Lithium Niobate-on-Insulator (LNOI) Overview

• LNOI is a thin-film LiNbO3 platform bonded on SiO2 that enables wafer-scale nanofabrication and high-index optical confinement, shifting device physics to the nanophotonic regime.
• It supports integrated passive and active photonic devices such as electro-optic modulators, nonlinear frequency converters, and quantum circuits through precise dispersion and quasi-phase matching engineering.
• The platform underpins diverse applications—from telecommunications to quantum photonics—while addressing challenges in manufacturability, sensitivity to fabrication errors, and thermal management.

Lithium niobate-on-insulator (LNOI) is a thin-film photonic platform in which a submicrometer LiNbO$_3$ device layer is bonded to an insulating layer, typically SiO$_2$, on a carrier substrate. In this geometry, the large electro-optic response and strong $\chi^{(2)}$ nonlinearity of LiNbO$_3$ are combined with high-index-contrast confinement, wafer-scale nanofabrication, and modal areas of order $0.5$–$1\,\mu\mathrm{m}^2$, rather than the $10$–$100\,\mu\mathrm{m}^2$ scale characteristic of Ti-diffused lithium niobate. Commercial LNOI is reported on $2$–$6$ inch wafers with $_2$0–$_2$1 films over $_2$2–$_2$3 buried oxide, and the platform now spans passive waveguides, high-speed electro-optic modulators, periodically poled nonlinear circuits, microresonator combs, acousto-optic devices, rare-earth-doped gain elements, and quantum-photonic subsystems (Zhu et al., 2021).

1. Material system, crystallographic cuts, and confinement

LNOI inherits the anisotropy of bulk LiNbO$_2$4, so the crystallographic cut is not a secondary fabrication detail but a primary design variable. X-cut and Z-cut films are both widely used. Representative stacks in the literature include a $_2$5 z-cut LiNbO$_2$6 device layer on $_2$7 buried SiO$_2$8 and a bulk z-cut LiNbO$_2$9 handle (Krasnokutska et al., 2017), and an X-cut $\chi^{(2)}$0 film bonded onto an approximately $\chi^{(2)}$1 buried SiO$\chi^{(2)}$2 layer on a silicon carrier for periodically poled nonlinear ridge waveguides (Kumar et al., 2022). Across these implementations, the index contrast between LiNbO$\chi^{(2)}$3 and SiO$\chi^{(2)}$4 is typically $\chi^{(2)}$5–$\chi^{(2)}$6, which is large enough to move LiNbO$\chi^{(2)}$7 from a weakly guiding integrated-optics material into the nanophotonic regime (Zhu et al., 2021).

This confinement changes both device physics and design methodology. Effective index, group index, bending loss, modal overlap with electrodes, quasi-phase-matching period, and microwave-optical velocity matching all become strongly geometry dependent. In practice, ridge width, etch depth, residual slab thickness, buried-oxide thickness, and cladding choice are co-optimized rather than treated independently. A representative partially etched geometry summarized in the literature is a $\chi^{(2)}$8–$\chi^{(2)}$9 film etched to leave a $_3$0–$_3$1 slab, with top widths from $_3$2 to $_3$3 and sidewall angles between $_3$4 and $_3$5 (Zhu et al., 2021).

A common oversimplification is to identify LNOI exclusively with shallow-etched ridge guides. The published record instead includes partially etched ribs, fully etched strips, air-clad and oxide-clad waveguides, suspended or pedestal-supported resonators, and heterogeneous stacks in which the LiNbO$_3$6 film is left unpatterned while passive routing is carried by another platform. This suggests that LNOI is better understood as a family of thin-film LiNbO$_3$7 integration strategies than as a single waveguide topology.

2. Nanofabrication, propagation loss, and passive circuit density

The central fabrication problem in LNOI is to obtain high confinement without converting sidewall roughness and etch-induced damage into dominant scattering loss. Several process families have achieved this with distinct trade-offs. Dry-etched monolithic waveguides fabricated by electron-beam lithography, hard-mask transfer, and mixed Ar/F plasma reactive-ion etching have reached TE propagation loss $_3$8, TM loss $_3$9, sidewall angle $0.5$0, and sidewall roughness below $0.5$1 at $0.5$2 (Krasnokutska et al., 2017). A different ion-free route based on chromium hard-mask patterning by femtosecond laser ablation followed by chemo-mechanical polishing produced $0.5$3-scale waveguides with $0.5$4 propagation loss and $0.5$5 top-surface roughness (Wu et al., 2018).

