Quasi-BIC metasurfaces enable rapid, localized singlet-oxygen generation

Abstract

Bound states in the continuum (BIC) provide a photonic route to concentrate and store electromagnetic energy with minimal radiative leakage. Here, we report Au–TiO₂ metasurfaces that couple q-BIC optical resonance to interfacial charge transfer, enabling hot-carrier–mediated generation of ¹O₂ within subwavelength optical paths. Photonic engineering acts through two cooperative effects: (i) increased optical absorption that boosts hot-carrier generation at the Au/TiO₂ interface; and (ii) ultralow noble-metal loading that suppresses electron–hole recombination at metal/semiconductor junctions, thereby prolonging carrier lifetimes. These effects jointly yield a synergistic enhancement of ¹O₂ production beyond either pathway alone. Under continuous-wave excitation, the metasurfaces reach a local molar-level concentration of ¹O₂ within seconds, corresponding to an approximately six-order-of-magnitude increase in local ¹O₂ concentration compared with conventional approaches. Tuning structural asymmetry and excitation wavelength enables wavelength- and position-selective cytotoxicity without additional molecular sensitizers. By decoupling strong absorption from nobel-metal usage while extending hot-carrier persistence in a single platform, BIC-engineered metasurfaces offer a general way to achieve efficient photon-to-chemical conversion at solid–liquid interfaces. This approach is of significant value to rapid-acting photodynamic therapy, selective oxidation, and flow microreactors where low dose and geometric precision are critical.

Introduction

Singlet oxygen (¹O₂), the lowest excited electronic state of molecular oxygen^1,2,3,4, serves as one of critical reactive species in photodynamic therapy^5,6, photocatalytic oxidation^7,8, and advanced environmental remediation⁹. Conventional ¹O₂ generation relies on molecular photosensitizers, such as porphyrins¹⁰ and xanthene¹¹ dyes, that undergo Type II energy transfer upon photoexcitation^12,13. While these organic systems achieve moderate quantum yields, they suffer from wavelength-nonspecific activation, rapid photobleaching, and poor biocompatibility—factors that necessitate recurrent dosing and limit clinical precision^14,15,16.

Recent advances in nanophotonics employ metallic and semiconductor nanostructures as alternative sensitizers^17,18,19,20, yet their ¹O₂ yields remain orders of magnitude below therapeutic thresholds due to fundamental limitations in carrier dynamics and photon management²¹. Metallic nanoparticles leverage localized surface plasmon resonances (LSPRs) to concentrate light into subwavelength volumes, generating hot carriers that can drive oxygen activation^22,23,24,25. However, ultrafast electron-hole recombination (<100 fs) in noble metals restricts catalytic turnover rates. Semiconductor nanoparticles such as titanium dioxide (TiO₂) possess a bandgap that reduces electron-hole recombination to some extent, yet their singlet oxygen (¹O₂) quantum yield still remains very low, and ultraviolet excitation is required, which is fundamentally incompatible with biological tissues^26,27,28. Hybrid metal-semiconductor architectures can enhance carrier lifetimes through the Schottky barrier, which effectively suppresses hot electron recombination^29,30,31. However, they face a challenging material trade-off: increasing the metal content enhances visible-light absorption and boosts the number of activated carriers, but it also introduces additional recombination centers, thereby reducing carrier lifetimes. On the other hand, designs with reduced metal content sacrifice photon absorption efficiency, thereby limiting the number of separated carriers. This trade-off confines practical ¹O₂ concentrations to sub-micromolar levels, which is generally insufficient for targeted therapeutic or catalytic applications^17,18,32,33.

Bound states in the continuum (BIC) represent a unique class of non-radiative electromagnetic modes that coexist with propagating waves in periodic photonic structures^{34,35,36,37,38}. Originating from symmetry-protected destructive interference and governed by symmetry- and topology-protected singularities in momentum space, these states exhibit exceptional light confinement capabilities by suppressing energy leakage through radiation channels³⁹. When implemented in dielectric metasurfaces, BICs focus optical energy into deeply subwavelength volumes, which sharply increases the interaction between light and matter^40,41,42, creating localized electromagnetic hotspots ideal for driving nanoscale photochemical processes^43,44,45,46. This light-trapping mechanism functions independently of material absorption, providing new opportunities to simultaneously enhance carrier generation and optimize carrier separation efficiency in catalytic systems⁴⁷.

