Membrane-Based Opto-Electromechanical Transducer

• The transducer is a hybrid device that co-localizes optical, microwave, and mechanical modes to enable coherent signal exchange.
• It relies on radiation-pressure and capacitive or piezoelectric coupling, using red-detuned pumps to linearize nonlinear interactions into effective state transfer.
• Various membrane geometries—including membrane-in-the-middle, integrated photonic crystal, and membrane-at-the-edge—offer tailored approaches for RF modulation and quantum networking.

Searching arXiv for the named paper and closely related membrane-based opto-electromechanical transducer work. A membrane-based opto-electromechanical transducer is a hybrid device in which a mechanical membrane, or a membrane-like nanomechanical resonator, serves as a shared dynamical degree of freedom that couples simultaneously to an optical cavity mode and to an electrical or microwave circuit mode. In this architecture, optical coupling is typically radiation-pressure or dispersive cavity optomechanical coupling, while electrical coupling is capacitive or piezoelectric; together they enable direct linkage or reversible conversion between electrical and optical signals through mechanical motion. Within Nano-Opto-Electro-Mechanical Systems (NOEMS), membrane platforms are described as a central route to coherent, bidirectional signal conversion between microwave and optical domains, spanning membrane-in-the-middle Fabry–Pérot cavities, integrated photonic-crystal membranes, membrane-at-the-edge geometries, and atomically thin resonators (Tian, 2014, Midolo et al., 2018).

1. Fundamental transduction model

The standard description is a three-mode system comprising an optical mode $a$, a microwave or electrical mode $b$, and a mechanical membrane mode $c$. In the review treatment of optoelectromechanical conversion, the full Hamiltonian is written as

$$\frac{H}{\hbar} = \omega_o a^\dagger a + \omega_e b^\dagger b + \omega_m c^\dagger c \;-\; g_o\, a^\dagger a (c + c^\dagger) \;-\; g_e\, (b + b^\dagger)(c + c^\dagger),$$

or, in a closely related NOEMS formulation,

$$H = \hbar \omega_c a^\dagger a + \hbar \Omega_m b^\dagger b + \hbar \omega_e c^\dagger c - \hbar g_0 a^\dagger a (b + b^\dagger) + \hbar g_{em} (b c^\dagger + b^\dagger c).$$

The membrane zero-point motion is

$$x_{\mathrm{zpf}} = \sqrt{\frac{\hbar}{2 m_{\mathrm{eff}} \omega_m}},$$

and the single-photon optomechanical coupling is

$$g_0 = \left(\frac{\partial \omega_o}{\partial x}\right) x_{\mathrm{zpf}}.$$

For capacitive electromechanics, the analogous coupling derives from $\partial \omega_e/\partial x$ through the displacement dependence of the circuit capacitance (Tian, 2014, Midolo et al., 2018).

Operation for coherent conversion is usually placed in the resolved-sideband regime, $\omega_m \gg \kappa_o,\kappa_e$, with both pumps red detuned by one mechanical frequency, $\Delta_o \approx -\omega_m$ and $b$0. Under strong red-detuned pumps, the nonlinear interactions linearize into beam-splitter terms,

$b$1

with $b$2. In this regime, the membrane mediates coherent state transfer $b$3, either in continuous-wave operation or through pulsed protocols such as double-swap and adiabatic dark-mode transfer. The corresponding dark mode,

$b$4

suppresses mechanical participation during intracavity transfer (Tian, 2014).

The central efficiency figure of merit is set by cooperativity. For itinerant conversion, the standard on-resonance internal efficiency is

$b$5

Impedance matching corresponds to $b$6, or equivalently $b$7 with $b$8. In the weak-coupling but sideband-resolved regime, the conversion bandwidth is set by the mechanically broadened linewidth,

$b$9

while in the strong-coupling regime it is bounded by $c$0 (Tian, 2014, Midolo et al., 2018).

