Two-dimensional (2D) materials are a promising solution for next-generation electronic and optoelectronic devices. The most investigated 2D materials range from semimetallic graphene, semiconducting transition metal dichalcogenides (TMDCs) and black phosphorus, to insulating hexagonal boron nitride (h-BN). 2D materials show significant advantages for the construction of optoelectronic and photonic devices. First, their optical responses cover a wide electromagnetic spectrum from ultraviolet to terahertz frequencies¹. Second, 2D materials possess intriguing optical properties. For example, graphene can support intrinsic plasmons that are tunable and adjustable, which promises new applications for conventional plasmonics². TMDC monolayers show valley pseudospins owing to broken spatial inversion symmetry, offering a new degree of freedom for information storage and processing^3,4. Black phosphorus possesses strong intrinsic in-plane anisotropy, which is promising for the realization of novel thin-film infrared polarizers and polarization-related sensors⁵. Hyperbolic phonon polaritons in h-BN have aroused intense interest in near-field optical imaging, guiding, and focusing applications⁶. Third, layer-by-layer stacking is allowed for 2D materials, without the lattice mismatch problem⁷. Van der Waals stacking enables multi-functionality and innovative device engineering for spintronics and opto-spintronics⁸. Therefore, light–matter interactions in 2D materials have become the research focus in the past decades.

Localized surface plasmon resonance (LSPR) arises from the collective oscillations of free charge carriers in bounded geometries like metallic particles and voids at the sub-wavelength scale. LSPR has been well known for synthetically controllable plasmon resonance wavelengths, large scattering/absorption cross-sections, and high electromagnetic field enhancement. It has been exploited as a powerful tool for enhancing light–matter interactions⁹. In the past decade, plasmonic nanostructures have been integrated with 2D materials for a range of applications such as solar energy harvesting¹⁰, sensing¹¹, optics/optoelectronics¹², and quantum nanophotonics¹³. In addition to commonly used individual plasmonic nanoparticles, plasmonic hybrid structures consisting of closely spaced metal nanoparticles are highly desired for the plasmonic modulation of 2D materials. The nanogaps in such plasmonic hybrid structures, such as plasmonic dimers, give rise to coupled plasmon resonances and “hot-spots” with higher local field enhancement and smaller mode volumes¹⁴. 2D materials embedded into plasmonic dimers can feel the strongly confined optical fields, which results in improved photoresponse¹⁵. However, it is generally difficult to place a 2D material inside the nanogap of a plasmonic dimer. The surface adatoms, cracks, and strain, which are unintentionally introduced during the preparation process, will cause degradation to the optical properties of 2D materials. Nanoparticle-on-mirror (NPoM) structures, which are another type of plasmonic hybrid structures supporting strongly confined optical fields, can be an elegant solution for the integration with 2D materials. The NPoM structure is made of a plasmonic nanoparticle on a smooth metal film with a nanospacer in between. The NPoM structure can confine optical fields with surprisingly low loss and with volumes below 1 cubic nanometers¹⁶, which is challenging to achieve with single nanoparticles. 2D materials are well-suited for integration into NPoM structures, forming subnanometer gaps¹⁷. The excitons in the gap region are coupled with the strong plasmon resonance, which can be tailored across weak to strong coupling regimes¹⁸. The ultraconfined plasmonic field can also lead to nonlinear optical effects¹⁹. In addition, the atomic nature of 2D materials pushes conventional plasmonics into the quantum regime²⁰. The integration of plasmonic NPoM structures with 2D materials is of great interest to both fundamental research and promising applications.

In this review, we highlight the research advances in the coupling between 2D materials and NPoM structures. We will start with a brief introduction to the plasmon resonances in NPoM structures as well as the preparation of 2D material-embedded NPoM structures. The imperfect factors that affect the optical properties of the hybrid structures will be discussed. We will then introduce the light–matter interactions in the hybrid system, with emphasis on the plasmon–exciton coupling. The possible quantum tunneling occurring in the subnanometer gap will be discussed. The potential applications will also be presented. We will finally point out unanswered questions and provide an outlook for future studies.

NPoM structures are composed of plasmonic nanoparticles deposited on mirror substrates made of Au or Ag, with a thin dielectric gap layer in between. Such structures exhibit strong confinement of optical fields, resulting in localized optical responses at the nanogap. To understand the optical confinement in such plasmonic nanocavities intuitively, the finite-difference time-domain (FDTD) simulation results of a nanosphere-on-mirror (NSoM) structure are provided ({L-End} Fig. 1). In the simulation, a 100 nm Au nanosphere (NS) is positioned on a Au film, separated by a 1 nm dielectric spacer. Such a spacer thickness is suitable for monolayer 2D materials. The Au NS is taken as an ideal sphere, with point contact on the substrate. The system is illuminated by a plane wave with wavevector k either parallel ({L-End} Fig. 1(a)) or perpendicular ({L-End} Fig. 1(c) to the mirror plane, corresponding to the out-of-plane and in-plane excitation, respectively. The simulated scattering cross-sections show pronounced peaks around the wavelength range of 500–Fig. 1000 nm in both cases. The near-field maps in {L-End} Fig. 1(b) and {L-End} Fig. 1(d) have two notable features. First, the electromagnetic field is highly confined in the gap region. Second, the out-of-plane field enhancement is significantly larger than the in-plane field enhancement for both excitation conditions. In addition, the field enhancement is much larger under the out-of-plane excitation condition. We will further look into the plasmon modes that generate substantial field enhancement.