The compactness question has evolved in parallel with the low-loss question. Partially etched waveguides long dominated LNOI because shallow confinement reduced bend radiation and fabrication sensitivity, but a fully etched $0.5$6-thick platform has demonstrated mean propagation loss $0.5$7, intrinsic $0.5$8, bend radii down to $0.5$9, and $1\,\mu\mathrm{m}^2$0 in compact microrings (Gao et al., 2023). In that study, a $1\,\mu\mathrm{m}^2$1 ring radius corresponded to an approximately $1\,\mu\mathrm{m}^2$2 free-spectral range, and the authors explicitly contrasted this regime with shallow-etched platforms requiring radii of at least $1\,\mu\mathrm{m}^2$3. The result directly counters the assumption that LNOI compactness and ultra-low loss are mutually exclusive.

Passive interfacing has also advanced. A bilayer inverse-taper edge coupler on X-cut LNOI with a SiON cladding waveguide reported fiber-chip loss of $1\,\mu\mathrm{m}^2$4 per facet for TE and $1\,\mu\mathrm{m}^2$5 per facet for TM at $1\,\mu\mathrm{m}^2$6, remaining below $1\,\mu\mathrm{m}^2$7 per facet for both polarizations above $1\,\mu\mathrm{m}^2$8 (Hu et al., 2020). At the circuit level, fabrication-aware inverse design has produced a $1\,\mu\mathrm{m}^2$9 spatial-mode multiplexer with experimental insertion loss around $10$0, a $10$1 waveguide crossing with approximately $10$2 insertion loss and approximately $10$3 crosstalk, and a $10$4 $10$5 bend with approximately $10$6 loss across $10$7–$10$8 (Shang et al., 2022).

The broader significance is that passive LNOI has moved from proof-of-principle nanofabrication to an engineering space in which low loss, tight bends, compact crossings, and sub-decibel fiber interfaces coexist. That transition is what makes the material credible as a full photonic integrated-circuit platform rather than a collection of isolated devices.

3. Quasi-phase matching, dispersion control, and nonlinear frequency conversion

The most distinctive feature of LNOI relative to many competing platforms is that modal geometry and $10$9 nonlinearity can be engineered simultaneously. For a three-wave process such as type-II sum-frequency generation (SFG) or spontaneous parametric down-conversion (SPDC), the source must satisfy

$100\,\mu\mathrm{m}^2$0

and

$100\,\mu\mathrm{m}^2$1

In nanophotonic LNOI, the group index

$100\,\mu\mathrm{m}^2$2

controls the spectral tilt of the phase-matching curve, with slope

$100\,\mu\mathrm{m}^2$3

By selecting ridge width and height, one can approach the limiting cases $100\,\mu\mathrm{m}^2$4 or $100\,\mu\mathrm{m}^2$5, corresponding respectively to $100\,\mu\mathrm{m}^2$6 and $100\,\mu\mathrm{m}^2$7, which are the extremal conditions for spectrally factorizable photon-pair generation (Kumar et al., 2022).

An experimentally realized design used $100\,\mu\mathrm{m}^2$8, $100\,\mu\mathrm{m}^2$9, and $2$0, giving degenerate signal and idler wavelengths at $2$1 and a pump at $2$2. The simulated group indices were $2$3, $2$4, and $2$5, implying $2$6. In experiment, the SFG map displayed an almost horizontal phase-matching ridge, and continuous-wave SPDC under a $2$7 pump yielded correlated photons near $2$8 with a coincidence-to-accidental ratio $2$9 (Kumar et al., 2022). The same study reported a predicted normalized SFG efficiency $6$0 when propagation losses were neglected, compared with an experimental value $6$1 including loss, coupling, and collection.

Cladding is an equally strong dispersion lever. Air-clad waveguides were found to exhibit higher index contrast and stronger geometry-induced dispersion, permitting the phase-matching slope $6$2 to be tuned over roughly $6$3 to $6$4 by varying width between $6$5 and $6$6 at fixed $6$7. Silica-clad guides were less sensitive to width errors and exhibited a global $6$8 optimum near $6$9, $_2$00, which makes fabrication tolerances more forgiving (Kumar et al., 2022). This is one of the clearest examples of LNOI’s central design logic: in thin-film LiNbO$_2$01, dispersion engineering is not an afterthought to phase matching but the mechanism by which phase matching is obtained.

Periodic poling remains essential because first-order quasi-phase matching permits direct use of the largest tensor element. For ideal $_2$02 duty cycle, the effective nonlinearity is

$_2$03

with $_2$04, so $_2$05 (Krasnokutska et al., 2021). Conventional electric-field poling in Z-cut LNOI is typically limited to periods $_2$06–$_2$07, but focused-ion-beam poling has demonstrated periods down to $_2$08, consistent over $_2$09, with period deviation no greater than $_2$10 and $_2$11 yield in piezoresponse-force-microscopy scans (Krasnokutska et al., 2021). Because the quasi-phase-matching condition is extremely sensitive to geometry and period, these advances are not merely lithographic refinements. A period error of $_2$12 at $_2$13 and $_2$14–$_2$15 can reduce conversion efficiency by more than $_2$16, and $_2$17 changes in ridge width or etch depth can shift the forward-SHG poling period by $_2$18–$_2$19 (Krasnokutska et al., 2021).

