Thin-Film Lithium Niobate

• Thin-Film Lithium Niobate (TFLN) is a photonic and phononic platform for integrated photonics with strong optical and electro-opto-magnetic confinement through a substrate, offering faster and more efficient data processing.
• The primary applications of TFLN include electro-optic modulation, nonlinear frequency conversion, optical parametric oscillation, frequency-comb generation, and more, making it suitable for high-bandwidth communication.
• The TFLN platform has demonstrated several key technological advances, such as optical waveguide creation with bend radii below 20 μm, travel electric-wave isolators with 39.50 dB isolation at 24 GHz, and non-linear frequency conversion ranging from transfer of wavelength to the molecular fingerprint region of 3–5 µm.

Thin-film lithium niobate (TFLN) is a photonic and phononic platform formed from a single-crystal lithium-niobate layer, generally integrated with an insulating or carrier substrate to provide high optical and acoustic confinement. In contrast with bulk lithium-niobate devices fabricated by titanium indiffusion or proton exchange, TFLN combines lithium niobate’s Pockels electro-optic response, $\chi^{(2)}$ and $\chi^{(3)}$ nonlinearities, piezoelectricity, photoelasticity, ferroelectric-domain engineering, broad optical transparency, and relatively high refractive index with submicrometre-scale waveguides, compact resonators, engineered dispersion, and wafer-scale processing. The resulting platform supports electro-optic modulation, nonlinear frequency conversion, optical parametric oscillation, frequency-comb generation, acousto-optics, microwave–optical transduction, quantum photonics, polarization processing, and millimeter-wave acoustic filtering (Zhu et al., 2021).

1. Material basis and platform architecture

Lithium niobate, $\mathrm{LiNbO_3}$, is a non-centrosymmetric ferroelectric crystal belonging to the $3m$ point group. Its crystallographic $Z$ axis is the ferroelectric $c$ axis and the extraordinary optical axis. The absence of inversion symmetry permits both second-order optical nonlinearities and the linear electro-optic, or Pockels, effect. Periodic reversal of the ferroelectric domains reverses the sign of relevant $\chi^{(2)}$ tensor elements and supplies quasi-phase-matching momentum.

Important platform properties include a transparency window of approximately $350~\mathrm{nm}$ to $5~\mu\mathrm{m}$, refractive indices near $1550~\mathrm{nm}$ of approximately $\chi^{(3)}$0 and $\chi^{(3)}$1–$\chi^{(3)}$2, a Curie temperature near $\chi^{(3)}$3, and a thermo-optic coefficient near $\chi^{(3)}$4 of approximately $\chi^{(3)}$5. Representative electro-optic coefficients include $\chi^{(3)}$6 and $\chi^{(3)}$7–$\chi^{(3)}$8. The dominant second-order nonlinear coefficient is approximately $\chi^{(3)}$9, while representative third-order behavior is characterized by $\mathrm{LiNbO_3}$0. Piezoelectric coefficients include $\mathrm{LiNbO_3}$1, $\mathrm{LiNbO_3}$2, and $\mathrm{LiNbO_3}$3 (Zhu et al., 2021).

The Pockels effect changes the inverse dielectric tensor according to

$\mathrm{LiNbO_3}$4

For a suitable principal-axis configuration, the induced extraordinary-index change is approximately

$\mathrm{LiNbO_3}$5

This direct, carrier-free index modulation distinguishes TFLN from silicon and III–V carrier-dispersion devices. The same crystal can therefore support fast phase modulation, resonant frequency tuning, electro-optic frequency shifting, electro-optic comb generation, and microwave–optical conversion.

A typical lithium-niobate-on-insulator (LNOI) wafer consists of a few-hundred-nanometre to approximately $\mathrm{LiNbO_3}$6 single-crystal LN film, a buried $\mathrm{LiNbO_3}$7 layer, and a silicon, LN, sapphire, or other handle substrate. The crystal cut—X-cut, Y-cut, or Z-cut—is selected according to the desired optical, electro-optic, nonlinear, or acoustic interaction. High LN–cladding index contrast enables compact ridge waveguides, microrings, microdisks, racetracks, photonic-crystal cavities, Mach–Zehnder interferometers, multimode-interference couplers, and arrayed-waveguide gratings.