2. Membrane geometries and device architectures

The most established architecture is the membrane-in-the-middle Fabry–Pérot system, in which a thin dielectric membrane is placed inside an optical cavity while the same membrane forms a position-dependent capacitance to a superconducting LC or CPW resonator. This configuration was identified as the membrane platform that achieved 47% classical conversion efficiency with matched opto- and electromechanical cooperativities of $c$1, with efficiency limited mainly by external coupling and optical mode matching (Chu et al., 2020).

Integrated membrane devices replace the centimeter-scale Fabry–Pérot cavity with nanophotonic resonators. A representative example is the all-integrated silicon-on-insulator transducer based on three released silicon membranes, where a slotted photonic-crystal cavity is formed by the middle and lower membranes and capacitive electrodes are patterned across the upper and middle membranes. The host photonic crystal uses a triangular lattice with lattice constant $c$2 nm and hole radius $c$3; the slotted waveguide has cavity slot width $c$4 nm, and the drive capacitor has a slot gap of 150 nm with electrode width of 600 nm. In that device, strong simultaneous electromechanical and optomechanical interactions enable efficient electrical excitation and optical transduction of in-plane bulk acoustic modes up to 4.196 GHz without piezoelectric materials (Sun et al., 2012).

A distinct membrane configuration is the membrane-at-the-edge geometry. Here a thin dielectric membrane is placed only microns from one end mirror of a Fabry–Perot cavity, producing a short input-side sub-cavity of length $c$5 and a long backstop-side sub-cavity of length $c$6. Passive alignment is achieved by flexing the silicon support frame against 21 $c$7m photoresist spacers on the input mirror. Gentle flexure with radius of curvature $c$8 m accesses the full range of optomechanical couplings, while aggressive flexure with radius $c$9 m yields membrane travel $$\frac{H}{\hbar} = \omega_o a^\dagger a + \omega_e b^\dagger b + \omega_m c^\dagger c \;-\; g_o\, a^\dagger a (c + c^\dagger) \;-\; g_e\, (b + b^\dagger)(c + c^\dagger),$$0 $$\frac{H}{\hbar} = \omega_o a^\dagger a + \omega_e b^\dagger b + \omega_m c^\dagger c \;-\; g_o\, a^\dagger a (c + c^\dagger) \;-\; g_e\, (b + b^\dagger)(c + c^\dagger),$$1m, tilt tuning over $$\frac{H}{\hbar} = \omega_o a^\dagger a + \omega_e b^\dagger b + \omega_m c^\dagger c \;-\; g_o\, a^\dagger a (c + c^\dagger) \;-\; g_e\, (b + b^\dagger)(c + c^\dagger),$$2 mrad, and a membrane–mirror separation $$\frac{H}{\hbar} = \omega_o a^\dagger a + \omega_e b^\dagger b + \omega_m c^\dagger c \;-\; g_o\, a^\dagger a (c + c^\dagger) \;-\; g_e\, (b + b^\dagger)(c + c^\dagger),$$3 $$\frac{H}{\hbar} = \omega_o a^\dagger a + \omega_e b^\dagger b + \omega_m c^\dagger c \;-\; g_o\, a^\dagger a (c + c^\dagger) \;-\; g_e\, (b + b^\dagger)(c + c^\dagger),$$4m. Unlike membrane-in-the-middle cavities, the optical spectrum evolves smoothly over wavelength-scale displacements, with none of the abrupt discontinuities typical of membrane-in-the-middle systems (Dumont et al., 2019).