The plasmon resonances in an ideal NSoM structure are dominated by the radiative antenna modes (l_n)²¹. The dipolar antenna mode (l₁) is pronounced. As the gap distance is reduced, higher-order antenna modes (n > 1) are also efficiently excited. When a circular facet is introduced at the nanosphere base ({L-End} Fig. 2(a)), which is closer to the experimental situation, gap modes emerge. The resonance modes in the gap are derived from an infinite planar metal-insulator-metal (MIM) model²². The finite gap nature is considered by use of a two-dimensional Fabry–Pérot resonator model, which describes the partial reflection of plasmons at the edge of the bottom facet of the NS. For a facet width of w, the resonance wavelengths are given analytically by^{22,2Fig. 3}

1λis≃λpwϵgdαi+ε∞,

where α_i are the zeros of the Bessel functions of the first kind; ε_∞ is the dielectric background; λ_p is the plasma wavelength of gold; d is the gap distance; and ε_g is the dielectric constant of the gap layer. The above equation defines the gap modes s_mn, where the indices m and n indicate the number of nodes in the azimuthal and radial directions. These gap modes are dependent on the facet width, spanning a range from the visible to near-infrared region ({L-End} Fig. 2(b)). The gap modes can be categorized as weakly radiative (solid lines in {L-End} Fig. 2(b)) or nonradiative (dashed lines in {L-End} Fig. 2(b)), showing distinct mode symmetries and charge distributions²¹. Their interactions with the antenna modes are therefore completely different. For example, the s_1n modes (azimuthal number m = 1) exhibit a transverse dipolar nature and swap the surface charge sign at each end of the gap ({L-End} Fig. 2(c)). These modes do not couple to the antenna modes (l_n), which are radially symmetric. In contrast, the s_0n modes (azimuthal number m = 0) have the same symmetry as the antenna modes ({L-End} Fig. 2(c)). They interact and form radiative hybrid modes (j_n). For instance, the l₁ mode meets spectrally with the s₀₂ mode, giving rise to the mode anticrossing feature with two hybridized modes j₁ and j₂ ({L-End} Fig. 2(d)). The antenna modes depend on the nanoparticle size, facet width, and gap distance. The gap modes, on the other hand, are insensitive to the nanoparticle size and are mainly determined by the facet width and gap layer thickness. Both the antenna and gap modes redshift as the gap distance is reduced ({L-End} Fig. 2(e)). These modes are usually accompanied with strong field enhancement in the gap, leading to diverse promising plasmon-enabled applications.

Gold and silver are commonly chosen as plasmonic materials for their tunable properties across a wide range of wavelengths, making them a top priority in plasmonics research. Wet-chemical synthesis techniques enable the production of Au and Ag nanoparticles with diverse shapes and sizes, which exhibit plasmon modes ranging from the ultraviolet to near-infrared region. Plasmonic Au NSs ({L-End} Fig. 3(a)), with high geometrical symmetry, have been extensively used to construct NSoM structures as model demonstrations^24,25. The resonance wavelengths of Au NSs, ranging from 520 nm to 800 nm²⁶, can be varied by adjusting the particle diameter. A point contact is formed between the NS and the mirror in the NSoM structure. NPoM structures composed of nanorods ({L-End} Fig. 3(b)) and nanocubes (NCs) ({L-End} Fig. 3(c)) have also been developed^27,28. The nanorod-on-mirror (NRoM) and nanocube-on-mirror (NCoM) structures show a line contact and a plane contact, respectively. Compared to the NSoM, the NRoM and NCoM structures have a larger field enhancement and a smaller mode volume owing to the lower symmetries and sharper features of the nanoparticles²⁹. Additionally, the {100} facets with steep inter-facet angles in the NCoM increase stability by avoiding atomic reconstruction under light illumination, while multiple atomic steps and shallow facet curvature on spherical nanoparticles can be problematic^{Fig. 30}. Recent research efforts have also been devoted to interesting nanoplate-on-mirror ({L-End} Fig. 3(d)) and nanodisk-on-mirror ({L-End} Fig. 3(e)) structures. The anapole states and plasmon-induced toroidal resonances observed in nanoplate-on-mirror structures facilitate the development of plasmonic anapole-based systems^{Fig. 31}.