The principal fabrication routes are ion slicing followed by wafer bonding, direct bonding followed by mechanical thinning, and heterogeneous transfer. Ion slicing uses $\mathrm{LiNbO_3}$8 or $\mathrm{LiNbO_3}$9 implantation to define a cleavage plane, bonding to a receiver substrate, annealing for splitting and damage recovery, and polishing. A separate wafer-bonding approach used a Z-cut bulk LN wafer, plasma activation, room-temperature flip bonding to silicon, and mechanical grinding and polishing to approximately $3m$0, followed by further thinning to approximately $3m$1 (Wang et al., 2014).

Waveguide fabrication commonly uses argon ion milling or argon-based inductively coupled plasma reactive-ion etching. Dry-etched waveguides historically exhibited losses above $3m$2, whereas optimized structures reached approximately $3m$3; losses below $3m$4 became routine in several laboratories (Zhu et al., 2021). Sidewall scattering remains a principal limitation because tightly confined modes overlap the etched boundaries. Alternative processes include chemomechanical polishing, diamond dicing, hybrid rib loading, and photolithography-assisted chemo-mechanical etching (PLACE). PLACE was used to fabricate a traveling-wave TFLN isolator with a $3m$5 X-cut film, a $3m$6 etch depth, and a $3m$7 top ridge width (Barrera et al., 2023).

2. Optical confinement, resonators, and dispersion engineering

TFLN waveguides provide much stronger confinement than bulk diffused LN waveguides. Conventional titanium-indiffused devices have index contrasts of approximately $3m$8–$3m$9, while proton-exchanged devices have approximately $Z$0. Their mode areas are roughly $Z$1–$Z$2, with millimetre-scale bend radii and centimetre-scale interaction lengths. TFLN instead supports submicrometre optical confinement, bend radii below $Z$3 in nanophotonic waveguides, and lithographically engineered modal dispersion (Zhu et al., 2021).

The propagation constant is

$Z$4

The group velocity and group-velocity dispersion are

$Z$5

Waveguide width, etch depth, film thickness, sidewall angle, cladding, crystal cut, and suspension state can modify $Z$6 and higher-order dispersion. This design freedom allows TFLN devices to compensate material dispersion, obtain near-zero group-velocity mismatch, engineer anomalous dispersion for Kerr solitons, and tailor phase matching for second-harmonic generation, sum-frequency generation, difference-frequency generation, and parametric down-conversion.

Whispering-gallery resonators

Microdisks, microrings, and racetracks support compact optical modes with high quality factors. Dry-etched microdisks have reached approximately $Z$7, while femtosecond-laser/FIB-processed and chemomechanically polished devices have reached approximately $Z$8. The highest values cited in the review include approximately $Z$9 for thin-film LN and approximately $c$0 for mechanically polished bulk LN (Zhu et al., 2021).

A released LN disk demonstrated an intrinsic optical quality factor of $c$1 in a $c$2-radius resonator. The device used a central silicon pedestal, an air clearance greater than $c$3, an $c$4 sidewall angle, and sidewall roughness below $c$5. The suspended geometry reduced leakage into the higher-index silicon substrate and allowed mechanical radial-breathing modes at $c$6 and approximately $c$7 (Wang et al., 2014).

For microrings and racetracks, intrinsic quality factor relates to propagation loss $c$8 through

$c$9

and the free spectral range is

$\chi^{(2)}$0

A principal limitation of traveling-wave resonators is that changing the circumference also changes curvature-dependent dispersion, group index, bending loss, and mode conversion. A further limitation is broad spectral support, which can enhance unwanted Raman scattering and parasitic nonlinear processes.

Straight photonic-crystal Fabry–Perot resonators

Straight photonic-crystal Fabry–Perot (PhC-FP) resonators decouple cavity length from curvature and provide standing-wave modes bounded by photonic-crystal reflectors. A representative device used a $\chi^{(2)}$1 LN film, $\chi^{(2)}$2 etch depth, $\chi^{(2)}$3 photonic-crystal period, and apodized reflectors. It achieved a maximum intrinsic quality factor of $\chi^{(2)}$4, propagation loss of $\chi^{(2)}$5, and photonic-crystal interfacial loss of $\chi^{(2)}$6 (Hwang et al., 19 May 2025).