We report Au-TiO₂ metasurfaces that couple q-BIC optical resonance to interfacial charge transfer, enabling rapid, hot-carrier-mediated generation of ¹O₂ within a subwavelength optical path and with ultralow noble-metal loading (Fig. 1a). Resonant field confinement boosts absorption while mitigating recombination; the metasurface with the optical thickness of ~100 nm achieves ~45% optical absorption at the wavelength of 532 nm. This enables a local molar-level ¹O₂ concentration within seconds, corresponding to an approximately six-order-of-magnitude increase over conventional approaches. Time-correlated single-photon counting of the 1270 nm phosphorescence with spectral filtering validates ¹O₂ emission, and complementary frequency-domain analyses together with thermal controls isolate its photochemical origin from photothermal backgrounds. Tuning structural asymmetry and excitation wavelength realizes wavelength- and pixel-level selectivity, enabling rapid, sensitizer-free cytotoxicity. Collectively, these design principles decouple strong absorption from nobel-metal usage and offer a general way to achieve efficient light-to-chemical conversion at solid-liquid interfaces.

Fig. 1: BIC-engineered Au–TiO₂ metasurface for high-density singlet-oxygen generation.

a Conceptual illustration of photoexcitation in aqueous media: a q-BIC metasurface concentrates green light to convert triplet oxygen (³O₂) into ¹O₂, enabling localized cytotoxicity. b Device architecture and unit cell. A square array of TiO₂ elliptical nanopillars (height h) on SiO₂ is capped with an ultrathin Au layer. An in-plane scaling factor S multiplies all lateral dimensions---major/minor diameters S × 2A, S × 2B and lattice periods S × P_x, S × P_y---while h remains fixed. Structural asymmetry is controlled by the relative axis rotation θ between neighboring ellipses (asymmetry parameter $\alpha =\sin \theta$). The factor S tunes the resonance wavelength without altering α. c Scaling of the bare TiO₂ q-BIC quality factor with asymmetry: Q = kα^m (fit m ≈ −2.109), validating the characteristic Q ∝ α⁻² behavior near the symmetry-protected BIC. d Angle-resolved reflectance map of the bare TiO₂ metasurface, showing a narrowband q-BIC dispersion that peaks at the Γ point, with its narrowest linewidth at 538 nm (green). e Radiative and absorptive quality factors as a function of the structural asymmetry $\alpha =\sin (\theta )$. The radiative quality factor Q_rad (blue) is obtained by fitting the simulated Q-α dependence with Q_rad = kα^m, whereas the absorptive quality factor Q_abs (red dashed line) is calculated from the integrated multilayer system. The intersection Q_rad = Q_abs indicates the critical-coupling condition, yielding θ ≈ 0.26 rad. f Simulated absorptivity of the Au–TiO₂ metasurface at λ = 532 nm as a function of the relative rotation angle θ. The absorptivity reaches its maximum at θ ≈ 0.26 rad (red dashed line), consistent with the predicted loss-matching (critical-coupling) condition in e

Results

Mechanism of singlet oxygen generation via BIC-enhanced metasurface

We adopt a critical-coupling strategy built on q-BIC control to maximize interfacial absorption in an ultrathin Au-TiO₂ metasurface. The unit cell (Fig. 1b) comprises TiO₂ elliptical nanopillars of height h = 100 nm on SiO₂, capped with a 7 nm Au layer. A weak in-plane symmetry breaking, implemented by a relative rotation θ between neighboring ellipses (asymmetry parameter $\alpha =\sin \theta$), lifts the symmetry protection and converts the BIC into a radiative q-BIC. The bare (Au-free) dielectric array exhibits the expected Q ∝ α⁻² scaling (Fig. 1c) and a narrowband dispersion with a maximum at the Γ-point at 538 nm (Fig. 1d). Introducing the ultrathin Au film opens an absorptive channel. We then tune the asymmetry (rotation angle θ) to continuously modulate the radiative leakage and identify the critical-coupling point by the condition Q_rad ≈ Q_abs^48,49, which occurs at θ ≈ 0.26 rad (Fig. 1e and Supplementary Text 1). At this point, the interfacial absorptance reaches its maximum, yielding ~45% at λ = 532 nm within an optical thickness of ~100 nm (Fig. 1f).

The optical response directly governs hot-carrier dynamics at the Au/TiO₂ interface. Near-field simulations confirm an E_x-polarized dipolar q-BIC mode (Fig. 2a), and the heterojunction exhibits a narrow absorptivity band pinned to the q-BIC wavelength, which exceeds that of a standalone 7 nm Au film (Fig. 2b). The near-field enhancement at the maximum absorption condition is shown in Fig. 2c, where the optical energy is strongly localized at the rims of the Au nano-ellipses and also penetrates into the underlying TiO₂ pillars (peak ∣E∣/∣E₀∣ ≈ 14). Consistent with this energy localization, two-temperature modeling further predicts rim-localized electron heating within the Au nano-ellipses (Fig. S2). In the Schottky-junction picture (Fig. 2d), BIC-enhanced absorption broadens the electron distribution and enables emission of hot carriers over the barrier Δϕ ≈ 0.8 eV from Au (E_F) into the TiO₂ conduction band (E_C)^50,51, thereby initiating interfacial redox steps that form ¹O₂. By design, BIC photonic engineering localizes the optical field and absorption at the metal rim, enabling the same radiative energy to be absorbed with a substantially smaller amount of metal (Fig. 2e and Supplementary Text 2).

