Main

Metasurfaces, periodic two-dimensional (2D) arrays of optically resonant nanostructures, have transformed the field of photonics by enabling flat optical components with vast design flexibility and applications from metalenses to polarization optics1,2. Originally introduced in the context of dispersion-engineered metasurfaces for phase control3, the term has since come to refer to a broader class of nanostructured surfaces incorporating resonant building blocks4. Their unique functionalities primarily arise from the precise engineering of the amplitude and phase of light interacting with each individual resonant nanostructure, typically at the subwavelength scale in terms of feature size and periodicity5. Additionally, metasurfaces can strongly confine light by designing optical resonances with high quality (Q) factors6—defined as the ratio of the resonance frequency to its linewidth—and strongly enhanced electromagnetic near fields, making them crucial for applications in sensing7,8 and nonlinear photonics9,10. Optical cavities with high-Q-factor resonances have been at the forefront of research in cavity quantum electrodynamics since the early 1990s11. However, in optical metasurfaces, cavity quantum electrodynamics concepts remain underexplored.

In parallel, 2D materials have revolutionized optoelectronics, enabling the deterministic assembly of nanostructures down to single atomic layers12. More broadly, van der Waals (vdW) materials are a family of crystalline solids in which atoms are covalently bonded along the in-plane direction, whereas weak vdW forces hold the planes together in the out-of-plane direction. Unlike conventional materials, this unique structure allows for the mechanical exfoliation of crystal planes and the precise manipulation of individual layers. A key technological advancement of vdW materials is their ability to stack 2D layers, overcoming constraints related to dangling bonds and lattice mismatch. This allows for the deterministic assembly of vertical vdW heterostructures composed of two or more layered materials, chosen from the extensive range of vdW crystals, enabling unparalleled control over the composition of artificial structures and the design of customized optoelectronic properties. Hexagonal boron nitride (hBN), a wide-bandgap dielectric, has found widespread use as a transparent substrate and encapsulation material for the fabrication of high-quality vdW heterostructures13,14, whereas 2D semiconductors such as transition metal dichalcogenide (TMDC) monolayers exhibit exceptional capabilities for light–matter coupling, ranging from stable excitons at room temperature15 to valley physics and single-photon sources16. Now, hBN encapsulation is standard in vdW heterostructure fabrication to preserve and improve the optical and electronic qualities of the fabricated 2D semiconductor devices17. This process has driven substantial progress in condensed-matter physics, enabling discoveries on moiré effects18,19, topological states20 and 2D magnetism21—all in submicrometre-thick devices.

For enhancing light–matter interactions, single-layer metasurfaces based on graphene or hBN have proven effective in the mid- and far-infrared ranges22,23. However, they do not provide sufficient optical confinement for exciton–photon coupling at visible wavelengths. Therefore, 2D semiconductors are typically integrated with metasurfaces by placing them into direct contact24, which maintains them as separate systems primarily interacting via the evanescent fields surrounding the resonant nanostructures. The transfer of 2D semiconductors on top of prefabricated nanostructures has been extensively studied. However, it remains a complex process that can introduce strain, intrinsic damage and unwanted substrate interactions to the transferred layer. This not only affects fabrication reproducibility but also impacts the optical properties of the transferred material. Although plasmonic nanostructures can generate strong electromagnetic fields at their surfaces and produce large Rabi splittings25, ohmic losses and broad resonance linewidths limit an efficient light extraction. Moreover, widely used particle-on-mirror geometries create mostly out-of-plane electric fields in the gap, posing challenges when coupling to the in-plane excitons of many common TMDCs. In dielectric nanophotonics, instead, the electromagnetic field is mostly confined inside the resonator material, thereby limiting the potential for strong light–matter coupling in the case of surface transfer. As a result, integrating a vdW heterostructure within dielectric resonators enables larger coupling strength and efficient cavity-mode overlap.