The preparation of individual NPoM structures commonly involves drop-casting, where an aqueous solution containing target nanoparticles is deposited onto substrates with a spacer layer, followed by nitrogen-blow drying for good dispersion. Spin-coating can be an alternative method for depositing nanoparticles evenly. The removal of ligands surrounding nanoparticles is also very important. Ethanol rinsing and thermal annealing are commonly used for this purpose to improve contact. The spacer layer made of 2D materials, such as single-layer graphene (SLG), can push the gap distance to a single-atom limit ({L-End} Fig. 3(f) and {L-End} Fig. 3(g))^{Fig. 17}. Such structures are highly stable in optical measurements ({L-End} Fig. 3(h)). 2D TMDCs are also widely adopted as the spacer layer and offer interesting optical properties and enhanced quantum effects, which will be discussed below.

The optical properties of a NPoM structure are highly sensitive to various factors, including the size and shape of the plasmonic nanoparticle, the properties and thickness of the gap layer, and the quality of the mirror substrate. For ideal NPoM structures, plasmonic nanoparticles with regular shapes, uniform sizes, smooth surfaces, and minimal residual ligands are preferred. Random facets of Au NSs have been reported to cause unexpected tilting of the nanoparticles when placed on mirrors, leading to nonuniform dark-field scattering spectra²⁵. The surface roughness of the mirror can also affect the resonance wavelength and the consistency of the optical response across different measurements. Gold and silver films fabricated through electron-beam evaporation exhibit low roughness, while inverted Au films produced through a template-exfoliation process exhibit even lower roughness³². Ultrasmooth Au mirror films result in narrower linewidths for the scattering peaks from NPoM structures compared to rough Au mirror films, particularly when the Au mirror is thin²⁷, as shown in {L-End} Fig. 3(i) and {L-End} Fig. 3(j). Gold and silver microplates obtained from wet-chemical synthesis can provide ultrasmooth and single-crystalline mirror substrates, which are highly desired for low-loss plasmonics^{Fig. 3Fig. 3}.

NPoM structures with strong field enhancement and ultrasmall mode volumes offer a good platform for the study of light–matter interactions in 2D materials. Among them, plasmon–exciton coupling in TMDC-monolayer-sandwiched NPoM structures has aroused intense interest. TMDC monolayers exhibit pronounced exciton emissions with large binding energies (up to 0.9 eV³⁴) and exciton transition dipole moments, which favor the interactions of the 2D excitons with the electromagnetic fields in plasmonic nanostructures. The interplay between plasmons and 2D excitons can be divided into two coupling regimes, strong and weak coupling. A coupled harmonic oscillator model ({L-End} Fig. 4(a)) is usually employed to describe the strong coupling effect^{Fig. 35,Fig. 36}. In this model, the coherent states are assumed to obey the oscillator-like motion

2iP˙=(ωx−iγ)P+gE,

3iE˙=(ωc−iκ)E+gP,

where P and E represent the excitonic contribution to the dielectric polarization and the electric field of the plasmonic resonance, respectively; P˙ and Ė are their time derivatives; ω_x and ω_c are the angular frequencies of the exciton transition and plasmonic resonance, respectively; the subscripts ‘x’ and ‘c’ refer to the exciton and cavity; γ and κ are the dissipations of the uncoupled exciton and plasmon mode; and g is the coupling strength. The general solution of the above equations gives two eigen-frequencies (ω_±) of the hybrid system

4ω±=ω0−i(γ+κ)/2±ΩR/2,

5ΩR=[4g2−(γ−κ)2]1/2,

where Ω_R denotes the Rabi frequency; ω₀ represents the resonant coupling of the exciton and plasmon, that is, ω_x = ω_c ≡ ω₀. Equations (4) and (5) provide the criteria for distinguishing between the two coupling regimes. In a strong coupling regime, the Rabi frequency should be real and there is g > |γ – κ|/2. Otherwise, the hybrid system is in a weak coupling regime. It should be noted that the damping of polariton modes (γ + κ)/2 can be comparable to Ω_R in actual cases, which makes the identification of Rabi splitting difficult. Another criterion, Ω_R > (γ + κ)/2, is also widely used to define the strong coupling regime^18,37. Such definition is established in a pragmatic way, that is, the strong coupling occurs whenever the Rabi splitting is experimentally observable. The coupling process can be understood as the competition between the energy exchange and the system loss. When the energy exchange rate between the plasmon resonance and the exciton transition overcomes the system dissipations, strong coupling occurs, which is featured by an emerging half-light–half-matter hybridized state. The detuning of the plasmonic cavity and exciton transition shows a distinct anticrossing behavior in the spectral domain, with two prominent branches associated with the high- and low-energy polaritons. In contrast, the weak coupling only affects the radiative decay rate of the excitons, which is examined through the Purcell enhancement.