Fig. 2: BIC-driven hot-carrier physics in a minimal-metal Au–TiO₂ heterojunction metasurface.

a Field localization at the Au/TiO₂ interface under x-polarized illumination. Normalized in-plane field on the yz-plane at x = 0, $| {E}_{x}/{E}_{x,\max }|$, shows dipolar hot spots that seed interfacial chemistry. b Absorptivity spectra for the q-BIC Au–TiO₂ heterojunction (dashed green), an ultrathin Au film (solid gray), and the bare dielectric q-BIC (dotted red); the heterojunction exhibits a narrow, resonantly enhanced band near the BIC wavelength. c Simulated near-field enhancement distribution (∣E∣/∣E₀∣) of the Au–TiO₂ metasurface under E_x-polarized illumination at the maximum-absorption condition (corresponding to θ ≈ 0.26 rad at λ = 532 nm). The map shows the field enhancement on the Au surface and within the underlying TiO₂ elliptical pillars; the peak enhancement reaches ~14×. d Schottky-junction picture: BIC-enhanced absorption raises T_e, enabling hot-carrier emission over the barrier Δϕ from Au (E_F) into the TiO₂ conduction band (E_C), thereby driving redox steps that generate ¹O₂. e Consequence of using less Au: for the same absorbed optical energy, concentration in a smaller metal volume yields higher T_e and, with reduced metal coverage, fewer interfacial recombination pathways---prolonging the catalytic effect. f Absorptivity versus Au volume per irradiated area for colloidal nanoparticle references (colored by diameter) and for our BIC metasurface (orange). Same absorption with orders-of-magnitude less Au should be viewed as a functional advantage---higher T_e and suppressed recombination---rather than an economic metric

The use of smaller metallic volumes to achieve higher light absorption for ¹O₂ generation shows two significant benefits: (i) By enhancing light absorption per unit volume of the metal, the system increases electron activity, which can be quantified by electronic temperature (T_e)⁵², thereby improving catalytic efficiency. (ii) Ultralow metal loading mitigates the recombination loss and extends carrier lifetime, thereby enhancing photocatalytic performance.

Our photonically engineered heterostructure outperforms conventional metal-semiconductor hybrids—including near-perfect absorbers⁵³, Au-TiO₂ nanospheres⁵⁴, Au-CdS nanoparticles⁵⁵, and arbitrary-radius nanostructures^56,57—in decoupling metal loading from optical absorptance (Fig. 2f).

For the equivalent absorbed laser power, the significant reduction in gold volume indicates strong light-matter interactions, leading to a substantial increase in electron temperature (T_e). The relationship between T_e and hot-carrier generation efficiency (J_th) is expressed as⁵⁸:

$${J}_{th}\propto {T}_{e}^{2}\exp (\frac{-q{\phi }_{B}}{{k}_{B}{T}_{e}})$$

(1)

where qϕ_B = 0.8 eV and k_B is the Boltzmann constant. When the 532 nm E_x-polarized illuminates the metasurface at the focused spot size of 10 μm and the power of 25 mW (S = 1.108, θ = 0.26 rad), the heterojunction generates a photocurrent of 2.6 pA—a 6.2-fold enhancement over the 0.42 pA produced by conventional Au-TiO₂ nanoparticles under identical excitation conditions (Supplementary Text 3).

While the photocurrent magnitude differs by a factor of 6.2 between the two approaches, the spatial distribution of current generation shows a fundamentally distinct behavior. In the nanoparticle system, the photocurrent constitutes the sum of microcurrents generated by individual particles dispersed throughout the macroscopic aqueous solution. In contrast, photocurrent generation in the metasurface is highly localized, occurring exclusively within the irradiated region. This spatial confinement enables rapid catalytic conversion of oxygen molecules within micrometer-scale proximity to the metasurface. Combined with the exceptional adsorption capacity of TiO₂, this configuration produces high ¹O₂ concentrations in the immediate vicinity⁵⁹. This q-BIC metasurface establishes synergistic interplay among photonic energy localization, hot electron dynamics, and surface adsorption processes to achieve catalytic enhancement.

Fabrication and parameter characterization of metasurfaces

The q-BIC metasurface is fabricated using micro-nano processing technology (Fig. 3a and Supplementary Text 4). (1) A layer of electron beam resist (polymethyl methacrylate) is coated on a glass substrate, and a thin layer of aluminum is deposited at the bottom of the resist layer to improve conductivity. (2) The resist layer is patterned and subsequently developed using electron beam lithography at 8 C/m², 0.8 nA, and 80 kV. (3) A 100 nm TiO₂ layer and a 7 nm Au layer are deposited on the imaged resist layer using electron beam evaporation. (4) During the stripping process, the sample is immersed in acetone for 12 hours to remove TiO₂ and Au in the unexposed areas. After completion, the metasurface is characterized by scanning electron microscopy (Fig. 3b–d).