Apart from their 2D form, bulk vdW materials (thickness 50–500 nm) have also garnered a strong interest in nanophotonics due to their high refractive indices26 and large optical anisotropies27. Nanostructuring vdW materials to engineer optical resonances, as in conventional semiconductor-on-insulator architectures, has emerged as a promising avenue in dielectric nanophotonics, offering a novel method for designing ultrathin photonic devices28,29. Beyond its encapsulation properties, hBN has also shown exceptional promise as a nanophotonic platform. Subdiffractional30 and high-Q31,32 cavities operating in mid-infrared ranges, photonic crystals in the visible range33, quantum emitters and surface phonon–polaritons34 have established hBN as a versatile photonic material. By leveraging quasi-bound states in the continuum (qBICs)35,36, hBN metasurfaces have been realized with high-Q resonances (Q > 200) spanning the visible range37. Such qBIC modes act as resonant optical cavities, strongly confining electromagnetic fields and efficiently enhancing light–matter interactions38,39, with the demonstration of the strong light–matter coupling regime with excitons in both monolayer and bulk TMDCs40,41. Integrating atomically thin vdW heterostructures with the photonic degrees of freedom provided by qBIC metasurfaces offers a promising pathway for developing ultrathin photonic devices. The robust light confinement enabled by qBIC metasurfaces can further address the limitations of traditional optical cavities, including the bulkiness and cost of Bragg mirrors42, as well as the losses and low Q factors of plasmonic particle-on-mirror cavities43.

In this work, we experimentally demonstrate the concept of vdW heterostructure metasurfaces (vdW-HM; Fig. 1a). By nanopatterning optical metasurface designs within stacked vdW materials, we show that the hBN encapsulating layers can function as dielectric resonators37, with qBIC resonances arising from the asymmetry of the unit-cell design6. In particular, this monolithic approach eliminates the need for bulky external cavity systems and enabling tunable high-Q resonances within the ultrathin vdW heterostructure. To demonstrate this concept, we fabricate a vdW-HM from a prototypical heterostructure composed of a tungsten disulfide (WS2) monolayer encapsulated in hBN. We observe strong light–matter coupling between the qBIC modes and 2D excitons, achieving Rabi splitting energies exceeding 30 meV at room temperature. We characterized the metasurface-coupled exciton–polaritons via reflectance and photoluminescence (PL) emission with angle-resolved spectroscopy. Strong nonlinearities under increasing excitation power are observed, leading to a saturation of the strong-coupling regime at fluences of <1 nJ cm–2, more than three orders of magnitude lower than those reported for TMDCs in standard optical cavities44. These findings highlight the potential of our platform for applications such as ultrafast switches, lasers and bosonic condensates45. Moreover, the rich physics and optical properties of 2D materials, such as spin-valley physics, single-photon sources and multibody exciton complexes, pave the way for advanced ultrathin photonics based on optical vdW metasurfaces.

Fig. 1: vdW-HMs.
figure 1

a, Illustration of a vdW-HM, composed of a WS2 semiconductor monolayer encapsulated between two thin hBN layers. The monolithic metasurface design is patterned into the heterostructure via top-down nanofabrication processes. The asymmetric unit-cell design, composed of two nanorods with different lengths, gives rise to symmetry-protected qBIC resonances, which strongly confine light and resonantly interact with the 2D WS2 excitons. Top right: schematic of the strong-coupling regime between exciton (EX) and qBIC (EqBIC) resonances, with the formation of the upper (UP) and lower (LP) exciton–polariton branches separated by Rabi energy ΩR. b, Strong coupling between the exciton in the WS2 semiconductor (left panel) and the photonic qBIC resonance (right panel) can be reached by maximizing the coupling strength g. Numerical finite-difference time-domain simulations of the electromagnetic-field enhancement (|E|/|E0|)2 for an hBN metasurface at the qBIC resonance show strong fields inside the resonators, ideally suited for boosting light–matter coupling. c, Numerical simulations of the transmission of an hBN metasurface (ΔL = 50 nm) demonstrate the effective spectral tuning of the qBIC resonance via a lateral scaling factor S applied to the unit cell. The PL spectrum of the room-temperature emission for a monolayer WS2 encapsulated in hBN is shown in black. d, Analytical calculation of the dispersion following a coupled harmonic oscillator model (Methods) and the formation of LP and UP branches in the strong-coupling regime, as a function of detuning between the cavity (qBIC) and emitter (excitons, X). e, Momentum dispersion of the qBIC–exciton system as a function of wavevector kx, parallel to the nanorod long axis. The solid line and dot–dashed line represent the exciton and qBIC modes, respectively. The dashed lines are the polariton branches. When the qBIC mode is positively detuned from the exciton, the negative angular dispersion makes the Rabi splitting appear at larger angles, where the two resonances cross.

