TECHNOLOGY EXPLAINER
Thin-Film Lithium Niobate (TFLN)
— slicing a half-century-old crystal thinner than 1 µm
Lithium niobate (LN) is the crystal that has carried optical communication modulators for decades. By slicing that crystal thin from a single crystal and bonding it onto another substrate, rather than depositing it as a film, modulators have become smaller, lower-voltage and faster. But LN is hard to etch, and lithium is a contaminant in CMOS fabs. Both the source of the performance and the barrier to volume production lie on the materials and processing side.
- Thin-film lithium niobate in three points
- The electro-optic effect — changing the refractive index with a voltage
- Why making it thin makes it better
- A materials engineer's view (1): sliced and bonded, not deposited
- Modulators — what has been achieved
- Our calculation: the trade-off between length and voltage
- A materials engineer's view (2): hard to etch, hard to bring into the fab
- What is still hard — a material in which charge moves
- Glossary / References / Claim-to-source audit
Sourced = stated in published material or a peer-reviewed paper (link given)
Our calculation = a figure this article derived, with the assumptions spelled out
Not yet confirmed = a plan or research-stage result with no confirmed production record
Structural readings and materials-design interpretations are marked separately as Commentary.
Performance figures are kept apart as either laboratory prototypes (peer-reviewed papers) or company products (samples or volume production).
1. Thin-film lithium niobate in three points
Thin-film lithium niobate (TFLN) is single-crystal lithium niobate (LiNbO₃, LN) made into a film of sub-micrometre thickness and placed on an insulating layer. The substrate is also called LNOI (lithium niobate on insulator).
- What kind of material: a review by Marko Lončar's group at Harvard University (Zhu et al., 2021) describes LN as a material that has served everyday life for decades, from high-speed optical communications to RF filters in mobile phones, and says its Pockels coefficient of r₃₃ ≈ 31 pm/V is why it has been the material of choice for electro-optic modulatorsSourced
- What changed: high-quality thin-film LN wafers made by ion slicing and wafer bonding became commercially available, and together with advances in nanofabrication, some thin-film devices (modulators and wavelength converters) have outperformed their conventional bulk-crystal counterpartsSourced
- What makes it hard: etching LN tends to leave redeposited non-volatile LiF, and lithium is a contaminant in CMOS fabsSourced
A comparison of the platform materials for photonic integrated circuits is in our explainer on photonic integrated circuits and indium phosphide, and silicon modulators (ring and Mach-Zehnder) are covered in our explainer on silicon photonics. Modulation formats themselves (intensity modulation, PAM4, coherent IQ modulation and so on) are covered in our explainer on optical modulation and modulation formats, and signal processing for coherent links in our explainer on coherent optical communication and DSP. This article focuses on the materials and processes that turn LN into a thin film.
2. The electro-optic effect — changing the refractive index with a voltage
The electro-optic effect (Pockels effect) is the phenomenon in which applying an electric field to a crystal changes its refractive index in proportion to the field. When the refractive index changes, the speed of light changes, and after a fixed length the phase of the light has shifted. A modulator uses an interferometer to turn that phase shift into changes in light intensity (our commentary).
Shekhar and colleagues write in a 2024 Nature Communications paper that “The most ‘pure’ electro-optic modulation relies on the Pockels effect, which provides an intrinsically very high E/O BW, even exceeding 100 GHz”Sourced. Whereas a silicon modulator changes its refractive index by moving carriers (electrons and holes) in and out, the Pockels effect is a response of the crystal lattice itself (our commentary).
| Property of LN | Value (table and text of Zhu et al.'s review) | What it means |
|---|---|---|
| Electro-optic coefficients | r₃₃ = 30.9 pm/V, r₅₁ = 32.6 pm/V and others | How much the refractive index changes per unit field |
| Refractive index (1550 nm) | 2.21 (ordinary) / 2.13 (extraordinary) | Higher than SiO₂ (1.44), so a thin film can confine light |
| Transparency window | 350 nm to 5 µm | Usable from the visible to the mid-infrared |
| Curie temperature | about 1210 °C | A stable ferroelectric phase that tolerates a wide range of processing and operating conditions |
| Thermo-optic coefficient | dnₑ/dT = 3.34×10⁻⁵ K⁻¹ | The refractive index moves with temperature |
| Pyroelectric coefficient | about 95 µC m⁻² K⁻¹ | Temperature changes produce surface charge |
Sourced (Zhu et al., Advances in Optics and Photonics 13, 242, 2021 [Reference 1]). The right-hand column is this article's explanation. Zhu et al. note that the thermo-optic, pyroelectric and photorefractive effects can serve as tuning mechanisms but can also limit device operation.