Other membrane realizations extend the concept rather than alter it. Vertically coupled Si$$\frac{H}{\hbar} = \omega_o a^\dagger a + \omega_e b^\dagger b + \omega_m c^\dagger c \;-\; g_o\, a^\dagger a (c + c^\dagger) \;-\; g_e\, (b + b^\dagger)(c + c^\dagger),$$5N$$\frac{H}{\hbar} = \omega_o a^\dagger a + \omega_e b^\dagger b + \omega_m c^\dagger c \;-\; g_o\, a^\dagger a (c + c^\dagger) \;-\; g_e\, (b + b^\dagger)(c + c^\dagger),$$6 nanomembranes form a monolithic optomechanical array with an intermembrane separation of $$\frac{H}{\hbar} = \omega_o a^\dagger a + \omega_e b^\dagger b + \omega_m c^\dagger c \;-\; g_o\, a^\dagger a (c + c^\dagger) \;-\; g_e\, (b + b^\dagger)(c + c^\dagger),$$7 $$\frac{H}{\hbar} = \omega_o a^\dagger a + \omega_e b^\dagger b + \omega_m c^\dagger c \;-\; g_o\, a^\dagger a (c + c^\dagger) \;-\; g_e\, (b + b^\dagger)(c + c^\dagger),$$8m and piezoelectrically controlled stress tuning, enabling strong intermode electromechanical coupling near degeneracy (Naserbakht et al., 2019). Atomically thin membrane resonators based on van der Waals heterostructures push the membrane concept to the GHz regime: a doubly clamped 2D membrane of dimensions $$\frac{H}{\hbar} = \omega_o a^\dagger a + \omega_e b^\dagger b + \omega_m c^\dagger c \;-\; g_o\, a^\dagger a (c + c^\dagger) \;-\; g_e\, (b + b^\dagger)(c + c^\dagger),$$9 can voltage-tune its fundamental flexural mode from about 1.83 GHz up to $$H = \hbar \omega_c a^\dagger a + \hbar \Omega_m b^\dagger b + \hbar \omega_e c^\dagger c - \hbar g_0 a^\dagger a (b + b^\dagger) + \hbar g_{em} (b c^\dagger + b^\dagger c).$$0 GHz and can host a single-photon emitter for microwave–phonon–optical transduction (Gao et al., 2017).

3. Coupling mechanisms and optical readout

Radiation-pressure optomechanical coupling is dispersive: membrane displacement modulates the optical resonance frequency $$H = \hbar \omega_c a^\dagger a + \hbar \Omega_m b^\dagger b + \hbar \omega_e c^\dagger c - \hbar g_0 a^\dagger a (b + b^\dagger) + \hbar g_{em} (b c^\dagger + b^\dagger c).$$1, producing a force proportional to intracavity photon number. In membrane platforms this is often quantified by the frequency-pull parameter $$H = \hbar \omega_c a^\dagger a + \hbar \Omega_m b^\dagger b + \hbar \omega_e c^\dagger c - \hbar g_0 a^\dagger a (b + b^\dagger) + \hbar g_{em} (b c^\dagger + b^\dagger c).$$2. In the slotted photonic-crystal membrane device, the TE-like cavity mode localizes strongly inside the 80 nm slot, giving a moved-boundary frequency-pull parameter

$$H = \hbar \omega_c a^\dagger a + \hbar \Omega_m b^\dagger b + \hbar \omega_e c^\dagger c - \hbar g_0 a^\dagger a (b + b^\dagger) + \hbar g_{em} (b c^\dagger + b^\dagger c).$$3

Mechanical displacement modulates the cavity frequency as

$$H = \hbar \omega_c a^\dagger a + \hbar \Omega_m b^\dagger b + \hbar \omega_e c^\dagger c - \hbar g_0 a^\dagger a (b + b^\dagger) + \hbar g_{em} (b c^\dagger + b^\dagger c).$$4

and when the laser is placed on the steep slope of the cavity transmission this frequency modulation is converted into amplitude or phase modulation of the transmitted light. A useful metric is the phase modulation index $$H = \hbar \omega_c a^\dagger a + \hbar \Omega_m b^\dagger b + \hbar \omega_e c^\dagger c - \hbar g_0 a^\dagger a (b + b^\dagger) + \hbar g_{em} (b c^\dagger + b^\dagger c).$$5; a 1 pm in-plane displacement yields $$H = \hbar \omega_c a^\dagger a + \hbar \Omega_m b^\dagger b + \hbar \omega_e c^\dagger c - \hbar g_0 a^\dagger a (b + b^\dagger) + \hbar g_{em} (b c^\dagger + b^\dagger c).$$6 for $$H = \hbar \omega_c a^\dagger a + \hbar \Omega_m b^\dagger b + \hbar \omega_e c^\dagger c - \hbar g_0 a^\dagger a (b + b^\dagger) + \hbar g_{em} (b c^\dagger + b^\dagger c).$$7 rad/s in that system (Sun et al., 2012).