The strength of the coupling is determined by the exciton oscillator strength and the mode volume³⁷. Specifically, the coupling strength shows an inverse square dependence on the mode volume: g∝1/V^1/2. In this regard, it is advantageous to exploit NPoM structures with ultrasmall mode volumes for achieving strong coupling. The thickness of the dielectric spacer can be adjusted to tailor the interaction between the nanocavity and emitters. For example, the interaction between the excitons and the plasmon resonance is pushed to the strong coupling regime by simply shrinking down the thickness of the spacer layer in a MoS₂-sandwiched NCoM structure ({L-End} Fig. 4(b–d))^{Fig. 18}. In this structure, a Ag NC is assembled onto a piece of MoS₂ monolayer, which is separated by a Al₂O_{Fig. 3} layer on the Au mirror ({L-End} Fig. 4(b)). When the thickness of the Al₂O_{Fig. 3} layer is reduced to 5 nm, two split peaks in the PL spectra are observed at room temperature, instead of a single A exciton peak from the MoS₂ monolayer on the gold film. Such energy splitting can also be found in the scattering spectra, giving rise to a larger Rabi splitting of 190 meV. The smaller splitting energy in PL is attributed to the faster polariton relaxation rate than the emission rate and the difference in splitting between absorption and scattering. When the thickness of the Al₂O_{Fig. 3} layer is further decreased, the mode volume becomes even smaller, which is, however, accompanied by a lower field enhancement ({L-End} Fig. 4(c) and {L-End} Fig. 4(d)). The suppressed field enhancement probably arises from the dramatic increase in absorption from the metal.

In contrast to NCoM, NSoM is found to be less effective for the strong coupling with the A excitons in TMDC monolayers^18,38. For a NSoM structure sandwiched with MoS₂ or WSe₂ monolayer ({L-End} Fig. 5(a)), the plasmon tuning always gives weak coupling with PL enhancement^{Fig. 38}. However, Rabi splitting exceeding 1Fig. 35 meV can be observed in multilayer structures ({L-End} Fig. 5(b) and {L-End} Fig. 5(c)). The above results indicate the importance of alignment between the polarization of the dipole movement and the plasmonic mode in the gap. Other important factors include the field enhancement and mode volume. As mentioned above, nanoparticles with sharp corners induce larger field enhancement and smaller mode volumes, compared to those with spherical shapes. The advantages of nanoparticles with sharp features for strong coupling are also revealed in Au nanoprism-on-mirror structures ({L-End} Fig. 5(d))^{Fig. 39}. A large Rabi splitting up to 16Fig. 3 meV in the dark-field scattering spectra and splitting features in the PL spectra are observed ({L-End} Fig. 5(e)). The estimated effective exciton number contributing to the coupling reaches the single-digit level (N < 10), which is close to single-exciton-based strong coupling. We would also make comparison on strong coupling between NPoM structures and individual plasmonic nanoparticles⁴⁰⁻⁴². Rabi splitting is only observed in the scattering spectra based on most studies using individual plasmonic nanoparticles for achieving strong coupling. In contrast, NPoM structures with ultrasmall mode volumes enable large coupling strengths to overcome the system loss. Therefore, not only scattering splitting but also PL splitting can be observed at room temperature.

In addition to strong coupling, weak coupling between plasmonic nanocavities and 2D excitons has also been widely studied. The large field enhancement in the NPoM significantly facilitates the excitation and radiative decay of 2D excitons⁴³⁻⁴⁶. This process can be approximately described using the formula <EF> = |E/E₀|² × (QY/QY₀), where <EF> is the PL enhancement factor; |E/E₀|² is the electric field intensity enhancement at the excitation wavelength; and QY/QY₀ is the change of the emission quantum yield, indicating the modulation of the spontaneous decay rate in the system⁹. When the quantum yield of the nanoemitter is low, the above formula can be simplified to <EF> = |E/E₀|² × γ_PF, where γ_PF is the Purcell factor regarding the radiative decay rate of a nanoscale emitter adjacent to a nanoantenna⁴⁷. A NCoM structure composed of a Ag NC on a Au mirror has been developed to provide both excitation and emission enhancement for the A excitons in MoS₂ monolayer ({L-End} Fig. 6(a))^{4Fig. 3}. In this structure, the MoS₂ monolayer is inserted between a 5 nm HfO₂ layer and a 1 nm polymer adhesion layer. The Ag NC is encapsulated with a Fig. 3 nm polyvinyl pyrrolidone (PVP) coating layer. The NCoM structure shows two tunable resonance modes, i.e., the dipolar mode located at 660 nm and the quadrupolar mode at 420 nm. These two modes are well separated in the spectral domain with good spatial overlap, enabling the simultaneous enhancement of the absorption and the emission quantum yield. As a result, a 2000-fold enhancement in the PL intensity of the MoS₂ monolayer has been obtained. Ag nanowires with higher-order plasmon resonances have also been demonstrated to achieve the goal of the resonant enhancement of both absorption and PL quantum yield at the same location⁴⁶. Moreover, polarization control of the exciton emissions can also be realized in such anisotropic systems⁴⁵.