3. Why making it thin makes it better
LN modulators have existed for a long time. What is different is how the waveguide is made. According to Zhu and colleagues, conventional LN waveguides have been made by thermal diffusion of titanium (Ti) or by proton exchangeSourced.
- Ti-diffused waveguides: thin Ti strips on a bulk LN substrate are heat-treated and diffuse a few micrometres in. The index increase is 0.001 to 0.04, depending on Ti densitySourced
- Proton-exchanged waveguides: the crystal is immersed in a hydrogen source at 150 to 400 °C, replacing Li⁺ with H⁺. The index changes only along the extraordinary axis, by about 0.1 or lessSourced
- The weakness: small index contrast means weak confinement. Mode areas of 10 to 100 µm² and bend radii on the millimetre scale make them ill-suited to dense integrationSourced
In thin-film LN, the LN film itself is etched into a narrow ridge and surrounded by a low-index material such as SiO₂. Zhu and colleagues explain that this makes the index contrast roughly an order of magnitude larger than in ion-diffused waveguides, confining the light tightlySourced.
When light is confined into a small space, the electrodes can sit right next to it. Zhu and colleagues say that in a typical X- or Y-cut modulator the gap between the (CPW) electrodes can be narrowed to about 5 µm, and that narrowing it further increases optical absorption and microwave lossSourced. For the same voltage, a narrower gap means a stronger field, so the same index change can be had at a lower voltage — which is why thin films lower the drive voltage (our commentary).
4. A materials engineer's view (1): sliced and bonded, not deposited
This is the part of the article most squarely aimed at materials engineers. Zhu and colleagues write that attempts to make thin LN films by sputtering, chemical vapour deposition, pulsed laser deposition, sol-gel and molecular beam epitaxy failed to reach high crystallinity, and the need for lattice matching severely restricted the choice of substrateSourced. What became the standard instead is “Smart Cut”, the technology used to make SOI wafersSourced.
Read as a process flow, Zhu and colleagues' account runs as followsSourced.
- A high-quality LN substrate is implanted with a high dose of He⁺ (or H⁺), creating a cleave plane at the depth of the target film thickness
- It is bonded to a support substrate (usually LN or Si) carrying an oxide layer
- Heat treatment splits it along the cleave plane, leaving a thin film behind
- A further anneal reduces crystal damage from the implantation, and polishing smooths the surface
- Bonding uses an adhesive such as BCB or direct bonding. Direct bonding is preferred for many applications because it allows higher-temperature anneals, which are needed to recover the nonlinear and electro-optic properties degraded by ion slicing
- Most LNOI is made from congruent rather than stoichiometric LN — that is, from lithium-deficient crystal
In other words, the quality of thin-film LN is set by three things: (1) the quality of the starting bulk single crystal, (2) how much of the implant damage the anneal can undo, and (3) whether the bonded interface survives a high-temperature anneal (our framing). It is not a matter of dialling in deposition conditions, but of the combined strength of an entirely different set of processes: single-crystal growth, ion implantation, bonding, annealing and polishing.
There is another route. NGK Insulators describes its TFLN composite wafer for optical communications as made by precision-polishing a composite wafer in which LN is bonded to a base substrate, giving thin-film LN free of crystal damageSourced. For its composite wafers in general, the company explains that the layers are directly bonded without adhesive and the functional layer is polished thin with high precision, and for TFLN it highlights high crystallinity and film-thickness uniformitySourced. Implant and split, or bond and then polish down — the difference is whether implant damage is repaired afterwards or never introduced in the first place, and in the second case it is polishing technology that can thin uniformly down to sub-micrometre thickness that decides performance (our commentary).