Full size image

Results

Principles of light–matter coupling in qBIC metasurfaces

Enhancing light–matter interactions with 2D semiconductors in optical metasurfaces requires strongly localized in-plane electromagnetic fields, necessary for the efficient mode coupling with 2D excitons confined in the atomically thin layers (Fig. 1b). Dielectric metasurfaces supporting photonic bound states in the continuum (BICs)35,36 excel at controlling light–matter interactions at submicrometre thicknesses by engineering radiation losses. The introduction of symmetry breaking in the geometry of the periodically arranged unit cells induces symmetry-protected qBIC resonances with Fano-like line shapes. These can be designed to optimize both spectral position and resonance linewidth, and providing strongly confined in-plane electromagnetic fields38. The control over the radiative properties of the qBIC mode is provided by the asymmetry parameter α in the unit-cell design6. A true BIC is a symmetry-protected state with a theoretically infinite Q at α = 0. When breaking the symmetry (α > 0), a radiative coupling channel opens, allowing the far-field excitation of the qBIC resonance, which can reach extremely high Q factors above 105 (ref. 46). By modifying α, the linewidth of the qBIC resonances can be precisely tailored to match the system’s critical coupling condition40. Moreover, qBIC resonances can be spectrally shifted across broad spectral ranges by simply scaling the unit-cell size using a lateral scaling factor S. This is applied to all in-plane dimensions of the unit-cell design except for height, which is typically given by the thickness of the active material layer.

We numerically designed hBN-based heterostructure metasurfaces covering the emission spectrum of monolayer WS2 (Fig. 1c). In particular, it is possible to adjust the qBIC resonance relative to the exciton spectral position via the scaling factor S, in analogy to the cavity–emitter detuning parameter δ = EqBIC − EX, where EqBIC and EX are the qBIC mode and WS2 exciton energies, respectively. The qBIC–exciton system can be described analytically using a coupled oscillator model (Methods). In the strong light–matter coupling regime, the coupled exciton–photon system hybridizes, leading to the formation of exciton–polariton branches (Fig. 1d). These inherit the potential for strong interactions from their excitonic component but offer wide tunability given by their photonic part, with promising applications in low-power optoelectronics and all-optical computation47,48. As the qBIC resonance is tuned across the exciton energy, an anticrossing is observed between the two resonances. At zero detuning (δ = 0), we determine the vacuum Rabi splitting energy (ΩR), directly proportional to the coupling strength (g = ΩR/2). The coupling strength, in turn, is proportional to the WS2 transition dipole moment and the magnitude of the quantized electromagnetic field of the qBIC cavity. Figure 1e illustrates the angular dispersion of the qBIC–exciton system as a function of the in-plane wavevector kx (oriented along the long axis of the nanorods). Unlike common optical cavities, with positive wavevector dependence, the qBIC cavity mode exhibits a negative dispersion (Supplementary Fig. 1), which supports saddle-like points in momentum space, a favourable configuration for condensation49. At non-zero detuning, the Rabi splitting shifts to higher angles, aligning with the crossing point of the two resonances.

Strong-coupling regime in vdW-HMs

The fabrication of vdW metasurfaces begins with the assembly of the vdW heterostructure, involving the mechanical exfoliation and deterministic transfer of individual layers. Figure 2a shows an optical image of the fabricated stack in which a WS2 monolayer is encapsulated between two hBN layers with a height of approximately 60–65 nm. We then performed electron-beam lithography, hard-mask deposition and reactive ion etching to transfer the metasurface design into the vdW heterostructure (Methods and Supplementary Fig. 2 provide more fabrication details). Even though this thickness is only around half the effective wavelength in hBN (~300 nm), the lateral features of the resonant rods are as low as 100 nm, resulting in clearly subwavelength metasurface building blocks. More than 15 metasurface fields with varying geometrical parameters were fabricated within the initial vdW heterostructure area of approximately 120 µm × 70 µm (Fig. 2a). The vdW-HM design, with a total height of approximately 125 nm, is composed of asymmetric hBN/WS2/hBN rod pairs, where L0 is the base rod length and ΔL defines the asymmetry factor for the generation of the radiative qBIC channel (Fig. 2b). The chosen double-rod design provides a favourable geometry for realizing the extremely small nanostructure feature sizes (down to 75 nm) required for visible-range qBIC resonances, and ensuring the generation of a single resonance peak within the spectral region of interest (Fig. 1c). Symmetry breaking is achieved by altering the length of one of the two rods, characterized by the asymmetry parameter ΔL. This difference in length translates into a change in the longitudinal dipole resonance of each nanorod, and the vectorial sum leads to an effective dipole moment that can couple to light, resulting in a Fano-like qBIC resonance6,37,38,40. Compared with other BIC-driven unit-cell geometries incorporating curved elements such as ellipses and rings, the double-rod geometry enables improved fidelity in nanoscale fabrication. An important consideration in the metasurface design process is the interaction between the qBIC resonance and the diffraction orders (or Rayleigh anomalies) imposed by the 2D periodic arrangement of the resonant elements. Since spectral overlap between these two modes can cause a suppression of the qBIC resonance, the periodicity of the system has to be adjusted to place the Rayleigh anomalies at higher energies. We analyse the impact of hBN thickness on the optical response of the metasurfaces, requiring heights above 100 nm to overcome the limit imposed by the diffraction orders (Supplementary Fig. 7). Finally, the asymmetry factor value (ΔL = 75 nm) was chosen to strike a balance between high Q factors and fabrication resolution limits37.

