Single-Crystal Silicon Membrane Metasurfaces
- Single-crystalline silicon membrane metasurfaces are nanophotonic platforms that combine ultra-thin, high-index Si films with precise patterning to achieve tunable optical resonances.
- They leverage diverse structural designs—such as posts, holes, and bridge-linked resonators—to support high-Q modes and enable active modulation in visible and mid-IR regimes.
- Mechanical reconfiguration and active tuning integrate with optical functionalities, enabling applications in beam steering, sensing, and nonlinear optics while offering near-field accessibility.
Single-crystalline silicon membrane metasurfaces are metasurfaces patterned directly into thin crystalline Si membranes—either fully free-standing or transferred onto transparent supports—so that the membrane itself provides the high-index photonic medium and, in some implementations, the mechanical degree of freedom. Within this platform class, recent work spans visible-wavelength transmissive wavefront control, mid-infrared quasi-bound states in the continuum (q-BICs) with air-accessible hotspots, transmission-mode biochemical sensing, mechanically reconfigurable resonances, nonlinear chiral third-harmonic generation, and actively tunable ultra-high- modulation (Zhou et al., 2016, Adi et al., 2024, Rosas et al., 2024, Gorp et al., 12 Feb 2025, Tonkaev et al., 18 Aug 2025, Brikh et al., 27 Sep 2025).
1. Platform definition and material basis
The defining feature of the platform is the use of single-crystalline silicon as a membrane rather than as a bulk substrate-supported photonic layer. A foundational fabrication study demonstrated ultra-thin, free-standing, single-crystalline flat silicon membranes with thickness down to $6$ nm, and introduced strain control through a tensile SiN perimeter frame characterized by the strain compensation ratio . In that work, membranes with thicknesses between $6$ and $54$ nm retained high crystalline quality by HRTEM, and the compensating frame removed buckling that otherwise produced about out-of-plane undulation in a $37$ nm membrane (Shchepetov et al., 2013).
Later metasurface implementations used thicker crystalline membranes chosen for optical functionality. In the visible, a $220$ nm-thick c-Si device layer was transferred from SOI onto fused quartz by wafer bonding, handle removal, and HF etching, yielding a thin crystalline silicon film that served as the metasurface platform for a transmissive beam deflector at $6$0 nm (Zhou et al., 2016). In the mid-IR, free-standing intrinsic single-crystal $6$1 or $6$2 silicon membranes with thickness $6$3 and millimeter-scale lateral dimensions were used to support q-BIC resonances in air-filled cavities, transmission-mode BIC-EIT resonances, and active modulation (Adi et al., 2024, Rosas et al., 2024, Brikh et al., 27 Sep 2025). A distinct nanoscale limit was reached in a nanopatterned $6$4 nm-thick silicon membrane in which the membrane simultaneously functioned as a mechanical metamaterial and an optical metasurface (Gorp et al., 12 Feb 2025).
This range of implementations suggests that “single-crystalline silicon membrane metasurface” denotes a platform family rather than a single canonical geometry. The common elements are crystalline Si, membrane-scale thickness, and metasurface patterning that exploits the optical advantages of a high-index material in a substrate-free or quasi-substrate-free configuration.
2. Structural archetypes and fabrication routes
Representative devices differ primarily in whether the membrane is patterned into posts, holes, cavities, or mechanically linked resonant elements. The geometric diversity is central to the field because the membrane is both the optical environment and, in several cases, the dominant mechanical or near-field-accessible region.
| Platform | Geometry | Representative dimensions |
|---|---|---|
| Visible transmissive metasurface (Zhou et al., 2016) | Square-lattice circular posts on transferred c-Si film | $6$5 nm height, $6$6 nm period |
| Mid-IR q-BIC membrane (Adi et al., 2024) | Double-hole square lattice with alternating radius perturbation $6$7 | $6$8-thick membrane, $6$9 window |
| Transmission q-BIC-EIT membrane (Rosas et al., 2024) | Pairs of oppositely tilted elliptical holes | 0-thick membrane, 1, 2 |
| Mechano-optical membrane (Gorp et al., 12 Feb 2025) | Bridge-linked resonant silicon “particles” in a nanopatterned membrane | 3 nm thickness, 4 nm-wide beams |
| Nonlinear chiral membrane (Tonkaev et al., 18 Aug 2025) | Square lattice of unit cells with L-shaped cavities | 5 nm-thick membrane, air gap 6 nm |
| Active ultra-high-7 membrane (Brikh et al., 27 Sep 2025) | Hexagonal lattice of elliptical holes | 8-thick membrane, 9 |
Fabrication routes are correspondingly varied but remain close to standard Si processing. Electron-beam lithography and deep reactive-ion etching recur across free-standing mid-IR membranes, with buried-oxide removal or HF vapor release used to free the membrane from SOI-derived stacks (Adi et al., 2024, Brikh et al., 27 Sep 2025). The nonlinear chiral platform similarly used e-beam lithography, deep Si reactive-ion etching, and buried oxide removal to form a free-standing membrane above the substrate (Tonkaev et al., 18 Aug 2025). In the visible, the transferred c-Si-on-quartz route used NOA61 bonding, 0 nm UV curing for 1 hours, baking at 2 for 3 days, DRIE handle removal, and HF etching of the buried oxide/silica (Zhou et al., 2016).