In recent years, plasmonic nanostructures with chiral morphologies have been designed to address the valley pseudospin in 2D TMDCs at room temperature⁴⁸⁻⁵². Chiral plasmonic nanostructures can generate chiral near-field with greater chiral asymmetry than that of circularly polarized light. TMDC monolayers inserted into the gap region experience both plasmonic field enhancement and optical chirality enhancement^48,49. Interestingly, the conventional NCoM structure has also been found to possess local optical chirality asymmetry when the mirror symmetry is broken by slightly tilting the NC ({L-End} Fig. 6(b))^{Fig. 50}. The left-handed (σ⁻) and right-handed (σ⁺) components of the PL emissions are found to be asymmetric ({L-End} Fig. 6(c)), showing a relatively high room-temperature degree of valley polarization (48.7%). The valley-polarized emissions can be attributed to the chiral Purcell effect resulting from the uneven spatial distribution and lifted degeneracy of the σ⁻ and σ⁺ photonic local density of states. The decay rate of the excitons with different valley indexes can therefore be separately modulated, leading to remarkably enhanced valley polarization ({L-End} Fig. 6(d)).

Plasmonic nanocavities with large field enhancement also offer a feasible tool for probing diverse excitonic complexes in addition to neutral A excitons, such as B excitons⁵³, dark excitons^54,55, localized excitons^13,56, and interlayer excitons⁵⁷. These types of 2D excitons are usually inaccessible under normal conditions owing to their low quantum yields. For instance, the NCoM structure is employed to create dominant B exciton emissions in MoS₂ monolayer ({L-End} Fig. 7(a))^{Fig. 5Fig. 3}. The B excitons in MoS₂ monolayer are associated with the optical transitions from the lower valence band to the bottom of the conduction band. A 6100-fold enhancement of the B exciton emissions is observed when spectral resonant matching is realized. More importantly, the gap plasmon modes of NPoM structures provide significant out-of-plane field enhancement, which is believed to be critical for the detection of 2D excitons with out-of-plane dipole moments. The probing of spin-forbidden dark excitons is a typical example of this aspect. The spin-forbidden dark excitons correspond to the optical transition from the higher-lying valence band to the split conduction band with an anti-parallel spin configuration. This type of dark exciton shows orthogonal transition dipole orientation with the A excitons and is generally hard to excite and detect in conventional optical measurements. The radiative decay rate of the dark excitons in WSe₂ monolayer has recently been shown to be enhanced by nearly 6 orders of magnitude through the Purcell effect in a NSoM structure ({L-End} Fig. 7(b))^{Fig. 55}. The PL intensity of the dark excitons clearly outperforms that of the spin-allowed bright excitons ({L-End} Fig. 7(c)), when the background emissions are removed through the etching of the uncoupled WSe₂ region. The above results indicate that the plasmonic field in the NPoM structure can extract electronic transitions before the spin flip under thermal excitation. This simple device structure offers a facile approach for manipulating the dark states by avoiding the use of extreme experimental conditions, such as cryogenic temperatures^{Fig. 58}, ultrahigh in-plane magnetic field^{Fig. 59}, and tip-enhanced techniques^{Fig. 54}. Interlayer excitons from layered TMDC heterostructures are another type of 2D exciton with out-of-plane dipole moments. The integration of the interlayer excitons in WS₂/MoS₂ stacking into plasmonic nanocavities based on Ag NCs gives rise to an order of magnitude PL enhancement at room temperature and a 5-time enhancement at cryogenic temperatures ({L-End} Fig. 7(d) and {L-End} Fig. 7(e))^{Fig. 57}. The plasmonic techniques also promise to manifest the fine structures of moiré excitons, which requires further investigation.

2D TMDCs are also an excellent platform for the observation of phonon-assisted upconversion owing to the enhanced optical transition strength and interaction of the 2D excitons with phonons. The anti-Stokes process is activated through the creation of resonant conditions, that is, the incident or/and emitted photon energy matches the resonance level of the material system. A series of excitonic states, including spin-triplet trions, spin-singlet trions, and intravalley (spin-forbidden) dark excitons, can be upconverted to the bright excitons under resonant conditions⁶⁰⁻⁶². Plasmonic nanocavities are integrated to change the local optical environment. For example, the broad plasmon resonance from a Au NCoM structure is coupled to the excitation laser light at 1.52 eV and the emissions at 1.65 eV in WSe₂ monolayer ({L-End} Fig. 8(a–c))⁶¹. An upconverted emission amplification exceeding Fig. 1000 times and a reduction of 2–Fig. 3 orders of magnitude in the saturated excitation power are achieved. The temperature-dependent spectra ({L-End} Fig. 8(d)) show evidence for the involvement of phonons in the upconversion process. In one most recent work, phonon-assisted upconversion in WSe₂ monolayer is obtained through the resonant excitation of the intravalley dark excitons at room temperature⁶². It is also proposed that the precise control of the nanoparticle geometry is important for reducing the variation in the anti-Stokes PL enhancement. The decahedral nanoparticle shape is found to provide a 10-fold narrower distribution of the enhancement factor for the anti-Stokes PL, compared to NSoM structures. The upconverted PL can be further modulated reversibly by electrochemical gating.