Fig. 3: Fabrication, imaging, and spectral multiplexing of Au–TiO₂ q-BIC metasurfaces.

Scanning electron micrographs of representative devices: a uniform array of Au-capped TiO₂ nano-ellipses (a); a stitched metasurface patch (b); and a pixelated pattern spelling “THU” (c). Scale bars as indicated. d Process flow. A PMMA/Al stack on SiO₂ is patterned by electron-beam lithography (EBL), developed, followed by sequential deposition of TiO₂ and ultrathin Au; lift-off defines the capped nano-ellipses. e Measurement schematic for normal-incidence reflectance under amplified-spontaneous-emission (ASE) illumination and microscope collection. f Photograph of the optical setup measuring metasurfaces immersed in aqueous media; g bright-field micrograph of the resulting colored “THU” pixels. h Spectral multiplexing by geometric scaling. Five metasurfaces with different global stretch factor S (uniformly scaling in-plane dimensions while keeping thickness fixed) are designed to target distinct resonances at 527, 532, 542, 550, and 562 nm (S = 1.104, 1.106, 1.108, 1.110, 1.112, respectively). i Normalized reflectance spectra for the five S values, showing a systematic red-shift with increasing S, indicating a positive correlation between the resonance wavelength λ and S. The metasurface of S = 1.106 resonates near 532 nm (vertical line), matching the operating wavelength used elsewhere in this work

BIC optical metasurfaces typically exhibit complementary spectral features that include absorption at a specific wavelength and a corresponding reflection peak at nearly the same wavelength. We utilize this phenomenon to precisely characterize the resonant behavior of our metasurface through reflection spectroscopy. The fabricated metasurface is immersed in deuterated water (D₂O), and a collimated amplified spontaneous emission (ASE) supercontinuum laser is focused on the patterned substrate (Fig. 3e). We optimize the interaction between the beam and the sample by vertically translating the substrate to align the beam waist with the metasurface plane. Since the optimized BIC mode is supported by the E_x polarization, a custom achromatic half-wave plate is used to precisely control the polarization direction perpendicular to the elliptical symmetry axis (transverse polarization), which is crucial for activating the q-BIC mode.

Our custom-built optical microscopy system enables precise positioning of the ASE beam on the metasurface. After optimized focusing, strong green reflectance is visually observed (Fig. 3f), with the fabricated THU-shaped pattern (Fig. 3d, g) confirming the designed resonance. To accommodate inevitable nanofabrication errors, we fabricate a sample comprising five metasurfaces with different structural parameters (S = 1.104–1.112, Fig. 3h). Hyperspectral imaging is performed using the same microscopy system to capture the reflection spectra of each sample (Fig. 3i). The measured reflection peaks show good agreement with simulations. The moderate broadening likely arises from a combination of (i) fabrication-induced dimensional variations among individual resonators and (ii) the finite angular spectrum introduced by the objective of high numerical aperture. The metasurface with S = 1.106 demonstrates optimal resonance at 532 nm and is subsequently selected for ¹O₂ generation experiments. The absorptance of the metasurface in aqueous solution is obtained from energy conservation, A = 1 − R − T, where R and T are the measured specular reflectance and transmittance of the metasurface, respectively. Experimentally, R and T are determined under 532 nm pulsed-laser excitation using a photodetector (PD) and an oscilloscope by reading the pulse peak amplitude as a linear proxy for optical intensity. For the reflectance measurement, a 50:50 non-polarizing beamsplitter is used and the reflected-signal increment is corrected by a factor of two, while background reflections from the deuterated water surface and the glass vessel are subtracted. For the transmittance measurement, the transmitted signal is recorded directly under the same focusing and collection conditions. The patterned region is compared against an unpatterned area measured in the same optical configuration as a reference. Using this procedure, we obtain A ≈ 0.45 at 532 nm, consistent with numerical simulations.

Steady-state singlet oxygen generation

To investigate the ¹O₂ production dynamics of the BIC metasurface in steady-state operation, we perform time-resolved phosphorescence measurements in oxygen-saturated deuterium oxide (D₂O)⁶⁰. A pulsed 532 nm laser (pulse width of 35 μs, repetition rate of 2 kHz) is used to activate the metasurface, and the ¹O₂ phosphorescence signal at 1270 nm is detected using time-correlated single photon counting. Long-pass filters (LPFs) and monochromators are employed to eliminate scattered excitation light, ensuring precise lifetime measurements (Fig. 4a). Unless otherwise stated, all kinetic curves are acquired under a fixed incident optical power (25 mW) at the sample plane (i.e., constant incident photon flux). We validate the phosphorescence wavelength by temporarily removing LPFs with different cutoff wavelengths (1250 and 1300 nm) and monitoring the photon counts; the pronounced change upon removing the 1300 nm LPF, together with the negligible change upon removing the 1250 nm LPF, indicates that the emission is primarily confined to 1250–1300 nm, consistent with the characteristic ¹O₂ phosphorescence near 1270 nm (Fig. 4b).

