Lithium Niobate on Sapphire (LNOS) Platform

• LNOS is a thin-film photonics and phononics platform combining single-crystal LiNbO3 with sapphire for enhanced optical, acoustic, and nonlinear performance.
• It enables efficient fiber-chip coupling, high-speed electro-optic modulation, and mid-infrared nonlinear optics through precise dispersion engineering and quasi-phase matching.
• The platform supports hybrid systems from cryogenic microwave-to-optical transduction to cavity quantum acoustodynamics by leveraging low-loss sapphire and high-χ(2) properties of lithium niobate.

Lithium niobate on sapphire (LNOS; also written LNSa or LNoS in parts of the literature) is a thin-film integrated photonics and phononics platform in which single-crystal LiNbO$_3$ is bonded directly to a sapphire substrate. It is used to combine the large electro-optic and nonlinear response of lithium niobate with sapphire’s wide optical transparency, high thermal conductivity, mechanical stability, high acoustic velocity, and ultra-low microwave dielectric loss. Across recent work, LNOS has been developed for fiber-chip interfaces, mid-infrared nonlinear optics, high-speed electro-optic modulation, gigahertz phononic integrated circuits, acousto-optic devices, cryogenic microwave-to-optical transduction, and proposed cavity quantum acoustodynamics architectures (Chen et al., 2 Oct 2025, Mayor et al., 2020, McKenna et al., 2020, Xu et al., 18 Sep 2025).

1. Material platform and physical basis

LNOS is described as a thin-film material system in which an X-cut LiNbO$_3$ layer is directly bonded to sapphire (Al$_2$O$_3$). Representative demonstrations use LiNbO$_3$ thicknesses of 200 nm, 400 nm, 500 nm, 525 nm, 620 nm, 630 nm, 940 nm, 0.9 $\mu$m, and 1.5 $\mu$m, depending on the intended optical, acoustic, or quantum function. Several studies explicitly note the absence of a buried oxide layer, which distinguishes LNOS from lithium niobate on insulator and is central to both its mid-infrared transparency and its acoustic behavior (Chen et al., 2 Oct 2025, McKenna et al., 2020, Lin et al., 2 Feb 2026).

The optical transparency window is one of the platform’s defining properties. Lithium niobate is reported as transparent from $\sim 350$ nm to 5 $\mu$m, while sapphire extends operation into the visible and, in mid-infrared studies, is described as transparent up to $\approx 4.5$ $_3$0m with negligible absorption up to $_3$1 $_3$2m. In practice, this underpins waveguiding from 400 nm to beyond 4 $_3$3m and enables devices in spectral regions that are inaccessible to SiO$_3$4-clad thin-film lithium niobate because of silica absorption beyond about 2.5–3.4 $_3$5m (Chen et al., 2 Oct 2025, Mishra et al., 2021, Didier et al., 29 May 2025).

The electro-optic and nonlinear figures of merit are inherited from lithium niobate. One study gives a Pockels coefficient $_3$6 pm/V along the extraordinary axis, while a mid-infrared modulator study uses $_3$7 pm/V in its half-wave-voltage estimate. For second-order nonlinear optics, LNOS work cites $_3$8 with $_3$9 pm/V, and the X-cut orientation is used to access tensor elements such as $_2$0, $_2$1, and, in acoustic-piezoelectric settings, $_2$2 or $_2$3 depending on propagation direction and device geometry (Chen et al., 2 Oct 2025, Didier et al., 29 May 2025, Mayor et al., 2020, Xu et al., 18 Sep 2025).

Sapphire also changes the acoustic and microwave context of the platform. Because sapphire is non-piezoelectric and has a higher acoustic velocity than LiNbO$_2$4, several works exploit a phononic analogue of index guiding, strong vertical confinement of surface and guided acoustic modes, and compatibility with superconducting microwave circuits fabricated directly on sapphire. This combination is used experimentally for phononic circuits and electro-optic transduction, and it is proposed as the basis for scalable qubit-phonon architectures (Mayor et al., 2020, McKenna et al., 2020, Xu et al., 18 Sep 2025).

