhancement^[36–38], immunoassays^[39], biochemical sensing^[40,41], photothermal therapy^[42–44], and surface-enhanced Raman spectroscopy^[45–48].

The progress of plasmonics is inextricably linked to the breakthroughs in characterization technologies. Recent developments in optics have facilitated the acquisition of detailed images of SPs with techniques such as leakage radiation microscopy (LRM)^[49,50] and scanning near-field optical microscopy (SNOM)^[51,52]. LRM transforms evanescent near-field SPPs to an observable far-field signal by quantifying the radiation leaked from substrates. However, as a classical far-field characterization technique, LRM cannot break through the optical diffraction limit. The requirement of higher spatial resolution urges the development of more advanced technologies. The success of the atomic force microscope (AFM) in super-resolution imaging inspires the upgrade of experimental instruments, resulting in the emergence of SNOM. AFM probe tips are involved to detect near-field electromagnetic signals with the spatial resolution beyond 10 nm^[53]. It is imperative to acknowledge the substantial background signal from optical component reflection and stray scattering, limiting the improvement of signal-to-noise ratio (SNR) and resolution^[54–56]. Special plasmon nanofocusing tips have been reported as an effective solution for high-quality measurements, but the challenge remains^[57–61]. High-energy electrons can be focused into a nanometer spot due to their short de Broglie wavelengths, providing another way for the characterization of SPs. The in-depth study of electron-photon interactions indicates the feasibility of the optical characterization with electron beams. Various electron nanoscopies with atomic-scale spatial resolution have greatly advanced the development of plasmonics^[62], including electron energy loss spectroscopy (EELS), cathodoluminescence (CL) nanoscopy^[63], photon-induced near-field electron microscopy (PINEM), and photoemission electron microscopy (PEEM). Two types of detectable signals for electron-photon interactions include the electron energy loss and gain in the inelastic scattering process^[64] and electron-induced radiation. Multi-dimensional measurements of the former form the technique of EELS, and the latter corresponds to CL nanoscopy. The energy, polarization, radiation angle, and temporal evolution of scattered electrons and emitted photons can be obtained, providing diverse near-field optical information^[65–68]. Both EELS and CL signals have been theoretically demonstrated to directly relate to the generalized local density of photonic states (LDOSs)^[69]. A special involvement of lasers in an EELS system forms the experimental setup of PINEM, which directly probes the subsurface plasmonic near-field dynamics of SPs via a laser-electron beam hybrid excitation^[70,71]. A quantum model has described the deep relationships among the above three technologies^[62]. PEEM captures the electrons emitted from the sample as a result of the high-energy optical incidence. Their intensity, momentum, and distribution reveal the comprehensive information of near-field electromagnetic fields^[72,73]. These techniques have been reported to be powerful tools for the fundamental research of SPs and the quantum behavior of free-electron-photon-matter interactions^[74,75]. They open up avenues for strong light–matter interactions^[76], quantum statistics^[77,78], quantum nonlinearities^[79], and the ultrafast dynamics of SPs^[80,81].

With the advancement of electron beam experiments, researchers have noted significant applications at the intersection of SPs and free electron physics. Free electrons can be utilized as probes to detect and manipulate SPs^[63,82,83]. The emission of photons with wide tunability over a broad spectrum, ranging from mm-waves to the ultraviolet (UV) band, has been demonstrated for SPs pumped by free electron beams^[84–86]. This tunability extends beyond the frequency range, encompassing energy^[87–89], polarization^[90–92], phase^[93–95], momentum^[96–98], and spatial distribution^[99,100] characteristics. The ability to manipulate plasmonic resonance properties is contingent on the meticulous design of nanostructures and materials, enabling the control of the radiation characteristics of free electrons. This prospect holds immense potential for the design of compact free-electron-driven light sources^[101–103], single-photon emitters^[104,105], medical imaging^[106], quantum information technologies, and device applications^[107,108].

In this review, we will discuss the characterizations and manipulations of SPs in recent years. These devices can be roughly classified into two types according to their fundamentals: (1) near-field optics and (2) electron nanoscopy. The review is arranged as follows: Section 2 outlines the fundamental physical attributes of SPs. Section 3 compiles techniques for characterizing SPs, encompassing optical and electron beam excitations, such as leakage radiation microscopy (Sec. 3.1.1), scanning near-field optical microscopy (Sec. 3.1.2), electron energy loss spectroscopy (Sec. 3.2.1), cathodoluminescence nanoscopy (Sec. 3.2.2), photon-induced near-field electron microscopy (Sec. 3.2.3), and photoemission electron microscopy (Sec. 3.2.4). Sections 4 and 5 demonstrate the near-field manipulation of SPs based on optical methods (discussed in Sec. 3.1) and electron beams (discussed in Sec. 3.2). Diverse properties of plasmons have been successfully controlled, including energy, polarization, phase, momentum, and spatial distribution. Section 6 envisions promising prospects for future research in this exciting field. Finally, Sec. 7 summarizes this review article.

Plasmons are collective excitations of electrons and are classified as bosonic quasiparticles associated with oscillations of the electron gas plasma. Specifically, when an electric field is applied to a metallic structure, the conduction electrons undergo a synchronized displacement from their equilibrium positions relative to the core ions. This displacement induces polarization within the sample and subsequently generates a depolarization field that acts as a restoring force. With a time-varying external field, this collective motion can be described as a Lorentzian oscillator, manifesting a characteristic peak in the displacement amplitude (polarizability) around the resonance frequency. Generally, there are two modes for plasmons, including bulk plasmons and surface plasmons.

Bulk plasmons are longitudinal-wave solutions of the Maxwell equations under the condition of vanishing permittivity and zero magnetic field. Since they are electrostatic and longitudinal in nature, excitation of bulk plasmons by the electric field of electromagnetic radiation is forbidden, which also results in that bulk plasmons only decay through the energy transfer to electrons, i.e., Landau damping. However, for the same reason, they couple efficiently to moving charges and play a conspicuous role in EELS.

SPs are confined to the interfaces between metals with a negative real and small positive imaginary permittivity and dielectric with a positive real permittivity, and they are actually the sources of interesting phenomena and applications that comprise the field of plasmonics. This section demonstrates the basic properties of these two subsets of SPs in detail.

SPPs are propagating modes at a given metal-dielectric interface, involving non-zero magnetic fields and transverse character, and they are evanescent in the direction perpendicular to the interface, as shown in Fig. 1(a). The mathematical deduction of SPPs starts by considering a surface wave propagating along the +x direction with a wave vector kspp at the interface between two materials with distinct dielectric constants [εm for metal (z<0) and εd for dielectric (z>0)], while the y direction is assumed to be homogeneous. Within this framework, an SPP wave can be derived as a solution of Maxwell’s equations, which has been elaborated in published textbooks, and the dispersion relation of SPPs is kspp=k0εmεdεm+εd,(1)where k0=ω/c is the wave vector of the light in the vacuum. Generally, SPPs exist at the metal-insulator (or vacuum) interface, and thus the dispersion relation strongly depends on the permittivity of metals. Notably, the excitation of conduction electrons in real metal inevitably suffers from both free electron and interband damping. Therefore, εm is complex, and so is the SPP wave vector kspp, compelling the traveling SPP to damp with an energy attenuation length, typically between 1 and 100 µm in the visible regime. Note that all dispersion models deduced from a microscopy description of charge motion (Drude model and Lorentz model) lead to an analytic form; it is convenient to model any dispersive metals with several Lorentz poles.

Taking Ag as an example, the permittivity is described as a single-pole Lorentz-Drude permittivity^[109]: εAg(ω)=ε∞(1−ωp2ω2−ω02+iγω),(2)with ε∞=1,ℏωp=3.78eV,^[110] and ω0=0. Figure 1(b) depicts the dispersion relation of SPPs confined in the air-Ag interface. SPPs exclusively exist as TM modes, requiring p-polarized incidences for excitation^[111]. Additionally, momentum matching between the in-plane wave vector component of incident light and the SPP propagation constant needs to be satisfied. However, the SPP dispersion curve always resides to the right of the light line in vacuum, creating a fundamental momentum mismatch that prohibits direct plane-wave excitation. It is therefore necessary to employ special phase-matching techniques in order to excite the waves. Common excitation approaches include: (1) prism coupling, utilizing evanescent waves generated during total internal reflection at dielectric interfaces to bridge the momentum gap^[112]; (2) grating coupling where periodic structures provide discrete momentum increments^[113]. Notably, swift electrons provide an alternative excitation pathway through direct momentum transfer via electron energy loss processes^[7].

For small wave vectors corresponding to low (mid-infrared or lower) frequencies, the SPP propagation constant kspp is approximately equal to k0 at the light line, and the waves extend over multiple wavelengths into the dielectric space. In this regime, SPPs assume the characteristics of a grazing-incidence light field, and are also known as Sommerfeld–Zenneck waves.

It is important to note that bound electromagnetic modes, including SPPs, are not supported in perfect conductors. In perfect conductors, the movements of conduction electrons are fast enough to respond to external electromagnetic stimuli. The electrostatic induction effect has the capacity to shield all electric fields, thereby maintaining a zero field internally. This effect precludes the existence of bound surface plasmon polariton modes^[114]. The prediction of numerical analysis is also in accord with this behavior in theory. When metals are approaching perfect conductors (εm→−∞), kspp tends to the wave vector in the dielectric (k0εd), as indicated by Eq. (1). Hence, the z-component of the wave vector tends to zero, leading to a free electromagnetic wave rather than the one confined to the surface^[115]. In the terahertz (THz) and radio-frequency regimes, the variation of the external electromagnetic field is significantly slower than the movements of conduction electrons. Therefore, the behavior of metals approximates perfect conductors at low frequencies, which hinders the development of plasmonic devices.

To address this issue, special structures have been proposed for the realization of tightly confined optical fields. They achieve the effective control of the field penetration in structured perfect conductors and establish innovative types of bound electromagnetic modes, which were termed spoof surface plasmons, or spoof SPs^[116]. As one of the typical structures, subwavelength apertures patterned on metallic surfaces induce effective field penetration depths analogous to those in lossy metals with finite skin depths^[114,117], thereby mitigating the inherent screening capability of perfect conductors. The structured surfaces exhibit effective permittivity profiles resembling Drude metal in the optical regime, enabling direct transfer of plasmonic concepts and geometries from the optical regime to longer wavelength applications^[109]. Representative implementations include gradient meta-couplers for efficient spoof SPs excitation and precise wavefront engineering in the THz regime^[118]. Graphene-integrated reconfigurable platforms have been reported to achieve multifunctional operations for slow-wave propagation with dynamic tunability via electrostatic gating^[119]. Spoof-SPs-based miniaturization techniques have been successfully implemented in microwave and millimeter-wave components including antennas^[7–19], filters^[120–122], and sensors^[123–125].

Multilayer systems, consisting of a thin metallic layer sandwiched between two thick dielectric claddings, an insulator-metal-insulator (IMI) heterostructure, or a thin dielectric core layer sandwiched between two metallic claddings, a metal-insulator-metal (MIM) heterostructure, have gained appreciation as routes to plasmonic devices unconstrained by the wavelength of light, yielding viable nanophotonics devices. In such heterostructures, with a small separation between adjacent interfaces compared to the decay length in the normal direction, SPPs sustained in each single interface give rise to coupling modes. Typically, the coupling modes can be classified into two categories: odd and even modes. With the IMI geometry, odd modes show an increased SPP propagation length upon decreasing metal film thicknesses, giving rise to long-range SPPs that play an important role in practical devices. With the MIM geometry, the most interesting part is that the wave vector of the fundamental odd mode, which is referred to as “gap plasmons”, increases with a decreased insulator film thickness, thereby allowing a large wave vector for excitation well below ωsp and a more localized electromagnetic field in the insulator layer^[126]. In addition, the “gap plasmons” can be found in various configurations, including nanopatch optical antennas^[127], metastructures^[128], and particles on surfaces^[129,130]. These gap modes exhibit remarkable field enhancement, enabling the trapping of optical fields with minimal loss and at a volume below 1 nm3^[131].

In contrast to propagating SPPs, LSPs represent localized oscillations of conduction electrons along the surface of metal nanoparticles (NPs)^[111]. The behavior of the LSP resonance is determined by the shape, size, and dielectric environment of the NP. In general, LSPs of noble metals (e.g., Ag and Au) are usually observed in the visible and near-infrared regions. Aluminum nanostructures exhibit LSPs in near-ultraviolet and visible regions. NPs supporting LSPs act as a “nanoantenna” driven by the energy resonant light wave, leading to greatly enhanced light absorption, scattering, and electromagnetic field at the surface.

In order to gain more insights into the resonant behavior of LSPs, the plasmonic properties of a spherical metal NP with dimensions much smaller than the wavelength of light can be derived analytically. The quasistatic approximation for NPs indicates that the surface plasmon resonance peak occurs when the polarizability is maximized. It is noteworthy that the natural frequency of plasmons in the material and the LSP resonance frequency are often different. The electric field penetrates the entire volume of the NP and polarizes it completely, thereby resulting in a dipolar response with a dipole momentum p(ω)=α(ω)E. The motion of conduction electrons in response to the electric field can be expressed by the metal polarizability: α(ω)=3εd(ω)VNPεm(ω)−εd(ω)εm(ω)+χεd(ω),(3)where ω is the frequency of the light, εd(ω) is the permittivity of the non-absorbing [Im(εd)=0] surrounding medium, VNP is the NP volume, χ is a geometrical factor (χ=2 in a sphere), and εm(ω) is the frequency-dependent complex permittivity of the metallic material, which has been discussed in Sec. 2.1. Then, the extinction cross-section of the NP can be expressed as σext=9ωεd32cVNPIm(εm)[Re(εm)+χεd]2+Im(εm)2.(4)

From Eq. (4), it is clear that the σext reaches the maximum when the denominator is minimized, indicating that the LSP is excited at the frequency meeting the Fröhlich condition, expressed as Re[εNP(ω)]=−χεd.(5)

It should be noted that for chiral metal NPs, the extinction cross-section in response to circularly polarized light becomes slightly different, as does the resonance condition^[132]. This can be interpreted by rewriting χ¯ to include the chiral term δχR=−δχL, which corresponds to right (R)- and left (L)-handed nanostructures. The extinction cross-section for a chiral plasmonic NP then becomes σext±=9ωεd32cVNPIm(εm)[Re(εm)+(χ¯±δχR)εd]2+Im(εm)2,(6)where + and − denote right-handed circularly polarized (RCP) light and left-handed circularly polarized (LCP) light. The above analytical expressions are based on the dipolar response of NPs. However, when NP dimensions are larger than twice the skin depth, retardation becomes non-negligible and the electric field acts primarily on the surface electrons, contributing to higher-order multipole resonances. It can be concluded that the LSPs with high-order multipole momentum are more susceptible to alterations in the dimensions and configuration of the NPs.

Besides the maximum optical extinction cross-section of NPs at the localized surface plasmon resonance (LSPR) frequency, the greatly enhanced electromagnetic field near the NP surface is another important effect of LSPs^[133]. Local electromagnetic fields exhibit a notable increase when pumped resonantly and a rapid decline with distance, which surpasses the resolution limit of ordinary far-field optics and localizes light on nanoscale. The feature sizes of common nanofabrication techniques, including nanoimprint and electron beam/focused ion beam lithography, have reached sub-10 nm^[134]. Therefore, it is totally controllable and reproducible to confine lights to volumes measuring 10–100 nm in dimensions for artificial metallic NPs featuring sharp edges or protrusions^[135–137]. Exciting advances in nanofabrication^[138] have made it possible, though still limited, for single NPs to attain a more tightly confined field (on the scale of 1 nm or less)^[139]. A feasible way to achieve heightened field localization and enhancement is to couple NP plasmons together, as light can be tightly confined to the gaps (on the scale of 1 nm) between NPs, where the gap plasmon manifests. A similar limitation in nanofabrication capabilities also leads to inconsistent control over gaps between NPs. The technique of nanoassembly has been proven to arrange single nanoparticles separated from a metal film by a thin dielectric layer, which exhibit a significant enhancement. The gap between the NP and the metallic film can be accurately controlled to a sub-nanometer scale by using molecular spacers^[140]. For example, it has been reported that the system composed of Au nanoparticles of 40 nm diameter and a 70-nm-thick Au film, separated by a 0.9 nm molecular spacer, leads to field enhancements of ∼103.

SPPs and LSPs are usually considered distinct phenomena: SPPs manifest as propagating surface waves sustained by collective structural coupling, whereas LSPs correspond to geometrically confined resonances localized around individual nanostructures^[141]. However, an attractive advance on purely plasmonic bound states in the continuum (BICs) has bridged these two concepts and established a unified framework^[142]. Theoretical analysis results in an inverse square root scaling law, which describes a universal scaling relationship between the dissipation quality factor Qdis and normalized detuning parameter Δ=(λ−P)/P: Qdis∝1Δ,(7)where λ and P denote the resonant wavelength and the lattice period of the metasurface, respectively. When Δ approaches zero (λ→P), the hot spots of LSP resonances evolve into collective SPP modes^[143]. At the critical condition λ≈P, the emergency of surface-wave-mediated long-range interactions effectively suppresses the radiative loss and results in the enhancement of the quality factor. The inverse square root law fundamentally facilitates understanding of localized versus delocalized plasmonic behaviors, especially promoting the regulation of lattice-wave vector interactions^[144].

Advances in characterization technologies have greatly promoted the development of plasmonics. Classical optical instruments, such as spectroscopes and optical microscopes, only realize far-field and macroscale detection of SPs, but hardly reveal the near-field mechanism and non-radiative properties in the nanoscale. The lack of nanoscale characterizations and fabrications limited the study of both SPPs and LSPs in the mid-twentieth century. The subwavelength nature of SPPs results in a momentum mismatch, which causes them to stay away from the radiation continuum. A compensation mechanism is required for the indirect detection of SPPs to couple with radiative optical modes. Unlike SPPs, LSPs behave as a resonant mode in open systems, whereby the energy can be emitted as free-space electromagnetic waves, thereby obviating the need for compensation. However, due to the intrinsic diffraction limit and the ultrafast dynamics at the femtosecond level, it is difficult to directly probe the spatial distribution and the temporal mechanism, emphasizing the necessity for innovative methods for plasmonic characterization. Following the developments of the 1980s, the advent of near-field optics facilitated the revelation of the surface dynamics of SPs, which were the consequence of the technological breakthrough of the traditional optical resolution limit. Concurrently, the combination of optical detection and electron microscopy gave rise to a multitude of innovative characterization technologies. Due to their high spatiotemporal resolution, these technologies also propelled the advancement of plasmonics and uncovered a substantial body of new phenomena. This section presents a comprehensive analysis of various characterization technologies and their key contributions to plasmonics in recent decades. They are divided into two types: optical characterizations (Sec. 3.1) and electron nanoscopies (Sec. 3.2). Section 3.3 summarizes this section and provides a direct comparison among these characterization technologies.

Classical optical instruments, such as spectroscopes and optical microscopes, only realize far-field and macroscale detection of SPs, but hardly reveal the near-field mechanism and non-radiative properties in the nanoscale. Advances in characterization technologies have greatly promoted the development of plasmonics. This section introduces the most common optical microscopies in plasmonic research, including LRM (Sec. 3.1.1) and SNOM (Sec. 3.1.2).