Fig. 2: Room-temperature strong coupling in hBN/WS2/hBN vdW-HMs.
figure 2

a, Optical image of the vdW heterostructure. The WS2 monolayer (black outline) is encapsulated between a bottom hBN layer (light-pink area) and a top hBN (purple area) using a multistep heterostructure fabrication process (top panel). Right panel: optical image of the final vdW-HM sample after lithography-based nanofabrication. The bright square is the unstructured reference heterostructure sample. b, Schematic and side view of a single qBIC metasurface unit cell, composed of hBN/WS2/hBN layers with a total height of 125 nm. Inset: geometry of the metasurface unit cell, where px = py is the lattice periodicity, w is the nanorod width, L0 is the base rod length and ΔL is the asymmetry factor. c, PL spectra of the reference vdW heterostructure and a positively detuned vdW-HM showing negligible changes in the PL emission spectra, confirming the preservation of monolayer quality after the fabrication step. d, Transmittance of the reference WS2 heterostructure sample, exhibiting the excitonic peak at 2 eV (multiplied 15 times), and that of the resonant qBIC metasurface, exhibiting a clear splitting at the corresponding exciton energy. e, Derivative of the transmittance (dT/dE) for the full set of optical metasurfaces for the sample shown in a, fitted with the coupled harmonic oscillator model described in the Methods. Inset: selected derivative trace for vdW-HM at close-to-zero detuning (in black) and negatively and positively detuned metasurfaces as comparison (in grey). f, PL emission as a function of scaling factor S and relative qBIC–exciton detuning δ. The white dots represent the PL peak maxima, and the data are fitted from the same coupled harmonic oscillator as that in e. g, Normalized WS2 PL spectra from different vdW metasurfaces with increasing scaling factor (bottom to top).

Full size image

The optical quality of the WS2 monolayer is preserved during the fabrication of the vdW-HM. This is demonstrated by the negligible differences between the PL emission spectra of the final metasurface samples and an unstructured hBN-encapsulated WS2 monolayer reference (Fig. 2c). Transmittance measurements from the reference heterostructure and a vdW-HM with δ = 15 meV reveal a clear splitting in the transmission spectrum at the exciton energy, providing evidence for a strong interaction between 2D excitons and qBIC modes (Fig. 2d). We further investigated the coupled exciton–metasurface system using optical transmission and PL emission measurements on the whole set of fabricated metasurfaces. We extracted Q factors above 100 for the detuned metasurfaces (Supplementary Fig. 3) and observed a decrease in Q values at the exciton resonance due to the broadening and splitting of the cavity mode. In Fig. 2e, we plot the derivative of the transmittance as a function of the scaling factor. The detuning of the qBIC cavity and the exciton resonance exhibits clear anticrossing at the WS2 exciton energy (Fig. 2e, inset), confirming the strong light–matter coupling regime in the ultrathin vdW-HM. When probing the PL emission from the same set of vdW-HM (Fig. 2f,g), we observe a PL peak at lower energies than the WS2 excitons, attributed to the lower polariton branch. From both datasets, we extracted a Rabi splitting energy of ΩR = 30 ± 1 meV at room temperature, using a coupled harmonic oscillator model (Methods), comparable with those found in TMDC monolayers in standard optical cavities45,50 and in dielectric metasurfaces covered with 2D layers24. Furthermore, the strong-coupling regime is replicated in a second sample (Supplementary Fig. 4).