At the device level, the review literature reports traveling-wave PPLN nano-ridge SHG efficiencies of $_2$20–$_2$21, resonator-enhanced $_2$22 QPM efficiencies up to $_2$23, and SPDC sources with heralded pair rate $_2$24 and $_2$25 (Zhu et al., 2021). Taken together, these results place LNOI among the few integrated platforms where quantum-state engineering, broadband classical frequency conversion, and dispersion-shaped nonlinear response can all be pursued using the same wafer technology.

4. Electro-optic modulation, microresonators, and frequency-comb dynamics

LNOI electro-optics rests on the same tensor that made bulk LiNbO$_2$26 the canonical modulator material, but thin-film confinement improves the overlap between guided optical modes and applied microwave fields. In the standard Pockels description,

$_2$27

and the half-wave voltage-length product follows the familiar scaling

$_2$28

Across thin-film LNOI modulators, reported $_2$29 values lie in the $_2$30–$_2$31 range, with $_2$32 bandwidths exceeding $_2$33 for $_2$34 devices and exceeding $_2$35 for $_2$36 devices (Zhu et al., 2021). These metrics are inseparable from the nanophotonic geometry: the same confinement that lowers optical mode area also lowers the electrode length needed for a target phase shift.

LNOI electro-optics is not restricted to lateral waveguide modulators. A vertical-cavity electro-optic modulator has been demonstrated by sandwiching a $_2$37 X-cut LN membrane between two TiO$_2$38/SiO$_2$39 one-dimensional photonic-crystal mirrors with integrated electrodes. The measured resonance occurred at $_2$40 with $_2$41, and under $_2$42 drive across a $_2$43 electrode gap the peak modulation depth reached $_2$44. The measured optical-intensity $_2$45 bandwidth was approximately $_2$46, while vector-network analysis in the supplement indicated intrinsic electro-optic bandwidth extending into the GHz regime (Liu et al., 4 Nov 2025). This architecture demonstrates that free-space modulation and thin-film LN are not disjoint categories; the LNOI membrane itself can serve as the electro-optic defect layer of a vertical cavity.

Microresonators add a second layer of functionality by combining the Pockels effect with cavity enhancement. The comb review reports resonant electro-optic comb generation with more than $_2$47 lines at $_2$48 spacing and approximately $_2$49 optical span in a high-$_2$50 ring driven with $_2$51 RF, as well as Kerr comb thresholds of approximately $_2$52–$_2$53, soliton bandwidths exceeding $_2$54, and supercontinuum generation spanning $_2$55 to $_2$56 in a $_2$57 LN waveguide pumped by $_2$58 pulses with pulse energy about $_2$59 (Yu et al., 2019). The coexistence of strong $_2$60, appreciable $_2$61, and electro-optic tuning in one material system is what makes these device classes unusually interoperable.

Raman scattering introduces a notable complication. In an X-cut LNOI microresonator, backward Raman oscillation was observed with continuous-wave pump threshold around $_2$62 and differential quantum efficiency $_2$63, while soliton modelocking in a $_2$64-FSR ring was achieved only after suppressing Raman gain through cavity-geometry control (Yu et al., 2019). This is an important corrective to the view that LNOI resonators can always be analyzed as pure Kerr or pure electro-optic systems. In practice, Raman, Kerr, and $_2$65 processes can compete, and resonator geometry serves as the arbitration mechanism.

5. Gain media, heterogeneous integration, and quantum photonic circuits

Although LiNbO$_2$66 is classically regarded as a passive or electro-optic material, LNOI has also become an active platform through rare-earth doping. A $_2$67-long erbium-doped thin-film lithium niobate waveguide amplifier demonstrated on-chip net gain greater than $_2$68 for $_2$69 signal light with $_2$70 pump power at $_2$71; the maximum reported value was $_2$72 at $_2$73, while $_2$74 gain reached $_2$75 (Chen et al., 2021). The device used a $_2$76 Er:LNOI film and a $_2$77 wide rib with $_2$78 etch depth and $_2$79 slab.

Microlasers extend the same logic into resonant active devices. Er-doped LNOI microring lasers with ring radius $_2$80 and loaded quality factors exceeding one million have been reported, with $_2$81 near $_2$82, $_2$83 near $_2$84, and on-chip lasing threshold approximately $_2$85 under $_2$86 pumping. At pump powers around $_2$87 on chip, a comb-like laser spectrum spanning $_2$88–$_2$89 was observed (Luo et al., 2021). A related Er$_2$90-doped microdisk implementation reached $_2$91, threshold approximately $_2$92, and pump-induced wavelength tuning efficiencies of $_2$93 below $_2$94 and $_2$95 above $_2$96, reflecting competing photorefractive and thermo-optic contributions (Wang et al., 2020).

