Buckled Dielectric Membrane Mirrors

• Buckled dielectric membrane mirrors are suspended dielectric films whose curvature is generated via controlled stress-induced buckling, enabling concave high-finesse surfaces.
• They are fabricated using multilayer dielectric stacks and precise lithographic patterning, with stress release triggering tunable radii of curvature.
• Their application in Fabry–Perot microcavities achieves ultra-low optical loss and high finesse, supporting compact and scalable photonic integration.

Buckled dielectric membrane mirrors are suspended dielectric films or multilayer dielectric stacks whose optical surface is formed by controlled out-of-plane deformation rather than by conventional polishing or direct micromachining. In the stress-driven implementations reported for Fabry–Perot microcavities, intrinsic compressive stress or guided delamination causes a freestanding dielectric membrane to assume a concave or dome-like profile while preserving the low-loss surface quality of the deposited mirror coating, enabling curved-mirror resonators in compact and scalable formats (Ding et al., 28 Sep 2025). Earlier on-chip work established curved-mirror Fabry–Perot resonators based on circular delamination buckles within multilayer a-Si/SiO$_2$ stacks (Bitarafan et al., 2015). Together, these studies define a class of mirrors in which mechanical buckling is not a defect to be suppressed but the mechanism by which optical curvature is created.

1. Definition and architectural scope

Buckled dielectric membrane mirrors are most precisely understood as dielectric mirror elements in which the reflecting surface is a suspended membrane and the mirror curvature emerges from buckling of that membrane. In the near-infrared microcavity platform reported in 2025, the mirror is a freestanding dielectric membrane produced from a high-reflectivity ion-beam-sputtered stack; after backside release, built-in compressive stress causes the membrane to self-buckle into a controlled concave shape (Ding et al., 28 Sep 2025). In the earlier monolithic platform, the top mirror of a Fabry–Perot resonator is formed by guided assembly of circular delamination buckles within a multilayer a-Si/SiO$_2$ stack, yielding dome-like curved mirrors directly on chip (Bitarafan et al., 2015).

Within this class, the central optical objective is to combine the low loss of high-quality dielectric surfaces with the geometric advantages of microscopic concave mirrors. The recent visible/NIR work frames this explicitly as an alternative to super-polished macroscopic substrates and precision dielectric coatings, which provide excellent performance but are large, hard to scale, and incompatible with compact photonic integration (Ding et al., 28 Sep 2025). The on-chip dome-cavity work emphasizes a related point: because the curved optical surface is produced by thin-film buckling rather than by sculpting, the optical roughness is set mainly by film deposition (Bitarafan et al., 2015).

A recurring distinction is that not every suspended dielectric membrane mirror is a buckled dielectric membrane mirror. Some membrane mirrors remain intentionally flat, and others generate effective curvature by patterned optical phase rather than by mechanical deformation. That distinction is essential for delimiting the term.

2. Stress-driven formation and fabrication routes

The highest-finesse buckled membrane platform begins with a commercially available 4-inch double-sided polished Si wafer whose as-received RMS roughness is about 0.3 nm (Ding et al., 28 Sep 2025). Surface preparation is performed by “wet thermal smoothing”: about 2.5 $\mu$m of wet thermal oxide is grown and then removed using 49% HF, with the oxidize-and-strip step repeated 2–3 times. After this treatment, AFM gives surface roughness as low as 50 pm down to 33 pm on a $2 \,\mu\text{m} \times 2 \,\mu\text{m}$ area. A high-reflectivity dielectric coating is then deposited by ion beam sputtering. The reported mirror design comprises 21 quarter-wave pairs of alternating $\text{SiO}_2$ and $\text{Ta}_2\text{O}_5$, plus a final SiO$_2$ capping layer described as a half-wave protective stack, with total thickness about 4.74–5 $\mu$m and vendor-reported transmission loss $T = 1.33$ ppm at 781 nm. Mirror geometry is defined lithographically on the backside, and the silicon underneath is removed using a Bosch deep reactive ion etch with SF$_6$ and CF$_2$0, leaving a freestanding dielectric membrane mirror (Ding et al., 28 Sep 2025).