The cavity resonance approximately satisfies

$\chi^{(2)}$7

where $\chi^{(2)}$8 and $\chi^{(2)}$9 are reflector phases. The cavity FSR is controlled primarily by the straight-cavity length:

$350~\mathrm{nm}$0

A separate $350~\mathrm{nm}$1-long suspended TFLN Fabry–Perot cavity with tapered photonic-crystal mirrors achieved a loaded quality factor of $350~\mathrm{nm}$2, a $350~\mathrm{nm}$3 linewidth at $350~\mathrm{nm}$4, a $350~\mathrm{nm}$5 FSR, and a finesse of approximately $350~\mathrm{nm}$6. The same device exhibited piezo-optomechanical resonances at $350~\mathrm{nm}$7 and $350~\mathrm{nm}$8 (Yang et al., 8 Oct 2025).

The straight geometry enables optical, microwave, acoustic, and crystal-axis orientations to be selected independently. The photonic bandgap can also restrict high-$350~\mathrm{nm}$9 enhancement to a designed spectral region. Suppression of Raman scattering and other parasitic nonlinearities is an expected benefit, although direct quantitative Raman-suppression measurements were not reported in the cited PhC-FP work.

3. Electro-optic, acousto-optic, and RF functionality

Electro-optic modulators

The Pockels effect enables phase and intensity modulation without carrier injection or depletion. Conventional traveling-wave TFLN Mach–Zehnder modulators have demonstrated $5~\mu\mathrm{m}$0 values of approximately $5~\mu\mathrm{m}$1–$5~\mu\mathrm{m}$2, greater than $5~\mu\mathrm{m}$3 electro-optic bandwidth, coherent data transmission up to $5~\mu\mathrm{m}$4, and extinction ratios exceeding $5~\mu\mathrm{m}$5 in short devices (Zhu et al., 2021).

A high-permittivity-cladding design used a conventional ground–signal–ground traveling-wave MZI with a $5~\mu\mathrm{m}$6 LN film, a $5~\mu\mathrm{m}$7 electrode gap, an $5~\mu\mathrm{m}$8 $5~\mu\mathrm{m}$9 upper layer, and a $1550~\mathrm{nm}$0 high-permittivity cladding. Glycerol, with relative permittivity approximately $1550~\mathrm{nm}$1 and refractive index approximately $1550~\mathrm{nm}$2, enhanced the RF field in LN while preserving optical confinement. The demonstrated device achieved

$1550~\mathrm{nm}$3

approximately $1550~\mathrm{nm}$4 excess optical loss, and an electro-optic 3-dB bandwidth greater than $1550~\mathrm{nm}$5 (Chen et al., 2023).

The high-permittivity strategy relies on the displacement-field boundary condition

$1550~\mathrm{nm}$6

Remote dielectric loading redistributes the microwave field toward the LN ridge without placing a high-index material directly over the optical core. The design illustrates the central electro-optic trade-off: stronger RF loading can reduce $1550~\mathrm{nm}$7, but excessive dielectric loading can enlarge the optical mode, alter microwave impedance and velocity, increase microwave loss, or introduce optical absorption.

Compact resonant modulators trade bandwidth for reduced footprint and capacitance. A photonic-crystal cavity IQ modulator used two electrically tuned one-sided cavities in a Fabry–Perot-enhanced Michelson interferometer. The device occupied approximately $1550~\mathrm{nm}$8, supported amplitude and phase control, and demonstrated 4-QAM at approximately $1550~\mathrm{nm}$9 with a $\chi^{(3)}$00-V peak-to-peak drive. The cavity 3-dB bandwidth was approximately $\chi^{(3)}$01, the estimated energy was $\chi^{(3)}$02, and the measured error-vector magnitude remained below $\chi^{(3)}$03 through approximately $\chi^{(3)}$04 (Larocque et al., 2023).

Polarization management

TFLN can implement polarization splitting, rotation, synthesis, scrambling, measurement, and feedback control using polarization splitter–rotators and Pockels-effect MZIs. A demonstrated platform achieved polarization extinction ratio up to $\chi^{(3)}$05, scrambling rate up to $\chi^{(3)}$06, and endless polarization tracking at $\chi^{(3)}$07. The MZI response time was below $\chi^{(3)}$08, the switching voltage was $\chi^{(3)}$09, MZI insertion loss was below $\chi^{(3)}$10, and waveguide propagation loss was approximately $\chi^{(3)}$11 (Lin et al., 2021).

The polarization measurement circuit reconstructed Stokes vectors using four electro-optically selected analysis states and reached sampling rates up to $\chi^{(3)}$12. Endless control used a four-phase-shifter MZI, integrated photodetector, ADC/DAC electronics, and an FPGA feedback loop. The demonstrated tracking rate was limited principally by the electronics and feedback algorithm rather than the intrinsic Pockels response.