A notable departure from purely optical design appears in the mechano-optical membrane. There, the resonant nanoparticles are connected by carefully designed 4 nm-wide beams chosen to emulate rotating-rectangle/kirigami motion, keep the maximum stress below the silicon failure limit, and minimize optical perturbation of the resonators. Simulated stresses remained below about 5 GPa at 6, below the cited failure threshold 7 GPa (Gorp et al., 12 Feb 2025).
3. Resonant physics, symmetry breaking, and field localization
The dominant physical mechanisms are q-BIC formation, Fano interference, Brillouin zone folding, EIT-like mode interference, and, in membrane-specific theories, strong quadrupolar participation. In the mid-IR double-hole membranes, Brillouin zone folding converts guided modes that are inaccessible at normal incidence in the unperturbed lattice into free-space excitable q-BICs by introducing an alternating hole-radius perturbation 8. The measured 9-factor follows the expected quadratic trend with the asymmetry parameter, and the resulting TE and TM q-BIC modes exhibit strong field localization in air voids and on membrane surfaces rather than primarily inside a dielectric slab. Reported peak electric-field enhancement values are roughly 0 for TE and 1 for TM, compared with about 2 and 3 for the corresponding guided modes; the reported effective mode volumes are approximately 4 for qBIC-TE and 5 for qBIC-TM (Adi et al., 2024).
A second route to narrow membrane resonances is interference between a broad surface lattice mode and an ultra-narrow q-BIC mode. In free-standing membranes with tilted elliptic holes, the broad resonance is associated with collective lattice effects and Rayleigh anomalies, while the narrow resonance appears once tilt angle 6 breaks in-plane symmetry and couples a BIC to free space. Their overlap produces a transmission peak analogous to electromagnetically induced transparency. At 7, the broad mode is dominated by electric dipole and magnetic quadrupole contributions; at 8, the narrow resonance is mainly driven by electric quadrupole and magnetic dipole moments (Rosas et al., 2024).
In mechanically reconfigurable membranes, the resonance is explicitly nonlocal. Stretching rotates and separates the resonant Si elements, tuning a q-BIC/Fano-like resonance whose quality factor in the ideal symmetry-breaking picture follows 9, although the actual device deviates from this textbook dependence because spacing changes simultaneously with rotation. The same study notes that the maximum spectral shift does not occur at the largest geometric displacement; in the idealized case it occurs around $6$0, because the effective dipole orientation evolves with strain and is not always aligned with the long axis of the ellipses (Gorp et al., 12 Feb 2025).
Membrane metasurfaces also motivated a distinct theoretical correction to standard dipole-based intuition. A semi-analytical dipole-quadrupole model showed that, unlike many solid-particle metasurfaces, membrane metasurfaces generally require electric and magnetic quadrupoles to reproduce reflection, transmission, Fano line shapes, generalized Kerker effects, lattice anapoles, and quasi-BIC behavior, especially under oblique incidence. The same work emphasized that multipole content depends strongly on unit-cell choice, and that an optimal symmetric unit cell can make the dipole-quadrupole truncation sufficient while poor choices artificially enhance higher multipoles (Allayarov et al., 13 Oct 2025).