Capacitive electromechanical coupling is equally direct. For a voltage

$$H = \hbar \omega_c a^\dagger a + \hbar \Omega_m b^\dagger b + \hbar \omega_e c^\dagger c - \hbar g_0 a^\dagger a (b + b^\dagger) + \hbar g_{em} (b c^\dagger + b^\dagger c).$$8

the electrostatic force is

$$H = \hbar \omega_c a^\dagger a + \hbar \Omega_m b^\dagger b + \hbar \omega_e c^\dagger c - \hbar g_0 a^\dagger a (b + b^\dagger) + \hbar g_{em} (b c^\dagger + b^\dagger c).$$9

with the near-resonant component

$$x_{\mathrm{zpf}} = \sqrt{\frac{\hbar}{2 m_{\mathrm{eff}} \omega_m}},$$0

The membrane response then follows

$$x_{\mathrm{zpf}} = \sqrt{\frac{\hbar}{2 m_{\mathrm{eff}} \omega_m}},$$1

This linear-in-$$x_{\mathrm{zpf}} = \sqrt{\frac{\hbar}{2 m_{\mathrm{eff}} \omega_m}},$$2 drive exploits the strong capacitance gradient of nanoscale electrode gaps; in integrated silicon membrane devices, 150 nm gaps were used specifically to maximize $$x_{\mathrm{zpf}} = \sqrt{\frac{\hbar}{2 m_{\mathrm{eff}} \omega_m}},$$3 (Sun et al., 2012).

The membrane-at-the-edge geometry broadened the coupling taxonomy by deriving analytical expressions not only for linear dispersive coupling but also for quadratic dispersive and linear dissipative couplings. A notable result is that the maximal linear coupling $$x_{\mathrm{zpf}} = \sqrt{\frac{\hbar}{2 m_{\mathrm{eff}} \omega_m}},$$4 is larger than in a membrane-in-the-middle system by a factor that tends to $$x_{\mathrm{zpf}} = \sqrt{\frac{\hbar}{2 m_{\mathrm{eff}} \omega_m}},$$5 in the $$x_{\mathrm{zpf}} = \sqrt{\frac{\hbar}{2 m_{\mathrm{eff}} \omega_m}},$$6 limit, while the maximum linear dispersive strong-coupling parameter $$x_{\mathrm{zpf}} = \sqrt{\frac{\hbar}{2 m_{\mathrm{eff}} \omega_m}},$$7 is unchanged because the linewidth $$x_{\mathrm{zpf}} = \sqrt{\frac{\hbar}{2 m_{\mathrm{eff}} \omega_m}},$$8 also increases. By contrast, the maximal quadratic dispersive coupling scales as $$x_{\mathrm{zpf}} = \sqrt{\frac{\hbar}{2 m_{\mathrm{eff}} \omega_m}},$$9, and the quadratic dispersive strong-coupling parameter is enhanced. At purely quadratic points, the membrane-at-the-edge geometry can reproduce the canonical membrane-in-the-middle quadratic strength while suppressing linear dissipative back-action by the small geometric factor $$g_0 = \left(\frac{\partial \omega_o}{\partial x}\right) x_{\mathrm{zpf}}.$$0 when the membrane is placed near the back mirror (Dumont et al., 2019). A common misconception is therefore that larger linear dispersive slope alone implies a proportionally improved strong-coupling figure of merit; in this geometry, it does not.