The monolithic dome-cavity route is materially and structurally different. There the curved top mirror is assembled from a 4-period quarter-wave stack of amorphous silicon and silicon dioxide, plus a half-wave a-Si capping layer, deposited by magnetron sputtering (Bitarafan et al., 2015). Circular delamination regions under built-in compressive stress buckle upward into dome-like structures. The typical total mirror thickness is about 1.6 $_2$1m, and the pre-buckling compressive stress is about 180 MPa. The resulting cavity is half-symmetric, with a flat bottom mirror and a buckled top mirror.

These fabrication routes differ in mechanism—released suspended membranes versus circular delamination buckles—but they share the same strategic feature: the optical surface quality is inherited from thin-film deposition. This suggests that buckling can serve as a curvature-generation step decoupled from the processes that set low optical loss.

3. Buckling mechanics, profile control, and geometric tuning

In released dielectric-stack mirrors, buckling is induced by built-in compressive stress from deposition. Once the silicon underneath is etched away, the suspended multilayer membrane relaxes by moving out of plane and forms a concave buckled profile (Ding et al., 28 Sep 2025). The reported deformation is highly controlled by geometry and stress rather than random wrinkling. For showcased arrays the mirrors are uniform circular mirrors with diameter around 600 $_2$2m, and the radius of curvature is tunable by lithographic diameter. COMSOL modeling uses a uniform compressive strain of 0.3% and a hard-clamped perimeter boundary condition, with nearly perfect agreement to measured profiles, $_2$3. The supplementary FEM workflow consists of a linear buckling study to find the buckling threshold, a stationary post-buckling study across strain values, calibration to profiler data, and a best-fit strain around $_2$4. The same study emphasizes that the profile is not well described by simple analytic functions like Gaussian or Bessel shapes, so numerical modeling is essential (Ding et al., 28 Sep 2025).

The curvature range is broad. A general fabrication survey reports ROC from 0.9 mm to 21 mm, while demonstrated cavity mirrors include ROC $_2$5 mm and ROC $_2$6 mm; the abstract highlights 1 mm to 10 mm as a representative device class (Ding et al., 28 Sep 2025). The fabricated mirror surfaces are rotationally symmetric, with the two orthogonal cross-sections essentially identical, $_2$7. In a cavity context, this matters because asymmetry or astigmatism would worsen mode matching and introduce excess loss.

The earlier dome-cavity work uses analytical buckling models more directly. For the idealized axisymmetric buckle, the elastic profile is written as

$_2$8

with $_2$9, and the real cavities lie between a purely spherical-dome approximation and a clamped-elastic-buckle model (Bitarafan et al., 2015). For a circular plate, the critical buckling stress is given as

$\mu$0

and, for compressive stress $\mu$1, the buckled peak height is approximated by

$\mu$2

Experimentally measured cross-sections for 100 $\mu$3m and 200 $\mu$4m diameter cavities fall between the elastic-buckle and spherical-dome limits, with smaller cavities closer to the elastic buckle prediction and larger ones more dome-like (Bitarafan et al., 2015).

A transferable control framework comes from voltage-induced buckling of clamped dielectric films. In the microfluidic actuator study, applied voltage generates Maxwell stress in a dielectric film; clamping prevents in-plane expansion, so compressive radial stress builds until the circular plate buckles out of plane (Tavakol et al., 2016). The classical plate description there uses

$\mu$5

with first-mode constant $\mu$6, and the scaling law

$\mu$7

Although that work is not an optical-mirror demonstration, it provides a design language for controllable buckling in clamped dielectric membranes (Tavakol et al., 2016).

4. Optical cavity behavior and performance

For Fabry–Perot resonators based on buckled dielectric membrane mirrors, finesse is treated as the central figure of merit because it directly measures resonator loss (Ding et al., 28 Sep 2025). The reported expressions are

$\mu$8

for round-trip loss $\mu$9, and, more practically for mirror-limited cavities,

$2 \,\mu\text{m} \times 2 \,\mu\text{m}$0

with $2 \,\mu\text{m} \times 2 \,\mu\text{m}$1 transmission, $2 \,\mu\text{m} \times 2 \,\mu\text{m}$2 scattering loss, and $2 \,\mu\text{m} \times 2 \,\mu\text{m}$3 absorption loss per mirror. The same study also emphasizes the scaling

$2 \,\mu\text{m} \times 2 \,\mu\text{m}$4

which makes ultrahigh finesse substantially harder at 780 nm than at telecom wavelengths (Ding et al., 28 Sep 2025).