Angle-resolved spectroscopy

To fully characterize the qBIC–exciton interactions in detuned metasurfaces, where the Rabi splitting occurs at non-zero angles (Fig. 1d), we performed numerical simulations as well as an experimental characterization of the metasurfaces’ angular dispersion via back-focal-plane spectroscopy (Methods and Supplementary Fig. 5 provide the experimental details). We performed rigorous coupled-wave analysis (RCWA) simulations of the angle-resolved reflectance of hBN vdW-HMs with a thickness of 125 nm, both with and without an embedded WS2 monolayer (Fig. 3a). Since the periodic qBIC metasurface functions as a 2D diffraction grating, the presence of the substrate induces a prominent contribution of the diffraction orders in the angular reflectance spectrum (Supplementary Figs. 6 and 7), described by the grating equation kq − ki = ±mΚ, where kq and ki are the wavevectors of the in-plane diffracted and incident waves, respectively; m is an integer; and Κ corresponds to the reciprocal 2D lattice set by the metasurface geometry. When incorporating the WS2 monolayer, the angular dispersion of the bare qBIC cavity is modified, exhibiting an anticrossing of the two resonances.

Fig. 3: Angle-resolved spectroscopy of vdW-HMs.
figure 3

a, Numerical results of RCWA simulations of an hBN vdW heterostructure without (negative wavevectors) and with (positive wavevectors) an embedded WS2 monolayer, where the diffraction order (±1, 0) is shown. In the presence of the monolayer, the dispersion is fitted using the model described in the Methods, with the resulting polariton branches depicted as the dashed white lines. The solid, flat line represents the WS2 exciton (X) resonance. b, Experimental back-focal-plane imaging of the reflectance from a vdW metasurface with detuning δ = 96 meV. The dispersion is overlaid with the calculated exciton–polariton branches energies (dashed white lines; Methods). c, Reflectance spectrum extracted from the data in b at kx = 0.09, exhibiting splitting at the corresponding exciton energy (grey-shaded area). d, RCWA numerical simulations of the reflectance from the vdW heterostructure, showing good agreement with the experimental results, and underscoring the effect of the diffraction orders. e, Angular dispersion of the WS2 PL emission from different vdW-HMs with different detuning values (from left to right: 96 meV, –55 meV and –94 meV). For negatively detuned cavities, the maximum of the PL emission is observed at kx = 0, corresponding to the qBIC-coupled lower polariton branch.

Full size image

To experimentally validate these effects, we measured the angle-resolved reflectance using a custom-made back-focal-plane imaging setup. Figure 3b shows the reflectance of a positively detuned vdW-HM (δ = 96 meV), revealing the presence of strong diffraction orders. We observe a splitting of approximately 44 meV at the exciton energy (Fig. 3c). Compared with the numerical simulations shown in Fig. 3d, there is excellent agreement between the experimental and numerical results (Supplementary Fig. 8). We then analysed the PL emission by exciting the sample from the substrate side and collecting the emitted PL signal along with the corresponding back-focal-plane image (Methods and Supplementary Fig. 5). The PL emission was collected from the vdW-HMs with varying detuning values, exhibiting the expected angle-resolved dispersion of the strongly coupled system (Fig. 3e). For positive detuning, the PL emission is predominantly redirected into the diffraction orders. By contrast, for negatively detuned cavities, the PL emission shifts to the maximum of the qBIC-coupled lower polariton branch at zero angle. This confirms that the additional PL peak observed in the vdW-HMs is not due to directional emission into diffraction orders but instead corresponds to the effective polariton emission arising from the negative dispersion of the qBIC cavity mode at kx = 0. Moreover, we note that this effect is accompanied by an overall enhancement in the collected PL intensity (Supplementary Fig. 9), demonstrating an increased radiative efficiency for negative detuning values.