Acousto-optics and microwave isolation

A released LN disk demonstrated acousto-optic modulation by driving a $\chi^{(3)}$13 radial-breathing mode with an external hovering RF probe. The mechanical quality factor was approximately $\chi^{(3)}$14, and the optical sideband power was $\chi^{(3)}$15 at a transmitted DC carrier power of $\chi^{(3)}$16 under a $\chi^{(3)}$17 RF drive (Wang et al., 2014). The measurement did not report an integrated electrical electrode or a vacuum optomechanical coupling rate.

Traveling-wave electro-optic modulation can also produce nonmagnetic optical isolation. A PLACE-fabricated TFLN isolator used a $\chi^{(3)}$18 electrode, a $\chi^{(3)}$19 X-cut LN film, and spot-size converters. Direction-dependent microwave–optical phase matching produced $\chi^{(3)}$20 isolation at $\chi^{(3)}$21 with a $\chi^{(3)}$22 microwave drive. Isolation exceeded $\chi^{(3)}$23 from $\chi^{(3)}$24 to $\chi^{(3)}$25, and fiber-to-fiber insertion loss was $\chi^{(3)}$26 (Gao et al., 2023).

Arrayed-waveguide routing

An electro-optically tunable TFLN arrayed-waveguide grating contained one input, eight outputs, twenty arrayed waveguides, and twenty independently addressable electrode pairs. The device provided approximately $\chi^{(3)}$27 channel spacing, corresponding to $\chi^{(3)}$28 near $\chi^{(3)}$29, and exhibited wavelength tuning of approximately $\chi^{(3)}$30. A $\chi^{(3)}$31 applied voltage shifted the center wavelength by approximately $\chi^{(3)}$32 (Wang et al., 2024).

For adjacent array arms, the AWG phase condition is governed by

$\chi^{(3)}$33

Electrically induced phase shifts modify the relative phase distribution across the array. The demonstrated experiment applied a common voltage to all signal electrodes, producing a common spectral shift; independent spectral shaping was proposed but not demonstrated.

4. Nonlinear frequency conversion and parametric processes

Traveling-wave second-harmonic generation

For second-harmonic generation, two photons at $\chi^{(3)}$34 generate one photon at $\chi^{(3)}$35. The phase mismatch is

$\chi^{(3)}$36

and efficient conversion requires $\chi^{(3)}$37. Quasi-phase matching adds a grating wavevector $\chi^{(3)}$38.

TFLN nanowaveguides provide strong modal confinement and lithographic dispersion control. In a $\chi^{(3)}$39-thick X-cut platform, uniform waveguides used modal phase matching between the fundamental TE mode at approximately $\chi^{(3)}$40 and a third-order transverse TE mode at approximately $\chi^{(3)}$41. The best measured normalized conversion efficiency was

$\chi^{(3)}$42

with approximately $\chi^{(3)}$43 propagation loss. The generated second harmonic exhibited a three-lobed spatial profile, confirming the third-order output mode (Wang et al., 2016).

A periodically grooved waveguide used a $\chi^{(3)}$44 period and an $\chi^{(3)}$45 groove depth to achieve quasi-phase matching between fundamental modes at both wavelengths. Its measured normalized efficiency was

$\chi^{(3)}$46

Modal phase matching provided higher conversion efficiency and simpler fabrication but generated a higher-order mode. Periodically grooved quasi-phase matching produced a fundamental output mode but introduced Bloch-mode radiation, fabrication complexity, and increased second-harmonic loss.

Optical parametric oscillation

A periodically poled LN microring used radial domain inversion with $\chi^{(3)}$47 periods and an effective poling period near $\chi^{(3)}$48. The device operated with a near-visible pump near $\chi^{(3)}$49 and generated signal and idler waves near $\chi^{(3)}$50–$\chi^{(3)}$51. Its loaded quality factors were $\chi^{(3)}$52, $\chi^{(3)}$53, and $\chi^{(3)}$54.

The measured on-chip threshold was approximately

$\chi^{(3)}$55

with a measured three-wave coupling rate of

$\chi^{(3)}$56

At an on-chip pump power of $\chi^{(3)}$57, the on-chip signal-plus-idler power was $\chi^{(3)}$58, corresponding to approximately $\chi^{(3)}$59 conversion efficiency. The OPO tuning range extended approximately from $\chi^{(3)}$60 to $\chi^{(3)}$61 as pump wavelength and chip temperature were varied. Degenerate OPO was reached near $\chi^{(3)}$62 (Lu et al., 2021).