Fig. 4: Time-resolved detection of ¹O₂ phosphorescence from q-BIC metasurfaces.

a Optical layout for excitation and near-infrared photon counting. A 532 nm laser (or ASE source) illuminates the BIC metasurface in aqueous media; reflected/elastic light is rejected by a 532 nm notch filter before the CCD/spectrometer, while the NIR arm directs photons through interchangeable long-pass filters (LPF, 1250 or 1300 nm) to a single-photon detector (SPD). A photodiode (PD) provides the timing reference. b Filter-removal test for spectral discrimination. Photon-count traces are recorded while sequentially removing the 1250 and 1300 nm LPFs (arrow). Removing the 1250 nm LPF causes no noticeable change in the photon count, whereas removing the 1300 nm LPF produces a pronounced change, confirming that the detected photons have a wavelength near 1270 nm and are attributed to ¹O₂ phosphorescence. c Top: Time-correlated photon-counting traces for (top to bottom) 30 μM RB, 10 μM RB, BIC-enhanced metasurface, and a negative control. Bottom: representative decay curves and single-exponential fits yield lifetimes of τ_RB ≈ 64 μs for the RB control and τ_MS ≈ 47 μs for the BIC-enhanced metasurface, whereas the negative control shows no long-lived component

The phosphorescence decay profiles further confirm that this metasurface with sub-micron optical path (100 nm) achieves ¹O₂ emission intensity comparable to macroscopic photosensitizer solutions (10 μM Rose Bengal aqueous solution with the optical path as long as 10 mm, Fig. 4c). We chose 10 μM RB as a representative benchmark concentration because increasing the concentration of conventional molecular photosensitizers does not increase singlet oxygen output, owing to concentration-dependent self-quenching/aggregation effects^61,62. Specifically, the ¹O₂ phosphorescence counts from 10 μM Rose Bengal (RB) solution (yellow curve) and from the metasurface (blue curve) exhibit identical peak amplitudes in temporal fitting (Fig. 4c). The measured ¹O₂ lifetime for the BIC metasurface is 47 μs, which is shorter than the 63 and 64 μs lifetimes observed for 10 μM and 30 μM RB solutions, respectively. This reduction in lifetime is attributed to the elevated local oxygen concentration near the TiO₂ surface, where oxygen adsorption significantly enhances the quenching of ¹O₂^63,64,65. The local ¹O₂ concentration at the metasurface interface reaches approximately 1.12 M as calculated (Supplementary Text 5), which is 10⁶ times higher than the ¹O₂ concentrations achieved by traditional photosensitizer-based methods (~0.87 μM from RB aqueous solution). This concentration is primarily due to TiO₂’s strong oxygen adsorption and the photonic enhancement provided by the metasurface.

The wavelength-specific nature of ¹O₂ nanogenerator is further confirmed by varying the structural parameter S. Metasurfaces with S = 1.104 and S = 1.106 exhibit significantly higher ¹O₂ photon counts relative to the baseline (negative control), while metasurfaces with S ≥ 1.108 show no detectable activity (Fig. S3c). This strong dependence on the q-BIC resonance highlights the critical role of photonic mode engineering in driving efficient ¹O₂ generation.

Temporal evolution of singlet oxygen generation

To investigate the temporal evolution of ¹O₂ generation, we employ singlet oxygen sensor green fluorescent probe (SOSG)⁶⁶, a highly selective fluorescent probe that reacts specifically with ¹O₂ to form SOSG-ep while emitting green fluorescence. Under identical conditions with constant incident photon flux, both the BIC metasurface and a 10 μM RB photosensitizer solution are illuminated while monitoring SOSG fluorescence changes to infer ¹O₂ generation (Fig. 5a, left). The metasurface exhibits a rapid fluorescence intensity rise within the initial 8 seconds, demonstrating enhanced ¹O₂ generation kinetics surpassing conventional photosensitizer solutions. This accelerated dynamics is attributed to the prioritized activation of oxygen molecules adsorbed on TiO₂ through photoinduced carrier transfer.