Optical transduction need not rely only on cavity-transmission slope readout. An interferometrically robust polarization transducer places a Si$$g_0 = \left(\frac{\partial \omega_o}{\partial x}\right) x_{\mathrm{zpf}}.$$1N$$g_0 = \left(\frac{\partial \omega_o}{\partial x}\right) x_{\mathrm{zpf}}.$$2 membrane in an asymmetric cavity and maps displacement onto the polarization of a reflected beam; near the operating point, the total reflectivity reduces to

$$g_0 = \left(\frac{\partial \omega_o}{\partial x}\right) x_{\mathrm{zpf}}.$$3

and balanced homodyne polarimetry measures the corresponding Stokes component with a common-path reference beam (Abbas et al., 2023). A different readout strategy uses a compound cavity formed by a Littrow external-cavity diode laser and a reflective trampoline membrane; membrane motion modulates the external optical feedback phase and amplitude re-injected into the diode, converting displacement into output-power fluctuations through laser self-mixing. In that platform, $$g_0 = \left(\frac{\partial \omega_o}{\partial x}\right) x_{\mathrm{zpf}}.$$4 in the linear regime, and the fundamental trampoline mode at $$g_0 = \left(\frac{\partial \omega_o}{\partial x}\right) x_{\mathrm{zpf}}.$$5 kHz was read out directly through the integrated photodiode of the laser (Baldacci et al., 2016).

4. Representative implementations and measured regimes

The membrane category spans MHz Fabry–Pérot devices, multi-GHz integrated nanophotonics, room-temperature RF sensors, and cryogenic few-photon transducers. The following examples illustrate the range of experimentally reported operating points.

Implementation	Membrane architecture	Reported metrics
Slotted photonic-crystal silicon transducer (Sun et al., 2012)	Three released silicon membranes on SOI; slotted PhC cavity plus 150 nm capacitive actuator	$$g_0 = \left(\frac{\partial \omega_o}{\partial x}\right) x_{\mathrm{zpf}}.$$6 GHz; mechanical $$g_0 = \left(\frac{\partial \omega_o}{\partial x}\right) x_{\mathrm{zpf}}.$$7–$$g_0 = \left(\frac{\partial \omega_o}{\partial x}\right) x_{\mathrm{zpf}}.$$8; $$g_0 = \left(\frac{\partial \omega_o}{\partial x}\right) x_{\mathrm{zpf}}.$$9; $\partial \omega_e/\partial x$0; $\partial \omega_e/\partial x$1 GHz/nm
Bidirectional membrane microwave–optical converter (Tian, 2014)	SiN membrane in Fabry–Pérot cavity with capacitive coupling to a superconducting microwave resonator	Mechanical mode near $\partial \omega_e/\partial x$2 kHz; optical frequency $\partial \omega_e/\partial x$3 THz; microwave $\partial \omega_e/\partial x$4 GHz; efficiency $\partial \omega_e/\partial x$5
Multi-mode RF-to-optical transducer (Haghighi et al., 2017)	1 mm $\partial \omega_e/\partial x$6 1 mm, 50 nm SiN membrane with 27 nm Nb coating, coupled to an LC circuit and optical homodyne interferometer	$\partial \omega_e/\partial x$7 nV/Hz$\partial \omega_e/\partial x$8 over $\partial \omega_e/\partial x$9 kHz in the presence of radiofrequency noise; optimal shot-noise limited sensitivity of $\omega_m \gg \kappa_o,\kappa_e$0 nV/Hz$\omega_m \gg \kappa_o,\kappa_e$1 over $\omega_m \gg \kappa_o,\kappa_e$2 kHz
OEMM for RF-to-optical transduction (Bonaldi et al., 2023)	Circular high-stress SiN nanomembrane partially metalized with TiN, intended for membrane-in-the-middle cavity integration	$\omega_m \gg \kappa_o,\kappa_e$3-factor above $\omega_m \gg \kappa_o,\kappa_e$4 at room temperature; optically measured steady-state frequency shift of $\omega_m \gg \kappa_o,\kappa_e$5 Hz with a polarization voltage of $\omega_m \gg \kappa_o,\kappa_e$6 V; average gap $\omega_m \gg \kappa_o,\kappa_e$7m
High-throughput cryogenic membrane transducer (Urmey et al., 14 Jul 2025)	Tensile Si$\omega_m \gg \kappa_o,\kappa_e$8N$\omega_m \gg \kappa_o,\kappa_e$9 membrane coupled to a superconducting microwave LC resonator and a Fabry–Perot cavity	$\Delta_o \approx -\omega_m$0 in both directions; $\Delta_o \approx -\omega_m$1 kHz upconversion and $\Delta_o \approx -\omega_m$2 kHz downconversion; $\Delta_o \approx -\omega_m$3 kHz; $\Delta_o \approx -\omega_m$4 photons upconversion and $\Delta_o \approx -\omega_m$5 photons downconversion