Another misconception is that advanced LNOI circuits must be monolithically etched in LiNbO$_2$97. Heterogeneous integration has shown otherwise. In a wafer-scale bonded lithium-niobate-on-silicon-nitride platform, the underlying Si$_2$98N$_2$99 PIC retained propagation loss below $\chi^{(2)}$00, fiber-to-chip coupling below $\chi^{(2)}$01 per facet, and adiabatic transition loss not exceeding $\chi^{(2)}$02 per LN-to-Si$\chi^{(2)}$03N$\chi^{(2)}$04 interface; ten transitions added less than $\chi^{(2)}$05 total loss (Churaev et al., 2021). Because the LN film in that platform is unpatterned, all critical passive features remain defined by foundry-grade Si$\chi^{(2)}$06N$\chi^{(2)}$07 lithography. The result is not a departure from LNOI’s objectives but a different solution to the same problem of combining low-loss routing with strong electro-optic functionality.

Quantum integration has progressed along the same trajectory. A low-loss LNOI waveguide network integrating a tunable Mach–Zehnder interferometer with two waveguide-integrated superconducting nanowire single-photon detectors reported straight-waveguide loss $\chi^{(2)}$08 at $\chi^{(2)}$09, $\chi^{(2)}$10, electro-optic $\chi^{(2)}$11 bandwidth about $\chi^{(2)}$12, bias-drift-free operation for $\chi^{(2)}$13 hours, and detector on-chip efficiencies of approximately $\chi^{(2)}$14 and $\chi^{(2)}$15, with dark counts around $\chi^{(2)}$16 counts/s and timing jitter down to $\chi^{(2)}$17 when a cryogenic amplifier was used (Lomonte et al., 2021). This is significant because it shows that LNOI is not only a source or modulator platform; it can host detection and reconfigurability in the same cryogenic circuit.

6. Acoustics, manufacturability, and open technical problems

LNOI also supports acoustic and acousto-electric devices because the thin-film stack confines not only light but also mechanical waves. For shear-horizontal surface acoustic wave resonators on X-cut LNOI, lithographically defined wavelengths between $\chi^{(2)}$18 and $\chi^{(2)}$19 produced series resonance frequencies between $\chi^{(2)}$20 and $\chi^{(2)}$21. The optimized $\chi^{(2)}$22 device achieved $\chi^{(2)}$23, $\chi^{(2)}$24, $\chi^{(2)}$25, and figure of merit $\chi^{(2)}$26 (Hsu et al., 2024). These values place LNOI directly in the discussion of centimeter-band 6G front-end components, extending the platform beyond optics in the narrow sense.

Several technical bottlenecks recur across otherwise disparate LNOI device classes. First, quasi-phase matching is exceptionally sensitive to fabrication errors. In periodically poled nanowaveguides, $\chi^{(2)}$27 variations in ridge width or etch depth can move the QPM condition across the telecom bands, while submicron duty-cycle errors and domain merging constrain the usable design space for backward processes (Krasnokutska et al., 2021). Second, thin-film domain inversion at $\chi^{(2)}$28 already requires careful electrode lithography, voltage control, and sometimes on-chip SHG monitoring to avoid incomplete inversion or merging (Kumar et al., 2022). Third, the same high confinement that reduces device footprint amplifies sidewall-scattering and substrate-leakage sensitivity, so process windows remain narrow even in mature dry-etch flows (Zhu et al., 2021).

Manufacturability is therefore not a peripheral issue. One response has been direct improvement of monolithic LN processing through smoother sidewalls, sub-nanometer polishing, and full-etch control. Another has been architectural: leave LN lightly processed or unpatterned and transfer complexity to a mature passive platform, as in bonded Si$\chi^{(2)}$29N$\chi^{(2)}$30 underlays with near-$\chi^{(2)}$31 bonding yield and wafer-scale uniformity (Churaev et al., 2021). A plausible implication is that the future LNOI ecosystem will remain plural rather than converging onto a single process stack.

The remaining system-level challenges are similarly multidimensional. The review literature identifies photorefraction, DC drift, and thermal management as persistent concerns; the oxide undercladding that enables optical confinement also limits heat extraction, particularly in suspended or high-power structures (Zhu et al., 2021). At the same time, the demonstrated combination of sub-$\chi^{(2)}$32 passive transport, $\chi^{(2)}$33 in the few-$\chi^{(2)}$34 range, quasi-phase-matched $\chi^{(2)}$35 conversion, cryogenic reconfigurability, rare-earth gain, and GHz-to-THz resonator engineering suggests that LNOI’s defining feature is not any single record metric. It is the unusually broad set of high-performance functions available within one thin-film ferroelectric platform.