On wafer size, Zhu and colleagues note that as of 2021 LNOI wafers of up to 6 inches were commercially available from suppliers including NanoLN, Partow Technologies, NGK Insulators and SRICOSourced. They also describe wafer-scale processing of 4-inch and 6-inch LNOI wafers by deep-ultraviolet lithography, with post-etch film thickness variation held below 6 nmSourced. NGK's web page likewise labels its photograph of a TFLN composite wafer as 6 inchSourced.
5. Modulators — what has been achieved
Research: an LN modulator running at CMOS-level voltages, 2018
In 2018 Wang and colleagues at Harvard University and elsewhere reported in Nature a monolithically integrated LN modulator operating at CMOS-compatible drive voltagesSourced. The abstract describes conventional LN modulators as “bulky, expensive, limited in bandwidth and require high drive voltages”, and says the new devices showed data rates of up to 210 Gbit/s and on-chip optical loss below 0.5 dBSourced. Zhu and colleagues' review restates, for this device, an example with a 20 mm length, a Vπ (the voltage needed to switch the light off) of 1.4 V and an extinction ratio of 30 dBSourced. These are values from laboratory prototypes, not product specifications.
| Metric | What Zhu et al.'s review says (a summary of research reports) |
|---|---|
| VπL (monolithic MZM) | Typically 2.1 to 2.5 V·cm at low frequency. An example of 1.8 V·cm with a 3.5 µm electrode gapSourced |
| VπL (hybrid) | Tends to be somewhat larger because only part of the light is in the LN, but an example of 2.1 V·cm in a SiNₓ-loaded LN MZMSourced |
| Bandwidth | Modulators with more than 100 GHz at CMOS-compatible voltages, and coherent modulators at up to 320 Gbit/s, have been reportedSourced |
| What limits speed | Velocity matching is easier than in bulk; loss in the transmission line (electrodes) is the main limit, dominated by metal lossSourced |
| Bonding to silicon | Travelling-wave modulators with 3 dB bandwidths above 100 GHz have been reported in a hybrid design that bonds an LN film onto processed SOI waveguidesSourced |
Source: Zhu et al. [Reference 1]. All are research-stage values reported in peer-reviewed papers, not the specifications of any particular product.
Products: Japanese modulator makers move to thin-film LN
On 3 March 2023 Sumitomo Osaka Cement announced that it had begun sample shipments of a high-bandwidth coherent driver modulator (HB-CDM) for 1.2 Tbps optical communication using thin-film LNSourced. The company says its thin-film LN modulator chip and driver support 130 Gbaud operation and that the product is planned to comply with OIF-HB-CDM-02.0 Type 3Sourced. According to the company's note, the widely deployed 100 Gbps coherent modulator runs at 32 Gbaud, so 130 Gbaud is about four times thatSourced. At the time of the announcement these were sample shipments, with sales planned to start in the second half of 2023Not yet confirmed.
In the same release, Sumitomo Osaka Cement says it has supplied a broad line-up of LN modulators from 10G to 400G in Japan and overseasSourced. In other words, a company that has made LN modulators for many years has itself adopted thin-film LN for the next generation. The material is still LiNbO₃, but the way the wafer is made and processed is entirely different, so both the material supply chain and the process change (our commentary).
6. Our calculation: the trade-off between length and voltage
Zhu and colleagues explain that Vπ is ideally inversely proportional to the length L, and VπL is used as a figure of merit for comparing the strength of the electro-optic interactionSourced. We use it to estimate the length required.
Assumption: VπL = 2.2 V·cm (a value this article picked from the typical range of 2.1 to 2.5 V·cm given by Zhu et al.)Our calculation
- Length 0.5 cm → Vπ = 2.2 ÷ 0.5 = 4.4 V
- Length 1 cm → Vπ = 2.2 V
- Length 2 cm → Vπ = 1.1 V
- For reference, Wang et al.'s device (Vπ of 1.4 V at 20 mm) corresponds to VπL = 1.4 × 2.0 = 2.8 V·cm
Assumptions and limits: this is a simple inverse-proportional conversion of low-frequency (near-DC) values. Zhu et al. note that longer modulators lose bandwidth to transmission loss in the electrodesSourced, so in practice the design is a three-way balance of voltage, length and bandwidth.
7. A materials engineer's view (2): hard to etch, hard to bring into the fab
Zhu and colleagues are specific about why LN is hard to processSourced.