The broader membrane-in-the-middle platform, as summarized in the 2020 perspective, remains notable because it was highlighted as the current record-holder for classical conversion efficiency: 47% with matched cooperativities of $\Delta_o \approx -\omega_m$6, limited mainly by external coupling and optical mode matching rather than by the underlying membrane interaction itself (Chu et al., 2020).

Recent cryogenic work also shifted emphasis from peak efficiency alone to throughput. In the 2025 doubly parametric membrane transducer, the efficiency–bandwidth–duty-cycle product

$\Delta_o \approx -\omega_m$7

reached approximately 7 kHz with continuous operation, $\Delta_o \approx -\omega_m$8, and few-photon added noise. The same work introduced an integrated quantum-capacity analysis for the transducer as a bosonic thermal-loss channel, showing that current measurements with $\Delta_o \approx -\omega_m$9–3 give zero two-way assisted quantum capacity, whereas future operation with $b$00 at the same throughput would enter the quantum-enabled regime (Urmey et al., 14 Jul 2025).

5. Noise, dissipation, and practical constraints

The dominant noise source in membrane transducers is usually the mechanical bath. Thermal occupancy

$b$01

enters the converted output unless suppressed by cryogenic operation and sideband cooling. A standard scaling for input-referred added noise is

$b$02

while the NOEMS review gives the complementary form $b$03, emphasizing the need for high quantum cooperativity to suppress thermal noise quanta (Tian, 2014, Midolo et al., 2018).

Internal optical and microwave losses are equally consequential. The 2020 perspective identifies thermal noise of the mechanical bath, optical absorption heating at millikelvin temperatures, microwave internal loss, and external-coupling inefficiency as the central bottlenecks in membrane-in-the-middle transducers. It also notes that the highest-efficiency membrane platform still faces the challenge that MHz mechanical modes carry large thermal occupancy unless aggressively cooled (Chu et al., 2020). In practice, transducers must therefore maximize $b$04 and $b$05, reduce $b$06 and $b$07, and balance the two cooperativities rather than increasing only one side.

Integrated membrane systems add additional constraints. In the slotted photonic-crystal transducer, mechanical quality factors were attributed to in-plane motion and reduced clamping loss, but losses still arise from clamping or anchor radiation, electrode-induced damping, optical absorption in metals, fabrication imperfections that reduced optical $b$08 to approximately five times below simulated expectation, and electrical stress that caused damage above $b$09 V across the capacitor slot (Sun et al., 2012). Vacuum below $b$10 mbar was used there to suppress viscous damping.

Photothermal coupling can also dominate the dynamical back-action of a membrane device. In suspended silicon photonic-crystal membranes with tunable membrane–substrate spacing, the platform operated in the sideband-unresolved regime with $b$11, and photothermal-mechanical feedback dominated the damping while optomechanical back-action contributed primarily to the spring at small gap. That work reported repulsive coupling as large as $b$12 GHz/nm at $b$13 nm, together with cooling to $b$14 K at $b$15 mW and regenerative oscillation at $b$16 $b$17W (Woolf et al., 2012). This shows that large $b$18 does not by itself imply a purely radiation-pressure-limited transducer; parasitic photothermal channels can set the observed linewidth and stability.

In the recent high-throughput cryogenic membrane transducer, residual Stokes leakage from finite sideband resolution produced a small parametric gain $b$19 that multiplied the ideal efficiency and contributed correlated noise terms. Optical pump noise was nevertheless made negligible, with optical phase noise near $b$20 of approximately $b$21 dBc/Hz after filtering and circulating optical power of approximately 3 mW. The remaining asymmetry between upconversion and downconversion was traced mainly to microwave bath occupation that depended linearly on pump strength, $b$22 (Urmey et al., 14 Jul 2025).