Raman scattering arises from the inelastic scattering of light by phonons, which are quantized states of vibrations in solids and molecules. Raman spectroscopy has become a fast and nondestructive method to characterize the physical and optical properties of 2D materials^63,64. Rich information about the lattice vibrations of 2D materials can be revealed through the analysis of the line shape, linewidth, peak position, and intensity, which helps to uncover microscopic processes such as electron–phonon coupling⁶⁵, strain dependence for elastic constants⁶⁶, and quantum interference in Raman scattering processes⁶⁷. Plasmonic nanostructures are highly active in enhancing Raman scattering on 2D materials based on two reasons. First, the efficiency of Raman scattering is very low, that is, typically one in every ~10⁵–10⁷ photons produces a Raman-scattered photon⁶³. Second, the light–matter interaction length in 2D materials is at the subnanometer scale, leading to low absorption of light compared to the bulk materials. Here we focus on SERS in (2D material)-sandwiched NPoM structures. The measured SERS enhancement factor is well known to follow the fourth power of the local electric field enhancement, which is also called the E⁴ law⁶⁸. The plasmonic cavity is thus a magnifier to monitor any subtle change in the phonon response of the 2D system. In turn, the vibrational modes of 2D materials can also be a probe to sense the local field enhancement in the nanocavity. For example, the quantitative SERS intensities of graphene and TMDC monolayer have been used to test the ultimate limit of plasmonic enhancement within a subnanometer gap^17,24,69. For a MoS₂-NPoM structure, the intrinsic out-of-plane (A_1g) and in-plane (E2g1) lattice vibrations of the MoS₂ monolayer are softened, with two new peaks (labelled as A_1g′ and E2g1′) showing up owing to the doping and strain effects from the metallic contact ({L-End} Fig. 9(a–c))²⁴. Both the out-of-plane components (A_1g + A_1g′) and in-plane components (E2g1 + E2g1′) are largely enhanced in the MoS₂-NPoM structure. The out-of-plane and in-plane vibrational modes are aligned with the perpendicular and horizontal local fields, respectively. The SERS enhancement of these two modes (EF_xy and EF_z) can therefore be used to quantitatively probe the local field enhancement in the two orthogonal orientations. Based on the spectroscopic analysis, the out-of-plane field enhancement (EF_z) in the gap region is much larger than the in-plane field enhancement (EF_xy). The value of EF_z is found to rapidly increase as the gap becomes narrower, reaching up to 5.1 × 10⁸ for a MoS₂ monolayer structure. The in-plane near-field enhancement when the gap distance is below 1 nm is, however, lower than expected. Electron tunneling is predicted in such a subnanometer gap. A recent work shows that the h-BN monolayer is a good dielectric spacer to block electron tunneling, extending the ultimate plasmonic enhancement limit in the classical framework in a single-atom-layer gap⁷⁰. TMDC monolayer has also been embedded at different positions in the gap region to show the field distribution in the nanocavity^71,72. For example, WS₂ monolayer is stacked in different orders with four layers of MoS₂ using the PMMA layer-by-layer transfer method and placed inside a NSoM structure⁷¹. The quantitative SERS intensity of the WS₂ monolayer located at different layer positions is used to measure the plasmonic field distribution. The plasmonic field in the nanocavity is found to be heterogeneous, with a maximal value close to the nanoparticle and a minimal value in the middle of the gap. The above investigations have demonstrated 2D materials as a good probe. They can be easily integrated into the “hotspot” and support vibrational modes with certain orientations.

Nonlinear effects can also be observed in the SERS measurements of the MoS₂-NPoM structure¹⁹. An anomalous superlinear power dependence of the SERS intensity of the A_1g mode appears when the excitation wavelength is blue-detuned to the plasmonic resonance ({L-End} Fig. 9(d–f)). This superlinear behavior is attributed to the dynamic backaction modification of the phonon population, that is, the pumping rate of phonons modulated by the plasmonic field is larger than the damping rate. Another phenomenon related to the nonlinear effect is the significantly large Raman scattering of the second-order Raman-inactive phonon mode 2LA (LA: longitudinal acoustic), which is round-trip scattering by two LA phonons ({L-End} Fig. 9(b)). The 2LA mode is extremely weak compared with the first-order Raman-active phonon modes A_1g and E2g1 if the plasmonic enhancement is absent. The 2LA mode also shows superlinear power-dependent behavior and surpasses the A_1g mode at large excitation intensities.