When the metal film where nanostructures are fabricated is thin enough and the refractive index of the substrate (typically glass) is higher than that of the superstratum medium (generally air), coherent SPPs leak into the far field through the substrate, which is a characterization technology called leakage radiation microscopy^[148,149]. In the past, the possibility of using rough surfaces or gratings to excite SPPs has been extensively studied, where these structures can provide a sufficient additional momentum for the light waves to match the SPP dispersion relation. As an inverse process, SPPs emit back into the substrate with a high refractive index as leakage radiation, while surface roughness, edges, particles, or gratings contribute to an increased leakage. As marked in Fig. 2(a), a reduced model analyzes electromagnetic modes propagating in a triple-layered system that typically contains air, metal, and glass layers. By applying boundary conditions to the Maxwell’s equations, the complex amplitude of the magnetic field can be written as^[50]H(x,z)={H0eikxxeik0z(z>0,in air)H0eikxx[ik0εmkmε0sin(kmz)+cos(kmz)](−D<z<0,in metal)γH0eikxxeik2(z+D)(z<−D,in glass),(8)where ε0, εm,ε2, and D are defined as the complex permittivity of air, the metal, and the glass, and the thickness of the metal layer, respectively. k0=−ε0ω2/c2−kx2, km=εmω2/c2−kx2, and k2=−ε2ω2/c2−kx2 denote the z-components of the wave vectors in different layers. SPs are evanescent waves in the air with the imaginary k0, leaking to the glass substrate with the real k2. The relationship among wave vectors determines the geometric relationship, including angles and positions, between the surface mode and the leakage radiation. The magnitude of the electric field of the light that leaks from a given point in a given direction is directly proportional to the magnitude of the electric field of the SPP excitation passing through that point in that direction. The coefficient γ=ik0εmkmε0sin(kmD)+cos(kmD) demonstrates the leakage rate of the SPP mode confined at the air-metal interface, with a notable decrease observed in the presence of decreasing D. Consequently, LRM necessitates a minimal thickness of the metallic film (i.e., below about 80–100 nm)^[148,149]. In brief, the spatial profile of the leakage radiation mirrors both the local intensity and the azimuthal distribution of the SPP on the surface.

In the experiment, the radiation in the substrate propagates above the critical angle and can be observed with a high numerical aperture (NA>1) oil immersion objective. Due to the fact that the radiation intensity is proportional to the amplitude of SP modes at the local point where LR occurs, by mapping the far-field radiation in direct images at the object plane, the near-field spatial intensity distribution of SP modes can be acquired^[49,50,150]. In the case of a measurement at the Fourier plane, the angular distribution of radiation can be determined and provide valuable information about the structure and SP modes^[151,152]. Recently, a Mach-Zehnder interferometer has been used to realize ultrafast leakage imaging^[145]. As shown in Fig. 2(b), a femtosecond laser pulse is split into two beams for the excitation and the reference. The plasmonic leakage radiation signal is collected by an oil immersion objective and then combined with the reference signal. The interferogram of these two pulses is captured at the image plane of the collection objective. Successive interferograms of the leakage radiation from SPs at different positions record the evolution of the phase and characterize SPs in the time domain.

LRM is a powerful tool for the analysis of SP modes and it has been successfully implemented in various plasmonic systems. As depicted in Fig. 2(c), LRM has been employed to illustrate that the near-field interference of circularly polarized dipoles leads to the unidirectional excitation of guided SPPs, even for spatially symmetric structures and without preferred far-field radiation direction^[146]. The directional launching of SPPs based on the spin-orbit interaction of light was experimentally demonstrated in both chiral and achiral nanostructures by angle-resolved LRM [Fig. 2(d)]^[147]. Using this method, researchers were able to probe the chirality of plasmonic nanostructures, observe SPP focusing, and study singular SPPs. Besides visualizing the directivity patterns of SPPs, in which the photon spin couples to its spatial motion, LRM is more applicable to study leaky plasmonic modes in metal nanowires^[153–155], which combine the physical properties of both “plasmonics” and “leaky radiation”^[156]. Recently, it has been reported that Ag nanowires serve as sensors for the simultaneous detection of spin angular momentum (SAM) and orbit angular momentum (OAM) of light^[157]. This method reconstructs SAM and OAM states of light beams unambiguously according to the transmitted scatter light intensity distribution in the Fourier plane in the time domain for measuring both group and phase velocities of near-field pulses with a high level of precision. LRM provides a simple and reliable approach to characterize SPs and their subtle phenomenon in both spatial and Fourier space, and the out-of-focal-plane imaging by LRM has been reported to link the object plane and Fourier plane images and obtain the spatial location and the angular distribution from the same emission spot^[158].

SNOM has been a pivotal technology in near-field optics, combining the advantages of scanned probe technology and optical microscopy to provide powerful tools for nanoscience, including plasmonics. The concept of SNOM can be traced back to the ideas of Synge, who proposed an experimental scheme that would extend optical resolution to the nanoscale^[162]. Synge’s concept involved the creation of a subwavelength emitter in close proximity to a sample, which was then scanned in a grid pattern to produce signals related to the near-field electromagnetic properties. In the 1980s, shortly after the invention of the scanning tunneling microscope, an optical microscope similar to Synge’s proposed scheme was reinvented and demonstrated^[163–165]. The key innovation is the introduction of a subwavelength optical aperture at the apex of the probe tip [Fig. 3(a)]. This approach is referred to as aperture-type SNOM (a-SNOM). An SNOM system collects the optical response while scanning the sample surface with the probe. The detection image of electromagnetic distribution is composed of measurement results for each scanning point, instead of geometric optical imaging. Therefore, the spatial resolution is mainly determined by the optical mode at the probe tip, and well-designed tips with deep subwavelength scale lead to ultra-resolution imaging beyond the optical diffraction limit. In the a-SNOM system, incident light traverses the aperture on the tip or the optical response is collected there. Another typical category of SNOM is the scattering-type SNOM (s-SNOM), also referred to as the apertureless SNOM, which was first reported experimentally in 1994^[166]. In the system of s-SNOM, an additional light beam irradiates the tip and the scattering signal is utilized to characterize the optical properties in the near field. In both configurations, the fundamental concept is to use a subwavelength tip to convert high-spatial-frequency components from the optical near field into detectable far-field propagating waves, and the optical properties can be observed with a lateral resolution beyond the diffraction limit. In conjunction with spectroscopic techniques, this approach offers a versatile tool for investigating a diverse array of phenomena across multiple scientific domains, including chemistry, biology, material science, and plasmonics. In the following section, we will provide a concise overview of the two principal configurations, a-SNOM and s-SNOM, and their applications in plasmonics.

In aperture-type SNOM, the near-field signals are detected by the tapered metal-coated probe tip with a nanometer-level aperture. An aperture represents a highly confined light source^[167], whereby the laser light of a suitable wavelength couples to it via a connected optical fiber. Alternatively, cantilevered hollow silicon AFM probes with nanoapertures can also be used to perform near-field optical measurements. The precise control of probe-sample interaction is based on shear forces measured with a standard AFM controller. The typical experimental configuration of a-SNOM is illustrated in Fig. 3(a). In this configuration, the incident light is directed towards the sample through the aperture, and the back-reflected signal is collected by the aperture. There are several other operation configurations: (1) illuminating the sample with the aperture and collecting the transmitted radiation through the sample with an inverted optical microscope; (2) illuminating the sample with the aperture and collecting the reflected signal with an external objective at oblique direction; (3) illuminating the sample from underneath with an inverted microscope and collecting the transmission with the aperture; (4) illuminating the sample with an external oblique source and collecting the reflection with the aperture. All of the above configurations lead to high-resolution imaging by the scanning of the tip.

The spatial resolution of a-SNOM is determined by the aperture diameter. Although the advances of nanofabrication allow for tips with a nanometer aperture, another limitation of the aperture diameter is induced by the concerns about the optical power throughput. As a type of tapered waveguide, the minimal transversal wave vector k⊥ (the component of the wave vector perpendicular to the tip axis) of the propagable light increases with the decrease of the aperture diameter. When k⊥>2π/λ, the longitudinal wave vector of the optical mode becomes an imaginary value and refers to an evanescent wave with an exponential decay. The transmission of the light passing through the aperture tends to be very low as the aperture diameter decreases into the deep subwavelength scale. Therefore, in order to maintain a considerable optical power throughput, the size of the aperture does not reach the limit of nanofabrications, which limits the spatial resolution of a-SNOM.

A-SNOM provides effective measurements for various optical signals in the study of plasmonics, including fluorescence^[168], luminescence^[169,170], elastically or inelastically scattered light^[171,172], and the second or third harmonic generation^[173]. It is noteworthy that a pioneering study employs a-SNOM to investigate plasmonic vortices^[174]. They captured the polarization-dependent near-field intensity distribution in a plasmonic microcavity and revealed near-field vortex modes with a spin-dependent topological charge. A two-tip near-field excitation/detection scheme of a-SNOM called dual-SNOM^[175] enables the synchronous excitation and nanoscale imaging of SPs^[176–178] with polarization-resolved near-field measurements^[179,180]. Furthermore, the magnetic field of light was successfully probed by coupling a split-ring resonator to the probe tip^[181]. Subsequently, the potential of a-SNOM to detect nanoscale electric and magnetic fields simultaneously was demonstrated^[182,183]. In the aspect of a complementary technique, experimental developments have highlighted that the combination with femtosecond laser excitation facilitates a-SNOM with visualization of the ultrafast dynamics of SP modes^[184,185]. The primary challenge for a-SNOM pertains to the optimal aperture size, which determines both the spatial resolution and the power transmission efficiency^[186]. It is estimated that a decrease of the aperture size by an order of magnitude will decrease the collected intensity by 104, necessitating a compromise between resolution and signal intensity. In addition, the damage threshold of the aperture tip is also an important factor^[187], which decides how much power can be coupled into the tip without damage.

As discussed above, the optimization of the spatial resolution and the optical power throughput are conflicting for a-SNOM systems. In order to improve the resolution, s-SNOM is proposed as an apertureless technique. The major difference between a-SNOM and s-SNOM is the shape of the probes as shown in Fig. 3(b)^{[166,188,189]}. In the configuration of s-SNOM, instead of metal-coated fiber optic probes, a metallic or metallized tip serves as a nanoscale scattering source to convert the near-field information into a far field and measurable signals^[190]. Under the irradiation of a laser light source, the evanescent field generated on the surface is enhanced by the needle-tip-sample interaction^[191,192]. The tip can be made very sharp to provide an image with better spatial resolution, making s-SNOM desirable for the study of plasmonic nanostructures and metamaterials below the scale of a unit cell.

The point-dipole model was first proposed for surface-enhanced Raman scattering^[193] and later applied to investigate the interaction between the needle tip and the sample in s-SNOM^[194]. In this model, the probe tip is simplified as a polarizable sphere, with a radius equal to the curvature radius a of the tip. As Eq. (3) indicates, the polarizability of the tip can be expressed as αsphere(ω)=4πa3εm(ω)−1εm(ω)+2.(9)

Notably, various theories have been established to provide more accurate descriptions of the polarizability. For example, the finite-dipole model replaces the point dipole with a slender ellipsoid^[195]. Due to the slenderness of the tip, it is more susceptible to be polarized by the z-component of the electric field. The sample surface is modeled as an infinite plane where the effect of polarization charges is equivalent to an image dipole. The polarizability of the tip becomes αeff(ω)=[1−βαsphere(ω)16πh3]−1αsphere(ω),(10)where h denotes the distance between the tip and the surface, and β denotes the reflectivity of the surface, which is determined by the dielectric constant of the sample εs, as β=εs−1εs+1. The intensity of the near-field scattering signal is proportional to the dipole moment of the tip.

The major problem for s-SNOM is its limited SNR. It should be noted that the signals captured by the detector also include the background scattering signals. A relatively intense far-field background composed of reflection and scattering is directly generated from the sample surface, which unavoidably creates a large background signal and reduces the discernibility of the system^[196]. In order to improve SNR, one strategy is the involvement of the mechanical oscillation of the tip, which is experimentally supported by tapping or dynamic non-contact AFM modes. The oscillation periodically modulates the effective polarizability of the tip. The near-field scattering signal and the background scattering signal exhibit completely different dependencies on the additional periodic regulation, enabling the separation and selective amplification of the signal. The near-field scattering signal contains significant high-order harmonic terms due to its tight relationship with the distance h. On the contrary, the background scattering signal mainly focuses on the low-order harmonic terms^[56]. Therefore, lock-in amplification technology extracts higher harmonics of the detection signals to suppress disordered background scattering signals, thus resulting in high-quality measurements with high SNR.

The structural improvement of the probe tip, which facilitates efficient nanofocusing, has been documented as another strategy for enhancing the performance of s-SNOM^[57–61]. Specific nanostructures promote the energy transformation from the external incidence into the localized resonant mode at the apex. They offer extra momentum compensations to satisfy the momentum matching conditions of excitations of SPPs. The phase of all SPPs propagating along the probe can be modulated to a constructive interference, generating a localized plasmonic hot spot with exceptional brightness^[197]. Besides the fabrication of additional nanostructures, tapered rods with special geometric parameters also realize plasmonic nanofocusing. When the radius of the rod shrinks slowly (the taper angle is sufficiently small), the reflection and scattering from the tip hardly influence the properties of SPPs. The characteristics of an electromagnetic field are mainly determined by the local distribution of materials, which is called adiabatic approximation^[198–200]. The adiabatic approximation facilitates the design of a nanofocusing tip^[201]. Adequately designed plasmonic nanofocusing probes possess the capability of concentrating electromagnetic fields into a focal point of a few nanometers in size^[202–204]. The field enhancement leads to an improvement in the transformation efficiency and a reduction in the mode volume at the tip. Consequently, plasmon nanofocusing on fabricated tips results in higher optical power throughputs and superior spatial resolutions^[205–209]. Typical configurations of the plasmonic nanofocusing tip include one-dimensional (1D) gratings^[58,210], annular gratings^[57,59], spiral gratings^[60,197], and quasi-adiabatic nanofocusing probes^[211,212]. Recent advancements have yielded significant improvements to the spatial resolution, which has been optimized to approximately 5 nm, while maintaining high throughput and SNR^[53,60,213].

Ultrafast time-resolved SNOM has been realized by the introduction of a pump-probe measurement. A femtosecond laser beam is split into the pump beam and the probe beam, whose incident time interval can be modulated by a precisely controlled delay line. Both of the beams are focused on the probe tip. In the pump-probe measurement, the sample is scanned by the probe tip for each delay, resulting in a spatiotemporal image of the surface electromagnetic field, as shown in Fig. 3(f). The spatial and temporal resolutions are theoretically determined by the tip size and the convolution between the pump and the probe^[161]. In the experiment, pulse broadening induced by optical nonlinearities and dispersion effects in fiber probes distorts ultrafast optical pulses, complicating time-resolved dynamic studies^[53]. A variety of dispersion compensation strategies have been implemented to mitigate fiber-induced spectral phase distortions. These strategies include grating pairs, chirped mirrors, and deformable-mirror-based pulse shapers^[185]. Furthermore, as discussed before, s-SNOM operates in AFM-tapping mode and involves tip oscillation with amplitudes of tens of nanometers to improve SNR^[214]. The lock-in detection of near-field signals generally necessitates laser repetition rates that far exceed the mechanical oscillation frequency of the tip, as stipulated by the Nyquist–Shannon sampling theorem^[215,216]. While high-repetition-rate ultrafast lasers improve SNR through signal averaging, they induce cumulative tip heating^[217]. However, spurious artifacts are generated by periodic thermal expansion/contraction of a fiber probe^[218] and far-field background scattering generates spurious artifacts, obscuring genuine near-field temporal signatures^[217]. This trade-off between SNR enhancement and heating at the tip-aperture ultimately degrades temporal resolution^[53]. Recent advancements integrate s-SNOM with low-repetition-rate (∼kHz) ultrafast lasers, reducing cumulative thermal effects while expanding compatible light sources^[219]. Phase-domain sampling techniques circumvent the limitation of repetition rate, enabling operation at repetition rates below the tip tapping frequency. This approach permits coupling of high-pulse-energy ultrafast lasers to the tip-sample junction, simultaneously enhancing detection sensitivity and localized field intensity^[217].

Another attractive advancement is the involvement of nonlinear optical effects. Nonlinear optics have shown great potential for highly tunable, non-perturbative and real-time characterization, which has been used to increase the contrast or resolution of near-field scattering ^[220,221] and to access evanescent modes in plasmonic nanostructures^[222–225]. In a nonlinear pump-probe measurement, the wavelength of the pump laser can be smaller than the target electromagnetic mode, which reduces the difficulties of the localization of optical modes and improves the spatial resolution. This method has been widely used in the nonlinear characterizations in the THz regime as an upgrade of laser terahertz emission microscopy (LTEM)^[226–228]. The complementarity of this technique and the conventional method has been experimentally proven^[227]. For instance, recent research has used blue light femtosecond laser pulses to generate THz emissions of bulk crystalline Si^[229]. When the AFM tip is irradiated by the high-energy laser, two-photon excitation above the wide direct bandgap causes THz emissions. The nanoscale resolution of the system has been demonstrated to be achieved by outcoupling a portion of the macroscopic photogenerated THz dipole. Moreover, an experimental setup without tips has been proposed as a nonlinear near-field optical microscopy (NNOM)^[230]. A partially degenerate four-wave mixing process produces nonlinearly generated waves whose wavelengths are much smaller than target SPPs, and the nonlinear wavelength is the decisive factor of the resolution limit. Hence, the spot size can be reduced over the diffraction limit of the target wavelength, resulting in a high spatial resolution. The characterization technique has realized real-time imaging of surface waves including propagating SPPs.

Due to its high sensitivity to the local out-of-plane electric fields and charge accumulation, s-SNOM is a powerful tool to visualize the field and charge distributions induced in plasmonic nanostructures, allowing the resolution of different LSPR modes. In pioneering research in 2001, s-SNOM was used to characterize the plasmonic resonance of Au nanoparticles with a spatial resolution of 10 nm^[231]. s-SNOM has then been widely applied to study LSP modes in metallic nanostructures, including nanoprisms^[159], nanoantennas^[232–235], nano-polymers^[236,237], and fractal structures^[238]. Additionally, due to the fact that the light-tip interaction causes a broad range of in-plane momentum compensation, s-SNOM is suitable for the direct visualization of propagating SPPs. For instance, the employments of an s-SNOM setup for the direct observation of intrinsic graphene plasmons were reported in 2012^[239,240], where the infrared nano-imaging revealed the drastic squeezing of the plasmon wavelength and the gate-tunable plasmonic properties. Another notable feature is that s-SNOM can be employed for both imaging and spectroscopy as it can simultaneously record amplitude^[241] and phase information^[160,242] by pseudo-heterodyne interferometric measurement [Fig. 3(e)]^[55].

As another approach for the breakthrough of the optical diffraction limit, electron microscopies have been widely used in the morphology characterization of materials. The short de Broglie wavelengths of high-energy electrons allow them to be focused onto the sample surface with a very small spatial extent, serving as a super-resolution probe. The spatial resolution is improving with the increase of the electron energy, especially down to sub-atomic scale in a transmission electron microscope (TEM) or scanning transmission electron microscope (STEM). Detections of diverse responses while scanning the sample surface with the electron probe result in innovative characterization technologies. The radiation generated by the electron beam and the energy loss of the electrons are collected and measured to image the near-field SPs beyond the diffraction limit of light. Compared to optical methods, electron microscopes have the capacity of selectively exciting and characterizing different modes of SPs over a broad spectrum, while maintaining high resolutions in multiple dimensions, including energy, momentum, and space. Moreover, the combination with ultrafast lasers allows electron microscopy to characterize the dynamics of SPs at the femtosecond level. The improvements in energy and time resolution of electron microscopes provide a new perspective for the characterization of SPs. In this section, we briefly review the advances in electron microscopy and highlight the latest techniques for plasmonic characterization.