The threshold scales approximately as $\chi^{(3)}$63 for comparable coupling conditions. The low threshold resulted from the combination of LN’s large $\chi^{(3)}$64, use of $\chi^{(3)}$65, strong fundamental-mode overlap, radial quasi-phase matching, and several-hundred-thousand quality factors.

Mid-infrared difference-frequency generation

TFLN-on-sapphire extends nonlinear conversion into the $\chi^{(3)}$66–$\chi^{(3)}$67 molecular-fingerprint region. A uniformly poled waveguide used a fixed $\chi^{(3)}$68 pump and a tunable $\chi^{(3)}$69–$\chi^{(3)}$70 telecom signal to generate idler wavelengths from approximately $\chi^{(3)}$71 to $\chi^{(3)}$72. The device used a $\chi^{(3)}$73 LN film, approximately $\chi^{(3)}$74 ridge width, $\chi^{(3)}$75 etch depth, and a poling period near $\chi^{(3)}$76.

Dispersion engineering placed the device near a zero group-velocity-mismatch point. The measured normalized efficiency was

$\chi^{(3)}$77

compared with a simulated value of approximately $\chi^{(3)}$78. The measured full width at half maximum was approximately $\chi^{(3)}$79, or approximately $\chi^{(3)}$80 near $\chi^{(3)}$81; the bandwidth between the first sinc zeros exceeded $\chi^{(3)}$82 (Mishra et al., 2022).

Sapphire was selected because silica has appreciable absorption in parts of the mid-infrared. Surface-adsorbed water on air-clad waveguides nevertheless introduced wavelength-dependent MIR loss, particularly near $\chi^{(3)}$83. Operation in dry nitrogen and above approximately $\chi^{(3)}$84 substantially reduced this limitation.

Quantum pulse gates and entangled photons

A TFLN quantum pulse gate used type-0 sum-frequency generation with co-polarized fields, $\chi^{(3)}$85, a $\chi^{(3)}$86 effective mode area, and a $\chi^{(3)}$87 periodically poled interaction length. The geometry produced near signal–pump group-velocity matching with a simulated phase-matching angle of approximately $\chi^{(3)}$88. The measured phase-matching width was $\chi^{(3)}$89, or approximately $\chi^{(3)}$90.

The device demonstrated temporal-mode selectivity of $\chi^{(3)}$91 and internal conversion efficiency of $\chi^{(3)}$92 at $\chi^{(3)}$93 pump power. The conservative normalized efficiency was

$\chi^{(3)}$94

approximately three orders of magnitude above previous QPGs based on titanium-indiffused LN (Babel et al., 10 Aug 2026).

An unpatterned $\chi^{(3)}$95-thick z-cut LN film generated polarization-entangled photon pairs around $\chi^{(3)}$96 from a $\chi^{(3)}$97 pump. The three-fold rotational symmetry of LN provided coherent pair-generation pathways through $\chi^{(3)}$98-related tensor components. Linear pump polarization produced $\chi^{(3)}$99 or $\mathrm{LiNbO_3}$00 Bell states, with measured concurrence values of $\mathrm{LiNbO_3}$01 and $\mathrm{LiNbO_3}$02, respectively. Circular pump polarization generated separable $\mathrm{LiNbO_3}$03 or $\mathrm{LiNbO_3}$04 states, while elliptical polarization continuously tuned the concurrence over approximately $\mathrm{LiNbO_3}$05 (Shinde et al., 30 Apr 2026).

The same material platform therefore supports both highly efficient guided-wave quantum frequency conversion and direct polarization-entangled-pair generation. The unstructured entanglement source did not require periodic poling, a second nonlinear crystal, a Sagnac interferometer, a crossed-crystal arrangement, or nanostructuring.

5. Acoustic, piezoelectric, and heterogeneous quantum platforms

Lithium niobate’s piezoelectricity enables acoustic resonators, surface-acoustic-wave devices, acousto-optic modulators, microwave filters, and phononic transducers. Relevant acoustic velocity is approximately $\mathrm{LiNbO_3}$06 for many surface or thin-film modes. Acoustic and optical fields can be co-confined in TFLN, while substrate selection determines acoustic leakage, power handling, and mode overlap.