2. Photonic interfaces and electro-optic devices

A practical LNOS system requires efficient coupling between external fibers and submicron on-chip waveguides. An experimentally demonstrated solution is a self-imaging apodized grating coupler with fixed period $_2$5 and a linearly tapered filling factor,

$_2$6

with $_2$7, $_2$8, and $_2$9 periods. The design uses first-order Bragg diffraction at a negative angle to match an angled SMF-28 fiber array and suppress higher-order diffraction. In the reported implementation, a 400 nm X-cut LN film is etched to 220 nm depth, and 2D and 3D FDTD simulations predict 42% peak coupling efficiency at 1550 nm with $_3$0 nm 3 dB bandwidth. Experimentally, the two-grating transmission peaks at 4.2% at 1545 nm, implying a single-grating efficiency of about 20.5%, with single-end efficiency exceeding 20% and bandwidth exceeding 25 nm (Chen et al., 2 Oct 2025).

The relevant phase-matching relation is written as

$_3$1

with $_3$2 for air. In the reported design workflow, 2D FDTD is used to optimize $_3$3, $_3$4, and $_3$5, after which 3D FDTD and Rayleigh–Sommerfeld propagation are used to recover the far field. The simulated far-field beam at $_3$6 $_3$7m has a 1/$_3$8 diameter of about 12 $_3$9m, close to the SMF-28 mode-field diameter of 10.4 $_3$0m. Fabrication is carried out by HSQ resist patterning, 100 kV electron-beam lithography, and CHF$_3$1/Ar ICP etching, with SEM showing smooth sidewalls, etch angle $_3$2, and groove widths from 154 nm to 616 nm (Chen et al., 2 Oct 2025).

At longer wavelengths, LNOS has also been used for high-speed electro-optic modulation in the mid-infrared. A Mach–Zehnder modulator operating from 3.95 to 4.3 $_3$3m is demonstrated on X-cut LNOS with 0.9 $_3$4m and 1.5 $_3$5m LiNbO$_3$6 films on sapphire. For the 1.5 $_3$7m film, the reported ridge cross section is $_3$8 $_3$9m and $\mu$0 $\mu$1m, supporting single-TE and TM modes at $\mu$2 $\mu$3m with mode area $\mu$4 $\mu$5m$\mu$6. The half-wave voltage is estimated from

$\mu$7

with $\mu$8 and $\mu$9–0.4. The measured device achieves $\mu$0 V$\mu$1cm at 4.0 $\mu$2m, extinction ratio 34.1 dB, fiber-to-fiber insertion loss 14.1 dB at 4.0 $\mu$3m and 17.5 dB at 4.3 $\mu$4m, on-chip loss of 4–5 dB, and 3 dB electro-optic bandwidth $\mu$5 GHz. System demonstrations include 10 Gbit/s OOK transmission with BER $\mu$6 and an 80 GHz comb generated by driving the modulator at 7 GHz with 40 dBm RF power (Didier et al., 29 May 2025).

Taken together, these results show that LNOS addresses two opposite ends of the photonic I/O problem: efficient coupling from standard telecom fibers into thin-film circuits, and high-speed phase and amplitude control deep in the mid-infrared. A plausible implication is that LNOS is unusual among thin-film LiNbO$\mu$7 variants in supporting both functions on the same substrate class.

3. Nonlinear mid-infrared photonics and dispersion engineering

Mid-infrared nonlinear optics is one of the clearest application areas in which sapphire changes the usable operating range of thin-film LiNbO$\mu$8. In periodically poled TFLN-on-sapphire waveguides, difference-frequency generation is demonstrated from a fixed 1.064 $\mu$9m pump and a tunable telecom signal, with continuous mid-infrared output from 2.81 $\sim 350$0m to 3.66 $\sim 350$1m and peak normalized efficiencies of approximately 200%/W$\sim 350$2cm$\sim 350$3. The first-order quasi-phase-matching condition is written as

$\sim 350$4

and in the undepleted-pump regime the normalized efficiency is defined by

$\sim 350$5

In the reported 4.1 mm poled devices, poling periods span 6.4–7.3 $\sim 350$6m, the generated light is TE-polarized, and the measured phase-matching bandwidth is about 8–15 nm, consistent with sinc$\sim 350$7 transfer functions (Mishra et al., 2021).