The central experimental result in the released-membrane platform is a record finesse

$2 \,\mu\text{m} \times 2 \,\mu\text{m}$5

at 780 nm (Ding et al., 28 Sep 2025). The best-performing cavity had linewidth $2 \,\mu\text{m} \times 2 \,\mu\text{m}$6 kHz and finesse $2 \,\mu\text{m} \times 2 \,\mu\text{m}$7; the highest-$2 \,\mu\text{m} \times 2 \,\mu\text{m}$8 point had linewidth $2 \,\mu\text{m} \times 2 \,\mu\text{m}$9 kHz and $\text{SiO}_2$0. Cavities were measured over

$\text{SiO}_2$1

with finesse decreasing at long length because diffraction loss grows as the mode samples more of the non-ideal mirror shape. Using the quoted mirror transmission $\text{SiO}_2$2 ppm per mirror at 781 nm, the measured finesse implies total nontransmission loss $\text{SiO}_2$3 ppm; assuming no absorption gives a bound $\text{SiO}_2$4 ppm, corresponding to an upper-bound roughness $\text{SiO}_2$5 pm. The same loss values extrapolate to $\text{SiO}_2$6 at telecom (Ding et al., 28 Sep 2025).

The measurement protocol uses two buckled mirrors face-to-face on translation stages, one mounted on a piezoelectric transducer, and a 780 nm laser locked to a rubidium resonance at 780.24 nm (Ding et al., 28 Sep 2025). A swept cavity ring-down technique is employed: as the cavity resonance sweeps past the fixed laser frequency, the transmission decays and is fit to a model of ring-down fringes written in the form

$\text{SiO}_2$7

The free spectral range is

$\text{SiO}_2$8

and the transverse-mode spectrum is described by

$\text{SiO}_2$9

Compact packaged devices are also demonstrated. The reported easy-to-assemble microcavity packages have total volume around 2 mm$\text{Ta}_2\text{O}_5$0 and 4 mm$\text{Ta}_2\text{O}_5$1 (Ding et al., 28 Sep 2025). For the short assembled cavity, $\text{Ta}_2\text{O}_5$2, $\text{Ta}_2\text{O}_5$3 MHz, $\text{Ta}_2\text{O}_5$4 THz, $\text{Ta}_2\text{O}_5$5, and $\text{Ta}_2\text{O}_5$6. For the longer assembled cavity with a spacer, the cavity stack length is 2 mm, $\text{Ta}_2\text{O}_5$7, and $\text{Ta}_2\text{O}_5$8. These devices are assembled by manual alignment and heat-cure glue.

The earlier dome-cavity platform operates in a different performance regime but clarifies the broader optical significance of buckled dielectric mirrors (Bitarafan et al., 2015). For a 100 $\text{Ta}_2\text{O}_5$9m diameter cavity, the dome peak height is about 2.4 $_2$0m and the mirror radius of curvature is about 270 $_2$1m. Using effective cavity length $_2$2 and the paraxial waist estimate

$_2$3

the predicted waist is $_2$4m, while imaging the TEM$_2$5 mode gives $_2$6m. The measured resonance linewidth is about 0.16 nm, corresponding to $_2$7; with longitudinal mode order $_2$8, the finesse is

$_2$9

The reported spectra are taken at low power because higher optical powers cause photo-thermal bistability and hysteresis (Bitarafan et al., 2015).

5. Thermomechanical response and active control

Buckled dielectric membrane mirrors are mechanically compliant optical elements, and their thermomechanical behavior is intrinsic to device operation. In the on-chip dome resonators, temperature changes alter the dome height and therefore the cavity length through differential thermal expansion between the mirror stack and the silicon substrate (Bitarafan et al., 2015). The reported temperature dependence of buckle height is

$\mu$0

with $\mu$1. For the 100 $\mu$2m dome with $\mu$3m, this predicts a resonance shift of about 0.7 nm/K; experimentally, the resonance red-shifts by about 1 nm/K. The same work notes that such strong thermal sensitivity is useful for tuning but can be a drawback for cavity QED, where resonance stability on the order of 1 pm may be required, implying temperature stabilization at the millikelvin level or better (Bitarafan et al., 2015).