At the beginning of this section is a general theory of the interactions between a free electron probe and a sample. When a high-speed free electron approaches the sample, the effect of the sample on the electron can be approximately neglected due to the electron’s high energy^[63]. This non-recoil approximation assumes that each electron is equivalent to a negative point charge −e moving at a constant velocity v=vz^. In classical scenarios, free electrons serve as external current sources for instantaneous electromagnetic fields: J(r,ω)=−ez^δ(R−R0)eiωz/v,(11)where R=(x,y) denotes horizontal coordinates of the electron. In consideration of the distribution of the dielectric constant of the sample ε(r,ω), a Green function G(r,r′,ω) is defined as follows to demonstrate the optical properties of the surface: ∇×∇×G(r,r′,ω)−ω2c2ε(r,ω)G(r,r′,ω)=−1c2δ(r−r′).(12)

It has been demonstrated that G(r,r′,ω) exhibits a direct relation to other electromagnetic properties, with the most significant being LDOS^[243], which is defined as the combined electric field intensity of all normalized photonic modes as a function of light frequency and position in space^[82]. The electromagnetic field induced by the electron on the sample can be expressed as Eind(r,ω)=−4πiω∫d3r′G(r,r′,ω)·J(r′,ω).(13)

The electromagnetic properties of the sample can be characterized effectively through a variety of measurements of Eind(r,ω). Two detectable physical quantities are the energy loss of the electron and the far-field radiation, which refer to EELS (Sec. 3.2.1) and CL nanoscopy (Sec. 3.2.2), respectively. As an innovative technology, PINEM (Sec. 3.2.3) represents a pioneering advancement that integrates quantum theory with pump-probe measurements. This technology has been proven to characterize the near-field nonequilibrium dynamics of physical systems with unparalleled ultrahigh spatial and temporal resolution. In contrast to other electron microscopes, PEEM (Sec. 3.2.4), which is supported by the photoelectric effect, provides near-field characterizations from a different perspective. The aforementioned technologies are discussed in this section.

The high energy of electrons enables the acquisition of superhigh spatial resolution; however, it also causes radiation damage in some cases. This damage manifests as alterations to the specimen induced by the beam, thereby constraining the applicability of electron microscopies. Before the detailed demonstrations of electron nanoscopies for plasmonic characterization, the discussion about radiation damage is necessary. The nature of the electron-scattering process dictates the category of radiation damage, namely, knock-on damage or ionization damage, resulting from elastic or inelastic scattering, respectively^[244].

The elastic scattering originates from the Coulomb effect of atomic nuclei on incident electrons. When great momentum transfers of high-energy electrons exceed the threshold of the material, the recoil causes significant displacements of atoms, resulting in knock-on damage. Its performance exhibits atomic dislocations and sputtering of atoms from its surface^[245]. Due to the low acceleration voltage employed in scanning electron microscope (SEM), the knock-on damage is generally not caused. Conversely, in the context of TEM, atoms with low or medium atomic numbers are susceptible to damage at elevated energies. For instance, carbon nanotubes and crystalline silicon exhibit a propensity for dislocation damage at an acceleration voltage of 200 kV^[246,247]. It has been reported that the implementation of a thin layer of heavy elements on the sample can result in a reduction of knock-on damage^[247].

The ionization damage involves a series of energy transfers after the incidence of electrons. The inelastic scattering of primary electrons produces secondary electrons, resulting in the accumulation of static charges in samples with poor conductivity. This charge accumulation can cause image drift or deformation, affecting the quality of secondary electron imaging. In addition, the excitation and de-excitation of electrons generate numerous phonons, leading to a substantial local temperature rise. The heating effect has been demonstrated to be more significant than the knock-on damage for 200 kV electron beams^[248]. Another result of the ionization damage is the rupture of chemical bonds, a phenomenon that typically manifests in semiconductor and insulator materials^[249,250]. For example, electron beam irradiation with about the kinetic energy of 300 keV induces the transition of graphene to a quasi-amorphous 2D membrane^[251].

While it is possible to enhance spatial resolution by increasing electron energy, it is imperative to thoroughly assess the trade-off between spatial resolution and the potential radiation damage to the sample. Dose limited resolution (DLR) is a parameter that characterizes this concept with the involvement of statistical noise in the signal, and Ref. [252] provides further discussion about this issue. Besides the decrease of the acceleration voltage, the optimal strategy to prevent electron-beam-induced radiation damage involves minimizing the radiation dose (the product of current density and recording time) to the level required for successful data acquisition. Cryogenic conditions also contribute to the reduction of radiation damage. Although atomic dislocations induced by electron excitation are temperature-independent, the suppression of subsequent atomic vibrations at lower temperatures has been demonstrated to be effective in diminishing structural damage to chemical bonds, particularly for organic crystals. At low temperatures (100 K), the relevant damage can be reduced by a factor of three to 10^[253]. Reasonable encapsulation has been shown to mitigate the radiation damage^[254]. The coating layer, composed of metal or a conductive adhesive, has been demonstrated to impede the charge accumulation and its subsequent damage^[255].

As stated above, EELS characterizes the properties of the surface based on the measurement of the energy loss of free electrons, which is associated with the force exerted on the electrons by the induced electric field Eind(r,ω). Specifically, it is linked to the work done by the equivalent current source, so the EELS probability, denoted by ΓEELS(R0,ω), can be expressed as^[82,243]ΓEELS(R0,ω)=eπℏω∫−∞∞dzRe{e−iωzvEzind(R0,z,ω)}.(14)

According to Eq. (13), the direct relation between ΓEELS(R0,ω) and the generalized local density of optical states ρz^(R,q,ω) is described as^[212]ΓEELS(R0,ω)=2πe2Lℏωρz^(R0,q,ω),(15)ρz^(R0,q,ω)=−2ωπIm{z^·G^^(R0,R0,q,−q,ω)·z^},(16)where wave vector q=ω/v, L denotes the length of the trajectory. G^^(R,R′,kz,kz′,ω) represents the Fourier transform of G(r,r′,ω) with respect to z and z′.ρz^(R,q,ω) is defined as the LDOS in a generalized space (x,y,qz), which is local in real space along the horizontal direction (perpendicular to the motion of incident electrons) and local in wave-vector space along the vertical direction (parallel to the motion of incident electrons). It has been experimentally proven that LDOS can be reconstructed from tomographic EELS measurements^[259]. Notably, ρz^(R,q,ω) encompasses both radiative and non-radiative modes. Consequently, EELS is capable of detecting not only “bright modes” that radiate to the far field but also “dark modes” that do not couple to the far field. For example, the radiation suppression of BICs in specific directions fundamentally aligns with the directional extinction properties of multipole radiation^[260]. For the symmetry-protected BICs, decomposition into multipole components with vanishing vertical radiation is possible, where the dominant multipole governs the modal characteristics and far-field profile. Conversely, accidental BICs emerge through the superposition of multipolar contributions that vanish in specific spatial directions^[261–263]. The broadband evanescent field associated with a swift electron spans the IR-vis-UV range, bypassing far-field selection rules and enabling simultaneous excitation of both bright and dark modes^[264–267]. Consequently, despite the fact that these dark BIC modes remain decoupled from free-space radiation, their existence can be verified through experimental observation of near-field interactions mediated by electron-induced evanescent coupling [Fig. 4(a)]^[256].

As shown in Fig. 4(b), EELS is typically integrated within a TEM/STEM system. The electron beam traverses the specimen, undergoing a loss of energy due to its interaction with the material. This process is monitored by an energy loss spectrometer, which quantifies the energy loss spectra of the electron beam. High-angle annular dark field (HAADF) detectors capture high-angle scattered electrons concurrently, providing the structure information. In the angle-resolved EELS system, the electron beam passes through the sample and the magnetic lens successively to form diffraction patterns on the diffraction plane. A movable round aperture^[268,269] or a slot aperture^[257,270] is positioned at the diffraction plane to collect phonon dispersion data serially. The former provides the angle-resolved information in both two dimensions of the entire angular space, while the latter produces a 2D intensity map versus both energy and momentum transfer in one dimension. In addition, time-resolved EELS^[258,271] has been achieved based on an ultrafast transmission electron microscope (UTEM). The normal spectrometer is replaced by the ultrafast detectors to measure the energy loss as a function of time, uncovering the surface dynamics. In summary, it has been reported that EELS realized imaging, spectral, angle-resolved, and time-resolved measurements with superhigh spatiotemporal resolutions.

In 2007, Colliex’s group overcame limitations on energy and spatial resolution and succeeded in directly imaging plasmonic modes in metallic nanostructures^[272]. Since then, the use of EELS to detect plasmonic modes has become increasingly popular. With angle-resolved imaging systems, EELS realizes low-energy excitations in 2D materials, ranging from atomic crystals to nanophotonic arrays^[273]. The methodology provides a comprehensive understanding of the complex optical phenomena exhibited by 2D plasmonic films. In addition to elucidating these phenomena, EELS offers momentum-resolved perspectives, facilitating a more profound and nuanced examination of the photonic density of states^[274]. Time-resolved EELS has been used to reveal electron dynamics in graphite^[258], bi-layered manganite^[275], and carbon nanotubes^[276]. As shown in Fig. 4(d), the demonstration of time-resolved EELS in a graphite specimen uncovers femtosecond nonequilibrium structural features caused by the direction of change from sp2 to sp3 electronic hybridization. The technique captures the evolution process of plasmonic modes that carry information about the electron density to reflect the contraction of the planar lattice, occurring over time lengths typically below 1 ps.

As a non-invasive detection method, CL nanoscopy provides invaluable insight into the study of materials and optical near-field dynamics for plasmonics. With the developments in electron beam manipulation and light collection techniques, the dimension of information accessible to CL has been extended to polarization, momentum, and time. This section summarizes the recent advances in CL nanoscopy technology for plasmonics.

As the name implies, CL nanoscopy directly measures the induced electric field Eind(r,ω) in the far field. Cathodoluminescence is exactly defined as the photoluminescence caused by the electron irradiation.

It has been proven theoretically and experimentally that the CL signal is also qualitatively related to LDOS^[83,243,283]. From Eq. (13), the far-field electric field induced by the electron can be represented as ECL(r∞,ω)=4πieω∫−∞∞dz′eiωz′vG(r∞,R0,z′,ω)·z^→ωr∞c→∞eiωr∞cr∞f(R0,ω),(17)f(R0,ω)=−4πiω∫dr′G(r∞,R0,z′,ω)·J(r′,ω),(18)where f(R0,ω) denotes the far-field amplitude. The CL probability is obtained through the integration of the far-field Poynting vector: ΓCL(R0,ω)=c4π2ℏω∫dΩ|f(R0,ω)|2,(19)ΓCL(R0,ω)=4ωce2ℏ∫dΩz^·[G^(r∞,R0,qz,ω)]TG^*(r∞,R0,qz,ω)·z^.(20)

G^(r,R′,kz′,ω) represents the Fourier transform of G(r,r′,ω) with respect to z′. Meanwhile, generalized radiative LDOS can be expressed as^[212]ρz^rad(R0,ω)=2ω02cπ∫dΩz^·[G(r∞,R0,ω)]TG*(r∞,R0,ω)·z^.(21)

A comparison between Eq. (20) and Eq. (21) implies that ΓCL(ω) is proportional to ρz^rad(R,ω). It should be noted that ρz^rad(R0,ω) demonstrates the localized state density of radiative modes. As shown in Fig. 4(a), CL spectra do not manifest emission peaks of non-radiative BIC modes. However, resonant wavelengths of quasi-BIC modes are distinctly discernible through CL nanoscopy^[256].

The experimental setup of CL nanoscopy has two types of optical paths for distribution [Fig. 5(a)] and spectral [Fig. 5(d)] characterizations. Meanwhile, this system enables both conventional (CL spectra and images) and angle-resolved (angular distributions and energy-momentum maps) measurements via different detections. The sample is mounted at the focus of a parabolic mirror that collects the CL signals and transforms them into a parallel light. In the imaging mode, while the electron beam scans the sample surface, the detector records the total CL intensity for each excitation position, which forms a CL image of the spatial distribution. It is noteworthy that both bandpass filters and spectrometers are available to realize the wavelength selection in imaging mode. This optical path also supports the measurement of angular distributions [Fig. 5(c)] without the displacement of the electron beam. The momentum information can be obtained directly from the intensity distribution that the detector (typically a CCD camera) captures due to the function of the parabolic mirror. Compared to distribution characterizations, an additional slit and a spectrometer are mounted into the optical path for spectral measurements. Besides basic CL spectra, the combination with spatial distribution leads to a comprehensive three-dimensional (3D) data called “hyperspectral image”, which demonstrates the CL intensity corresponding to different wavelengths and excitation positions^[82]. Another advanced angle-resolved measurement selectively observes specific areas in the Fourier plane with a designated horizontal momentum by controlling the position of the slit. As the CL signal passes through the slit, it retains the information of the vertical momentum. The spectrometer then disperses the signal in the horizontal direction to yield a 2D energy-momentum map that demonstrates the relationship between the vertical momentum and the energy^[98,278,284]. Furthermore, the incorporation of specific wavelength filters and polarization elements into the optical path prior to the CCD enables measurements of the full set of full Stokes parameters of the CL spectra, images, angular distribution, and energy-momentum maps^[285,286].

In recent years, significant progress has been made in time-resolved CL using ultrafast electrons, owing to advances in ultrafast electron microscopy techniques^[287,288]. There are generally two methods for generating electron pulses in electron microscopes, as shown in Fig. 5(g). The first method involves an electrostatic deflector placed inside the electron microscope, which splits the continuous electron beam generated by the electron gun into discrete electron pulse sequences. It is available to reduce pulse duration to hundreds of picoseconds^[289]. Another method directly drives the electron gun with UV femtosecond laser pulses; then electron pulses are generated due to the photoelectric effect. In general, the temporal broadening of pulse electrons produced by femtosecond laser pulses can be less than 1 ps, which is significantly better than that achieved by ultrafast electron beam choppers^[290].

The temporal dynamics of photon emissions can be measured using two primary time-resolved CL techniques: decay trace measurements and time correlation measurements. In decay trace measurements, a brief electron pulse excites the material to a higher-energy state. A detector, coupled with a time-correlator, is employed to detect photons emitted from the sample and record the delay between the excitation pulse and photon arrival. The characteristic emission lifetime can be extracted from the decay trace^[291]. The time correlation measurements characterize the second-order correlation function g(2) of CL radiation^[292], which demonstrates the quantum property in time-domain statistics. The light beam from a quantum emitter passes through a dichroic filter. The time delay between the two successive photon arrivals constructs a histogram of occurrences demonstrating g(2). Time correlation measurements can exhibit three distinct behaviors: coherent (g(2)=1), antibunching (g(2)<1)^[293,294], and bunching (g(2)>1)^[295,296]. Furthermore, the combination of scanning electron beams and time-resolved CL measurements has been demonstrated to be a viable method for achieving the high-resolution spatiotemporal mapping^[297]. It is worth noting that time-correlated electron and photon counting microscopy has been recently proposed, which statistically evaluates single-coincidence events of the excitation electron and generated photons^[298].

Conventional CL nanoscopy provides a basic spectral and imaging characterization for plasmonics at the nanoscale. CL nanoscopy directly characterizes the spectra of LSPs in single nanorods with different lengths^[299]. The detection of coupling modes of nanowires and nanoantennas contributes to effective modulation of the propagation of SPPs^[89]. In addition, the polarization-resolved CL nanoscopy provides essential near-field information. For instance, it reveals the significant optical chirality of nanostructures with six-fold symmetry [Fig. 5(b)], which is attributed to a prominent plasmonic Fano resonance^[277]. Similarly, this technique uncovers “hidden chirality” in achiral nanostructures that are difficult to access with conventional optical methods^[300]. The directionality of CL emissions also significantly complements the characterization of optical modes in nanostructures. As shown in Fig. 5(c), the angular distribution demonstrates the band structure of photonic crystals^[278]. It is reported that the directional CL emission is the result of the coherent superposition of different multipoles in the far field^[97], and the analysis of the angular pattern uncovers the distribution and symmetry of plasmonic modes^[301]. Due to the limitation of temporal resolution, it is quite challenging to characterize the dynamic behavior of SPs within sub-optical cycles. However, recent breakthroughs have been made in revealing the luminescent dynamics of fluorescent materials at the nanoscale^[282] [Fig. 5(h)]. By performing time-correlated measurements of inelastic scattering events and photon emission events during the interaction between free electrons and the sample, it is possible to separate the photon emission processes and energy loss processes in the CL-EELS joint setup. This resolves the long-standing issue in CL technology of being unable to analyze the electron energy transfer mechanisms during incoherent CL emission. It showcases the broad prospects of time-resolved CL technology in applications such as energy upconversion^[302] and quantum emitter optimization^[303]. In short, CL nanoscopy has been a valuable tool for the research on plasmonics, providing technical support for the control over specific quantum emitters at the nanoscale^[304].

The ultrafast electrons in UTEM can be inelastically scattered not only by the electrons confined in the material, but also by the laser-excited near field in nanostructures. PINEM is a pump-probe time-resolved spectroscopy tool that involves a femtosecond laser as the pump and pulsed electrons as the probe. PINEM characterizes the near-field dynamics of nonequilibrium systems with ultrahigh spatial and temporal resolution^{[305,308–310]}.

PINEM is predicated on the quantum theory of the interaction between electromagnetic fields and free electrons. When a free electron with the charge −e and the mass m is modulated by a classical electromagnetic potential, the general time-dependent Schrödinger equation is expressed as [(P+eA)22m−eV]Ψ=iℏ∂Ψ∂t,(22)where P=−iℏ∇ denotes the momentum operator, and V and A are the scalar and vector potentials of the electromagnetic field, respectively. The nonrecoil approximation implies that the initial electron wave function Ψi(r,t) can be written as^[62]Ψi(r,t)=eiq0z−iE0t/ℏϕi(r−vt).(23)

E0 and q0 are defined as the initial central energy and the wave vector. ϕi(r−vt) describes a slowly varying envelope of the incident electron. As a result of the time evolution, Ψ(r,t)=Ψi(r,t)exp⁡[−iℏ∫−∞tdt′H^int].(24)

When the electron is irradiated by a monochromatic light of frequency ω, the Hamiltonian H^int is described by the vector potential A(r,t)=2cωIm{E(r)e−iωt}.(25)

E(r) contains both the incident femtosecond laser and the components scattered by the sample. In a PINEM system, the electron energy is typically sufficiently substantial to neglect the A2 term, which describes the ponderomotive force. Subsequent to an extended interval of interaction, Ψ(r,t) can be expressed as^[62]Ψ(r,t)=Ψi(r,t)eiφ(R)∑l=−∞∞Jl(2|β(R)|)eilarg⁡{−β(R)}eilωz/v,(26)β(R)=eℏω∫−∞∞dzEz(R,z)e−iωz/v,(27)where Jl denotes the lth Bessel function of the first kind. Equation (26) provides a comprehensive explanation of the energy comb [Fig. 6(b)] observed in PINEM, in which each term demonstrates that the electron absorbs or radiates l photons. As Eq. (27) implies, the correlation coefficient β(R) describes the electromagnetic distribution on the surface.

It is worth noting that a natural correspondence exists between PINEM and CL^[62]. A free electron undergoes a loss of energy, resulting in the CL emission of the sample. Conversely, in the case of PINEM, an external light source exchanges energy with the electrons. Therefore, the correlation coefficient β(R) can be obtained from the far-field amplitude of the CL emission due to the reciprocity^[62]: β(R)=ic2ℏω2f˜CL(R,ω)·Eext,(28)where Eext denotes the electric field amplification of the external plane-wave source. f˜CL(R,ω) represents the far-field CL distribution from an electron moving with opposite velocity. As depicted in Fig. 6(c), the relationship has been partially validated by exploring the spatial characteristics of EELS, CL, and PINEM on the same Au nanostar^[306].