4. Spectral regimes and reported performance
The platform spans visible through mid-IR operation, with markedly different optical targets in each regime. In the visible, the transferred c-Si metasurface demonstrated polarization-independent beam deflection at $6$1 nm. Using eight distinct post diameters on a square lattice, the device achieved full $6$2 phase control, a measured beam deflection angle of $6$3 versus a simulated $6$4, and a measured deflection efficiency of $6$5. Efficiency varied by only about $6$6 as the input linear polarization was rotated. The same work identified approximately $6$7 absorption in c-Si, $6$8 interface reflectivity, and about $6$9 power in other diffraction orders as the main loss channels, while a redesigned $54$0 nm-thick structure with $54$1 nm period was predicted to reach $54$2 efficiency (Zhou et al., 2016).
| Regime and function | Representative platform | Reported performance |
|---|---|---|
| Visible beam steering | c-Si posts on quartz (Zhou et al., 2016) | $54$3 measured efficiency at $54$4 nm; $54$5 deflection |
| Mid-IR q-BIC in air voids | Free-standing double-hole membrane (Adi et al., 2024) | Highest measured $54$6-factor of $54$7 |
| Mid-IR transmission resonance | Tilted elliptic-hole membrane (Rosas et al., 2024) | $54$8 at $54$9 |
| Mechano-optical visible tuning | Beam-linked 0 nm membrane (Gorp et al., 12 Feb 2025) | Realistic tuning 1–2 nm; 3–4 |
| Active mid-IR ultra-high-5 | Elliptical-hole membrane (Brikh et al., 27 Sep 2025) | Measured 6 at 7; measured up to 8 |
The mechano-optical case is unusual because both idealized and realistic performance were reported. In the idealized kirigami metasurface, the resonance tunes from about 9 nm to $37$0 nm, corresponding to a shift of $37$1 times the resonant linewidth, and the quality factor decreases with strain to a minimum of about $37$2 at $37$3. In the realistic beam-linked membrane, optical absorption and beam-induced perturbations reduce the tuning range to $37$4–$37$5 nm and the quality factor to roughly $37$6–$37$7, but the resonance remains strong and tunable (Gorp et al., 12 Feb 2025).
In the mid-IR, the single-crystalline membrane platform supports both high-$37$8 passive and high-$37$9 active operation. Free-standing air-cavity membranes reached a highest measured TE q-BIC $220$0-factor of $220$1, with TM q-BIC reaching $220$2 (Adi et al., 2024). Transmission-mode tilted-elliptic-hole membranes reached $220$3 around $220$4 or $220$5, while also showing group delay up to $220$6 ns and group index as high as $220$7 (Rosas et al., 2024). Actively tunable elliptical-hole membranes then extended the measured $220$8-range from several hundred up to $220$9, including a measured transmission dip with $6$00 and amplitude contrast close to $6$01 at $6$02 (Brikh et al., 27 Sep 2025).
5. Functional modalities
Mechanical reconfiguration is one of the clearest examples of membrane-specific functionality. In the nanopatterned $6$03 nm silicon membrane, in-plane strain bends the $6$04 nm beams, inducing counter-rotation-like motion and increasing inter-particle spacing. In the idealized design, $6$05 varies from $6$06 to about $6$07 at strain $6$08, giving an inter-particle distance increase of up to $6$09. The membrane therefore acts simultaneously as a compliant kirigami-like structure and as a resonant metasurface, enabling large programmable tuning for tunable filters, reconfigurable lenses, reflective displays, and dynamic wavefront control (Gorp et al., 12 Feb 2025).
Open-cavity vibrational strong coupling is another major modality. In the mid-IR double-hole membranes, the accessible air voids enable coupling between q-BIC modes and the PMMA C=O vibration near $6$10. The experimentally observed minimum splitting at zero detuning is about $6$11, essentially at the onset of the strong-coupling threshold given by $6$12, and thicker PMMA layers exceed this threshold more clearly. The same work found that the Rabi splitting scales approximately as $6$13, consistent with collective strong coupling (Adi et al., 2024). In transmission-mode BIC-EIT membranes, coating with $6$14, $6$15, and $6$16 nm PMMA films produced Rabi splittings of $6$17, $6$18, and $6$19, respectively, all exceeding the stated combined loss rate $6$20; the same platform also detected protein A/G monolayers through perturbations in the amide I and amide II regions using transmission spectroscopy (Rosas et al., 2024).