Fig. 5: Chemical readouts and frequency-domain discrimination of singlet-oxygen generation.

a Temporal traces of reporter fluorescence under continuous 532 nm excitation. Left: SOSG probe; right: MB bleaching. In both reporters, the Au–TiO₂ q-BIC metasurface yields a markedly stronger response than a 10 μM RB reference, confirming efficient ¹O₂ production. b Illumination on/off test (blue-shaded window indicates the initial light-on period). The photon-count signal rises promptly upon excitation and remains well above the negative control, confirming a light-triggered process. c Bode analysis separating photothermal effects from ¹O₂ phosphorescence. The normalized transfer amplitude versus chopper frequency shows a flat, high-bandwidth response for the NIR phosphorescence channel (characteristic lifetime τ_SO ≈ 46.8 μs), while the photothermal signal exhibits a strong low-pass roll-off with a much longer thermal time constant (τ_th ≈ 0.04 s). The distinct bandwidths demonstrate that the detected ¹O₂ emission is not a thermal artifact

Under laser illumination of the RB solution, ¹O₂ is produced throughout the entire illuminated region. In contrast, when the laser is focused on the metasurface, the generation of ¹O₂ is confined to within a diffusion length of only a few hundred nanometers (~800 nm) above the patterned region. Although the total number of ¹O₂ molecules generated in the RB solution and metasurface may become comparable over extended irradiation times (Fig. 5a, left), the distinct volumes in which they exist result in significantly different concentration profiles (Fig. S4). The fluorescence intensity from BIC metasurface then gradually decreased after the peak at 8 s, which can be attributed to the conversion of ¹O₂ into superoxide anions (${O}_{2}^{-}$) that subsequently desorb from the TiO₂ surface⁶⁷. This transformation signifies the completion of the catalytic cycle, where ¹O₂ activation leads to oxygen molecule dissociation and regeneration of active sites.

To further confirm the photocatalytic activity of the metasurface, methylene-blue (MB) degradation experiments are conducted (Fig. 5a, right)¹². Under identical illumination conditions, the metasurface demonstrates faster MB degradation compared to the 10 μM RB solution, validating its superior photocatalytic performance. The enhanced degradation rate correlates with the accelerated ¹O₂ generation observed in the SOSG experiments. A direct comparison of the temporal behavior of SOSG fluorescence and MB degradation (Fig. 5b) reveals that the two processes exhibit consistent kinetic profiles under illumination. This temporal alignment reflects the ultrafast generation of ¹O₂ by the metasurface immediately after the onset of illumination, providing valuable insights into the dynamics of photonic-enhanced reactive oxygen species (ROS) production in the early stages of light activation.

To exclude the possibility that photothermal effects produce ¹O₂, we performed a compact frequency-domain discrimination by modulating the pump light and comparing the normalized transfer amplitudes of the NIR phosphorescence (1270 nm band) with those of the metasurface reflectance measured by a photoelectric detector (Fig. 5c). The phosphorescence exhibits a flat, high-bandwidth response consistent with the microsecond lifetime of ¹O₂, whereas the reflectance shows a strong low-pass roll-off with a thermal time constant on the order of 10⁻² s. The clear bandwidth separation confirms that the detected 1270 nm emission is photochemical rather than photothermal.

Evaluation of the ability of metasurfaces to induce death in cells

Under short-term laser irradiation, the concentration of ¹O₂ generated by photosensitizers dispersed in aqueous environments is generally insufficient to induce cell death, as the threshold concentration required for cell death is approximately in the millimolar range²¹. Achieving higher ¹O₂ concentrations requires increasing photosensitizer dosage and extending laser exposure times to several hours. However, excessive photosensitizer levels pose significant toxicity to biological tissues, limiting their biomedical applications^68,69, and prolonged irradiation significantly increases patient pain^70,71. Notably, BIC metasurfaces introduced here effectively overcome these limitations, having the ¹O₂ concentration rapidly reaching to molar level within 10 s.

To investigate the cytotoxicity of q-BIC metasurfaces through ¹O₂ generation, we conducted in vitro experiments using U2OS human osteosarcoma cells cultured directly on metasurface substrates⁷². The substrate comprises a circular SiO₂ wafer with three identical metasurface modules, each containing parametric arrays defined by geometric parameters (Fig. S6).

A 532 nm expanded laser beam (diameter: 2.5 mm, power: 25 mW) uniformly irradiates each metasurface module. Post-exposure samples are incubated for 6 h to allow death progression, with cell death assessed by Calcein-AM/PI dual staining. Apoptotic cells are mainly localized around metasurface-patterned regions S = 1.104 and S = 1.106 (Fig. 6, Panels V and VIII). Dose-response analysis reveals time dependence: increasing exposure duration from 75 to 125 s enhances ¹O₂ generation, correlating with elevated cell mortality.

Fig. 6: Wavelength-selective, pixel-resolved photocytotoxicity on Au–TiO₂ q-BIC metasurfaces.

