OFC 2026 - TFLN Hits Its Inflection Point: Volume Scaling and the 1.6T/3.2T Deployment Path - HyperLight, Broadcom, Ciena, Eoptolink, Jabil

2 days ago

4 min read

At this high-profile OFC 2026 panel, thin-film lithium niobate (TFLN) formally declared its move from lab "technology demos" to the inflection point of large-scale commercialization. As AI compute pushes bandwidth and power to the limit, TFLN's ultra-high bandwidth and low drive voltage make it the key material for overcoming silicon photonics (SiPh) performance bottlenecks in the 1.6T and even 3.2T generations.

Core Technical View: From Academic Breakthrough to Industrial Volume

Harvard professor and HyperLight co-founder Marko Loncar reviewed TFLN's history, noting that the technical key was overcoming redeposition during the etching process.

Loss performance evolution: labs reached a low loss of 3 dB/m in 2017, the current lab record is 1 dB/m, and the material's limit on certain vendors' oxide platforms is about 0.1 dB/m.
Wavelength compatibility: TFLN has a wide bandgap covering the visible through near-infrared bands. Below 1 µm, its Vπ efficiency improves sharply, scaling inversely with the square of wavelength.

Full Panel Recap by Company: Strategic Positioning for the 1.6T Generation

1. Ciena: Coherent Optics' Absolute Craving for "Low Loss"

Ciena pointed out that the coherent driver modulator (CDM) is TFLN's immediate product opportunity.

Pain point: modulation loss: in a typical coherent transmit chain, modulator loss reaches as high as 26 dB, severely limiting optical signal-to-noise ratio (OSNR).
The 1.6T challenge: moving to 1.6T doubles signal bandwidth and cuts SNR by 3 dB. TFLN's low Vπ and high bandwidth (measured at only 6–8 dB roll-off at 100 GHz) can offset this degradation.
Direction for improvement: TFLN today is mostly single-ended drive; Ciena hopes to see it evolve toward differential drive to save power.

2. Broadcom: The "Physics First" Principle for the 425G Single-Lambda Era

Broadcom just launched its Taurus-generation 400G DSP.

Fix the physical layer instead of leaning on the DSP: Vasu stressed that the DSP is no cure-all. CMOS scaling improves analog front-end (AFE) bandwidth by only about 5% per generation, so jitter and crosstalk must be solved at the signal source.
The FEC gain trap: powerful forward error correction (FEC) brings huge overhead and power. If TFLN improves physical link quality, there is no need to rely on overly complex FEC.

3. Eoptolink: Measured 1.6T/3.2T Module Performance

Eoptolink showcased a 1.6T DR4 module based on Broadcom's Taurus DSP.

Direct drive: the DSP output can drive the TFLN modulator directly, eliminating the cost and power of an external driver.
Key data: at 425 Gbps per lambda, direct drive achieves an extinction ratio (ER) of 3.5 dB and TDECQ of 2–2.5 dB, with a bit error rate (BER) of 1E-6 at -1 dBm received power.

4. Jabil: Getting Manufacturing Ready for "Industrialization"

As a major manufacturer, Jabil assesses that TFLN is now ready for large-scale volume production.

Compatibility: TFLN packaging flows (such as fiber coupling and wire bonding) are highly compatible with existing silicon photonics (SiPh) and indium phosphide (InP) infrastructure.
Power benchmark: Jabil's measured 1.6T LRO (linear receive optics) module consumes only 11.5 W.

5. HyperLight: Supply Chain Stability and Yield Breakthroughs

HyperLight is focused on turning R&D into production capacity.

Foundry partnership: in deep collaboration with UMC, the 6-inch line has entered volume production with annual capacity of 1 million known-good dies (KGD); the 8-inch line is in the final stage of pilot production.
Yield and reliability: end-to-end yield has reached 90%, and the devices have passed Telcordia GR-468 reliability qualification.
Power advantage: a 1.6T fully retimed module consumes roughly 20–21 W.

Consensus and Points of Divergence

Dimension	Industry consensus	Potential divergence / challenges
Performance value	TFLN is the only technology at 1.6T and above that balances bandwidth with low voltage.	Silicon photonics modulators are still evolving; TFLN must keep proving its cost advantage.
Drive architecture	Moving to differential drive is an inevitable trend to cut power further.	Mainstream products are still mostly single-ended; PCB design needs to be rethought.
Integration	TFLN needs to integrate more components (such as photodetectors, PDs).	Stray light and control complexity from integrated PDs remain unsolved.
Market timing	The 400G single-lambda standard is just getting started, with a breakout expected in 2027.	Early deployments may bridge with gearbox solutions rather than pure single-lambda designs.

Simple Tech Trend View: TFLN Holds a "Generational Lead"

This panel shows that TFLN is no longer far-off exotic technology. STT's editor-in-chief sees three factors shaping the landscape over the next two years:

AI-driven power dividend: HyperLight says about 1 billion 1.6T-equivalent ports will be needed by 2028. TFLN can save roughly 20% of system power, a decisive financial consideration for gigawatt-scale data centers.
Ecosystem maturity: UMC joining as a foundry and Jabil's manufacturing assurance remove Tier 1 customers' doubts about "supply stability." This stands in sharp contrast to BTO and polymer modulators, which are still in R&D.
Packaging is the key: TFLN's performance may be unbeatable, but its integration with the DSP (e.g., 3D packaging, differential drive) will decide the ultimate winner.

As the 425G-per-lambda standard firms up toward the end of 2026, TFLN will be first to take mainstream share in high-end 1.6T/3.2T pluggable modules, and from 2027 will begin to threaten the position of traditional EML and silicon photonics.