The same resonators also exhibit mechanical resonances in the MHz range (Bitarafan et al., 2015). For the 200 $\mu$4m cavity, shell and buckled-plate models predict a fundamental frequency near 1.6 MHz, whereas a flat-plate estimate gives about 430 kHz; the measured lowest-order resonance is in the MHz range and agrees well with the shell/buckled-plate models, especially for the larger cavities. The paper further reports stiffness estimates around $\mu$5 N/m across cavity sizes. This mechanical behavior is relevant for sensing, optomechanics, and any application in which the optical mode and the lowest mechanical mode are co-localized to the central dome region (Bitarafan et al., 2015).

Voltage-induced buckling offers an explicit actuation pathway. In the microfluidic dielectric-film study, the membrane is a clamped circular PDMS plate of radius $\mu$6 and thickness on the order of $\mu$7, with one sample at $\mu$8 (Tavakol et al., 2016). The electrostatic loading is described by

$\mu$9

and, for one sample with $T = 1.33$0 and $T = 1.33$1 MPa, the estimated critical buckling voltage is 3.0 kV, while the measured value is $T = 1.33$2 kV. The observed buckled mode is axisymmetric and spherical-cap-like, and the membrane returns quickly to its undeformed state when voltage is removed (Tavakol et al., 2016). A plausible implication is that electrically induced buckling can be used as a shape-control mechanism for dielectric membrane mirrors, although that paper does not address optical quality, wavefront control, hysteresis, or long-term dielectric reliability in mirror service.

6. Related membrane-mirror platforms, distinctions, and misconceptions

A common misconception is to equate all suspended dielectric membrane mirrors with buckled dielectric membrane mirrors. The large-area suspended photonic crystal mirrors reported in 2017 are a clear counterexample (Moura et al., 2017). Those devices are free-standing silicon nitride photonic crystal membranes fabricated from LPCVD SiN, with high intrinsic stress of about 1 GPa; the authors state that this stress should guarantee the membrane’s flatness. Their reflectivity is produced by photonic crystal resonances in a periodic lattice of holes, not by mechanical curvature, and the headline result is greater than 90% reflectivity at 1550 nm on a 56 nm thick suspended mirror, with lateral sizes up to 10 $T = 1.33$3 10 mm. These are suspended dielectric membrane mirrors, but not buckled mirrors in the geometric sense (Moura et al., 2017).

A second boundary case is the focusing membrane metamirror (Agrawal et al., 2024). This device is a suspended, high-reflectivity focusing metamirror realized by non-periodic photonic crystal patterning of a 200-nm-thick Si$T = 1.33$4N$T = 1.33$5 membrane. Its function is to impose a spherical reflection phase

$T = 1.33$6

so that the reflected wavefront is focusing even though the membrane itself is ultrathin. The effective radius of curvature is

$T = 1.33$7

The reported device has $T = 1.33$8 cm and design reflectivity $T = 1.33$9, and it enables a stable, short cavity with $_6$0m and $_6$1 (Agrawal et al., 2024). From the measured transverse mode spectrum, the splitting is consistent with effective radius of curvature $_6$2 cm, corresponding to $_6$3 cm. This is a curved dielectric membrane mirror in optical function, but its curvature is optical rather than mechanically buckled (Agrawal et al., 2024).

These distinctions clarify the scope of the field. In the strict sense, buckled dielectric membrane mirrors are defined by mechanical buckling of a dielectric membrane or dielectric multilayer into a curved optical surface. In a broader landscape of membrane mirrors, flat photonic crystal membranes and focusing metamirrors address some of the same system-level goals—low mass, high reflectivity, compact cavity integration, and mechanical compliance—but they do so through different physical mechanisms. The stress-buckled platforms remain distinctive because they directly convert thin-film stress engineering into mirror curvature while retaining ultra-smooth dielectric surfaces, making them especially relevant where visible/NIR ultrahigh finesse, small radii of curvature, mirror arrays, and compact cavity packages are required (Ding et al., 28 Sep 2025).