Cells were cultured directly on patterned metasurfaces and imaged in three channels: White LED (bright field, top row), PI (propidium iodide; dead, middle row), and Merged (Calcein-AM/PI overlay; live/dead, bottom row). Columns (left to right): I–III 10 μM RB control, 125 s exposure; IV–VI 532 nm irradiation for 75 s; VII–IX 532 nm for 125 s; X–XII 562 nm for 125 s. Dashed boxes mark representative metasurface pixels used for comparison. Under 532 nm illumination (IV–IX), PI-positive cells concentrate in resonance-mapped zones corresponding to designs scaled to S = 1.104 and S = 1.106; surrounding, off-resonant regions retain Calcein fluorescence. When the excitation is shifted to 562 nm (X–XII), cell death localizes to pixels red-tuned to S = 1.112, consistent with spectral matching. Scale bars, 10 μm. Together, these data demonstrate spatially selective, resonance-matched photodynamic cytotoxicity arising from localized singlet-oxygen generation

This spatial selectivity aligns with spectral response mapping (Fig. 3h), where regions of S = 1.104 and S = 1.106 exhibit stronger 532 nm resonance than regions of S = 1.108–1.112, demonstrating the localized cytotoxicity. To validate wavelength specificity, the 532 nm laser is replaced with a 562 nm light source (25 mW, 125 s exposure), which induces significant cell death near region of S = 1.112 (Fig. 6a, Panel XI), corresponding to the 562 nm reflectance peak. In contrast, using 10 μM RB in Hank’s balanced salt solution (HBSS) buffer under equivalent photodynamic conditions (25 mW, 125 s exposure, Fig. 6a, Panels I–III) shows no patterned cell death in laser-irradiated regions.

The above results demonstrate that q-BIC metasurfaces enable precision-targeted photodynamic therapy, with spatial resolution governed by metasurface patterns and spectral selectivity defined by engineered optical resonances. This on-demand ¹O₂ generation technology achieves unprecedented spatiotemporal control, representing a significant advancement over traditional photosensitizer-based cell-targeting approaches.

Discussion

We have established a metasurface strategy that unites q-BIC resonance control with critical coupling to maximize interfacial absorption in a minimal-metal Au-TiO₂ architecture. By balancing radiative leakage with intrinsic loss within a subwavelength optical path, the q-BIC metasurface concentrates optical energy at Au/TiO₂ heterojunctions while maintaining ultralow noble-metal loading. The small metal volume concentrates the same absorbed power at high density, thereby elevating the electron temperature and strengthening the hot-carrier population; simultaneously, the reduced density of metallic recombination centers suppresses carrier recombination and extends carrier persistence at the interface. This design yields strong absorption at 532 nm and promotes hot-carrier generation and longevity, enabling local molar-level ¹O₂ within seconds and an approximately six-order-of-magnitude increase in local ¹O₂ concentration compared with conventional approaches, providing a material-efficient route to activate interfacial redox chemistry.

Under continuous-wave excitation, near-interface photochemistry rapidly yields pixel-confined, high-density ¹O₂, enabling on-chip spatially programmable cytotoxicity. The ¹O₂ signal is validated by time-resolved detection of the 1270 nm phosphorescence with spectral filtering, and complementary chemical readouts (SOSG activation and MB bleaching) exhibit temporally coherent kinetics, confirming a common reactive-oxygen flux. The metasurface thereby enables wavelength- and pixel-level control of reactivity without molecular sensitizers, yielding rapid, spatially resolved cytotoxicity in cell assays.

Beyond the specific Au–TiO₂ implementation, the principles demonstrated here—q-BIC mode engineering, impedance matching via critical-coupling, and minimal-metal hot-carrier activation—can be applied to other metal/semiconductor pairs and spectral bands. In stark contrast to conventional dye sensitizers and many nanoparticle photocatalysts that often require higher dosages and prolonged illumination and tend to be inadvertently activated by ambient light due to broadband absorption, our metasurface enables rapid build-up of highly localized ¹O₂ with wavelength-addressable, patterned reactivity. We anticipate immediate opportunities in rapid photodynamic therapy, selective oxidation, and on-chip or flow microreactors, where sensitizer-free operation, reduced cross-talk, and low noble-metal usage are desirable. Future work will focus on extending operation to NIR excitation, optimizing interface chemistry and stability in complex media, and integrating multiplexed resonators for spatial/spectral programmability of photon-to-chemical conversion at solid–liquid interfaces.

Materials and methods

Device fabrication

Au–TiO₂ q-BIC metasurfaces are fabricated on BF33 float-glass substrates. Substrates are sonicated in acetone and ethanol for 10 min at 200 W, rinsed with deionized water, dried with N₂, and baked at 120 °C for 30 min. A 950 PMMA A4 resist is spin-coated at 3000 rpm for 45 s and baked at 180 °C for 90 s. Patterns are defined by electron-beam lithography (NB5, 80 kV, 0.8 nA, dose 8 C/m²) under high vacuum (<2.7e-8 Torr), developed in MIBK:IPA = 1:3 for 60 s, rinsed in IPA for 60 s, and dried with N₂. Films are deposited by e-beam evaporation in a PVD75 (<1e-7 Torr, 1 Å/s): 100 nm TiO₂ followed by 7 nm Au. Lift-off in acetone for 48 h yields Au-capped TiO₂ elliptical nanopillars.