Millimeter-wave acoustic filters

Transferred LN on amorphous silicon and silicon or sapphire enables acoustic filters beyond $\mathrm{LiNbO_3}$07. A third-order ladder filter on an LN–aSi–sapphire platform achieved $\mathrm{LiNbO_3}$08 insertion loss and $\mathrm{LiNbO_3}$09 3-dB fractional bandwidth at $\mathrm{LiNbO_3}$10. An LN–aSi–silicon device achieved $\mathrm{LiNbO_3}$11 insertion loss and $\mathrm{LiNbO_3}$12 fractional bandwidth at $\mathrm{LiNbO_3}$13, with a footprint of approximately $\mathrm{LiNbO_3}$14 (Barrera et al., 2023).

The filters use thin transferred single-crystal LN, aluminum interdigital transducers, thickness-shear or antisymmetric acoustic modes, two LN thicknesses for series and shunt resonators, and XeF$\mathrm{LiNbO_3}$15 release of the a-Si sacrificial layer. The sapphire stack exhibited a $\mathrm{LiNbO_3}$16-arcsec HRXRD rocking-curve width, compared with $\mathrm{LiNbO_3}$17 arcsec for the silicon stack. Substrate electrical conduction, crystal quality, thickness variation, routing inductance, and electromagnetic feedthrough contribute to the platform differences.

A separate $\mathrm{LiNbO_3}$18 Y-cut LN acoustic filter used an approximately $\mathrm{LiNbO_3}$19 LiNbO$\mathrm{LiNbO_3}$20 platform on amorphous silicon and silicon. Its $\mathrm{LiNbO_3}$21 mode operated at $\mathrm{LiNbO_3}$22 with $\mathrm{LiNbO_3}$23 insertion loss and $\mathrm{LiNbO_3}$24 fractional bandwidth. A bilayer periodically poled piezoelectric film on a-Si and sapphire produced a $\mathrm{LiNbO_3}$25 acoustic resonance with $\mathrm{LiNbO_3}$26, $\mathrm{LiNbO_3}$27, and figure of merit $\mathrm{LiNbO_3}$28. The same device exhibited a $\mathrm{LiNbO_3}$29 resonance with $\mathrm{LiNbO_3}$30, $\mathrm{LiNbO_3}$31, and figure of merit $\mathrm{LiNbO_3}$32 (Kramer et al., 2023).

LiNDa and spin–phonon coupling

A heterogeneous thin-film lithium-niobate-on-diamond platform, termed LiNDa, combines a $\mathrm{LiNbO_3}$33 x-cut TFLN film with diamond-hosted silicon-vacancy centers. Aluminum interdigital transducers generate surface acoustic waves that propagate in LN and extend strain into diamond. The platform operates near $\mathrm{LiNbO_3}$34, with measured microwave-to-acoustic transmission of $\mathrm{LiNbO_3}$35 over $\mathrm{LiNbO_3}$36, intrinsic acoustic quality factor approximately $\mathrm{LiNbO_3}$37, and low-temperature propagation loss of $\mathrm{LiNbO_3}$38.

Simulations predicted $\mathrm{LiNbO_3}$39 and single-spin coupling rates of $\mathrm{LiNbO_3}$40–$\mathrm{LiNbO_3}$41, compared with approximately $\mathrm{LiNbO_3}$42–$\mathrm{LiNbO_3}$43 for an AlN-on-diamond comparison. Coherent control of a single $\mathrm{LiNbO_3}$44 electron spin produced an acoustic Rabi frequency of $\mathrm{LiNbO_3}$45 at $\mathrm{LiNbO_3}$46 microwave power, more than twice the comparable AlN-on-diamond result (Xu et al., 13 May 2025).

The platform illustrates the value of combining LN’s piezoelectricity with diamond’s high acoustic velocity and strain-sensitive quantum centers. Remaining limitations include acoustic propagation loss, bidirectional IDT operation, finite transferred-membrane size, nondeterministic spin placement, and environmental contamination.

Heterogeneous integration

TFLN does not intrinsically provide lasers, high-performance detectors, memories, or all electronic functions. Heterogeneous integration with silicon, silicon nitride, III–V materials, amorphous silicon, sapphire, diamond, superconducting nanowire detectors, rare-earth ions, and quantum emitters is therefore central to complete systems.