6. Applications, materials directions, and outlook

The immediate application space is microwave photonics. The integrated slotted photonic-crystal membrane system was explicitly presented as a platform for narrowband RF–optical modulators, filters, and signal processors, for coherent frequency conversion, and for low-phase-noise optoacoustic oscillators, with superhigh-frequency operation up to 4.20 GHz broadening the usable spectrum for integrated RF photonics (Sun et al., 2012). At lower frequencies, multi-mode membrane transducers offer a different operating point: by engineering constructive interference between two electromechanical pathways, a membrane device can increase bandwidth at fixed sensitivity, as demonstrated by the 15 kHz room-temperature bandwidth obtained in the presence of radiofrequency noise (Haghighi et al., 2017).

Quantum networking motivates the most stringent version of the problem. The 2025 cryogenic membrane transducer frames performance through both throughput and added noise, because the relevant target is entanglement distribution between superconducting processors at rates competitive with memory lifetimes of approximately $b$23 $b$24s–1 ms. That work shows that once $b$25, further reductions in noise give diminishing returns while throughput increases improve the rate linearly in the small-$b$26 limit. This suggests that membrane platforms able to tolerate strong pumping without optical pump-induced noise are strategically attractive for network demonstrations (Urmey et al., 14 Jul 2025).

The membrane category also includes platforms designed for QND-like sensing rather than maximal linear conversion. In the membrane-at-the-edge geometry, purely quadratic dispersive points give the same quadratic strength as membrane-in-the-middle operation, while placing the membrane near the back mirror suppresses linear dissipative back-action by $b$27. That geometry was explicitly proposed as useful for position-squared conversion and QND-like energy sensing (Dumont et al., 2019).

Materials development is becoming increasingly central. High-stress Si$b$28N$b$29 remains the baseline because it combines low mass, high tensile stress, and high mechanical $b$30, but conductive or superconducting membranes reduce the penalty of adding metal for capacitive coupling. Ultra-high-stress crystalline TiN membranes provide a prominent example: a square 100 nm thick TiN membrane of side length $b$31 $b$32m achieved tensile stress exceeding 2.3 GPa below 100 K and a quality factor $b$33 at 2.2 K, with $b$34 for the fundamental mode. Because TiN is electrically conductive and superconducting below about 5 K, it can serve directly as the moving electrode of a microwave capacitor without added metallization, and the authors explicitly identify such membranes as a powerful tool for opto- and electromechanical systems (Matsuyama et al., 3 Sep 2025).

At the opposite geometric extreme, atomically thin 2D membranes illustrate how far the membrane concept can be pushed. In the proposed GHz transducer based on a freestanding heterostructure hosting a single-photon emitter, the electromechanical coupling reaches $b$35 MHz for $b$36 V, the Stark-induced emitter–phonon coupling reaches $b$37 MHz for $b$38 V, and numerical simulations show transfer fidelity larger than 0.95 including realistic dark counts within $b$39–200 ns windows (Gao et al., 2017). This does not replace the high-$b$40 MHz membrane paradigm; rather, it indicates that membrane-based transduction encompasses both ultracoherent low-frequency mechanical mediators and genuinely high-speed single-photon interfaces.

Taken together, the literature portrays membrane-based opto-electromechanical transducers not as a single device class but as a design family organized around one principle: co-localizing optical, electrical, and mechanical interactions in a membrane degree of freedom. The empirical frontier is defined by balanced cooperativity, high external coupling, low internal loss, suppression of thermal and pump-induced noise, and membrane engineering that preserves mechanical coherence while increasing optical and electrical participation. The trajectory from 10% bidirectional conversion and $b$41 added quanta, through 47% classical efficiency, to continuous-operation throughput of $b$42 kHz with few-photon added noise indicates that the membrane platform has moved from proof-of-principle transduction toward quantitatively optimized, application-specific interfaces for microwave photonics and hybrid quantum networks (Midolo et al., 2018, Chu et al., 2020, Urmey et al., 14 Jul 2025).