Numerical modeling

A single unit cell is modeled in COMSOL Multiphysics using Floquet periodic boundary conditions in the in-plane directions and perfectly matched layers along the propagation axis. The top and bottom faces use scattering boundaries. The superstrate is deuterated water and the substrate is SiO₂. The geometry consists of a 100 nm TiO₂ pillar capped by a 7 nm Au film.

Time-correlated single-photon counting of ¹O₂

Singlet-oxygen phosphorescence at 1270 nm is measured with a single-photon detector (Aurea Technology, SPD_NIR, 900-1700 nm) time-tagged by a time-correlated single-photon counting module (PicoQuant TimeHarp 260 NANO). Samples are excited with a pulsed 532 nm laser (pulse width of 35 μs, repetition rate of 2 kHz). Emission is collected through a Nikon TU PLAN ELWD 50×/0.60 long-working-distance objective in the same microscope path; LPFs at 1250 and 1300 nm, together with a 532 nm notch filter, block the 532 nm excitation. Removal of the 1300 nm LPF verifies that the detected signal lies within 1250–1300 nm.

Frequency-domain discrimination

The pump light is mechanically modulated with an optical chopper (Thorlabs MC2000B; modulation frequency up to 10 kHz). The amplitude transfer response of the 1270 nm phosphorescence channel is compared with a simultaneously recorded reflectance channel detected by a photodiode amplifier (Thorlabs PDA36A2; 350–1100 nm). The phosphorescence channel shows a flat, high-bandwidth response consistent with microsecond-lived emission, whereas the reflectance exhibits a pronounced low-pass roll-off with a thermal time constant of about 0.04 s, confirming a photochemical origin of the NIR signal.

Photon-flux normalization

To avoid ambiguity between incident and absorbed photon budgets, we explicitly distinguish the incident photon flux Φ_inc and absorbed photon flux Φ_abs in data analysis. The incident optical power P_inc is measured at the sample plane using a calibrated power meter. The corresponding incident photon flux at wavelength λ is calculated as

$${\Phi }_{{\rm{i}}{\rm{n}}{\rm{c}}}(\lambda )=\frac{{P}_{{\text{inc}}}}{hc/\lambda }=\frac{{P}_{{\text{inc}}}\lambda }{hc}$$

(2)

where h is Planck’s constant and c is the speed of light. When photon-flux density is needed, we have

$${\phi }_{{\text{inc}}}(\lambda )=\frac{{\Phi }_{{\text{inc}}}({\rm{\lambda }})}{S}$$

(3)

where S is the illuminated area at the sample.

The absorbed photon flux is obtained from the measured absorptance A(λ) as

$${\Phi }_{{\text{abs}}}({\rm{\lambda }})={\Phi }_{{\text{inc}}}({\rm{\lambda }})\times A({\rm{\lambda }})$$

(4)

and the absorbed optical power is P_abs(λ) = P_inc(λ) × A(λ). The absorptance is determined from reflection and transmission measurements via

$$A(\lambda )=1-R({\rm{\lambda }})-T({\rm{\lambda }})$$

(5)

where R(λ) and T(λ) are the total reflectance and transmittance (measured, e.g., with a spectrophotometer equipped with an integrating sphere). Unless otherwise stated, kinetic traces (e.g., SOSG activation, MB bleaching, and 1270 nm phosphorescence counting) are acquired and plotted under fixed incident power P_inc (i.e., constant Φ_inc). In contrast, any quantity intended to represent an intrinsic per-photon efficiency (e.g., apparent quantum-yield-type metrics) is normalized to the absorbed photon flux Φ_abs using the corresponding A(λ).

Chemical reporters

SOSG stock is prepared by dissolving 100 μg reagent in 33 μl methanol (about 5 mM), then diluting to 10 μM with deionized water before applying to the metasurface under a water layer. A non-patterned area with 10 μM RB serves as a blank, and four samplings are averaged. An MB working solution is prepared at 10 μM. MB bleaching and SOSG fluorescence are recorded under continuous 532 nm excitation in the same microscope to track ¹O₂ dynamics.

Cell culture and phototoxicity

U2OS human osteosarcoma cells are cultured directly on the metasurface at 37 °C in a CO₂ incubator. Experiments start after the cell adhesion. Live/dead staining reagents (Calcein-AM and propidium iodide, PI) are added, and the cells are incubated for 6 h. The entire patterned area is then irradiated by a 532 nm (or 562 nm) laser for the indicated duration, after which the cells are returned to the incubator for an additional 30 min. Cell death is assessed on a fluorescence microscope (Olympus BX53M) using an Olympus U-HGLGPS 130 W fiber-coupled fluorescence light source for excitation.