Nonlinear symmetry engineering has been demonstrated in a different free-standing crystalline membrane architecture. A $6$21 nm-thick membrane with $6$22-symmetric L-shaped cavities was designed to be effectively achiral in linear optics, with ideal linear relations $6$23 and $6$24, yet it exhibited pronounced nonlinear circular dichroism in third-harmonic generation. The unperturbed metasurface showed nonlinear CD of $6$25 experimentally, while intentional in-plane symmetry breaking activated a co-polarized THG channel and reversed the sign, yielding nonlinear CD as large as $6$26 experimentally. The measured linear CD remained comparatively small, with maximum values about $6$27 for the unperturbed metasurface and $6$28 for the perturbed one (Tonkaev et al., 18 Aug 2025).
Dynamic modulation extends the platform beyond passive resonance engineering. In a $6$29-thick single-crystalline Si membrane perforated on a hexagonal lattice, electro-thermal Joule heating with thin Al electrodes produced a temperature increase of only $6$30 yet shifted the resonance by up to $6$31 FWHM, about $6$32 nm, yielding an absolute modulation amplitude of $6$33. The device reached more than $6$34 modulation depth at $6$35 V and a maximum modulation speed of $6$36 kHz for a $6$37 membrane with $6$38 mW dissipated power. Under $6$39 nm, $6$40 fs optical pumping, the same platform exhibited a resonance-shift decay time of $6$41 ns, an inferred carrier lifetime of $6$42 ns, estimated sub-GHz modulation rates, and operation in a low-fluence regime below $6$43 (Brikh et al., 27 Sep 2025).
6. Constraints, classification issues, and design outlook
Several limitations recur across the literature. In realistic mechanically tunable membranes, silicon absorption and beam-induced perturbations reduce $6$44 and prevent unity peak reflection, while finite-size clamping produces nonuniform deformation near handholds over a length scale $6$45 nm, about three unit cells, slightly broadening and weakening the reflectance peak relative to the infinite-array prediction (Gorp et al., 12 Feb 2025). In visible transmissive devices on quartz, performance is limited by c-Si absorption, interface reflectivity, and fabrication tolerances; the measured average diameter error of about $6$46 nm was sufficient to reduce predicted efficiency from $6$47 to $6$48, essentially matching the measured $6$49 (Zhou et al., 2016). In optimized freestanding nanohole deflectors operating around $6$50, the figure of merit drops to half its maximum for roughly $6$51 nm membrane-thickness variation, indicating strong thickness sensitivity, although the provided text for that study does not explicitly state single crystallinity (Ong et al., 2018).
A classification issue also follows from the literature itself: not every freestanding silicon membrane metasurface should automatically be labeled single-crystalline unless crystallinity is explicitly specified. Some core studies do explicitly use “single-crystalline,” “single-crystal,” or “c-Si” (Shchepetov et al., 2013, Zhou et al., 2016, Adi et al., 2024, Rosas et al., 2024, Tonkaev et al., 18 Aug 2025, Brikh et al., 27 Sep 2025), whereas related work on freestanding nanohole deflectors refers more generally to a silicon membrane and is best treated as adjacent rather than definitive evidence for single-crystalline implementations (Ong et al., 2018).
A common misconception is that membrane metasurfaces are merely substrate-removed versions of conventional substrate-supported metasurfaces. The literature does not support that reduction. In free-standing mid-IR membranes, the air environment increases effective index contrast and makes near-field hotspots directly accessible to molecules (Adi et al., 2024). In the mechano-optical membrane, removal of the substrate is part of a deeper co-design in which the same nanopatterned single-crystalline Si layer is both mechanical metamaterial and optical metasurface (Gorp et al., 12 Feb 2025). Another misconception is that high-$6$52 membrane resonances are intrinsically reflection-only; transmission-mode BIC-EIT membranes with $6$53 and protein-monolayer sensing directly contradict that assumption (Rosas et al., 2024).
A plausible implication is that the field is moving toward tighter integration of membrane mechanics, resonance engineering, and active control. The ingredients already exist separately: flat and strain-tunable single-crystalline membranes (Shchepetov et al., 2013), mechanically reconfigurable q-BIC metasurfaces (Gorp et al., 12 Feb 2025), air-accessible strong-coupling cavities (Adi et al., 2024, Rosas et al., 2024), nonlinear symmetry-engineered functionality (Tonkaev et al., 18 Aug 2025), and wafer-scale active ultra-high-$6$54 modulation (Brikh et al., 27 Sep 2025). Taken together, they define single-crystalline silicon membrane metasurfaces as a distinct nanophotonic platform in which optical confinement, near-field accessibility, mechanical compliance, and semiconductor process compatibility are co-located in the same crystalline membrane.