A complementary direction uses dispersion engineering rather than chirped poling to broaden the transfer function. In uniformly poled TFLN-on-sapphire with $\sim 350$8 nm, $\sim 350$9 nm, and $\mu$0 nm, the waveguide is designed so that the group-velocity mismatch between signal and idler approaches zero near $\mu$1 $\mu$2m. The phase mismatch is written as

$\mu$3

and the normalized conversion efficiency is given by

$\mu$4

The measured on-chip normalized efficiency is approximately 102%/W$\mu$5cm$\mu$6 at $\mu$7 $\mu$8m, within about 20% of the simulated 167%/W$\mu$9cm$\approx 4.5$0, while the instantaneous bandwidth reaches 18.5 THz FWHM, extending from about 2.8 $\approx 4.5$1m to about 3.8 $\approx 4.5$2m between the first zeros of sinc$\approx 4.5$3 (Mishra et al., 2022).

These mid-infrared results are closely tied to the LNOS substrate choice. Sapphire is cited as having low absorption in the 3–5 $\approx 4.5$4m band and negligible absorption up to about 5 $\approx 4.5$5m, whereas silica exhibits strong overtone absorption beyond 2.5 $\approx 4.5$6m. This permits mid-infrared propagation without a substrate-limited loss channel and enables tightly confined waveguides with index contrast $\approx 4.5$7 even near $\approx 4.5$8 $\approx 4.5$9m (Mishra et al., 2022, Mishra et al., 2021).

The nonlinear-optics literature on LNOS also identifies specific non-idealities. In the 2.8–3.8 $_3$00m difference-frequency-generation study, surface-adsorbed water can produce up to about 25% drop in DFG power at the short-wavelength edge near 2.8 $_3$01m, with loss peaks of about 1.5–2 dB/cm, while purging with dry N$_3$02 and heating above 150 $_3$03C restores the full sinc$_3$04-shaped transfer function. In the earlier 2.81–3.66 $_3$05m study, effective loss rises from about 0 cm$_3$06 at 2.95 $_3$07m to about 1.7 cm$_3$08 at 3.325 $_3$09m, consistent with residual OH absorption in LN rather than substrate absorption (Mishra et al., 2022, Mishra et al., 2021).

4. Phononic integrated circuits and acousto-optic modulation

LNOS has also been developed as a gigahertz phononic platform. In a representative implementation, a 500 nm X-cut LN film is direct-bonded to sapphire and rib-etched by 300 nm, leaving a 200 nm slab. The fundamental description uses the anisotropic elastic wave equation

$_3$10

and the physical picture is the phononic analogue of index guiding: the lower wave speed in LN relative to sapphire confines guided acoustic modes by total internal reflection of elastic waves. For a 1 $_3$11m-wide rib, finite-element simulation yields a guided quasi-Love mode around 3.4 GHz with $_3$12 km/s, $_3$13 km/s, and effective mode area $_3$14 $_3$15m$_3$16; the displacement magnitude decays by more than $_3$17 over a few microns into sapphire (Mayor et al., 2020).

The same work demonstrates practical building blocks. An IDT with 1 $_3$18m period, 50% duty cycle, and $_3$19 finger pairs reaches $_3$20 and is matched to 50 $_3$21 over a bandwidth of about 20 MHz. Measured reflection reaches $_3$22 dB at 3.42 GHz, and a 200 $_3$23m IDT–waveguide–IDT delay line gives $_3$24 dB. A racetrack resonator with round-trip length 871 $_3$25m and bend radius 40 $_3$26m shows intrinsic quality factor $_3$27 at 300 K and $_3$28 at 4 K. Nonlinear operation is also reported: phononic four-wave mixing yields $_3$29–20 (mW)$_3$30 and $_3$31 mW$_3$32 mm$_3$33, with significant gain expected for on-chip pump powers of order 0.3 mW and below 0.1 mW at 4 K (Mayor et al., 2020).