- Fluorine-based reactive ion etching: niobium forms a volatile fluoride and can be removed, but lithium forms non-volatile LiF, which causes severe redeposition. The LiF is even harder to etch and increases sidewall roughness and scattering loss
- Ar⁺ etching (physical milling): redeposition keeps the sidewalls from being vertical. However, the redeposited LN is smooth and can be removed by a wet treatment after the dry etch
- Waveguides made by CMP: using a Cr mask and chemical mechanical polishing, an example with sub-nanometre surface roughness and propagation loss of 0.027 dB/cm. But the sidewalls are shallow-sloped, so bend radii cannot be made small
- A history of improvement: optimising lithography, etching and post-etch cleaning brought the loss of dry-etched LNOI waveguides down from more than 6 dB/cm in 2007 to 0.027 dB/cm in 2017, and below 0.1 dB/cm is now routinely achieved by several groups
- The main source of loss: in dry-etched LNOI waveguides it is sidewall roughness. Material absorption measured in bulk congruent LN is below 0.4 dB/m, so a gap remains
So most of the optical loss comes not from the material itself but from the surfaces created by processing (our commentary). Lithium-containing compounds produce by-products that do not volatilise during etching — a property that will be familiar to materials engineers who handle lithium-based materials.
And the same lithium makes it hard to bring the material into the fab. Shekhar and colleagues write that “Lithium, a contaminant in CMOS foundries, restricts FEOL integration”Sourced. The paper therefore treats hybrid assembly of etched LNOI modulators onto silicon PICs as the practical solution, and says that bonding unetched LN onto silicon has been achieved through back-end-of-line (BEOL) integration and packaging techniquesSourced. Leaving the LN unetched avoids the problems that come with etched LNOI — formation of structural defects, Nb depletion, and accumulation of heat and pyroelectric charge — but makes strong optical confinement harderSourced.
8. What is still hard — a material in which charge moves
(1) Photorefraction and bias drift
Zhu and colleagues call effects caused by mobile charge — photoconductivity, the photorefractive effect and dielectric relaxation — “long-standing problems” of LN, saying they often limit the stability and power handling of LN devices and appear to be stronger in LNOI than in bulkSourced.
- Photorefractive effect: light excites charge out of crystal defects, and the field that charge creates changes the refractive index. A stronger and faster response than in bulk has been reported in LNOI; it can be reduced by annealing or by waveguides without cladding, but more research is needed to understand how material properties affect itSourced
- Dielectric relaxation: in a material with permittivity ε and resistivity ρ, charge moves with a relaxation time τ = ερ and screens the applied field, which can cause bias voltage drift in electro-optic devices. There are still few direct studies in LNOI, but drift consistent with dielectric relaxation has been observedSourced
The expression τ = ερ says that a modulator's long-term stability is set by the resistivity of the crystal — that is, by impurities, defects and lithium deficiency (our commentary). As section 4 showed, most LNOI is made from lithium-deficient congruent crystal and goes through an anneal to undo ion implantation damageSourced. Zhu and colleagues also cite reports that the field needed to reverse the polarisation (the coercive field) is higher in thin-film LN than in bulk, with the bonded interface and out-diffusion of Li⁺ during annealing suggested as possible causesSourced.
So how the film was made (implantation, bonding, annealing) may reach, through composition and defects, all the way to a modulator's drift and optical power handling. That quantitative relationship, however, has not been established in the primary sources this article consulted (our commentary).
(2) The distance from the CMOS fab
Since lithium is a contaminant in CMOS fabs, building thin-film LN directly into the silicon front end is restrictedSourced. Whether hybrid assembly, back-end bonding or a dedicated line becomes the mainstream route to volume production was not settled as of this article's research (September 2026)Not yet confirmed.
(3) Only limited information on how far commercialisation has gone
Sumitomo Osaka Cement's thin-film LN modulator was announced in March 2023 as being in sample shipment, with sales planned to start later that yearSourced. Its subsequent production volume, and how widely thin-film LN has been adopted in short-reach data centre applications (optical transceivers), could not be confirmed in the primary sources this article consulted.