Reported integrations include LN bonded to silicon or silicon-nitride photonic circuits, amorphous-silicon photodetectors with responsivities of approximately $\mathrm{LiNbO_3}$47–$\mathrm{LiNbO_3}$48, InAs quantum dots coupled to LN circuits with approximately $\mathrm{LiNbO_3}$49 efficiency, Er-doped LNOI resonators with loaded $\mathrm{LiNbO_3}$50 near $\mathrm{LiNbO_3}$51, and Er-doped LN waveguides with approximately $\mathrm{LiNbO_3}$52 net gain in a $\mathrm{LiNbO_3}$53 spiral (Zhu et al., 2021).

6. Cryogenic operation, manufacturing constraints, and future directions

Cryogenic TFLN operation is relevant to superconducting detectors, microwave–optical transducers, solid-state quantum emitters, and quantum communication systems. Measurements from room temperature to approximately $\mathrm{LiNbO_3}$54 show that TFLN remains functional but is not temperature invariant. A racetrack resonance shifted by approximately $\mathrm{LiNbO_3}$55 during cooling from $\mathrm{LiNbO_3}$56 to $\mathrm{LiNbO_3}$57. An MZI half-wave voltage increased from $\mathrm{LiNbO_3}$58 to $\mathrm{LiNbO_3}$59, a $\mathrm{LiNbO_3}$60 increase. Type-0 SHG phase matching blue-shifted by $\mathrm{LiNbO_3}$61, from $\mathrm{LiNbO_3}$62 to $\mathrm{LiNbO_3}$63, while Type-II SFG phase matching red-shifted by $\mathrm{LiNbO_3}$64, from $\mathrm{LiNbO_3}$65 to $\mathrm{LiNbO_3}$66 (Kuttner et al., 31 Jul 2026).

These changes arise from temperature-dependent refractive indices, electro-optic coefficients, permittivity, thermal contraction, waveguide geometry, and poling-period contraction. Consequently, cryogenic TFLN designs require operating-temperature phase matching, on-chip thermal monitoring, cryogenic electro-optic parameters, and geometry-dependent nonlinear simulations.

The principal technical challenges are:

• Optical scattering: etched sidewall roughness remains a dominant residual loss mechanism.
• Photorefraction and charge effects: resonance drift, hysteresis, bias drift, and instability are especially relevant at visible wavelengths and in high-$\mathrm{LiNbO_3}$67 cavities.
• Periodic poling: thin-film poling faces high coercive fields, leakage through oxide and substrate, nonuniform domain depth, duty-cycle errors, and short-period limits.
• Fiber coupling and packaging: high optical confinement produces mode mismatch; inverse tapers, spot-size converters, and grating couplers reduce but do not eliminate coupling loss.
• Acoustic loss and power handling: suspended devices improve acoustic confinement but reduce robustness and RF power tolerance.
• Wafer-scale uniformity: film-thickness variation, etch nonuniformity, bonding defects, and release variation affect optical, nonlinear, and acoustic performance.
• CMOS compatibility: lithium diffusion and process residues create contamination concerns, motivating heterogeneous integration and dedicated LN foundries.
• Environmental stability: mid-infrared devices are sensitive to surface-adsorbed water, while resonant electro-optic devices require thermal and photorefractive stabilization.

Future directions include lower-loss etched waveguides, stoichiometric or doped LN with reduced photorefraction, higher-quality and shorter-period poling, integrated lasers and detectors, higher-order mmWave acoustic filters, LN-on-sapphire and LN-on-diamond phononic systems, low-noise quantum transducers, electro-optic and Kerr soliton sources, programmable frequency-domain processors, integrated $\mathrm{LiNbO_3}$68-$\mathrm{LiNbO_3}$69 systems, compact quantum pulse gates, entangled-photon sources, and heterogeneous photonic–electronic packaging.

The broad significance of TFLN lies in the coexistence of several functions in one material system: low-loss routing, high-$\mathrm{LiNbO_3}$70 optical confinement, fast Pockels modulation, strong $\mathrm{LiNbO_3}$71 and useful $\mathrm{LiNbO_3}$72 nonlinearities, ferroelectric-domain engineering, piezoelectric actuation, acoustic filtering, and compatibility with quantum emitters and detectors. The platform’s principal limitations are predominantly fabrication, packaging, environmental, and thermal-engineering constraints rather than the absence of a fundamental optical or electro-acoustic mechanism.