The experimental setup of PINEM is illustrated in Fig. 6(a). As a pump-probe method, a femtosecond pump laser radiates the surface of the sample to excite diverse optical modes including SPs; then a pulsed electron beam impacts the sample to interact with the surface electromagnetic field. The EELS detector captures the electrons and usually demonstrates spectrally the energy comb, which directly indicates the correlation coefficient β(R). The scanning of the electron beam leads to a spatial characterization with a spatial resolution of 10 nm^[71]. By manipulating the time interval between the arrival of the pump light pulse and the electron pulse, ultrafast dynamic imaging of the sample at various moments after being excited by the pump light can be achieved. As discussed in Sec. 3.2.2, the ultrafast pulsed electron beam in PINEM can also be generated by an extra beam of UV femtosecond laser pulses, due to the photoelectric effect^[70]. The temporal resolution has been reported to exceed 500 fs for this configuration. However, a more advanced technique has been proposed recently to realize attosecond-resolution electron microscopy^[311,312]. Additional modulation membranes are mounted into the electron microscopy and radiated by a continuous-wave laser. As the electrons pass through the membrane, they are modulated by the external light field, which accelerates and decelerates electrons periodically. The electron beam becomes a series of ultrashort electron pulses of attosecond duration. The continuous-wave laser is also employed to pump the sample instead of the original pulsed laser to utilize the entire electron beam. This maximization of the repetition rate is meaningful for the accuracy and efficiency of the detection. Consequently, PINEM is capable of providing a video recording of near-field dynamics with an exceptionally high spatiotemporal resolution. The complete decoupling of the light excitation and electron detection mechanisms in PINEM enables the execution of tailored studies of specific SPs across a comprehensive energy range.

PINEM has been employed as a powerful tool for the study of nanoscale electron-photon interactions, with applications including plasmonic imaging of nanoparticles^{[70,313–315]}, carbon nanotubes^[308], rough metallic films^[316], and biological structures^[252,317]. As recently reported, in ultrafast energy-filtered PINEM imaging, the optically excited SPP Fabry-Perot modes on Ag nanowires can be simultaneously imaged for the energy spectra and spatial distribution of single-electron interactions with discrete photons, reconstructing the resonant optical near field^[70]. Exciting developments also include the manipulation of an electron wave function in spatial and temporal dimensions [Fig. 6(d)], where a PINEM experiment on Au nanowires has generated electron pulses composed of a series of attosecond pulses, observing free electron Rabi oscillations. Free electrons perform quantum walks on an energy ladder with the spacing set by the laser frequency^[318].

The advent of PINEM has established a possible platform for investigating nanoscale nonlinear optical responses. The nanoscale characterization of nonlinear optics is an extremely challenging task for EELS and CL, because the higher-order nonlinear response is generally considered unattainable due to the weak interaction between individual electrons and sample excitations. The excitation yields of photons by high-energy electron beams are very limited (e.g., <10−4 per electron with an acceleration voltage of about 100 kV), and nonlinear interactions hardly occur between photons generated by different electrons. Some theories propose the possibility of the nonlinear excitation with low-speed electrons, which significantly reduces the spatial resolution^[319,320]. However, the optical excitation in PINEM enables the detection of higher-order nonlinear responses. Under intense laser irradiation, effective excitations of material nonlinearities have been theoretically predicted^[321] and experimentally discovered^[322] through asymmetric gain-loss features in EELS spectra. This configuration enables electrons to emulate high-fluence optical pulses^[62], inducing characterizing nonlinear spectral modifications in graphene nanostructures^[320] and facilitating Jaynes-Cummings-type interactions with dielectric microcavities^[323]. Such nanoscale ultrafast probing techniques open new avenues for exploring quantum nonlinear phenomena, including the realization of deterministic single-photon sources^[324] and generation of a squeezed-vacuum state^[325].

Recently, PEEM has been used as a powerful tool to characterize SPs^[329,330]. In contrast to other electron microscopes, PEEM operates on the “photon in, electron out” principle^[331]. When a pulsed laser beam is focused on the sample, electrons can be excited from below the Fermi level to the vacuum level, thereby escaping from the surface. PEEM captures these emitted electrons to characterize electromagnetic properties of the sample^[332]. As the photoelectric effect indicates, the total energy of absorbed photons needs to exceed the work function of the material to realize linear or nonlinear excitations. The majority of metals exhibit work functions within the range of 3–6 eV (Au: 5.1–5.5 eV, Ag: 4.2–4.7 eV, Cu: 4.5–5.0 eV, Al: 3.8–4.4 eV, Ni: 4.6–5.2 eV)^[333]. Consequently, the nonlinear photoemission is generally required to characterize SPs in visible and infrared bands. The involvement of femtosecond lasers realizes PEEM based on the nonlinear photoemission^[334]. Due to the ultrahigh instantaneous field intensity of the ultrafast laser, recent experimental advances have achieved the simultaneous absorption of two to five photons to produce photoelectrons^[329,335]. PEEM images directly mirror the spatial distribution of hot electron populations excited in the sample to provide nonlinear maps of the optical fields.

A simplified theoretical model explains the principle of PEEM semi-quantitatively. The surface electromagnetic field of a plasmonic sample radiated by an ultrafast laser contains both the incident light (PLight) and the excited SP mode (PSP): Ptotal=PLight+PSP. The polarization field of the external laser beam is determined by the incident intensity and the optical properties of the material. The initial phase of the SP mode is determined by the interactions between the incident light and the coupling structure. Due to the clear phase relationship between the two contributions, the total polarization field Ptotal demonstrates their constructive and destructive interferences^[80]. This coherent signal is enhanced by the nonlinear photoemission, and the PEEM spatial intensity distribution is proportional to the integration of Ptotal2m^[72]: IPEEM(x,y)∼∫−∞+∞Ptotal(x,y,t)2mdt,(29)where m is the photoemission order (typically m = 2–5 for nonlinear photoemission, and m=1 for linear photoemission), which corresponds to the number of photons required to excite electrons above the vacuum level. As depicted in Fig. 7(b), the earliest spatially distinct interference patterns were observed on an Ag film containing a single-line defect, resulting from the interference between propagating SPPs excited by the laser pulse and localized polarization^[326]. Similar phenomena have been observed in nanorods and nanowires^[336,337]. Therefore, PEEM images reveal the distribution of surface optical responses through the detection of photoelectrons.

Conventional PEEM focuses photoelectrons with electron optics and collects them with a 2D electron imaging detector [Fig. 7(a)]. The spatial resolution of PEEM imaging has been reported to be better than 10 nm^[338,339]. UV lamp and synchrotron radiation light sources have been mounted to provide high-energy photons for single-photon photoemission. Femtosecond lasers are a more advanced type of source available for PEEM. They greatly enhance nonlinear effects and enable time-resolved pump-probe detections. Tunable radiation sources generate incident light of different wavelengths to selectively excite surface electromagnetic modes in different bands, realizing spectral measurements. The polarizing optical elements can be mounted into the optical path to uncover the electromagnetic modes under excitations with different polarizations. Additional information can be obtained from energy-resolved photoelectron detection^[340]. The detection at the Fourier plane after an electron lens achieves the angle-resolved PEEM with a high 2D momentum resolution^[341]. As shown in Fig. 7(c), the angle-resolved PEEM characterizes relationships among the energy and different components of momenta of photoelectrons. Unlike other characterization technologies based on electron microscopy, the momentum resolution and the spatial resolution of PEEM cannot reach high performance simultaneously, because the detections utilize electron optics instead of a scanning electron beam. It is important to note that positions and momenta are Fourier conjugate parameters for photoelectrons collected by PEEM. Conventional and angle-resolved PEEM provide high-precision photoelectron images in real and momentum space.

Pump-probe methods are involved to realize interferometric time-resolved photoemission electron microscopy (ITR-PEEM) with a temporal resolution of 10 fs^[328]. When a femtosecond laser pulse radiates a plasmonic structure, the SPs’ dynamics experience a fast evolution and return to the equilibrium in a short time. This process is significantly shorter than the time interval between two consecutive pulses and the PEEM image acquisition time. Therefore, conventional PEEM actually measures an integration of the entire photoemission process, as Eq. (29) indicates. In pump-probe detections, two phase-locked excitation pulses are employed to obtain PEEM images at varying time delays. The interference between excited optical modes and the probe light visualizes diverse electromagnetic dynamics of the surface in the space and time domains. The temporal resolution of time delays has been promoted to the attosecond scale^[342], but it is controversial to claim that the temporal resolution of PEEM exceeds the femtosecond scale, because the duration of the laser pulses (typically ∼10fs) also limits the resolution^[72]. Undeniably, relevant technologies have shown potential for true attosecond time-resolved PEEM.

In recent years, there have been notable plasmonic advances supported by PEEM, particularly in the imaging of the ultrafast spatiotemporal evolution and the manipulation of plasmonic nanostructures^{[330,343–345]}. For example, ITR-PEEM reveals the decoherence process between the LSPs and the incident laser pulse in the femtosecond scale^[328]. As shown in Fig. 7(d), the interference pattern diminishes when the time delay exceeds 20 fs, due to the detuning of the coherent LSP. Further research indicates that the dephasing time can be controlled by the size and the shape of nanoparticles and the polarization state of the incident light^[346,347]. Moreover, by utilizing high-intensity and few-cycle XUV pulses, the emission wave functions of individual electron wave packets in solids can be significantly controlled through quantum path interferometry^[348]. PEEM effectively promotes the development of plasmonics by revealing the near-field evolution of SPs.

It is crucial to select a suitable method that aligns with the specific objectives of the intended application. As illustrated in Table 1, a concise overview of the techniques is provided as suggestions on the selection, including distinctive advantages and disadvantages of the characterization technologies discussed in Sec. 3.

Recent advancements suggest that electron microscopies exhibit superior spatiotemporal resolutions compared to optical technologies^{[63,72,264,281]}. The spatial resolutions of far-field optical detections as LRM are significantly constrained due to the optical diffraction limit^{[146,151,152]}. SNOM employs scanning nanoscale tips to generate nano-resolution images, a process that diverges from the conventional optical imaging methods^{[164,239,349,350]}. Despite the success of near-field optics technologies in surpassing the optical diffraction limit, electron microscopies, particularly in conjunction with TEM systems, have achieved spatial resolutions that surpass those of optical methods. This enhancement is attributed to the short de Broglie wavelength of high-energy electrons. On the other hand, the temporal resolution is principally determined by the duration of excitation pulses, irrespective of whether the excitation is optical or electron-based. Femtosecond lasers generate optical pulses of about 10 fs, in accordance with the temporal resolution of LRM^[145], SNOM^[185], and PEEM^[328]. A sophisticated technique for generating ultrafast electron pulses utilizes femtosecond lasers to irradiate the electron gun, yielding pulse durations measured in hundreds of femtoseconds. The technique has been widely used in EELS, CL nanoscopy, and primary PINEM^[351,352]. An exciting advance is that the creation of attosecond electron microscopies improves the temporal resolution to below 1 fs, which has been proven in a PINEM system^[318]. Although there are no extant reports on this, it is hypothesized that the technique of attosecond electron microscopy can be imported to promote EELS and CL nanoscopy. Consequently, the most advanced electron microscopies perform better than optical technologies on the aspect of spatiotemporal resolutions.

As the technical development of characterizations, basic detections including imaging, spectral, and time-resolved measurements have been achieved in all types of technologies discussed above. Angle-resolved measurements that directly acquire information in the momentum space have also been widely reported with these technologies except SNOM and PINEM^{[151,257,268–270,274,278,280,327,341]}. In order to satisfy the momentum-matching condition, the tip of SNOM provides an additional momentum for the transformation between near-field and far-field signals, which impedes momentum detections. The absence of angle-resolved PINEM is partly attributed to the short study time. As a nascent technology, the potential capacity of PINEM remains to be further developed. It is actually possible that the replacement of an EELS detector with an angle-resolved EELS detector can realize angle-resolved PINEM. Despite the scarcity of direct measurements, indirect detection methods for the propagation of SPPs are attainable through the utilization of SNOM and PINEM systems^{[71,239,313,353]}. The acquisitions of phase images and time-resolved videos of the surface electromagnetic field have been effective methods for the effective demonstration of the propagation properties.

Despite the fundamental differences in their underlying physical principles and the detection of disparate physical quantities, SNOM and electron microscopy exhibit a high degree of similarity and complementarity in the quantitative analysis of nanoscale near-field mode distributions. SNOM is capable of collecting near-field optical information mixed with background scattering signals. In the most advanced SNOM system, the involvement of AFM tips confines the electromagnetic field to the nanoscale, enabling near-field optical detection with deep subwavelength spatial resolution. The multifaceted electrodynamics of the near-field interaction between the needle tip and the sample contain tip-induced electromagnetic fields, and the tip’s scattering of electromagnetic modes^[354], which is more significant. Polarized far-field illumination can selectively excite propagating electromagnetic modes on the surface of the sample^[355–357] and LSPs of nanoresonators^[159,358]. As discussed in Sec. 3.1.2.2, common models for s-SNOM describe the electromagnetic response of the tip as multiple dipoles arranged perpendicularly to the surface. Through numerical analysis of these dipoles, the theoretical value of the induced polarization can be ascertained. The far-field recording of the scattering signal, proportional to the induced polarization, enables the imaging of the near-field distribution of the vertically polarized component of the electric field^[359]. On the other hand, different types of electron nanoscopies are able to elaborate on the interaction between photons and electrons from different perspectives. The electron beam couples with the electromagnetic field distribution of all possible plasmonic modes in the nanostructure; therefore, what is captured by electron nanoscopes is their superposition^[360,361]. The high momentum and energy of the electrons facilitate the observation of the spatial distribution of diverse optical modes^[362] including dark mode^[363,364] at the nanoscale. Significant advancements in energy and time resolution over the past decade have precipitated substantial progress of electron nanoscopies, leading to revolutionary advances in the imaging of metal plasmon resonance and photonic crystal Bloch mode^[301,365]. Similar to the theory for s-SNOM, the electron beam serves as a linear current density, which is usually modeled as a series of dipoles with a temporal phase delay in simulations^[366,367]. The dipoles are positioned parallel to the movement of the electrons. These dipole approximations, as demonstrated above, enable the application of theoretical methods, such as the time-domain finite difference method (FDTD) and the finite element method (FEM), to analyze the near-field interaction between the dipole and the sample. This analysis serves to mirror the near-field interaction between the needle tip or free electrons and the sample, offering a high degree of convenience for experimental interpretation^[368].

The experiment yielded consistent and complementary results from optical methods and electron nanoscopies. These methods demonstrated optimal performance for diverse experimental requirements, such as wavelengths and materials. For example, the study of phonon polaritons in the far-infrared band has historically relied on the use of free electron lasers or synchrotron radiation light sources^[369,370]. It has been reported recently that EELS exhibits a higher efficiency of excitation and detection within this spectral range. The experimental observations included interference fringes of propagating surface phonon polaritons in zinc oxide nanowires and revealed ultra-slow group velocity characteristics. These results are highly consistent with the s-SNOM measurements of mid-infrared surface phonon polaritons in 1D hBN nanotubes^[371] and nanowires^[372].

In a noteworthy development, researchers mounted an SNOM system within the vacuum chamber of an SEM, establishing new possibilities for the characterization of high-resolution transport phenomena^[373]. This integrated system combines the functions of local excitations from SEM and optical scanning acquisition from SNOM. The electron beam is selectively focused on a specific point to generate a particular optical mode, while the optical probe scans the surface and collects signals. This configuration enables the acquisition of local surface morphology and near-field electromagnetic distribution of the sample, which exhibits the capacity of dual-beam imaging (comparable to AFM modified with focused ion beams^[374,375]). For instance, this technique has achieved the imaging of the carrier diffusion process inside GaN/InGaN core-shell nanowires^[376]. The major technical challenge lies in ensuring the mechanical stability of SNOM probes under the electron beam excitation, as well as the dynamic coupling between electron spontaneous emission and plasmonic electromagnetic modes. The prospect of enhanced performance in characterization through the development of a multi-probe collaborative system and 3D tomographic imaging technology is promising.

The aforementioned advanced optical characterization technologies have been employed to conduct a comprehensive investigation into the diverse properties of SPs, thereby facilitating multi-dimensional manipulations of SPs. It has been demonstrated that traditional optical methods can be used to control SPs, particularly at the scale beyond the wavelength. The recent advancements in optical characterization have enabled numerous groundbreaking discoveries in the field of plasmonic manipulations at the nanoscale. These characterization techniques, including LRM and SNOM, have been instrumental in experimental demonstrations of the effectiveness of SP manipulation, paving the way for further research with enhanced regulatory capabilities. This section introduces the near-field manipulation of SPs and focuses on the in-plane regulations. Despite the practical limitations imposed by material loss, fabrication precision^[396], and theoretical errors arising from approximations such as weak-coupling assumptions, the flexible control of SPs promotes the flourishing of plasmonic devices. The interference effect is a crucial fundamental principle in optical control, and phase manipulation (Sec. 4.1) provides the basis for diverse multifunctional regulations of plasmonic wavefronts. These wavefronts determine the propagation direction (Sec. 4.2) and distribution in real space (Sec. 4.3). This section also introduces the wavelength- and polarization-multiplexed techniques as a supplement (Sec. 4.4).

As one of the most significant challenges associated with plasmonic devices, the issue of loss must be addressed prior to a discussion of manipulations. In the interaction between light and a plasmonic metal, the inherent ohmic damping originates from the scattering of the intraband transition of electrons, including electron-phonon scattering, electron-electron scattering, and electron-impurity scattering^[397]. The ohmic damping reduces the efficiency of plasmonic devices and generates Joule heating of the metallic structure, limiting practical applications of plasmonics. Significant progress has been made in recent years to mitigate the impact of these intrinsic losses.

The judicious selection of materials can effectively reduce losses in plasmonic applications. Ag and Au are widely used materials offering excellent performance in the visible and near-infrared ranges. Their low rates of scattering of free electrons lead to high conductivity and high reflectivity^[143,398]. Ag has been demonstrated to exhibit lower loss characteristics in comparison with Au, while Au possesses the advantage of enhanced chemical stability. Al, on the other hand, exhibits distinct material properties that enable strong plasmon resonances with low loss, spanning from the visible region to the UV region^[399]. Cu has garnered attention as an alternative to noble metals due to its affordability. However, it should be noted that Cu plasmons experience significant dampening at photon energies that exceed its interband transition (2.1 eV). Nevertheless, Cu exhibits a low-loss window ranging from 620 to 750 nm^[400]. It is noteworthy that the low-loss window can be extended through the rational design of nanostructures^[401]. However, it is crucial to acknowledge the propensities of Al, Cu, and Ag to undergo oxidation, which significantly complicates the processes of preservation and nanofabrication due to the rapid formation of an oxide layer when exposed to the atmosphere. The plasmonic resonance exhibits a remarkable sensitivity to the presence of oxide within the metal^{[399,400,402,403]}. Besides common metals discussed above, alkali metals such as Na exhibit electronic transport properties that more closely resemble the ideal free electron gas model, accompanied by reduced interband transition losses. As a result, Na has been considered as a promising candidate for achieving extremely low optical losses. Relevant sodium-based plasmonic devices with state-of-the-art performance have been realized at the near-infrared region^[397]. However, the highly reactive chemical nature of alkali metals results in the stringent conditions required for their fabrication, which significantly limits their experimental explorations. After a thorough examination of the relevant factors, Au and Ag remain the metals of choice for plasmonic devices.