Nonlinear metasurface-coupled exciton–polaritons

Strong polaritonic interactions are fundamental for low-threshold lasing and Bose–Einstein condensates51, requiring high nonlinearities or large polariton densities. To further characterize the nonlinear nature of the exciton–polariton states in our vdW-HM, we investigated the PL as a function of the excitation intensity. Figure 4a shows the angle-resolved PL emission of a negatively detuned vdW metasurface in both low- and high-power regimes. At high fluences, the PL maximum at zero angles shifts to higher energies, accompanied by the disappearance of the excitonic peak. As the polariton density increases under optical pumping, their mutual dipole–dipole interactions induce a nonlinear response. This translates to a blueshift of the polariton energy52, potentially leading to population inversion for lasing or to condensation in the case of even stronger interactions.

Fig. 4: Nonlinear exciton–polaritons and saturation of the strong-coupling regime.
figure 4

a, Comparison of the PL angular emission for a vdW-HM (δ = –94 meV) at low (10−13 J cm−2) and high (10−10 J cm−2) excitation fluences. The grey dashed line represents the exciton energy, whereas the black dashed lines mark the peak position of the LP emission peak. b, Transmittance for a vdW-HM (δ = 20 meV) excited using a supercontinuum laser with a repetition rate of 2.05 kHz. Similar to the emission, we observe—even in absorption—a transition to weak coupling at fluences above 10−9 J cm−2. c, Normalized PL emission for a metasurface, δ = 20 meV, as a function of the excitation fluence. The exciton peak in the emission doublet is prominent at low power, but it gradually loses its weight, leaving the lower polariton branch to dominate the PL emission until fluences approaching 109 J cm−2, where we observe a collapse to a single optical transition. d, Normalized PL emission of a vdW metasurface on a second sample (δ = –55 meV) under increasing fluences. We observe the same saturation of the strong-coupling regime when approaching 10−9 J cm−2 excitation fluences. e, Peak maxima positions for the exciton (X) and LP peaks for increasing the excitation fluence. Both peaks exhibit a redshift with increasing power, eventually leading to the collapse of both peaks into a single emission channel. f, Integrated PL intensity as a function of excitation fluence, exhibiting no saturation up to values approaching 109 J cm−2. g, Full-width at half-maximum (FWHM) values extracted for the exciton peak and LP peak as a function of excitation fluence. The exciton exhibits the expected broadening due to the increased non-radiative processes, whereas on the contrary, for the LP, we observe a reduction in the linewidth.

Full size image

The optical transmission of a vdW-HM with small detuning (δ = 20 meV) shows a splitting in the transmission spectrum at the corresponding WS2 exciton energy. The system is excited with a low-repetition-rate supercontinuum pulsed laser (Methods provides the experimental details). Remarkably, increasing fluences lead to a saturation of the strong-coupling regime and at fluences exceeding 10−9 J cm−2, the two absorption peaks collapse into a single, weakly coupled resonance (Fig. 4b). The use of a low repetition rate of 2.05 kHz minimizes thermal effects and rules out long-lived dark states as contributors to the observed saturation behaviour. This transition is attributed to strong interactions driven by the increased polariton density and a reduction in the oscillator strength. These observations are corroborated by PL emission measurements (Fig. 4c). Similar to the absorption behaviour, an emission doublet is observed at low densities, which merges into a single emission peak as the excitation fluence increases.

The same phenomena are reproduced in a second sample, as shown in the normalized PL emission spectra for a negatively detuned vdW metasurface (δ = –55 meV) in Fig. 4d. When comparing the relative spectral position of the two emission peaks (Fig. 4e), both polariton branches shift to lower energies. Because a stronger interaction is expected to shift the polariton to higher energies, this apparent redshift is driven by the unique response of the WS2 excitons53, as also observed in the reference heterostructure (Supplementary Fig. 10). The relative blueshift of the lower polariton branch and redshift of the upper one confirms that the main process in our metasurfaces is governed by the phase-state-filling mechanism50. This mechanism, analogous to the Pauli exclusion principle, leads to a reduction in the oscillator strength due to fewer available excited states. In particular, the PL intensity does not saturate even at the maximum excitation fluence (Fig. 4f). This lack of saturation indicates a transition to a lasing regime54, as previously analysed in vdW-HM with theoretical models55. To confirm this effect, Fig. 4g shows the linewidths of both peaks, extracted from the data in Fig. 4d, revealing opposite behaviour: the exciton peak broadens under higher densities due to predominant non-radiative effects, whereas the lower polariton peak exhibits a reduced linewidth, indicating the onset of amplified spontaneous emission56.