As the plasmonic nanocavity comes to the subnanometer range, the plasmon resonance enters the quantum regime. The quantum mechanical effects of plasmon resonance arise from the nonlocal screening of electrons and electron tunneling through the gap when the gap distance continues to shrink⁷³. Classical electromagnetic treatments fail at this regime and the plasmonic properties of the cavity can be completely modified by electron tunneling. Quantum tunneling induces blueshifts in the plasmon resonance, reduces the local field enhancement, and generates charge-transfer plasmons^73,74. The tunneling plasmonics not only pushes fundamental research to a new frontier but also brings potential applications in the ultimate miniaturization of photonic components for quantum optics⁷⁵, nanoreactors⁷⁶, and biosensing with single-molecule sensitivity⁷⁷. However, it has been challenging to fabricate quantum plasmonic nanocavities with subnanometer gaps in a reliable and repeatable manner with state-of-the-art technologies. 2D materials with atomic thicknesses can provide solutions to the challenges of fabricating quantum plasmonic devices and controlling plasmons by electronic means. Electron tunneling mediated by MoS₂ monolayer has been predicted in a work where the in-plane near-field enhancement limit is probed by SERS²⁴. Simulations based on classical electromagnetics have been found to be able to precisely estimate the plasmonic enhancement until the gap distance is narrowed to ~ 1 nm ({L-End} Fig. 10(a)). The measured field enhancement is ~ Fig. 38.4% lower than the calculated one when the gap distance is below 1 nm. After the reliability of the experimental design is taken into account, the quenching of the field enhancement is believed to arise from electron tunneling across the MoS₂ monolayer. In addition to MoS₂ monolayer, graphene has also been reported to change the plasmonic response of the nanocavity^20,78. Spectral doublets from the coupled dimer modes are observed with single-layer graphene as a spacer and disappear for increasing numbers of layers ({L-End} Fig. 10(b) and {L-End} Fig. 10(c)). The spectral features are believed to correlate with the charger-transfer-sensitive gap plasmons. However, there have been only a few experimental demonstrations of using 2D materials as the conductive gap, especially compared with their molecular counterparts. The problems probably lie in the large out-of-plane resistance of 2D materials and the unavoidable van der Waals gap, which can form a tunnel barrier to block electrons⁷⁹. For example, an air gap is observed between the graphene layer and the mirror possibly owing to the gold film roughness, relaxed van der Waals forces, or the wrinkling of the graphene monolayer^{Fig. 17}. The exploration of a process or a structure to form a good electrical contact between 2D materials and metallic nanostructures is therefore very important.

Hybrid nanostructures composed of 2D materials and plasmonic nanocavities have been demonstrated to be an excellent platform for the fundamental investigation of light–matter interactions, as discussed above. Their intriguing characteristics can be used for promising applications, especially in nanoscale luminescent sources. Single-photon sources based on 2D TMDCs have attracted great interest for future quantum optical applications^13,80−82. Plasmonic nanocavities have been employed to enhance single-photon emissions in TMDC monolayers. The effect of a plasmonic nanocavity on TMDC single-photon emitters can be twofold: strain localization and electromagnetic enhancement. For example, a lithographically defined Au NC array is fabricated and pressed onto a WSe₂ monolayer, with the sharp corners of the Au NCs to both provide large electric field enhancement and apply local strain on 2D TMDCs ({L-End} Fig. 11(a) and {L-End} Fig. 11(b))^{Fig. 1Fig. 3}. The Purcell factors of up to 551, with an average of 181, are obtained for a Fig. 3 × 4 array of quantum emitters. As a result, the brightness of the single-photon emissions is largely enhanced, without affecting the single-photon purity. A representative zero delay time value of g⁽²⁾(0) = 0.15 ± 0.0Fig. 3 ({L-End} Fig. 11(c)) and an average g⁽²⁾(0) well below 0.5 from 15 tested structures ({L-End} Fig. 11(d)) confirm pronounced single-photon antibunching. The single-photon emission rates of up to 42 MHz and a narrow exciton linewidth as low as 55 μeV are further demonstrated at the cryogenic temperature of Fig. 3.8 K. A few recent works have further shown the possibility for room-temperature observation of strain-localized emissions through the optimization of plasmonic enhancement^{8Fig. 3,84}. However, the single-photon emissions from 2D TMDCs at room temperature have not yet been demonstrated. Negatively charged boron vacancy (V_B⁻) defects in h-BN with optically addressable spin states are also good candidates for quantum optical applications such as ultrathin quantum sensors^85,86. Integrating V_B⁻ defects in h-BN with plasmonic NPoM structures raises the overall intensity enhancement up to 250 times, along with preserved optically detected magnetic resonance contrast⁸⁶. NPoM structures have also found promising applications in electroluminescent devices. A van der Waals quantum tunneling device has been combined with a Ag NC-on-Au mirror structure⁸⁷. The photon generation is created from the inelastic electron tunneling through the van der Waals heterostructure. It exhibits a broad emission peak. The plasmonic nanocavity enables a resonant enhancement of the photon emission rate in a narrow frequency band by four orders of magnitude. The NPoM nanocavities in these devices serve as a medium to amplify the weak signals from the nano-emitters through the weak coupling effect, which benefits in avoiding strict application requirements. The strong coupling between plasmonic nanocavities and 2D materials is also expected in applications such as tresholdless lasing⁸⁸, which requires further study.