Alongside efforts to develop advanced plasmonic materials, the special design of nanostructures is another promising solution to loss suppression. One typical approach is to optimize optical power flow in nanostructures, including the construction of some innovative electromagnetic modes. For example, ultrathin spoof SPP waveguides (discussed in Sec. 2.1) have exhibited the capacity for low-loss propagation in terahertz and microwave regimes^[404,405]. Well-designed periodic structures can generate novel electromagnetic waves with behaviors analogous to those of traditional SPPs. In comparison with conventional microstrip lines, the strategy has been shown to significantly reduce the optical losses due to its tight field-confinement feature^[406–410]. Meanwhile, the low-loss properties of metals in terahertz and microwave regimes facilitate enhanced efficiency and augmented propagation length when compared to conventional SPP devices. In addition, dissipative and radiative losses in plasmonic nanostructures can be mitigated by manipulating their optical spectra through mechanisms such as a combination of constructive and destructive interference of SPP resonances. The effect has been found in various systems, including finite-sized clusters of plasmonic nanoparticles, a linear dimer chain of Au nanoparticles, and nonlocal high-Q plasmonic metasurfaces^[142,411]. The optimal results occur when electromagnetic energy is re-circulated outside the metallic volume and stored in the magnetic field rather than in the kinetic energy of electrons.

Another strategy involves hybrid dielectric-plasmonic devices, where plasmonic nanostructures are coupled to resonant optical elements, enhancing performance while minimizing losses. Hybrid nanostructures, combining metallic and dielectric components, have been proposed to synergize the strengths of both material systems: the strong electromagnetic confinement near the metal and the high scattering directivity and low losses provided by the dielectric. Examples of such configurations include metallic [or high-refractive-index dielectric (HRID)] nanoparticles on HRID (or metallic) substrates^[412,413], metal-dielectric core-shell nanoparticles^[414], plasmonic-dielectric gratings^[415], and the hybrid plasmonic waveguide^[416,417]. These hybrid systems have demonstrated enhanced performance over standalone metallic or dielectric structures in diverse applications, including photovoltaics, quantum light sources, photoluminescence emission, lasing, sensing, and surface-enhanced Raman scattering (SERS), among others. However, it is important to note that HRID materials exhibit negligible losses only within specific spectral regions, typically at wavelengths longer than their bandgap. This necessitates careful selection of HRID materials tailored to the target application.

These advancements about loss mitigation establish foundations for plasmonic manipulations, yet it is imperative to address these limitations more effectively in order to ensure future progress and successful real-world implementation.

As the base of the control over plasmonic wavefronts, phase manipulation plays a fundamental and crucial role for all types of optical modulations. Three main types of phase modulation have been investigated deeply and applied in diverse optical devices: resonant phase, geometric phase, and propagation phase^[418].

Figure 8(a) illustrates the typical metallic nanoantennas with several different shapes, which are representative of modulation units for resonant phase. The phase response of a multi-resonance nanoantenna usually covers the entire range of 2π with the change of geometric parameters. When external optical incidences excite surface electromagnetic waves on the nanoantenna surface, a damped harmonic oscillator model can be employed to describe oscillations of free electrons in the system, whose movements result in the formation of LSPRs^[425]. A charge q [mass m, displacement x(ω,t)] on a spring with spring constant κ is driven by an external field at frequency ω:^[418]md2xdt2+Γa,xdxdt+κx=qE0eiωt+Γs,xd3xdt3,(30)where Γa,x and Γs,x are the internal damping coefficient and the radiative loss, respectively. The last term, denoted by Γs,xd3xdt3, is employed to delineate the recoil effect in the emission process, formally designated as the Abraham–Lorentz force. The harmonic solution of Eq. (30) can be expressed as x(ω,t)=(q/m)E0(ω02−ω2)+iωm(Γa,x+ω2Γs,x)eiωt.(31)

Equation (31) describes the behavior of a solitary resonant mode. However, the range of tunable phase is constrained when solely a single antenna resonance is considered^[426]. Multiresonant nanoantennas have been employed to extend the phase response to the full range of 2π, leading to flexible regulation of electromagnetic waves. A paradigmatic example is provided by V-shaped nanoantennas composed of two nanorods of equal length, which exhibit both symmetric and antisymmetric modes. These modes originate from two types of coupling between the resonant modes of the two nanorods, in which the electromagnetic fields on the two rods exhibit the same or the opposite phases. Symmetric or antisymmetric modes can be driven by incident lights with polarization aligned with or perpendicular to their symmetry axis, respectively. Both lengths and included angles of nanostructures can regulate the resonant phase effectively^[427]. The first-order approximation describes the resonant wavelength as λ0,s≈2Lneff for the symmetric mode and λ0,a≈4Lneff for the antisymmetric mode, where neff is the effective refractive index. When the length deviates slightly from the resonance, the antenna impedance undergoes a transition from a capacitive state to a resistive state and subsequently to an inductive state as the length of the antenna increases. The variation in the included angle of two arms leads to the alteration of the overlap between the incident field and resonant modes, consequently resulting in the variation of radiative losses of the two resonant modes. The aforementioned alterations collectively induce changes in the resonant behaviors, thereby enabling the manipulation of the resonant phase through structural parameter tuning. The array composed of V-shaped nanoantennas has been experimentally demonstrated to implement 2π phase coverage [Fig. 8(d)]^[419,426].

Due to the complexity of the plasmonic resonance, the manipulation of resonant phase depends on numerical simulations without analytic expressions, and requires separate calculations for every unit. Consequently, the design based on resonant phase has been demonstrated to be ineffective for non-periodic multifunctional arrays, which require substantial computing resources for numerous nanostructures. The combination with geometric phase has been reported to reduce the computational cost substantially^[15,428,429]. It is noteworthy that the involvement of advanced algorithms including machine learning also saves the computing resources and expands the parameter space to promote the device performance^[430–433].

A further prevalent strategy for phase modulation refers to geometric phase, also designated the Pancharatnam-Berry phase (PB phase), which is directly correlated with the rotation angle of functional structures. The geometric phase arising during adiabatic cycling can be easily achieved for scattered light by spatially rotating each meta-atom. As depicted in Fig. 8(b), the phase difference between the scattered beams from two units with distinct rotation angles is equal to half the solid angle enclosed by their respective paths on the Poincaré sphere. Therefore, given a circular polarized incidence EiR/L=22(ex±iey), the transmitted light of a θ-rotated structure can be written as^[418]EtR/L=tx+ty2EiR/L+tx−ty2e±i2θEiL/R,(32)where tx and ty are the complex transmissivities for the incident linearly polarized light along the two primary axes. The circularly polarized transmitted component opposite to the incident polarization state has an additional geometric phase of 2θ, and the sign of the geometric phase is inverted for reverse circularly polarized incidence.

In the case of in-plane manipulations, another simple model utilizes dipole approximations, which has been widely applied in the arrays composed of nanoslits^[434]. The model supposes that an excited nanostructure can be described as a secondary SP dipole source with a specific orientation, which means that only one given linear polarized component of the incidence is effective. This approximation is applicable for nanoslits due to the fact that an electric field oriented perpendicularly to the nanoslits can result in their excitation. In the case of circularly polarized normal incidence, the SP field at the target point can be expressed as^[435]EpR/L=ηiλSPrei(kSPr±θ)cos(θ−ξ),(33)where θ and ξ denote the orientations of the dipole and the target point, respectively. η is defined as the complex conversion efficiency of SP excitation. kSP and λSP are the wave vector and the wavelength of the SPP. The geometric phase is seemly equal to once the rotation angle instead of twice when the factor cos(θ−ξ) is considered as an amplified factor. On the other hand, the field distribution is usually predicted through the integration of all unit cells of the array. The factor e±iθcos(θ−ξ) is equivalent to 12{e±i(2θ−ξ)+e±iξ}. In some cases, the integration of the second term can be disregarded; however, this is not universally applicable^[435]. In fact, the description of both once^[436,437] and twice^[438–440] the rotation angle has been employed in successful manipulations of SPs. Both of them are one type of approximation and have their own limitations. Simulations like FDTD and FEM methods can provide a more accurate analysis for the optical response of the single nanostructure. In either description, the strategy of geometric phase exhibits its independence from the dimensions of the structure, optical resonance, and intrinsic material dispersion. The recent development of the geometric phase has promoted various devices for wavefront manipulations^[420], as shown in Fig. 8(e). The modulation of geometric phase is contingent upon the rotation angle of nanostructures. Building upon this fundamental principle, researchers proposed a phase-gradient metasurface composed of U-shaped nanoapertures for helicity-dependent beam steering in theoretical studies^[441]. Subsequent experimental validation employed Au nanorod arrays to demonstrate the anomalous refraction phenomenon^[442]. Due to the succinct expression of the PB phase, the design can be easily extended to other different frequency regimes^[443]. Moreover, the manipulation efficiency can be further enhanced through several approaches including reflection array^[444], plasmonic hybridization^[445], and multilayer nanoantennas^[446].

The third type of phase controllers with the characteristic of the high aspect ratio of the unit element has been proposed to realize the propagation phase. As the name indicates, the propagation phase originates from the light propagation in the nanostructure [Fig. 8(c)], and it depends on the geometric size of the structure. The accumulated phase can be considered as the product of the height and the effective index with the neglect of Fabry-Perot effects, which is determined by the cross section of the structure. As an analytical instance, in MIM multilayers^[447] with intermediate layer width w, the wave vector of the SPP βSPP can be analytically determined by solving the Maxwell’s equations:^[418]tanh(w2βSPP2−εdk02)=−εdβSPP2−εmk02εmβSPP2−εdk02,(34)where k0 denotes the wave vector in free space. Hence, the width w provides direct control over the wave vector of the SPP, enabling precise phase engineering. Notably, the interfacial reflection coefficient is also regulated by the width, which involves an additional phase. Another classical phase shifter is depicted in Fig. 8(f); the propagation phase is well controlled by dielectric ridge waveguides and high-refractive-index scatterers^[421].

All of the aforementioned strategies of phase modulation have been corroborated by the near-field optical characterizations discussed in Sec. 3.1. Here, we take optical vortex generations as an example, which require a helical phase distribution on the surface of the sample. As shown in Figs. 8(g)–8(i), elaborate rings composed of nanostructures generate vortex beams with different topological charges. The first instance uses the strategy of combining resonant and geometric phases. The meta-atom is composed of two metallic arcs that are connected by a central bar. This configuration regulates the resonant phase by altering the length of the arc. The ring is fabricated on an SiO2 substrate, which enables the collection of leakage radiation from SPPs by LRM. The bottom right panel of Fig. 8(g) depicts the intensity distributions collected by the LRM system, which involves a Michelson interferometer for phase measurements^[422]. The interference pattern determined by the phase difference characterizes the topological charge of the vortex beam. Secondly, the rotation of nanoslit pairs regulates the geometric phase to generate THz vortex beams [Fig. 8(h)]. The frequency of these beams is determined by the structural parameters of the nanoslit. In this case, the THz near-field scanning system is applicable for amplitude and phase measurement. The distribution of both phase and amplitude can be obtained through the Fourier transform of the signal in the time domain. As shown in the left panel of Fig. 8(i), a spiral slit is a typical vortex generator that employs the strategy of the propagation phase. When SPP waves propagate from the slit pattern to the center, they undergo distinct phase retardations, which are proportional to the distance from the slit to the central region. Consequently, this results in the formation of a spiral phase profile, thereby generating a vortex beam. The vortex beam in the visible and infrared bands can be characterized by a specialized s-SNOM system. In this system, the incident laser beam is separated into two optical pathways for interferometric pseudo-heterodyne detection. The photodetector measures the interference between the scattered wave of the tip and the reference wave, which induces an additional periodic variation due to the vibration of the mirror. The phase of SPs can be obtained by analyzing the complex Fourier coefficient of the interference signal. The phase distributions measured through the pseudo-heterodyne interferometric detection manifest the OAM properties in the near field.

The plasmonic propagation direction can be flexibly manipulated based on the aforementioned phase modulation methods. The control over the propagation direction of the generated SPPs is essential for the high-efficiency powering of on-chip systems. Both simple nanostructures and arrays are proposed to regulate the propagation of SPPs.

Subwavelength nanostructures take advantage of their high compactness and maintain effective directionality manipulation of SPPs. The phase difference between two resonators generally depends on their distance, thereby enabling directional propagation of SPPs due to the interference. For instance, a pair of MIM nanostructures has been proposed for directional excitation of SPPs^[151]. The directionality of SPPs can be modulated by strategically adjusting the separation between the two nanoresonators, which is characterized by an LRM system. The excited SPPs at the air-metal interface radiatively leak through the thin Au film. The leakage radiation is then collected by the bottom immersion oil objective with a high numerical aperture. The imaging of the Fourier plane and the image plane provides the measurement of the directionality and the distribution of the SPP fields. As depicted in Fig. 9(a), the experimental results demonstrate that SPPs propagate predominantly along a desired direction. Manipulation of resonant modes for more significant directionality is required for single complex nanostructures. The defect aperture in a subwavelength waveguide achieves unidirectional and bidirectional plasmonic coupling^[448]. Figure 9(b) demonstrates the geometry and field distribution of the waveguide. The defect aperture breaks the structural symmetry and causes polarization-free unidirectional propagation of SPPs under illumination from the substrate side. A compact 3D interlayer nanorouter has been proposed to manipulate the propagation of SPPs^[451]. The incident SPP wave propagates along the air-Au interface and the scattered wave of the well-designed slot is determined by the resonant phase difference, which depends on the distance between the two slots. This strategy enables arbitrary routing in the range from 0° to 360° by appropriately modifying the interference of the SPPs.

The small size of single-subwavelength nanostructures represents a limitation in terms of their emission efficiency. Consequently, in order to meet the requirements of practical applications, arrays composed of nanostructures have been deeply investigated for directional propagation with a higher performance. With the improvement of grating couplers, coupler devices transfer the normal incident light to a directional SPP wave. A common design strategy estimates coupling coefficients for different directions through the overlap integral between the SPPs along the flat interface and scattered waves. In this way, a unidirectional launcher comprising 11 nanoscale grooves with different widths and depths realizes efficient directional regulation of SPPs^[452]. The excitation pattern can be tailored flexibly through variation of the lateral offset between the incident wave and the device. The geometric parameters are optimized with the aperiodic Fourier models to realize both high efficiency (>50%) and high extinction rate. This approach has been extended to easily fabricated rectangular grooves of equal depth^[453] and two cascaded sets of subwavelength grooves^[454]. 2D periodic arrays are high-performance schemes for in-plane directional routing, due to the multiple-beam interference. As a typical unit cell, nanoslits with various shapes^[450,455] have been reported to realize flexible SPP launchers^{[10,456–458]}. The interference between unit cells of a 2D periodic array results in SPPs propagating along two orthogonal directions. The resonant mode of a single structure determines the selection of the directionality and the polarization. The resonant eigenmode of a structure composed of multiple substructures can be considered as a superposition of the individual eigenmodes from its various components. The coupled-mode theory provides a comprehensive framework for the description of the proportion and phase of these various modes^[459–461]. The regulation of coupling coefficients among substructures can result in effective manipulation of the resonant eigenmode, potentially enabling directionality selection. A typical instance is illustrated in Fig. 9(d)^[450]. The unit cell consists of a bar-shaped slit resonator (BSSR) and a split-ring slit resonator (SRSR). Under x-polarized irradiation, the eigenmode of BSSR leads to SPPs propagating along the x-axis. Conversely, the eigenmode of BSSR cannot be excited by x-polarized incidences, and the y-polarized excitation results in the SPPs propagating along y-axis. The gap between BSSR and SRSR has been experimentally proven to regulate effectively the coupling coefficient, enabling the manipulation of the hybrid structure’s behavior. When the two resonators are positioned closely, the electromagnetic energy excited in BSSRs significantly couples to the SRSR, resulting in SPPs propagating along ±y directions. In addition to the control of eigenmodes, interference between different eigenmodes has been reported as another strategy. To illustrate, an L-shaped nanoantenna exhibits both a symmetric and an asymmetric mode. The phase retardation of these modes, under the influence of circularly polarized excitations, is determined by geometrical parameters^[455]. The interference between these two eigenmodes gives rise to a phenomenon of polarization-dependent unidirectional propagation of SPPs in two perpendicular directions.

Besides periodic arrays, non-periodic arrays of resonant subwavelength elements^[462] have shown unprecedented capabilities over manipulating optical fields. Gradient phase metasurface with interfacial phase discontinuity have been proposed to implement SPP directional couplers^[439]. As demonstrated in Fig. 9(e), the spatial rotation of metal slit resonators offers a geometric phase gradient of σΔφ/s in the x direction, where s is the period of the unit cell, and σ denotes the incident helicity. The chirality of the incident light switches the propagation direction of the excited SPPs, because the geometric phases are intrinsically opposite for the two spins. In addition to the geometric phase, the resonant phase strategy facilitates unidirectional SPP excitation^[449]. Figure 9(c) depicts the geometry of the metasurface, where the shape of each unit cell varies along the x direction. The incidences with different polarizations excite various optical modes, which exhibit distinct resonant phases. These distinct phase gradients lead to different electromagnetic waves. In the case that the phase gradient is smaller than the light wavenumber in free space, the radiated wave is a propagating wave. However, when it exceeds this threshold, the radiated wave becomes an SPP wave confined on the interface^[463]. Eventually, unidirectional SPP excitation under x-polarization and free-space beam steering at y-polarization are simultaneously achieved.

To summarize, this subsection discusses two types of devices for directionality manipulation: single subwavelength nanostructures and arrays. More specifically, the ultra-small size of single nanostructures leads to on-chip applications, including plasmonic sources^[151], fully optical information processing^[448], and other integrated plasmonic circuits. By exploiting polarization-dependent constructive or destructive interference, asymmetric nanoantenna pairs have the capacity of generating SPPs directionally propagating on-chip under plane-wave illumination with specific polarization. As an ultracompact instance, the defect-aperture-waveguide structure (scale < 1 μm) transforms the incident energy to the plasmonic waveguide while preserving polarization-selective control.

Arrays generally exhibit superior efficiency and greater manipulation flexibility. These arrays can be further classified into two categories: periodic arrays and aperiodic arrays. 1D periodic arrays primarily support light manipulation along one single axis for waveguide coupling applications. 2D periodic arrays offer more flexible in-plane regulations of SPPs, including polarization-selective directional propagations. These regulations fundamentally rely on the coupling and interference effects of resonant modes of unit cells. The design simplicity and fabrication convenience of periodic arrays render them particularly attractive in scenarios where additional functionalities are not required. Aperiodic arrays provide superior flexibility in multifunctional wavefront engineering when compared with periodic arrays. Such structures have been utilized in integrated near-field and far-field plasmonic launchers, multiple-channel polarization-multiplexed modulations, and other applications. Nonetheless, the design processes in question are contingent upon resonant/geometric phase modulations, the approximations of which are discussed in Sec. 4.1. The involvement of weak-coupling assumptions also inevitably affects design accuracy and efficiency due to the lack of translational symmetry. In order to improve the design productivity and the device efficiency, some sophisticated computational methods, including artificial intelligence algorithms, have been introduced to realize structural freedom for intricate functions.

The regulation of directionality and phase provides a significant base for plasmonic manipulation in real space. This subsection mainly concentrates on the arbitrary spatial control of plasmonic profiles, including arbitrary wavefront manipulation, integrated near- and far-field launching, and the excitation of specific electromagnetic modes.