In particular, the observed nonlinear polaritonic behaviour in our vdW-HM appears at excitation fluences on the order of nanojoules per square centimetre, which is three to four orders of magnitude lower than previous works on TMDC monolayers strongly coupled to optical cavities41,44,57,58,59, indicating extremely large exciton–polariton interactions at low polariton densities. The origin of these large nonlinearities could be the considerable kinetic energy stored within the metasurface exciton–polaritons, confined to a single unit cell at the hundreds-of-nanometres scale, and affected by both collisional losses and confinement length60. Although our results demonstrate strongly nonlinear exciton–polaritons at room temperature, achieving phenomena such as lasing or condensation typically requires cryogenic temperatures to reduce thermal dephasing and enhance coherence, which are not yet met by our design. Future work will focus on addressing these challenges and enable high-Q cavity integration to further enhance nonlinear phenomena.

Discussion

We have experimentally realized nonlinear exciton–polaritons in vdW-HM, merging the design flexibility of optical metasurfaces with the deterministic assembly of vdW heterostructures. We demonstrated strong light–matter coupling at room temperature by using excitons in a WS2 monolayer embedded within a nanostructured hBN-based vdW heterostructure. Remarkably, this was achieved without the need for an external optical cavity, relying instead on carefully designed qBIC metasurfaces to generate intrinsic optical resonances within the stacked vdW materials. The characteristic exciton–polariton physics of the embedded WS2 monolayer observed in both absorption and emission demonstrates exceptionally strong polariton interactions, over three orders of magnitude larger than in other 2D cavity systems41,44,57,58,59. Such reduced thresholds can pave the way for low-power optical applications, from low-threshold polariton lasers that minimize thermal loading and facilitate on-chip integration to high-speed, low-energy optical switching for neuromorphic computing and integrated photonic circuits. Furthermore, the enhanced light–matter coupling at these reduced intensities may enable room-temperature quantum phenomena, such as few-particle entanglement or single-polariton nonlinearities, broadening the scope of polaritonic devices within quantum photonics. Finally, our approach seamlessly integrates 2D materials with flat optics and can be applied to various vdW heterostructures, including those based on TMDC hetero-bilayers, black phosphorus or other vdW semiconductors. Building on these principles, our rational vdW-HM design methodology can be extended to more exotic vdW materials, ranging from superconducting to topological systems, opening new avenues for light–matter interactions in layered quantum materials. Overall, vdW metasurfaces opens to advances in both nano-optics and condensed-matter physics, fostering the development of next-generation devices with unparalleled structural control and tailored functionalities, for applications in optical communications, quantum technologies and sensing.

Methods

Fabrication

The vdW heterostructures were fabricated by mechanical exfoliation and a combination of dry transfer techniques from commercially available single crystals (HQ Graphene). Initially, hBN was exfoliated onto commercial silicon wafers. Flakes with thickness ranging from 60 nm to 80 nm were selected using optical microscopy and a Stylus profilometer (Bruker Dektak) to achieve a total height of approximately 120 nm to 160 nm of the final heterostructure. The first hBN flake was transferred onto a fused silica substrate using a polydimethylsiloxane and poly(bisphenol A carbonate) stamp. Subsequently, WS2 was mechanically exfoliated onto a separate polydimethylsiloxane stamp. Suitable monolayers were detected via optical microscopy and then transferred on top of the first hBN flake. The second hBN flake was then picked up using a polydimethylsiloxane and poly(bisphenol A carbonate) stamp and transferred on top of the WS2 monolayer and base hBN-flake complex, thereby completing the vdW heterostructure. Stacks were further processed and patterned following the methods described elsewhere37. As a metasurface unit-cell design, a double-rod system with length asymmetry was chosen. At a scaling factor S = 1, the pitch size was set to px = py = 410 nm, and the resonators were parameterized by length L0 = 360 nm, width w = 100 nm and asymmetry factor ΔL = 75 nm. A double layer of poly(methyl methacrylate) (Allresist, 80 nm of 950k on top of 100 nm of 495k) was spin coated onto the substrate, and each poly(methyl methacrylate) layer was baked at 170 °C for 3 min. Electron-beam lithography was then used to define the drafted metasurface pattern in the poly(methyl methacrylate) layer on top of the monolayer region. After development, 2-nm titanium as an adhesion promoter and 60-nm gold were deposited using electron-beam evaporation. Bathing the sample over 8 h in a resist remover solution (MICROPOSIT REMOVER 1165) allowed for a clean lift-off process, creating the hard-mask pattern of choice. The unprotected regions of the heterostructure were etched away using inductively coupled reactive ion etching with sulfur hexafluoride and argon gases under 6.0-mtorr pressure with 300-W HF and 150-W inductively coupled plasma power. The remaining gold hard mask was then removed inside a potassium-based gold etchant.