Acousto-optic modulation on LNOS uses the coexistence of guided optical and acoustic modes in the same X-cut film. In a 525 nm LN layer on c-axis sapphire, a 1.25 $_3$34m-wide optical ridge waveguide is placed next to aluminum IDTs. The electromechanical coupling coefficient is strongly orientation dependent. At 8 $_3$35m acoustic wavelength, the Rayleigh-like mode reaches $_3$36 near $_3$37, while the SH mode is about 0.3% at that pitch; at 2 $_3$38m pitch the SH mode exceeds 10% above about 2 GHz. The acousto-optic overlap is expressed through

$_3$39

with reported peak coupling coefficients of about $_3$40 m$_3$41 W$_3$42 for Rayleigh modes and $_3$43 m$_3$44 W$_3$45 for SH modes. For $_3$46 $_3$47m, the measured modulation efficiency is of order 0.1 rad/$_3$48, and the electrical bandwidth is about 50 MHz for Rayleigh modes and about 60 MHz for SH modes (Sarabalis et al., 2020).

Reconfigurability has been demonstrated thermally rather than piezoelectrically. In thin-film lithium niobate on sapphire with a 620 nm X-cut film, gigahertz SAWs are guided in 7 $_3$49m-wide rib waveguides and controlled by integrated NiCr microheaters. The phase shift obeys

$_3$50

The measured phase-shift efficiency is about $_3$51/mW at 2.46 GHz, and an acoustic Mach–Zehnder interferometer achieves maximum extinction ratio of about 10 dB at heater power around 0.45 W with small-signal modulation bandwidth of about 2 kHz. The thermal 3 dB cutoff of the LNSa phase modulator is about 20 Hz, faster than the comparable bulk-LN implementation because of sapphire’s higher thermal conductivity (Shao et al., 2022).

5. Cryogenic transduction and quantum-acoustic architectures

LNOS has been used experimentally for cryogenic microwave-to-optical conversion. In the triply resonant transducer of McKenna et al., a 500 nm X-cut LN film is directly bonded to C-cut sapphire and etched to form optical racetracks, while a quasi-lumped-element Nb microwave resonator is patterned on the same chip. Two optical supermodes are split by about 6.8 GHz and tuned so that

$_3$52

matching a microwave resonance at $_3$53 GHz. In the diagonal basis, the interaction Hamiltonian is

$_3$54

and the measured single-photon coupling is $_3$55 kHz. The device operates in a dilution refrigerator with on-chip microwave-to-optical conversion efficiency of $_3$56 and measured conversion bandwidth of about 20 MHz; the off-chip efficiency is about $_3$57 because of two 12.2 dB grating couplers (McKenna et al., 2020).

A central engineering issue in that transducer is acoustic radiation into sapphire driven by the piezoelectric $_3$58 response of LN under the electrodes. The mitigation strategy is to remove the LN slab beneath the electrodes while leaving a 6 $_3$59m pedestal beneath the optical waveguides. COMSOL is reported to show a more than tenfold reduction in radiated shear displacement, and the microwave quality factor improves to about 2000–3000, whereas no resonance is seen with an unetched slab. This is an example of LNOS being advantageous because sapphire itself is non-piezoelectric outside the LN pedestals (McKenna et al., 2020).

A later theoretical proposal extends the platform into cavity quantum acoustodynamics. In that architecture, a 200 nm X-cut LiNbO$_3$60 film on c-plane sapphire supports unsuspended phononic waveguides and microring resonators, while superconducting transmon qubits are fabricated on the sapphire. Straight waveguides and rings operate with quasi-Love or quasi-Rayleigh modes near 6 GHz, and narrow waveguides with $_3$61 nm are preferred because the quasi-Love mode maintains polarization purity around the ring. The effective piezoelectric coupling extracted from the IDT model is $_3$62 for the quasi-Love branch and $_3$63 for the quasi-Rayleigh branch. The quantized interaction is written as

$_3$64

and numerical FEM evaluation gives $_3$65 MHz for a single finger-pair IDT at $_3$66. Because $_3$67 while the IDT capacitance remains much smaller than the total qubit shunt, $_3$68 is projected to yield $_3$69 MHz, entering strong coupling (Xu et al., 18 Sep 2025).