The geometric phase is the most common strategy for wavefront manipulation. The SPP field at the target point can be considered as the result of the interference of SPPs generated from the resonators, whose phase retardations can be tailored by rotation of the structures. For instance, an array of aperture pairs realizes anomalous manipulation of the wavefront^[469]. Diverse shaping of the SPPs, including tilted, focused, and divergent wavefronts, has been achieved with the geometric phase control. Similar manipulation has been extended to plasmonic fields with arbitrary curvatures [Fig. 10(a)]^[464]. As Eq. (32) indicates, the geometric phase is highly related to the chirality of lights, which hinders independent modulations of the two circularly polarized components. In order to address the problem, the combination of geometric phase and resonant phase realizes arbitrarily high-efficiency chirality-delinked surface-plasmon wavefront control for both LCP and RCP components^[118]. The relevant research will be discussed in Sec. 4.4 in detail. This independent regulation has been achieved in terahertz^{[118,470,471]}, microwave^{[440,472,473]}, and near-infrared^{[429,474–476]} ranges. The design approach also generates other typical forms of SPP fields including plasmonic vortices^{[423,477–480]}, Bessel beams^{[472,481–483]}, and Airy beams^{[464,465,484]}.

In addition to pure near-field plasmonic wavefront manipulation, an integrated near-field and far-field plasmonic launcher has been proposed to achieve active control of the SPP transmission direction and the real/virtual focus of the far-field light [Fig. 10(d)]^[466]. The geometric phase of the nanoslit is designated as σθ for in-plane waves as SPPs, while a geometric phase 2σθ is generated for free-space waves. The integration of both the near- and far-field functionalities within a single structure is enabled by the utilization of distinct geometric phases of the nanoslit. In a similar manner, researchers have designed a bi-channel vortex beam generator in both the near and far fields, whose topological charges of the vortices are independent^[437].

Additionally, well-designed nanostructures are capable of generating distinctive near-field electromagnetic modes as toroidal and anapole modes. Toroidal multipoles are terms of the Taylor expansion of charge and current densities. The current density distribution of a toroidal dipole consists of a series of tiny current loops, and a toroidal dipole is generated from the poloidal currents flowing along the meridians of a torus^[485]. The direct discernment of this unique multipole mode is often hindered by the influence of more prominent electric and magnetic multipoles. By manipulating the near-field coupling between plasmonic nanostructures, it is possible to achieve complete destructive interference for major multipole modes. To address this, an array of split-ring resonators has been proposed as a means of suppressing the electric and magnetic dipoles, thereby revealing the toroidal response^[486]. Actually, the single resonator supports magnetic dipoles, and the array exhibits two types of resonances (in-phase and antiphase). When the resonator couples to the antiphase resonances, four adjacent magnetic dipoles form a head-to-tail configuration and compose a toroidal dipole. Besides, metasurfaces with toroidal dipole resonances facilitate numerous applications, including the enhancement of the third harmonic generation^[487], image edge detections^[488], and others.

Electric anapole modes are formed by the combination of electric and toroidal dipoles with the appropriate amplitude and phase relationship, which allows for destructive interference. These modes possess strong fields near the sample and a minimal far-field emission^[489]. A plasmonic metasurface consisting of dumbbell apertures and vertical split-ring resonators has exhibited transverse toroidal moment and resonant anapole behavior upon excitation with normal incidence. The existence of magnetic anapole modes has been demonstrated in gap surface plasmon resonators, validating the efficacy of the cancellation of a magnetic dipole with a magnetic toroidal dipole^[490]. Subsequent research employs magnetic anapole modes in plasmonic nanocavities to enhance and manipulate local second harmonic generation in lithium niobate^[491]. Results of spectral measurements are highly consistent with the theoretical predictions of multipole decomposition, validating the theory of plasmonic anapole modes. Meanwhile, a specific SNOM system has exhibited the potential for characterizations of this unique electromagnetic mode^[468], which has succeeded in the direct observation of anapole modes in dielectric nanodisks. As shown in Fig. 10(f), the sample is illuminated from the substrate side and collected on the top with the metal-coated tip in the transmission configuration. It is imperative to acknowledge that the signal collected by the a-SNOM probe does not directly correspond to any of the near-field components in isolation. Instead, it is the result of a convolution integral of both electric and magnetic near fields. The experimental results display the progression of the near-field mode in a silicon nanodisk and clearly demonstrate the appearance of a near-field maximum intensity at the anapole wavelength, which is consistent with the maximum of electric near fields in simulations. Since nanostructures with anapole responses possess electric hot spots in the field distributions, leading to a strong interaction between the hotspot and sensing medium, they can be readily used in refractive index sensing applications^[492].

As a basic freedom for information storage, polarization-multiplexed manipulation promotes the efficiency and the capacity of optical devices significantly. As Eq. (32) demonstrates, geometric phases for different circular polarizations are chirality-linked, which greatly restricts flexible polarization-multiplexed manipulation. However, the strategy of geometric phase has still been employed to realize 2D in-plane holography. The arbitrary closed-loop-shaped holograms inside circular areas are usually manipulated by regulating the surrounding multiple rings of structures. Each nanostructure serves as a secondary wave source with opposite geometric phases for LCP/RCP incidence. The structure designed for the LCP-excitation target field profile produces the complex conjugate profile of that under RCP incidence. The target profile can be conceptualized as a series of inwardly radiating (focused) point sources situated within the confines of the circular hologram region. Its conjugate version corresponds to an imaginary point source that radiates outwardly from the ring, which can usually be neglected without interrupting the target SPP profiles. Hence, the orientation profile α(x,y) of the nanoslits can be designed as the superposition of the normal profile for one spin and the complex conjugate profile for another spin^[493]: α(x,y)=f+[Ez++(Ez−)*]=f−[Ez−+(Ez+)*],(35)where Ez± is defined as the target SPP profiles of the LCP/RCP incidence; f± describes the relationships between orientations and the target SPP profiles of the LCP/RCP incidence. As shown in Fig. 11(a), the results of LRM manifest the patterns of “Δ” and “Y” and exhibit a high degree of consistency with the target profile, which substantiates the aforementioned theory. In order to improve the efficiency and the quality of the holography, the amplitude information is required to be modulated simultaneously. This necessitates additional degrees of structural freedom; hence more intricate unit structures are employed to achieve such high-quality holographic imaging. For instance, the angle of the nanoslit pair allows for amplitude control, and the corresponding device also facilitates simultaneous manipulation of amplitude and phase^[498].

The combination of geometric phase and propagation/resonant phase has been reported to realize on-chip polarization-controlled plasmonic vortex lenses. A subwavelength nanoslit in an Archimedes spiral-shaped arrangement^{[423,478,499–501]}, which simultaneously manipulates the propagation phase and geometric phase, is a typical design method for vortex generation. The opposite geometric phases for circularly polarized incidence result in extra contrary topological charges σm (incident photonic spin σ=±1 for LCP/RCP). Nanostructures are arranged in spiral for the generation of topological charge l: ρ(θ)=ρ0+l·θ·2π/λSPP,(36)where ρ0 is the inner radius in polar coordinates and λSPP represents the SPP wavelength. The whole structure yields a vortex beam with the topological charge of q=l+σm. The spin-independent propagation phase and spin-dependent PB phase work together to form arbitrary topological charges. Besides, harnessing the resonant mode facilitates the manifestation of intriguing effects. A single asymmetric crossed air-slit has been proposed to generate counterclockwise (clockwise) localized SP vortex fields (topological charge q=±1) by itself under −45° (45°) polarized excitation^[480]. When 20 slits are arranged geometrically as an Archimedes spiral, another topological charge is involved due to the propagation phase. The sum of these two topological charges determines the topological charges of vortices under distinct polarizations. SNOM characterizes the field profiles to uncover the topological charges, as demonstrated in Fig. 11(b).

Polarization-dependent directional transmission represents another significant area of research. One classical strategy utilizes the principle that only one linearly polarized component (i.e., magnetic field parallel to the slit and electric field perpendicular to the slit) contributes to the SPP excitation for nanoslits. Hence two series of slits with orthogonal orientations work independently under the orthogonally polarized incidences. As revealed in Fig. 11(c), four identical nanometric slits are arranged to form a pin cushion structure for polarization-sensitive SPP focuses^[494]. The polarization state of the illuminating beam determines the focal spots that appear along the polarization direction. Besides, the interference of multiple slit pairs also enables polarization-dependent directional transmission^[456,502]. The design incorporates three distinct degrees of freedom: the spin-dependent geometric phase (rotation angle), spin-independent resonant phase of two slit resonators with different geometries, and spin-independent propagation phase (relative distance between slits and distance the two slit-pairs). These parameters can be meticulously controlled to achieve polarization-dependent constructive and destructive interferences in different directions, facilitating the unidirectional propagation for a given polarization [Fig. 11(d)]^[15]. The interference of SPPs also allows the generation of polarization-controlled structured plasmonic fields^[458]. The formation of plasmonic fringe patterns is observed when two SPP beams oppositely propagate, and the shift of a plasmonic pattern depends on the incident polarization. This property leads to applications such as illumination microscopy and super-resolution imaging.

Wavelength-multiplexed devices play an important role in optical interconnections, due to their high capacity for information. The key point for wavelength-multiplexed technology is how to deal with the dispersion of nanostructures to simultaneously satisfy the requirements of different wavelengths. The geometric phase, being inherently independent of frequency, results in an achromatic phase control. In scenarios where stringent regulation efficiency is not a paramount concern, such as in the domain of holography, a viable approach entails the direct superimposition of the requisite complex field distribution for multiple frequencies^[435]. When the incident frequency deviates from the target frequency, the corresponding contribution of the wavefront profile becomes disordered. This results in a field intensity that is significantly lower than that of an effective hologram at the target frequency. Consequently, the influences from contributions designed for other frequencies are very limited compared to the target frequency. For example, as displayed in Fig. 11(e), a THz time-domain spectroscopy system validates the functions of a multiplexed focusing lens at four frequencies. However, this strategy actually ignores the crosstalk issue of multiple beams at different wavelengths. To further enhance device efficiency, it is necessary to mitigate the crosstalk between different target wavefronts. Due to the lack of relevant quantitative physical models, the design process often relies on massive numerical simulations. Diverse high-performance algorithms have been involved to provide solutions efficiently, such as the gradient-based optimization algorithm^[16]. A common strategy employed is iterative optimization based on the Huygens-Fresnel principle. This method involves repeatedly adjusting the structural parameters until the desired intensity distribution is obtained. In the case of multiple foci, the simulated field patterns after 40 iterations are shown in Fig. 11(f), which indicates that SPP waves are focused at different locations for different wavelengths.

In addition to metasurfaces, topological photonic crystals (TPCs) have also been reported to realize broadband and wavelength-multiplexed devices^[495]. They support propagating spoof SPP topological edge states and confine the SPP at the interference. By connecting the two different TPCs, the capacity of wavelength selections of the topological edge states results in a wavelength-multiplexed transportation. Figure 11(g) demonstrates that a meticulously designed valley topological photonic crystal displays wavelength-multiplexed unidirectional propagation when subjected to different excitation locations.

Besides polarization- and wavelength-multiplexed devices, joint-multiplexed modulations have also been achieved by combination of various strategies discussed above. For instance, a metalens composed of nanoslits has been proposed for switchable focusing^[496]. The orientation of nanoslits determines the effective incident polarization, while nanoslits are positioned at intersections of equiphase lines of the desired wavefronts of two wavelengths—in detail, two series of nanoslits with rotation of 0° or 90° function under two linearly polarized incidences, respectively. Each series generates focused SPPs when illuminated with two wavelengths. The SNOM images shown in Fig. 11(h) display its joint-multiplexed functions. Figure 11(i) demonstrates another typical example for spectral and polarimetric directional couplers^[497]. The spin-dependent and frequency-dependent interferences have been realized based on the geometric and propagation phase simultaneously, with the rotation and spatial arrangement of the nanoslits. The phase difference between adjacent nanoslits for different incident polarizations and wavelengths can be regulated independently, resulting in a spin-frequency-dependent directional transmission.

With the advancement of experimental techniques, electron microscopy is no longer limited to characterizing metallic nanostructures. The combination of nanophotonics and free electron radiation has made groundbreaking progress in the field of regulating electron-photon-matter interactions. The electron beam provides extra freedoms for the near-field multifunctional modulation of diverse optical modes including SPs. This section introduces representative achievements in the manipulation of SPs with the electron beam, including wavelengths (Sec. 5.1), polarizations (Sec. 5.2), directionalities (Sec. 5.3), spatial distributions (Sec. 5.4), and the interaction between SPs and transition metal dichalcogenides (TMDCs) (Sec. 5.5). In the end of the section, Sec. 5.6 discusses the electron-driven light sources as an example for practical applications of the technology based on electron beams.

At the beginning of the section is a brief introduction to the electromagnetic fields excited by free electrons. The radiation from free electrons can be categorized into two types: grazing-angle interaction zone and impact interaction zone, depending on the different scenarios of interaction between the free electrons and the sample^[503]. In the case of grazing incidence of a free electron, the excited optical modes are determined by the phase-matching condition^[504]ω(k)=v·k,(37)where ω and k denote the frequency and the wave vector of the photon, respectively. This general principle demonstrates that the mode phase velocity is required to coincide with the electron velocity. It can be applied to localized or extended optical modes in diverse material systems, including free space, gratings, metasurfaces, photonic crystals (PhC), free-form structures, and thin films. As a simple instance, when free electrons move at a velocity that exceeds the phase velocity of light in a uniform medium, they emit a “shock wave” that satisfies the phase-matching condition. This phenomenon is known as Cherenkov radiation (CR) ^[505,506]. A certain threshold velocity of CR is determined by the refractive index of the medium. In the wave vector space of periodic media, the electron plane described by Eq. (37) folds and intersects photonic bands at various locations in the Brillouin zone^[503]. Therefore, diverse optical modes, such as BIC modes^[507,508] and Bloch modes^[509,510], can be excited by the electrons and produce a far-field radiation as Smith–Purcell radiation (SPR)^[503,509].

In another scenario, the incident free electrons move in the normal direction to the surface and directly collide with the sample. The electron kinetic energy directly transfers to the material, leading to subsequent photon emission processes. The phase-matching condition cannot describe the coupling between free electrons and light, due to the inelastic collisions with the sample. The free electron radiation can still be controlled through nanophotonic structures, including transition radiation (TR)^[511,512], plasmon-induced radiation^[67,513], bremsstrahlung^[514], and electron scintillation^[515,516]. As a common theoretical model, free electrons are understood as fluctuating current sources for electromagnetic fields and thus emit photons in various ways^[516]. The current sources can be approximated as a series of dipole sources arranged normally to the interface between a sample and air^[277]. The LSPs excited by such dipoles exhibit a tight relationship to the excitation position. When a free electron impacts the “hotspot” of a specific LSP mode, the maximized overlap integral between the free electron near field and the electromagnetic field of the mode^[517] leads to a high intensity.

From a more general perspective, all free electron radiation under grazing incidence and the TR, plasmon-induced radiation in the case of direct impact share a common physical origin: they all arise from the coherent interaction between the intrinsic optical modes of nanostructures and the free electron current source. The light radiation maintains a stable phase relationship with the free electrons and thus can be collectively referred to as coherent CL radiation. Conversely, incoherent CL radiation (ICL) exhibits a random phase difference with respect to the electrons, primarily stemming from the presence of electron-hole pairs, defect states, and certain quantum systems. They originate from the transition of energy levels, and the lifetimes of high-energy states induce a random phase retardation. As the Purcell effect points out, the spontaneous emission rate is not only related to the atomic properties but also to the electric field strength and LDOS^[518]. Despite that direct regulations of the quantum states are complex and difficult, it is possible to enhance the spontaneous emission rate by the so-called photonic engineering^[88,519] (i.e., by controlling the intrinsic optical modes of nanostructures). The electron beam selectively excites a nanoscale region of the sample, rather than a micrometer spot. Modulating the LDOS of such a small region is significantly more feasible than regulating the average value of the micrometer area. It is noteworthy that the LDOSs of nearly all nanostructures manifest as ununiform distributions. This characteristic enables the electron beam to selectively excite quantum emitters at distinct locations, thereby enhancing their emissions through the modulation of the surrounding LDOS. In the context of a composite structure comprising quantum emitters and plasmonic nanostructures, electron beam irradiation initiates the excitation of both the emitter and a specific optical mode of the plasmonic nanostructure. The ICL from the quantum emitter typically predominates the total CL emission, attributable to its discernible intensity superiority. This ICL is modulated by the surrounding LDOS from the optical mode directly excited by the electron beam, because LDOS describes the total intensity of excitable optical modes and consists of contributions from various optical modes. In other words, the presence of quantum emitters functions as an auxiliary light source at the nanoscale level. The electron beam has the capacity to either excite the optical mode independently or to stimulate the quantum emitters, whose radiation is modulated by the optical mode. In this way, diverse properties of ICL have been successfully controlled, such as radiation wavelength, direction, polarization, and OAM^[520,521].

In order to regulate light-matter interactions, the displacement and energy control of the electron beam realize active wavelength manipulations of the SPs. As depicted in Fig. 12(a)^[522], the distance between the excitation position and the edge of the triangle decides the energy of the excited plasmonic resonant mode. When the electron beam excites positions 1–3, the edge mode is excited. As the excitation position shifts from 4 to 7, a transition occurs from edge mode to tip mode. The super high spatial resolution of CL nanoscopy enables the precise modulation of peak wavelength with a 19 nm displacement of the electron beam. In the same way, Fig. 12(b) shows two plasmonic resonance modes obtained from a nonamer structure under selective electron beam excitation. The lower-energy mode corresponds to the central disc excitation, and the higher-energy mode corresponds to the outer annulus excitation^[523]. Due to the lack of specific polarization of the electrons, the experimental results do not show the Fano resonance mode revealed by optical excitation. Instead, the electron beam excites discrete non-intrinsic modes formed by the coupling of the bright and dark modes of the entire structure, resulting in a central particle mode and a ring mode. The Purcell effect provides a theoretical foundation for the modulation of quantum emitters with energy-level transitions. When the electron beam impacts on aluminum nanodisks fabricated on a ZnO film, selective excitation of two LSPR modes can be realized via the displacement of the electron beam, whose energy satisfies the resonant conditions corresponding to the band-edge and defect transitions of ZnO [Fig. 12(c)]^[88]. The electron beam causes the energy-level transition of ZnO in a specific region, where the LDOS of the Al nanodisk exhibits a significant wavelength-dependent difference. The emissions from different energy levels of ZnO are enhanced differently due to the Purcell effect. In addition, recent studies have systematically investigated the phenomena under different excitation energies of 4–30 keV and revealed the coupling between electrons and the near-field optical modes [Fig. 12(d)]^[524]. The experimental results indicate that the strongest coupling occurs when the electron velocity is in phase with the phase velocity of the mode, thereby satisfying the so-called phase-matching condition. This provides fundamental insights into the strong coupling of plasmon-electron interactions.

Notwithstanding the electron beam’s inherent lack of polarization, the excitation of the electron beam can effectively regulate the polarization of the electromagnetic field it generates. This section will delve into the pertinent research concerning both scenarios of incidence.

When an electron beam grazes the surface of a periodic nanostructure, the electrons couple to the Bloch modes of the nanostructure and produce far-field radiation, known as SPR. Both theories and experiments demonstrate that the SPR of a simple grating exhibits linear polarization, but specially designed nanostructures achieve effective regulations of the polarizations of the SPR. By carefully adjusting the cross-coupling between electric and magnetic dipoles in a THz Babinet metasurface [Fig. 13(a)], researchers break through the restriction that the far-field radiation of a 1D grating always polarizes in one direction^[525]. This discovery not only allows for precise manipulation of the SPR polarization but also provides unparalleled control over its directionality. A well-designed holographic grating successfully generates vortex SPR beams with certain OAM [Fig. 13(b)]^[526]. Recently, controllable chiral SPR has been achieved in a light-well structure consisting of stacked metallic and dielectric layers^[527] [Fig. 13(c)]. The movement of free electrons through the light well generates synchronous surface currents on the two nearest sidewalls, which couple to the phase-matched Bloch modes. Two energy-degenerate and polarization-orthogonal Bloch modes compose a pair of perpendicular light sources with a stationary phase difference and result in chiral SPR^[517].