- Thin-film LN is single-crystal LN sliced thin (or polished down) and bonded onto an insulating layerSourced
- Because it confines light tightly, electrodes can come to within about 5 µm, lowering the drive voltageSourced
- At a VπL of 2.2 V·cm, 1 cm gives 2.2 V and 2 cm gives 1.1 V — voltage and length (and bandwidth) trade off against each otherOur calculation
- Better processing has cut loss to about 1/220, but it is still about seven times the material limitOur calculation
- The challenges are LiF redeposition in etching, lithium contamination in CMOS fabs, and photorefraction and bias drift — all of them materials and process problemsSourced
9. Glossary
- Lithium niobate (LN)
- LiNbO₃. A ferroelectric single crystal with large electro-optic, piezoelectric and nonlinear optical effects.
- TFLN / LNOI
- Thin-film lithium niobate / lithium niobate on insulator. A substrate with a thin LN film on SiO₂ or similar.
- Pockels effect
- A change in refractive index proportional to the applied field. The linear electro-optic effect.
- Electro-optic coefficient r₃₃
- How much a field along the crystal Z axis changes the refractive index for light polarised along the Z axis.
- Vπ (half-wave voltage)
- The voltage needed to shift the phase of light by π (switching the light off in an MZM). The lower it is, the less power is needed.
- VπL
- The product of Vπ and electrode length. Used to compare modulation efficiency independently of length.
- MZM
- Mach-Zehnder modulator. Splits light into two paths, adds a phase difference and recombines them to produce intensity changes.
- Push-pull
- A configuration that applies opposite fields to the two arms, producing opposite phase changes.
- CPW
- Coplanar waveguide. A planar RF transmission line with the signal line between two grounds.
- Velocity matching
- Making the electrical signal on the electrodes travel at the same speed as the light, so that modulation builds up in the same direction.
- X-cut / Z-cut
- The orientation in which the crystal was cut. In X-cut, the Z axis lies in the plane.
- Smart Cut / ion slicing
- Creating a cleave plane by ion implantation, then bonding and splitting with heat to transfer a thin film. Also used for SOI wafers.
- Congruent composition
- The composition at which the crystal grows with the same composition as the melt. In LN, lithium is below the stoichiometric ratio.
- Ti-diffused waveguide
- The conventional waveguide made by thermally diffusing Ti into bulk LN. Its index contrast is small.
- Proton exchange
- A waveguide-forming method that raises the refractive index by replacing Li⁺ in LN with H⁺.
- Photorefractive effect
- A change in refractive index, via the electro-optic effect, caused by the field of charge excited by light.
- Dielectric relaxation
- Charge in a material moving over time to screen an applied field.
- BEOL / FEOL
- Back end of line (the wiring steps) / front end of line (the steps that make the transistors) in semiconductor manufacturing.
- Gbaud
- A unit for the number of modulation events (symbols) per second.
- HB-CDM
- High-bandwidth coherent driver modulator. An OIF implementation specification for a wideband coherent modulator with integrated driver.
10. References (primary sources)
- D. Zhu, L. Shao, M. Yu et al., M. Lončar (Harvard University and others) “Integrated photonics on thin-film lithium niobate”, Advances in Optics and Photonics 13(2), 242–352 (2021, DOI: 10.1364/AOP.411024). The text was checked in the authors' version (arXiv:2102.11956) — arxiv.org
- C. Wang, M. Zhang, X. Chen et al. “Integrated lithium niobate electro-optic modulators operating at CMOS-compatible voltages”, Nature 562, 101–104 (2018, DOI: 10.1038/s41586-018-0551-y) — nature.com
- S. Shekhar, W. Bogaerts, L. Chrostowski, J. E. Bowers et al. “Roadmapping the next generation of silicon photonics”, Nature Communications 15, 751 (2024, DOI: 10.1038/s41467-024-44750-0) — nature.com
- NGK Insulators “Next-generation wafers contributing to ultra-high-speed communications and power devices for EVs (under development)”, product and technology page (in Japanese) — ngk.co.jp
- NGK Insulators “NGK takes part in ECOC 2025, exhibiting composite wafers and ceramic packages for optical communications”, 17 September 2025 (in Japanese) — ngk.co.jp