The experimental transducer and the proposed QAD platform emphasize different operating regimes, but they rely on the same material logic: a piezoelectric LiNbO$_3$70 layer for optical and electromechanical coupling, and a low-loss sapphire host for superconducting circuitry and acoustic confinement. This suggests a coherent LNOS roadmap from classical acousto-optics to hybrid quantum networks.

6. Loss mechanisms, limitations, and research directions

Loss in LNOS is mode-specific and application-specific rather than uniformly low. For microwave-frequency acoustics, a systematic study of a 400 nm x-cut LiNbO$_3$71 film on sapphire reports monotonic temperature dependence from 4 K to 400 K for both Rayleigh and shear-horizontal modes. At 300 K, the measured propagation loss is $_3$72 dB/mm for a Rayleigh mode at about 1.84 GHz and $_3$73 dB/mm for an SH mode at about 2.35 GHz; at 4 K, these fall to $_3$74 dB/mm and $_3$75 dB/mm, respectively. The data are interpreted in the Akhiezer regime,

$_3$76

and, unlike LNOI, no low-temperature TLS-like loss peak is observed because there is no buried oxide layer. Cross-sectional HAADF-STEM additionally shows a $_3$77 nm bright contrast layer on the LN side and a $_3$78 nm darker layer on the sapphire side of the interface, indicating a transition zone that may contribute to residual temperature-independent loss (Lin et al., 2 Feb 2026).

Optical loss mechanisms are likewise diverse. In the 4 $_3$79m modulator, intrinsic waveguide propagation loss in the 1.5 $_3$80m film is measured by cut-back as about 2.24 dB/cm at 4 $_3$81m and 2.80 dB/cm at 4.3 $_3$82m. In fiber-chip grating couplers, the measured efficiency is below the 42% simulation partly because, without a bottom reflector, a fraction of about 20% of the power inherently couples into sapphire. The same grating-coupler study identifies fabrication tolerances in $_3$83, $_3$84, and etch depth, sidewall roughness, residual mask irregularities, and imperfect alignment to the self-imaging focal plane as additional contributors (Didier et al., 29 May 2025, Chen et al., 2 Oct 2025).

A recurring misconception is that replacing silica with sapphire removes all substrate-related limitations. The published record is more specific. Sapphire eliminates the buried-oxide loss channel that dominates low-temperature acoustic loss in LNOI, and it removes the mid-infrared absorption ceiling associated with SiO$_3$85 claddings; however, substrate leakage can still limit optical grating couplers, surface-adsorbed water can suppress short-wavelength mid-infrared performance, and nm-scale interfacial disorder can remain relevant for acoustic propagation (Lin et al., 2 Feb 2026, Mishra et al., 2022, Chen et al., 2 Oct 2025).

The reported research directions are correspondingly concrete. For grating couplers, proposed improvements include a thin SiO$_3$86/SiN bottom reflector, improved electron-beam dose control, and polymer overcladding (Chen et al., 2 Oct 2025). For microwave-to-optical conversion, proposed improvements include removing PECVD oxide, using NbTiN and high-impedance kinetic-inductor circuits, increasing optical quality factor toward $_3$87, and suppressing photorefraction via MgO doping or active stabilization (McKenna et al., 2020). For mid-infrared photonics, future directions include visible-wavelength gratings for atom-chip integration, broadband mid-infrared frequency combs via dispersion engineering in microresonators, on-chip up-conversion detectors, and integration with on-chip Brillouin and piezo-optomechanical devices (Chen et al., 2 Oct 2025, Mishra et al., 2021).

In aggregate, LNOS is best understood not as a single optimized device technology but as a substrate-level platform linking photonics, nonlinear optics, phononics, acousto-optics, superconducting microwave circuits, and hybrid quantum transduction. The literature shows that its distinctive advantage lies in the simultaneous availability of a high-$_3$88, strongly electro-optic, strongly piezoelectric thin film and a transparent, high-velocity, low-loss crystalline handle wafer.