In the case of normal incidence, the involvement of the electron beam disrupts inherent symmetry and provides a new platform for controlling chiral nanophotonics^[91,301]. On the contrary, normal incident lights cannot excite the near-field chiral effects of symmetric metallic nanoantennas, so it is difficult for traditional optical characterizations to characterize them^[300]. Polarization-resolved CL nanoscopy reveals hidden chirality in non-chiral nanostructures. As shown in Fig. 13(d), when the electron beam impacts the endpoint of the metallic nanoantenna, it excites the corresponding circularly polarized electromagnetic mode, which reflects the near-field chiral radiative LDOS distribution. Therefore, changing the electron beam excitation position realizes the active switch of the CL chirality at the sub-nanoscale, whose maximum circular polarization degree reaches 90%. The hidden chirality originates from the interference between the symmetric and asymmetric modes of the V-shaped nanoantenna. The helicity of the optical transitions of quantum emitters determines the chiral-selective interaction between the emitter and the photon mode^[83]. Hence, the electron beam modulates chiral radiation of quantum emitters via controlling the distribution of LDOS [Fig. 13(e)]^[69]. The CL image of the Au nanoantenna exhibits four hotspots of LDOS at the corners, and each diagonal pair of them corresponds to some circularly polarized component. Due to the chiral distribution of LDOS, the chirality of the optical response of WSe2 exceeds 93%.

As with the optical case, the regulation of phase serves as the foundation for directionality modulation in electron beam excitation. However, due to the complexity of the process in which a high-energy electron impacts the sample, the methods for phase control via electron beams are relatively limited. As a consequence of the image charge, TR exhibits a well-defined phase in accordance with Maxwell’s equations. Meanwhile, the propagation phase of SPPs generated by an incident electron beam represents another phase that can be readily controlled. Recent interference measurements in the SEM-CL system reveal coherent behaviors of different types of CL^[94]. When the electron beam impacts the metallic substrate, both TR and SPPs are generated. The SPP propagates along the surface and excites optical modes of the Ag sphere. The interference between the TR and the light scattered by the Ag sphere results in spectral, angular, and excitation position-dependent fringe patterns in the CL intensity [Fig. 14(a)]. A comparison between the reference fields and the aforementioned fringe patterns allows for the construction of a model that represents both the intensity and the phase of the spatial SPP near field. Meanwhile, researchers have investigated the SPP scattering in helical nanoholes of single-crystal Au^[95], observing significant interference patterns along the ky direction and a phase singularity with a topological charge of −1 in the far-field phase distribution. When rotating clockwise around the phase singularity, the phase evolves from 0 to 2π [Fig. 14(b)]. These studies about phase regulations support the effective modulation of radiation directions.

Nanophotonic engineering has been proven to successfully regulate the directionality of luminescence produced by free electrons. In the case of periodic optical structures, the radiation power is proportional to the overlap integral between the electron trajectory and the Fourier components of the optical eigenmodes, which enables selective enhancement of specific angular components^[503,529]. The angle distribution of the emission can be flexibly manipulated by well-designed nanoantennas, as the near-field plasmonic modes determine the far-field properties of the radiation^[530]. The interactions between electron beams and plasmonic nanostructures have emerged as a promising system to investigate the control of light^[531], and the angle-resolved CL nanoscopy provides the main approach for the above research^[66,532].

Localized electron excitation has been proven to successfully realize directional emission of a single resonator and plasmonic antenna arrays^[97,528,533]. The former is attributed to the complex far-field interference of different multipole moments in the structure^[97]. As shown in Fig. 14(c), under the irradiation of an electron beam, the spectral and angular response of the metallic nanodisk strongly depends on the excitation position. In nanodisks with a size much smaller than the wavelength, the electric dipole mode is the dominant contributor to the near-field electromagnetic distribution. As the size increases, higher-order multipoles become more significant, and the directionality of the radiation is determined by the coherent superposition of all multiple modes. Other than a single resonator, it is necessary to consider the coherent near-field interactions among the basic units of the plasmonic nanoarray. For example, the array composed of Au nanoparticles has been demonstrated to control the radiation direction [Fig. 14(d)]^[528]. A concise theory employs dipole approximation and posits that far-field radiation is the consequence of the interference of a multitude of dipoles. Dipole moments of NPs are attributable to the field directly induced by the electron beam and the electromagnetic waves radiated by other NPs in the array. A set of linear equations that describe electromagnetic inductions among NPs can be solved in a self-consistent manner to determine the induced dipole moments. As with multi-beam interference, the radiation of nanoarrays is observed to exhibit improved unidirectionality in comparison to that of a single resonator.

When the electron beam grazes the surface of the sample, both CR and SPR exhibit significant wavelength-angle dependence due to the momentum-matching condition. Beyond the intrinsic unidirectionality, well-designed gratings serve as concave and convex lenses that regulate the far-field radiation caused by electron irradiation^[534,535]. Recent experiments have illustrated that chirped gratings realize focusing and defocusing of multi-wavelength SPR [Fig. 14(e)]^[99]. The structure transforms the conventional SP plane-wave emission into cylindrical wavefronts with a convex or concave curvature for light in the visible and near-infrared spectral regimes. Furthermore, van der Waals materials have been induced to manipulate the SPR far-field radiation in the X-ray. Atomic-level chirped grating effects can be achieved by stacking different types of materials or by stressing the multilayers to adjust the interlayer spacing^{[309,536,537]}.

Another noteworthy advancement is the incorporation of a rotating plate polarimeter into the angle-resolved CL configuration, thereby attaining comprehensive spatial and angular regulation over the full range of CL polarization states^[285,538]. Recently, the deep-subwavelength-scale optical spin Hall effect induced by electron beams has been realized in Au rectangular nanoantennas^[108]. Impacting different positions of the nanoantenna regulates the phase and amplitude of the electric dipole and quadrupole moments, thus enabling directional separation of photons with different spins [Fig. 14(f)]. A similar phenomenon has been discovered in 3D Au nanohelices. The LCP and RCP components radiate in different directions, which depends on the excitation position^[92]. According to the results of EELS measurements, the electron beam modulates longitudinal plasmonic modes, which induces dipolar transverse modes at resonant energy.

The advances of electron microscopies experimentally reveal and spatially map the degenerate states of multipole plasmon modes with nanometer spatial resolution, providing insights into their symmetry and degeneracy. As a nanoscale probe, electron beams also selectively produce given plasmon modes via the regulation of incident parameters including the electron energy and the excitation position^{[301,539,541–544]}. For instance, the plasmonic properties of chemically synthesized individual Au nanodecahedra have been investigated by EELS^[539]. Figure 15(a) depicts the experimental EELS spectra in excellent agreement with boundary-element method simulations, with a particular focus on the azimuthal plasmon modes in a 65 nm side-length Au decahedron supported by a mica substrate. Two basic dipolar modes polarize along the particle’s pentagonal symmetry axis (polar mode) and the plane perpendicular to that axis (azimuthal mode). The asymmetric interaction between the nanoparticle and the substrate results in the splitting of degenerate modes, and these distinct hybrid modes emerge selectively when the electron beam impacts different corners of the decahedron. Similarly, when scanning the electron beam along the bisector of a nanoprism, the experimental CL spectral images reveal the tight relationship between the excitation position and the excited plasmon modes^[301]. For example, the dipole mode is efficiently excited when the electron beam is located at a corner of the nanoprism, and its calculated photon emission probability maps are in good agreement with experimental counterparts for all emission polarizations [Fig. 15(b)]. As the electron beam moves to the center of the nanoprism and the midpoint of its bottom edge, higher-order modes, including hexapolar and quadrupolar modes, are observed correspondingly.

Notably, electron nanoscopies provide an advanced platform that illustrates the optical modulation of the near-field spatial electromagnetic distribution, particularly in the evolution over time. The introduction of lasers has enabled the full control of optical pulses, including their intensity, duration, and coherence^[93]. Recently, ultrafast transmission electron microscopy has been advanced to attosecond time resolution within one plasmonic optical cycle by applying a continuous-wave laser to modulate the electron wave function into a rapid sequence of electron pulses. The PINEM system is capable of capturing femtosecond movies of the electromagnetic near field in both space and time^[312]. The chirality of the incident light determines the propagation of SPPs on the surface of a needle tip. The left-handed excitation generates SPPs that propagate almost entirely along the underside of the tip, whereas the right-handed excitation produces SPPs on the upper side. Another available approach is to modulate the coherent amplitude or phase of the free electron wave function^[392]. As the group velocity of electrons differs from the phase velocity of light, modifying the relative distance between the sample and the reference membrane allows for the alteration of the phase difference, thereby facilitating coherent detection. Both magnitude and complex amplitude are calculated via an interferogram analysis. As shown in Figs. 15(c), (d), different near-field electromagnetic distributions are observed under the incidence with distinct polarizations. In addition to PINEM, the interferometric pump-probe technique of PEEM directly reflects the dephasing process of LSPR^[72,73,545]. It has been reported that p-polarized and s-polarized lasers with an oblique incident angle (74°) respectively excite dipole and quadrupole modes of the Au nanoblocks [Fig. 15(e)]^[540]. The experimental results prove that the quadrupole mode has a longer dephasing time than the dipole mode due to the smaller net dipole moment. Moreover, time-resolved PEEM has been highlighted as the optimal tool for capturing the evolution of plasmonic vortices. Archimedean spiral coupling structures impart a defined OAM to the wavefront of SPPs^[342,545]. The experimental results reveal three stages of the evolution: the formation of converging spiraling wavefronts, the creation of a circulating vortex field with bright lobes, and the eventual outward propagation of the vortex and loss of lobe structure. The primary spin-orbit conversion of the vortex is attributed to the annihilation of photons during the emission of SPPs from the boundaries of the Archimedean structure, which results in the formation of converging spiraling wavefront threads. Subsequently, the SPPs that counterpropagate in inward and outward directions interfere with each other, resulting in a concentrated rotating vortex field in the central region. Moreover, as illustrated in Fig. 15(f), plasmonic vortex cavities have been demonstrated to produce a succession of vortex pulses with increasing topological charge as a function of time^[309]. The reflection from structural boundaries serves as a new degree of freedom for the generation and control of plasmonic orbital angular momentum.

TMDC materials, comprising layered structures with strong in-plane bonding and weak out-of-plane interactions, facilitate delamination into 2D sheets. In particular, TMDCs such as MoS2, MoSe2, WS2, and WSe2 exhibit substantial bandgaps that transition from indirect to direct in their monolayer form, enabling a wide range of applications including transistors, photodetectors, and electroluminescent devices^[549–551]. In TMDC-metal hybrid structures, the interaction between SPs and excitons is highly enhanced due to that SPs tightly confine an electromagnetic field much below the diffraction limit. Hence, the TMDC-metal hybrid system provides a potential platform for further manipulation of light-matter interaction. Especially, the electron beam with nanoscale spatial resolution opens up a unique sight for relevant investigation. It has been reported that CL nanoscopy directly detects optical signals of monolayer TMDCs encapsulated with h-BN flakes^[552]. Moreover, a subwavelength manipulation of valley polarization has been realized in Au antennas and h-BN/WSe2/h-BN hybrid nanostructures [Figs. 16(a)–16(c)]^[546]. When the electron beam impacts the corner of a rectangular rod, it excites the chiral near-field plasmonic dipole modes, which transfer energy with the excitons in the WSe2 monolayer through a dipole-dipole interaction. As the valley-dependent optical selection rules^[553] indicate, mere excitons from the valley with the corresponding chirality couple to the circularly polarized near fields. Therefore, the displacement of the electron beam manipulates the valley polarization of WSe2. In addition, a similar system also achieves far-field valley-separated emission of WS2 [Figs. 16(d), 16(e)]^[530]. The simulated and experimental results demonstrate the separation of LCP and RCP components with the incident electron beam at the midpoint of the short bar. This separation process can be explained by the multipole mode interference effect in nanostructures. It is worth mentioning that routing emission from different valleys supports the base of valleytronic information reading, which requires isolated measurement for one valley state. Besides the control over valley polarizations in TMDCs, the valley degree of freedom in plasmonic modes can also be achieved and regulated^[547]. The STEM-CL system visualizes a valley-polarized plasmonic edge mode and manipulates its propagation [Figs. 16(f), 16(g)]. The strong coupling between TMDCs and SPs tailors the band diagram^[548]. As depicted in Figs. 16(h), 16(i), the TMDC-metal hybrid structure regulates the quasi-propagating Bloch modes in photonic crystal coupling to excitons in the WSe2 flake. The strong light-matter interaction leads to the formation of plexcitons and significant changes in the band diagram, which generates new flat bands.

In contrast to the discrete emission wavelengths of laser sources, free-electron-driven radiation sources have attracted considerable attention due to their wide tunability and available wavelength ranges. The high energy carried by free electrons allows for nanoscale manipulation from ultraviolet to X-ray sources, enabling high-energy-level transitions^{[514,554–556]}. The recent development of electron-driven light sources has benefited from advances in nanotechnology. With the progress of nanotechnology such as photolithography, electron beam lithography (EBL), and focused ion beam (FIB) etching, it is possible to fabricate grating structures with smaller spatial dimensions^[557,558]. The latest research has illustrated that the emission caused by electron irradiation can be fully regulated, regardless of whether the incidence is grazing or normal.

As mentioned before, electron grazing incidence usually generates coherent CL emissions as CR and SPR. The first on-chip integrated threshold-less Cherenkov radiation source was reported in 2017^[102]. The hyperbolic dispersion relation supports electromagnetic modes with large wave vectors, which couple to low-energy free electrons for CR generation. In the experiments, the interaction between free electrons and the hyperbolic metamaterial formed by alternating Au and SiO2 films successfully eliminates significantly the electron velocity threshold required to generate CR. The device exhibits a two-order reduction in the threshold of electron energy and a two-order enhancement in the output power. In the hyperbolic metamaterial with optical topological transitions, the excellent photothermal properties and the nonlocality of graphene provide an extra effective approach for CR caused by low-energy electrons [Fig. 17(a)]^[559]. As the electron velocity approaches the Fermi velocity of graphene, the plasmonic properties of the graphene-based hyperbolic metamaterial are regulated by nonlocality-induced optical topological transitions, resulting in the emergence of a new lower velocity threshold for hyperbolic CR. Additionally, Cherenkov surface waves have also been observed^[560], whereby free electrons emit narrow-bandwidth photonic quasiparticles that propagate in two dimensions. The 2D Cherenkov interaction has the potential to achieve strong coupling effects with photons, which could lead to significant advances in the field of quantum optics.

Compared to the stringent experimental conditions and material limitations of CR, SPR supports an optimal platform for constructing on-chip ultra-compact light sources, due to its intrinsic threshold-less characteristics and flexible control over the spectrum, emission direction, polarization, and even chirality^[561]. Theoretical studies have indicated that when electrons are confined to a region smaller than the wavelength, the radiation becomes coherent. In this case, the emitted energy does not increase linearly, but quadratically with the number of electrons, when the electromagnetic fields of the electrons are added linearly^[563]. In the THz range, electron beam superradiance can be achieved by incorporating a low-energy spread, low-emittance electron beams, and a diffraction grating into SEM^[86]. Notably, a “light-well” structure based on a multilayer metamaterial composed of periodic Au and SiO2 supports the first theoretical foundation for nanoscale SPR sources^[561]. The emission of the light well originates from an oscillating dipole source created as electrons undergo periodic potential modulation within the well [Fig. 17(c)]. As a consequence of the interaction with the 1D photonic bands of the periodic structured well, the wavelength of the discrete emission is determined by the wave vector of the guided mode. Researchers have developed an ultra-compact chiral light source with the capability of tuning the degree of circular polarization (DOCP) from −0.45 to 0.45 by adjusting the excitation position [Fig. 17(d)]^[527]. Similarly, theoretical studies have demonstrated controllable THz SPR generated by low-energy electron bunches moving atop the graphene layer on a grating^{[562,564,565]}. Due to low losses and high confinement, the localized field enhancement significantly increases the radiation intensity [Fig. 17(e)], and the Fermi level of the graphene layer modulates effectively the radiation frequency. Besides conventional periodic structures, a disordered plasmonic structure has also been demonstrated to produce free electron radiation in control^[87]. As shown in Fig. 17(f), the SPR from a 2D plasmonic crystal composed of random metallic rods contains two polarized components corresponding to two mechanisms including the collective effect of conventional SPR and the localized plasmonic resonance. The disordered photonic crystal has a periodicity in two directions, which leads to a phase resonance and generates conventional SPR with the polarization parallel to the electron incidence. The geometry of a single rod determines that the plasmonic resonant mode is polarized perpendicular to the direction of electron motion.

The miniaturization and integration of free-electron-driven sources is an emerging area of research^[566]. Considerable efforts have been made to achieve stronger electron radiation at lower electron velocities, including the use of highly mobile carriers in 2D quantum materials to generate SPR^[567]. This paves the way for the development of compact, room-temperature, high-intensity free-electron-driven radiation sources for further applications in spectroscopy and sensing.

When electrons normally impact the sample, they generate surface optical modes that directly produce far-field radiation or propagate along the surface, as LSPs and SPPs. The latter modes may subsequently be scattered by nanostructures, resulting in the energy coupled to the far field. The effective multidimensional modulation of SPs facilitates the development of several electron-driven light sources for normal incidence, which can be utilized in a range of advanced optoelectronic experiments^[570]. For example, a well-designed plasmonic metasurface has been demonstrated to convert the localized electron beam excitation to a free-space light beam^[568]. As illustrated in Fig. 18(a), the spectra with components measured separately on an unstructured part of the sample subtracted reveal a strong correlation between the CL peak emission wavelength and the plasmonic absorption peak. In contrast to conventional CL radiation, the phenomenon indicates that square arrays of asymmetrically split rings form a collective oscillation of the whole metasurface other than a few units near the excitation position. Therefore, many 2D structures have been investigated as novel electron-driven light sources.

It has been reported that incident electrons generate chiral SPPs on a photon sieve, which are subsequently scattered by nanoholes to the far field^[101]. The wavefront of outcoupled radiation can be entirely tailored by the geometry and the distribution of holes on the Au film. Both TE- and TM-polarized Laguerre–Gauss light beams with a chiral intensity distribution are detected via the interference with transition radiation. Figure 18(b) illustrates another function of the radiation modulation that the number of arms in the photon sieve directly determines the OAM of the vortex light. The detected angle-resolved CL exhibits the interference pattern of the vortex light and TR. The holographic optical design of a plasmonic metasurface has also been applied as a general method for electron-driven light sources^[569]. Multiple properties of the emission, including wavelength, direction, divergence, and topological charge, can be flexibly regulated via the holographic method [Fig. 18(c)]. Additionally, the focusing effect has been realized in a photo sieve composed of subwavelength etched holes on metal films. The optical resonance modes of each small hole produce coherent CL that interferes constructively at the specific far-field location [Fig. 18(d)]^[100]. Researchers manipulate the effective refractive index determined by the geometrical manifestation to obtain the configuration of the photonic sieve. Investigations of electron-driven light sources facilitate their incorporation inside an electron microscope for further applications as coherent detections^[280].