- Sumitomo Osaka Cement “Sample shipments begin of an optical communication component (LN modulator) for 1.2 Tbps transmission”, 3 March 2023 (in Japanese) — soc.co.jp
11. Claim-to-source audit
| Claim in the text | Basis | Label |
|---|---|---|
| That LN has been used for decades from high-speed optical communications to RF filters in mobile phones; that high-quality thin-film LN wafers made by ion slicing and wafer bonding became commercially available and some thin-film devices outperformed bulk ones; r₃₃ ≈ 31 pm/V; the properties in the table (r₃₃ = 30.9 and r₅₁ = 32.6 pm/V, refractive indices 2.21 / 2.13, SiO₂ 1.44, transparency 350 nm to 5 µm, Curie temperature about 1210 °C, dnₑ/dT = 3.34×10⁻⁵ K⁻¹, pyroelectric about 95 µC m⁻² K⁻¹) and that these can limit operation; Ti diffusion (index increase 0.001 to 0.04, a few micrometres of diffusion) and proton exchange (150 to 400 °C, Δnₑ about 0.1 or less), mode areas of 10 to 100 µm² and mm-scale bend radii; that the index contrast of thin-film LN is roughly an order of magnitude larger than ion-diffused waveguides, with sloped sidewalls and a slab; that the CPW electrode gap can be narrowed to about 5 µm; that sputtering, CVD, PLD, sol-gel and MBE failed to reach high crystallinity; the Smart Cut steps (He⁺ / H⁺ implantation, bonding, thermal splitting, annealing, polishing) and support substrates, BCB and direct bonding, the preference for direct bonding because a high-temperature anneal can recover properties, and that most LNOI is congruent (Li-deficient) LN; that LNOI of up to 6 inches was commercially available in 2021 from NanoLN, Partow, NGK and SRICO; wafer-scale processing of 4- and 6-inch wafers with film thickness variation below 6 nm; that a push-pull MZM in CPW electrodes is the most common layout, with the Z axis in plane for X- or Y-cut; the example of Vπ 1.4 V and 30 dB extinction ratio at 20 mm; typical VπL of 2.1 to 2.5 V·cm, 1.8 V·cm with a 3.5 µm gap, larger values for hybrids and 2.1 V·cm with SiNₓ loading; reports of above 100 GHz and 320 Gbit/s; that velocity matching is easier than in bulk and transmission-line loss (mainly metal loss) is the limit; above 100 GHz in LN / SOI hybrids; that Vπ is inversely proportional to L with VπL as a figure of merit, and that longer devices lose bandwidth; LiF redeposition and sidewall roughness in fluorine RIE, redeposition and non-vertical sidewalls with Ar⁺ etching and their wet removal, CMP waveguides at 0.027 dB/cm with shallow sidewalls, the improvement of dry-etched waveguide loss from over 6 dB/cm in 2007 to 0.027 dB/cm in 2017 with below 0.1 dB/cm now routine, sidewall roughness as the main loss in dry-etched waveguides, and material absorption in bulk below 0.4 dB/m; that mobile-charge effects are “long-standing problems” that appear stronger in LNOI, the mechanism and mitigations of the photorefractive effect, and dielectric relaxation τ = ερ with bias drift; and reports of a higher coercive field in thin-film LN than in bulk, with the bonded interface and Li⁺ out-diffusion as suggested causes | Zhu et al., Advances in Optics and Photonics 13, 242 (2021), authors' version, Reference 1 https://arxiv.org/abs/2102.11956 | Sourced |
| A monolithically integrated LN modulator operating at CMOS-compatible drive voltages, with data rates up to 210 Gbit/s and on-chip optical loss below 0.5 dB; that conventional LN modulators are “bulky, expensive, limited in bandwidth and require high drive voltages”; and publication in Nature in 2018 | Wang et al., Nature 562, 101 (2018), Reference 2 https://www.nature.com/articles/s41586-018-0551-y | Sourced |
| “The most ‘pure’ electro-optic modulation relies on the Pockels effect, which provides an intrinsically very high E/O BW, even exceeding 100 GHz”; “Lithium, a contaminant in CMOS foundries, restricts FEOL integration”; that hybrid assembly of etched LNOI modulators is the practical solution, that bonding of unetched LN has been achieved through BEOL integration and packaging, and that leaving LN unetched avoids structural defects, Nb depletion and accumulation of heat and pyroelectric charge but makes confinement harder | Shekhar et al., Nature Communications 15, 751 (2024), Reference 3 https://www.nature.com/articles/s41467-024-44750-0 | Sourced |