Optical spectroscopy

The optical spectroscopy of vdW-HMs was carried out in a custom-built optical setup working in the transmission and reflection geometries (Supplementary Fig. 5). For the reflection measurements, the collimated emission of a tungsten–halogen source (Thorlabs SLS201L) is sent through a 50:50 beamsplitter and into the aperture of a 0.95-numerical-aperture ×60 objective. The reflected white light signal is collected through the same objective and focused on the slit of a spectrometer with 75-mm focal length (Princeton Instruments) for spectroscopic analysis. For angle-resolved spectroscopy, the addition of a Bertrand lens (Supplementary Fig. 5) in the collection path, focusing at the back-focal plane of the objective and in a 4f system with the lens in front of the spectrometer, allows to project the angle-resolved signal collected from the sample on the slits. To achieve spectral dispersion, the sample is mounted on a rotating stage to align its axis with the spectrometer slits. By utilizing the vertical slits as a spatial filter, the reflected signal is projected onto the grating, dispersing the spectral components onto the spectrometer charge-coupled device camera. The dispersion with the spectrometer optical grating allows the retrieval of the angular dispersion of the light emitted by the metasurface. For the PL measurements, the excitation is performed with the output of a frequency-doubled, 80-MHz-repetition-rate, 150-fs pulsed titanium:sapphire laser (Coherent Chameleon Ultra II) tuned to 405 nm. The laser is aligned in the transmission geometry and focused on the sample through a 0.25-numerical-aperture ×10 objective. The resulting focused laser resulted in a spot size of 8.15 µm in diameter, which is large enough to efficiently excite the optical metasurfaces (approximately 10 µm × 10 µm in size). The transmitted PL emission is then directed to the detector via the same setup described for the reflection measurements. For the measurements in the transmission mode, the sample is either excited with a halogen lamp or with the output of a supercontinuum ~35-ps pulsed laser (NKT Photonics), at a repetition rate of 2.05 kHz.

Numerical methods

Numerical simulations are carried out via a commercial software package (Ansys Lumerical 2024) using the RCWA method for the angular dispersion of the metasurfaces and a finite-difference time-domain package for the calculation of the electromagnetic fields. The monolayer is simulated with a mesh of 0.1 nm, ensuring a high number of mesh cells, with a nominal thickness of 1 nm.

Coupled harmonic oscillator model

The qBIC mode dispersion at low angles can be represented as the diffractive coupling of two symmetry-protected counterpropagating in-plane modes49. EqBIC(θ), the qBIC spectral position as a function of the emission angle θ, is modelled as

$${E}_{{\rm{qBIC}}}\left(\theta \right)={E}_{{\rm{qBIC}}}\left(\theta =0\right)+U-\sqrt{{U}^{2}+{v}^{2}{k\left(\theta \right)}^{2}},$$
(1)

where EqBIC (θ = 0) is the value at zero emission angle, U is the off-diagonal coupling term and v represents the group velocity of the BIC mode at high angles. Here k(θ) = (2π/a)sinθ is the in-plane wavevector and a denotes the lattice periodicity. The Hamiltonian of the coupled harmonic oscillator model describing the coupling between the qBIC mode (EqBIC) and the 2D excitons (EX), takes the form of

$$H=\left(\begin{array}{cc}{E}_{{{\rm{qBIC}}}}-{\rm{i}}{\gamma }_{{{\rm{qBIC}}}} & g\\ g & {E}_{{\rm{X}}}-{\rm{i}}{\gamma }_{{\rm{X}}}\end{array}\right),$$
(2)

where γqBIC and γX are the damping rates of the qBIC mode and exciton, respectively, and g is the coupling strength. Solutions of the model for different qBIC–exciton detuning values are shown in Supplementary Fig. 1. A value of γX = 14 meV is extracted from the fit of the optical transmission from the encapsulated sample and we extract γqBIC = 6.5 meV from the numerical simulations and γqBIC = ~15 meV from the experimental data fitting.