The (2D material)-sandwiched NPoM structures can also find potential applications in sensing. We have mainly discussed the intensified SERS signals of 2D materials above. The SERS signals are amplified by the electromagnetic enhancement from the plasmonic nanocavities in these structures. The 2D materials themselves can also serve as SERS substrates to provide chemical enhancement for molecular sensing⁸⁹. The principle is based on the charge transfer effect, which is the formation of charge-transfer complexes between adsorbed molecules and substrates. The utilization of (2D material)-sandwiched NPoM structures is believed to be able to combine charge transfer enhancement and electromagnetic enhancement for molecular sensing. In addition, NPoM structures have been employed to achieve chiral sensing of single molecules in a recent work⁷⁷. Interestingly, tunneling electrons are found to provide an additional enhancement to the chiral sensitivity. The (2D material)-sandwiched NPoM structures can also function as a good platform for chiral sensing. In the (2D material)-sandwiched NPoM structures, active control can possibly be realized by switching on and off the tunneling channel through the doping of the intermediate 2D material⁹⁰. Hot electrons that are generated from the plasmonic nanoparticle and penetrate through the 2D material can also be used to trigger photocatalytic reactions¹⁰. The utilization of a graphene monolayer as the gap material has been found to increase the hot-electron transfer efficiency by over three times.

NPoM structures can support much stronger optical confinement compared to individual plasmonic nanoparticles, making them a good platform for the investigation of light–matter interactions in 2D materials. In this review, we have overviewed the important research progress in the integration of NPoM structures with 2D materials. First, we have briefly introduced the cavity modes and the important factors that affect the optical properties of NPoM structures. Second, we have examined in detail plasmon-modulated light–matter interactions in 2D materials. The important structural parameters to modulate 2D excitons from the strong coupling to weak coupling regimes have been elucidated. The SERS measurements as well as the quantitative study of the near-field enhancement in the nanogap have been presented. Third, we have discussed the possible quantum effects of the extreme gap. Fourth, we have shown the potential applications, especially for nanoscale light sources. We can see that the plasmonic strategy helps to reveal the rich microscopic processes in the 2D system. 2D materials, in turn, provide new functionalities for plasmonic control. For example, graphene has been employed as a mirror substrate to achieve plasmon confinement down to the ultimate limit of the length scale of one atom in the mid-infrared region⁹¹. The use of graphene is attractive for suppressing optical loss, which is large in metals due to Landau damping.

We finally would like to point out unanswered questions and future directions for the combination of NPoM structures and 2D materials in four aspects. (1) Strong coupling. Mode splitting in the PL spectra has been observed in TMDC monolayers embedded into NPoM structures, indicating the emergence of ultrastrong coupling. The ultrastrong coupling can be a good probe to reveal coherent processes in 2D TMDCs. For example, the formed exciton–polaritons in a microcavity have been demonstrated to maintain the valley properties of their excitonic component⁹²⁻⁹⁴. Moreover, valley-coherent exciton–polaritons can be addressed^95,96, with the degree of the valley coherence of the quasiparticles exceeding 90% upon excitation through a two-photon absorption process⁹⁶. A coherent phonon scattering process in MoS₂ monolayer has also been demonstrated under the strong coupling regime⁹⁷. The strong coupling between the phonon modes and the plasmonic cavity modes enables a drastically enhanced valley polarization (2.5 times) in the Raman scattering spectra. These findings provide a good reference for future exploration of more fundamental polaritonic processes in the 2D system, which will certainly boost an in-depth understanding of the valley-polarized properties of the exciton–polaritons modulated by the strong plasmonic field. In addition, more effort should be made to clarify the critical structural parameters for achieving ultrastrong coupling. (2) Weak coupling. The large plasmonic field enhancement in NPoM structures has been successfully employed to detect various excitonic complexes. More efforts are encouraged to achieve a new dimension of control. For example, a tilted NCoM structure with chiroptical response has been coupled to rare-earth-doped nanoparticles to achieve up-converted PL chirality⁹⁸. Compared to tilted NCoM structures, (chiral NC)-on-mirror structures provide a more controllable way to modulate and amplify chiroptical signals⁹⁹. Future studies are encouraged to integrate versatile 2D materials with the new type of chiral nanocavities for plasmonic modulation of valley excitons and moiré excitons. The cavity-dependent circularly polarized single-photon emissions in 2D TMDCs¹⁰⁰ is also very interesting and more research efforts are desired. (3) Quantum tunneling. There has still been a lack of experimental evidence for confirming the electron tunneling in the ultimate gap formed by single-layer 2D materials. As discussed above, the large out-of-plane resistance and the air gap are the possible reasons for hindering electron tunneling. Further investigations on the quantum tunneling enabled by 2D materials and its active control are highly desired. (4) Exploration for more possibilities. Most of the current research efforts focus on the study of individual NPoM structures. Plasmonic superlattices composed of arrays of nanoparticles can provide lattice resonance modes for plasmonic modulation. For example, Au superlattices have been found to support multiple band-edge modes. The integration of Au superlattices with appropriate gain materials has enabled population inversion at plasmonic hotspots and multi-modal nanolasing¹⁰¹. Moreover, the development of robust and reproducible fabrication techniques for plasmonic nanostructures will be crucial for practical applications.

In all, we believe advances in this field will provide more insight into the plasmonic control of 2D materials and facilitate new nanophotonic applications.

This work was supported by the National Natural Science Foundation of China (62205183) and the Research Grants Council of Hong Kong (ANR/RGC, Ref. No. A-CUHK404/21).

The authors declare no competing financial interests.