| That the TFLN composite wafer for optical communications is described as made by precision-polishing a composite wafer in which LN is bonded to a base substrate, giving thin-film LN free of crystal damage; and that the photograph is labelled 6 inch | NGK Insulators, next-generation wafers (under development) page, Reference 4 https://www.ngk.co.jp/rd/wafer/ | Sourced |
| That the composite wafers are directly bonded without adhesive and the functional layer polished thin with high precision; that high crystallinity and film-thickness uniformity are highlighted for TFLN; and the presentation at ECOC 2025 | NGK Insulators news (17 September 2025), Reference 5 https://www.ngk.co.jp/news/20250917_2.html | Sourced |
| That on 3 March 2023 the company began sample shipments of an HB-CDM for 1.2 Tbps using thin-film LN; 130 Gbaud operation, planned compliance with OIF-HB-CDM-02.0 Type 3, and about four times the 32 Gbaud of 100 Gbps coherent modulators; and that it has supplied LN modulators from 10G to 400G | Sumitomo Osaka Cement news release (3 March 2023), Reference 6 https://www.soc.co.jp/news/69155/ | Sourced |
| That sales were planned to start in the second half of 2023 | A plan as of the announcement; the outcome was not confirmed by this article, Reference 6 https://www.soc.co.jp/news/69155/ | Not yet confirmed |
| Vπ assuming VπL = 2.2 V·cm (4.4 V at 0.5 cm, 2.2 V at 1 cm, 1.1 V at 2 cm); converting Wang et al.'s device to a VπL of 2.8 V·cm; converting 0.4 dB/m to 0.004 dB/cm; 6 ÷ 0.027 ≈ 222 (about 1/220) and 0.027 ÷ 0.004 ≈ 6.8 (about 7 times) | Our calculation. 2.2 V·cm is an assumption this article picked from Zhu et al.'s typical range. Near-DC values converted by inverse proportion, without regard to bandwidth. The loss values are representative figures from different fabrication methods and structures set side by side | Our calculation |
| Which route to integrating thin-film LN with silicon in volume production (hybrid, BEOL bonding, dedicated line) becomes mainstream; the production volume of thin-film LN modulators and their adoption in short-reach data centre applications; and the quantitative relationship between film fabrication conditions and drift or optical power handling | Not stated because no primary source could be confirmed as of this article's research (September 2026). This article's own judgement | Not yet confirmed |
| The explanation that the Pockels effect is a lattice response, unlike carrier-based modulation; the explanation that a narrow electrode gap lowers the voltage; the framing that thin-film quality is set by single-crystal quality, recovery from implant damage and the bonded interface; contrasting ion slicing with bonding plus polishing as repairing damage afterwards versus never introducing it; the reading that most loss comes from processed surfaces; the point about the non-volatility of lithium compounds; the reading from τ = ερ that long-term stability is set by resistivity and defects; the point that the same LN comes with a different way of making and processing the wafer; and how the figures are drawn | Our own framing and commentary based on the published material. Not views expressed by the companies or institutions | Commentary |
| That Figs. 1 to 4 and Fig. 6 are explanatory drawings rather than real cross-sections, design drawings or measured data; that the hero image and Fig. 5 are AI-generated images; and that “leftover LN may be reused” in Fig. 2 is our addition, not stated in the source | Our note | Commentary |
Last updated 25 September 2026. Sources are limited to primary material (peer-reviewed papers and reviews, and official announcements and product pages from companies). Because the article includes numerical conversions and materials-design readings, those are marked as “Our calculation” or “Commentary” and kept separate from sourced fact. Performance figures from research papers (Vπ, bandwidth, data rate, loss) are values for prototype devices, not product specifications. The production volume of thin-film LN modulators, their adoption in data centre applications, and the quantitative relationship between film fabrication conditions and drift are not stated here because no published primary source could be confirmed. All figures are explanatory concept graphics. Figs. 1 to 4 and Fig. 6 are vector drawings, and the hero image and Fig. 5 are AI-generated images; none of them shows a real cross-section photograph, micrograph or physical product.