As previously discussed, significant progress has been made in plasmonics. Near-optical methods and electron nanoscopies have revealed the underlying mechanism of SPs, validating relevant theories based on classical Maxwell’s equations. From an application perspective, the flourishing of plasmonic devices based on optical modulations has yielded remarkable outcomes in a multitude of practical fields, with promising prospects for the future. It is another important research field to explore the interaction mechanism between SPs and other systems, including nonlinear materials, biomolecules, photocatalysis, and others. Interdisciplinary research greatly expands the scope of plasmonic devices. On the contrary, electron-beam-based plasmonic devices are predominantly in the laboratory phase at present. One of the primary reasons for this is that high-energy electrons continuously collide with gas molecules in a low-vacuum environment, leading to a rapid decay of their kinetic energy. In addition, the significant radiation damage resulting from the inelastic collision between the high-energy electron beam and the sample is also a major limitation of related devices. The inherent ultrahigh spatial resolution of electron beams makes this technology approach one of the potential solutions for highly integrated devices in the future, but it is difficult to see large-scale applications in the short term. On the theoretical side, the main focus of future research on pure plasmonics will be on the quantum model, which remains a substantial unknown and of great research value. Although existing technologies have already partly revealed quantum effects, measurements of numerous quantum properties and behaviors are still lacking. Electron nanoscopies exhibit greater potential for quantum detections, especially the scheme of PINEM. Recent studies demonstrate the effectiveness of PINEM in characterizing some quantum effects associated with photon-electron interactions^[307]. Some innovative upgrades of electron nanoscopies have exhibited their prospects for more advanced characterization technologies, which probably support a new way for quantum plasmonics. Due to space constraints, this section does not delve into the details of optical applications and interdisciplinary research but rather focuses on the quantum plasmonics (Sec. 6.1) and the upgrade of characterization technologies (Sec. 6.2).

In the realm of classical electrodynamics, we disregard the electronic genesis of plasmons and primarily concentrate on the electromagnetic field they propagate. Surface plasmons are merely perceived as electromagnetic field modes confined to the boundary between metals and dielectrics. Numerous plasmon applications stem from their profoundly localized, intense electric fields, and classical electromagnetic field calculations have yielded remarkable success in plasmonic research, accurately portraying numerous experimental phenomena. However, due to the rapid advancement of nanofabrication techniques, optoelectronic device structures have become increasingly intricate. When the structural dimensions approach the mean free path of electrons in the material, the dielectric functions measured from macroscopic materials cannot describe the properties of the materials accurately, because it is necessary to consider the behavior of the electronic states that form the SPs. Therefore, there has been an escalating enthusiasm to explore the quantum properties of SPs and develop plasmonic devices that function reliably at the quantum scale. As one kind of quasi-particle, it is significant to research the quantum nature of SPs from a fundamental perspective, but many of their quantum properties remain unknown^[571].

The quantization of SPs has been investigated for a long time. In the 1950s, Bohm and Pines carried out the first model for quantization of SPs^[572]. Several years later, Hopfield proposed the concept of “plasmon-polaritons” and their basic quantum model as bosons^[573,574]. Another macroscopic method based on Green’s function has been established recently^[571,575]. Despite the longstanding history of theoretical research on quantum plasmonics, their experimental advancements have been constrained by the evolution of experimental technologies. However, with the breakthroughs in technologies such as CL, EELS, and PINEM in the past decades, there has been a significant and rapid surge in their experimental progress. As one of the fundamental properties, the wave-particle duality of single SPPs has been experimentally observed in 2009^[576]. A complex system composed of a single-photon source and an Ag nanowire exhibits both wave and particle properties of single surface plasmon polaritons. The antibunching experiments illustrate the particle behavior of SPPs and the wave figures are proved by the self-interference. Recently, ultrafast PINEM successfully imaged both the spatial interference and the quantization of SPPs^[70]. An intense femtosecond laser has been utilized to excite an isolated metallic nanowire to form an SPP standing wave that could be modulated by the polarization of input light. The new ultrafast-imaging approach provides a snapshot capturing the interaction between the electrons and the SPP standing wave. This picture demonstrates a 2D projection of one spatial coordinate versus electron energy. Although the quantum models of SPs have been proven primarily for simple systems, the underlying quantum parameters are hardly measured, which are meaningful for the confirmation of quantization theories. Therefore, it is expected that the advanced characterizations will illustrate the diverse quantum systems of SPs directly and specifically in the future.

Interference holds paramount significance in enhancing our comprehension of the wave-like attributes of light. Similarly, quantum interference plays a crucial role in the illustration of the quantum nature of SPs. It is reported that two-plasmon quantum interference has been observed in both the free-space and plasmonic waveguides^[577]. Diverse quantum interferometers are realized in plasmonic systems, such as the Hong-Ou-Mandel (HOM) interferometer^[578,579] and the N00N state interferometer^[580]. These experimental results reveal the feasibility of generating indistinguishable single plasmons in plasmonic devices, which is one of the fundamental requirements for all quantum information technologies. In addition, they also prove that the intrinsic losses involved by metallic materials do not break the quantum features of the particles. Once the theoretical foundation of optical computing is firmly established, it becomes imperative to delve into the quantum dynamics of SPs and devise strategies for the manipulation and measurement of quantum states. Recently, the near-field dynamics of a plasmonic system have been measured through the projection into its constituent multiparticle subsystems^[581]. It is experimentally confirmed that the classical near-field dynamics of SPs can be described by nonclassical processes of scattering among their constituent multiparticle subsystems. Another 2D nanoscopy with a plasmon-polariton-assisted electron emission as a signal channel spatially resolves quantum coherences of the quantum states of SPPs^[582]. Researchers observe a plasmon quantum wave packet via the coherence oscillation at the third harmonic of the plasmon frequency. Although there are numerous exciting discoveries, how to probe and manipulate on the nanoscale in the time domain remains a great challenge. As the base of optical computing, the corresponding investigation is still in its nascent stages, and further research in this rapidly evolving field would be highly significant.

It is widely acknowledged that SPP preserves the quantum statistical information of the excitation light field. This assertion has been corroborated by a substantial body of experimental evidence^[583–585]. Nevertheless, it has been demonstrated that multi-particle scattering between SPs and photons can result in alterations to the quantum statistical characteristics of plasmonic systems. The probability of SPP propagation along the surface can be regulated by the ratio of horizontal to vertical polarization in the incident light. Moreover, the potential for utilizing coherent or incoherent boson scattering to alter the quantum fluctuations of multi-particle systems has been corroborated^[586]. The input light exerts control over the quantum statistical properties of the SPPs, and joint conditional detection can be performed using multiple photon number-resolving detectors at the output. Recently, the near-field dynamics of SPP have been measured by projecting plasmonic systems onto subsystems of quantum light fields, with the objective of obtaining their quantum dynamic properties^[581]. In contrast to classical scattering processes, the experimental results indicate that the quantum dynamics of plasmonic systems are governed by coherent processes of bosons or fermions.

Fundamentally, the spontaneous light source driven by free electrons unveils quantum effects that transcend the classical Maxwell’s equations. Classical theory determines the energy of emitted photons by assuming that the particle trajectory remains unchanged after electron interactions, a premise that contradicts quantum electrodynamics. Consequently, the energy of photons spontaneously emitted by electrons may deviate from classical predictions, a phenomenon known as quantum recoil. Although this phenomenon has been extensively discussed and studied theoretically, experimental advancements have only recently been made. In this experiment, the SEM was employed to measure Smith–Purcell radiation emitted by electrons with kinetic energies ranging from 10 to 15 keV in the X-ray band^[587]. The scattering of electrons from multilayer sub-nanometer gratings crafted from graphene and boron nitride exhibits a shift in the energy of the emitted photons. This is fundamentally distinct from the electron diffraction experiment conducted in a TEM. The former phenomenon can be attributed to quantum recoil, which occurs as a consequence of the emission of photons by electrons. In contrast, the latter phenomenon is caused by the interaction between the electronic matter wave and the crystal lattice. Further experimentation should be conducted with the objective of observing quantum recoil in plasmonic systems, specifically within the visible light spectrum. Although the quantum wave function has been observed in electron spontaneous emission within a semiclassical framework, the experimental findings can be elucidated through probability distribution methods, with electrons continuing to interact with the metal grating as classical point charges^[588]. The description of the interaction between free electrons and metal plasmons within the framework of quantum electrodynamics remains a significant challenge for physicists.

Other than traditional optical methods, electron nanoscopies provide an excellent approach for the underlying physical mechanisms, especially quantum effects. Although there has been remarkable progress for characterization technologies, it requires more multifunctional experimental methods to explore plasmonics. Besides the advances in engineering, involving new freedoms or sources gives electron microscopies abundant abilities to measure more complicated plasmonic properties.

The modulation of electron wavefronts provides specialized electron beams for unique detections, because of the additional freedom offered by them. Distinct from conventional electron beams with plane-wave phase fronts, the wave function of structured electron beams is precisely crafted in either the temporal or spatial domain. Recent experiments have proven numerous techniques for producing such beams, thereby enabling their utilization as quantum electron probes. For instance, in the temporal domain, a solitary electron pulse can be fashioned into a sequence of attosecond pulses through the coherent interplay with optical near fields^{[312,318,322]} and compressed via enhanced resonator-assisted pulsed THz fields^[589,590]. In the spatial domain, the phase front of electron beams can be sculpted into a singular spiraling phase employing methods such as rotated superimposed graphene sheets^[591], holographic gratings^[592], magnetic charges^[593,594], and plasmonic near fields^[595]. This vortex electron beam, carrying orbital angular momentum, underscores the parallels between electron optics and photonics, stemming from the functional equivalence of the Helmholtz equations and the time-independent Schrödinger equation. Additionally, various beam geometries beyond vortices have been realized^[596], demonstrating a versatile modulation of the phase front. The corresponding technique has been used in quantum plasmonics. For example, the electron beam with the symmetry broken artificially has been employed to detect the special inside plasmon modes of nanoparticles in a TEM-EELS system^[597]. Their experiments revealed the coupling mechanism between the plasmonic potential and the phase of electron beams in inelastic scattering. It is proposed that vortex electron beams can be used to map the plasmonic chiral modes^[598]. There are numerous other applications of structured electron beams^[599–602], and their combination with diverse characterization technologies is meaningful for plasmonic measurements. In light of the complexity and significant energy demands associated with the generation of functional electron beams, we also anticipate a more flexible and compact experimental manipulation setup to involve it in electron microscopies.

Involvement of an extra source is another way to upgrade the characterization methods. Due to the high quality of coherence between electron beams and excited SPs, an electron-driven photon source has been installed into CL nanoscopy systems to generate naturally synchronized light, offering an interference-based approach to investigate the temporal dynamics of sample responses^[280]. As a significant supplement for time-resolved characterizations, this interferometric measurement heralds vast future prospects for the analysis of the nanoscale dynamics. Nanophotonic engineering of electron-driven photon sources can yield tailored beams for diverse measurement needs, and the entanglement between electrons and emitted photons may serve as a platform to explore quantum aspects of electron-photon interactions. Furthermore, this interferometric method offers an alternative for minimizing electron-induced specimen degradation, amplifying weak signals of reduced electron interaction with a calibrated reference light^[603]. This feature paves the way for expanded application in non-destructive characterization, which is crucial for biomolecules and strongly correlated materials.

In summary, this review discusses advanced characterization technologies and their applications in plasmonics, from perspectives of near-field optics and electron nanoscopies. Over the past two decades, numerous advanced characterization technologies have been proposed and developed, driving the rapid growth of plasmonics. Optical methods, such as LRM and SNOM, are utilized to study the near-field properties of SPs, enhancing our understanding of plasmonic nanostructure theory and design. In comparison, electron nanoscopies offer distinct advantages, including high excitation energy and additional excitation freedoms. Relevant characterization technologies, including EELS, CL, PINEM, and PEEM, provide insights into the intrinsic nature and time-resolved dynamics of plasmonic processes. Researchers use both optical and electron-induced methods to comprehensively manipulate the properties of SPs. Precise control of classical properties has been achieved, ranging from distribution, wavelength, propagation, phase, and polarization. In light of these findings, a variety of nanostructures have been proposed with the goal of realizing diverse applications.

SPs are distinctive types of electromagnetic modes that are primarily generated by the collective oscillation of electrons within metals^[1,2]. Notable phenomena include SPPs capable of propagating along the interface and LSPs localized on the surface of nanoparticles. Due to its extremely high field enhancement effect and environmental sensitivity, plasmonic devices have found wide-ranging applications in various fields, including biology, medicine, catalysis, information processing, and sensing. The pursuit of a more profound comprehension of the underlying mechanisms governing SPs has consequently given rise to higher expectations for the characterization techniques. However, classical optical methods face significant challenges in achieving near-field characterization at subwavelength scales. The characterization of the near-field distribution of LSPs necessitates instruments capable of overcoming the optical diffraction limit, while the momentum mismatch problem caused by SPPs also hinders their far-field optical detections. LRM technology, operating within the framework of classical optics, has effectively addressed the challenge posed by momentum mismatch in SPPs, establishing itself as an effective method for far-field plasmonic optical characterization^[148,149]. It provides optical properties of SPPs, including spatial distribution, momentum distribution, wavelength, and polarization, by measuring leakage from high-refractive-index media and deducing them in reverse. However, its resolution remains constrained by the optical diffraction limit. The advent of AFM technology has engendered novel approaches for near-field optical characterization^[163–165]. Specifically, the adoption of needle-tip scanning, as opposed to geometric optical imaging, has been instrumental in overcoming the limitations imposed by the optical diffraction limit, facilitating the emergence of SNOM technology. a-SNOM was first proposed, and it subsequently achieved a breakthrough in improving instrument resolution. This pioneering technique involves the utilization of a subwavelength optical aperture attached to the needle tip, which serves to detect or inject electromagnetic waves^[167]. It then employs the collection of radiation signals from diverse positions to ascertain near-field electromagnetic distribution information. However, due to the conflicting optimization of the optical throughput and the spatial resolution, the mainstream position of a-SNOM is gradually being replaced by s-SNOM. s-SNOM diverges from the a-SNOM approach by eliminating the needle-tip apertures, enabling high-resolution detection of near-field distributions through the detection of scattered signals induced by the needle tip. To enhance the SNR, the AFM tapping mode is typically employed to facilitate a locked-in amplification measurement^[56]. The incorporation of pump-probe detections has been instrumental in the development of time-resolved s-SNOM technology^[161]. Recent studies have shown that plasmon nanofocusing tips can further effectively improve their spatial resolution, light transmission efficiency, and SNR^[57–61]. SNOM has successfully achieved spatial imaging at the nanoscale and temporal dynamic characterization at the femtosecond scale, greatly promoting the development of plasmonics.

Electron microscopy has provided another technological route for characterizing SPs. The ultrashort de Broglie wavelength of high-energy electrons endows electron microscopes with the capability to overcome the optical diffraction limit, facilitating the acquisition of high-resolution images. In conjunction with certain optical detection methodologies, electron microscopy techniques have been developed for the purpose of characterizing SPs. Electrons in high-speed motion approach a sample, interacting and exchanging energy and momentum. In the experimental setting, researchers can characterize the products of the system, and common detection methods include measuring the emitted electrons and radiated photons. The detection of electron energy loss has led to the development of EELS technology, which indirectly detects near-field electromagnetic distribution through the energy and momentum lost by electrons in their interaction with the electromagnetic field on the sample surface. The measurement of electron-induced radiation has led to the development of CL nanoscopy. Theoretical research has demonstrated a strong correlation among the strengths of EELS, CL, and LDOS, suggesting their effectiveness in near-field characterization^[83,243,283]. CL nanoscopy serves not only as a quantitative measurement method for analyzing LDOS, but also as a direct characterization of CL signals themselves. The advent of ultrafast electron microscopy has facilitated the development of ultracold electron beams, endowing EELS and CL nanoscopes with femtosecond-scale temporal resolution capabilities^[258,271]. PINEM represents an emerging experimental technology. Quantum theory provides a framework for understanding the intrinsic correlation between EELS and CL nanoscopy^[306]. By integrating the pump-probe detection with EELS technology, PINEM facilitates the acquisition of video recordings of near-field dynamics with an exceptional spatiotemporal resolution that exceeds 1 fs^[312,392] and 1 nm^[71]. PEEM, in contrast, adopts a distinct approach to characterize the electromagnetic modes on the surface of the sample. This approach entails measuring the electron emission induced by strong light irradiation. This technology possesses capabilities for energy-momentum- and time-resolved measurements.

The development of near-field characterization techniques has enabled experimental verification and guidance for the design theory of SP-controlled devices. Consequently, substantial advancements have been made in the multidimensional in-plane manipulation of SPs. This manipulation principally relies on phase modulation, and its primary mechanisms comprise resonant phase^[419,426], geometric phase^[418], and propagation phase^[421]. The geometric phase, which is related to the chirality of the incident light and the rotation angle of the structure, is independent of the specific shape. The convenience and unique performance of this strategy make it a mainstream phase control method. The interference effects that result from phase manipulation can be used to control the propagation behavior of SPPs, including directional propagation, unidirectional transmission, the generation of vortex beams, and holographic imaging, among others. The integration of diverse phase modulation mechanisms has enabled the development of multifunctional devices. The research of wavelength- and polarization-multiplexing devices has led to a substantial enhancement in the degree of control freedom, exhibiting remarkable application value.

SPs generated by electron beam irradiation can also be effectively manipulated. In comparison with photoexcitation, electron beams offer augmented excitation degrees of freedom, encompassing excitation position and incident electron kinetic energy. This additional degree of freedom facilitates the realization of multifunctional devices. For instance, the excitation position can selectively excite different plasmon modes. Furthermore, electron beam incidence can be categorized into two distinct types: grazing incidence and normal incidence. Through rational structural design, effective control of wavelength, direction, polarization, and spatial distribution has been achieved in both scenarios. Electron-induced radiation can be directly detected using techniques as CL nanoscopy and EELS, guiding the design of related devices. In hybrid systems comprising metal structures and luminescent materials, the radiation properties of quantum emitters can be achieved by manipulating the plasmon mode of the metal structure through electron beam manipulation, as evidenced by exciton luminescence in 2D TMDC materials.

Plasmonic in-plane manipulations with optical methods have become a well-established field of study, with related devices demonstrating a gradual transition towards practical applications. Through interdisciplinary research with other disciplines, related technologies can be widely applied in fields such as integrated circuits^[31,32] and biosensing^[40,41]. The involvement of innovative physics, including nonlinear effects, special electromagnetic modes, phonon-polaritons, and quantum effects, results in attractive phenomena that retain significant research value. In contrast, plasmonic manipulations based on electron beams remain at the laboratory stage. It is challenging to envision significant-scale applications emerging in the near term due to the necessity of a vacuum environment and the radiation damage. From a theoretical standpoint, classical Maxwell’s theory offers a comprehensive explanation of the fundamental behavior of surface plasmons, and future research will prioritize exploring quantum effects. Electron nanoscopies have shown significant potential for enhancement, particularly in the realm of quantum effects detection. The present study demonstrates the successful characterization of quantum effects in the context of photon–electron interactions by PINEM. Despite advances accomplished in quantum plasmonics, substantial quantum properties remain unknown. It is expected that more advanced characterizations will be developed to directly illustrate the diverse quantum systems of SPs. As potential solutions, some valuable improvements of electron nanoscopies have been reported, including specialized electron beams and built-in electron-driven light sources. The researchers propose potentially more advanced experimental techniques that are expected to promote further development of plasmonics.

In conclusion, advanced characterization technologies promote the study of plasmonic manipulations and provide a solid foundation for the application of plasmonic devices. It is our aspiration that this review will not only provide a summary of recent advances on plasmonic characterizations but also serve as a catalyst for further discoveries.