Mapping the TFLN Track: Players in the Year Mass Production Began

C114 News, September 4 (Shui Yi) — Large models have fully entered the trillion-parameter era this year, and AI agent applications are spreading rapidly, driving explosive growth in computing power demand. To meet this need, the industry is not only expanding computing infrastructure but also exploring how to fully unlock computing efficiency. Networking, as one of the key variables, has made metrics such as bandwidth, latency, power consumption and transmission efficiency a focus of industry attention.

With multiple advantages including high bandwidth, low latency, low power consumption and interference immunity, optical interconnect has taken center stage. It is no longer merely supporting infrastructure in the data center, but has become the critical foundation underpinning the continued expansion of AI computing power.

As the core building block of optical interconnect, optical modules perform key functions including optical signal generation, modulation, transmission and optoelectronic conversion. Their performance directly determines the transmission speed, quality and reliability of optical networks. Today, 800G is deployed at scale, 1.6T is ramping up rapidly, and 3.2T is accelerating technical validation, while emerging technologies such as NPO and CPO are also evolving quickly.

Each step up in optical module speed pushes the modulator toward its physical limits. As per-channel rates continue to climb, established materials such as indium phosphide (InP) and silicon photonics (SiPh) are beginning to fall short, and thin-film lithium niobate (TFLN) is gradually coming into industry view, moving quickly from the laboratory to volume production.

Why is TFLN needed?

In a recent article published in Semiconductor Digest, UK photonics company G&H detailed the electro-optic modulation performance of three major platforms: indium phosphide (InP), silicon photonics (SiPh), and thin-film lithium niobate (TFLN).

Silicon photonics modulates light by injecting or depleting charge carriers, and its modulation speed is fundamentally limited by the RC time constant of the doped silicon and the PN junction. Today's most advanced silicon modulators achieve a 3 dB bandwidth of about 70 GHz.

Indium phosphide uses electroabsorption, which gives it better performance at high modulation speeds but also introduces thermal management issues. Compared with SiPh devices, devices based on mature InP remain costly to manufacture at scale because of complex epitaxial stacks and small wafer sizes; however, they benefit from monolithic integration with other active devices.

Thin-film lithium niobate modulates light via the Pockels effect: applying an electric field directly changes the refractive index of the lithium niobate crystal. No carriers move during modulation, so no heat is generated, and the response is very fast. Commercially available TFLN modulators today operate at drive voltages below 2 V with electro-optic bandwidths exceeding 110 GHz.

Arguably, as optical modules upgrade from 800G to 1.6T and 3.2T, the large-bandwidth advantage of thin-film lithium niobate modulators will become more pronounced. Future optical communications will demand ever-higher cost, power consumption, and performance, and with its multi-faceted advantages in power consumption, cost, and performance, thin-film lithium niobate is poised to gain market share rapidly as data rates scale up.

This is not a zero-sum game for market share, but a hard constraint at the physics level. CITIC Securities noted in a research report that the 3.2T era is approaching, pure-silicon solutions are nearing the physical bandwidth limit, and thin-film lithium niobate is expected to become the mainstream modulation solution for the next generation. CITIC Securities believes that at the 400G/lane node, the material system will replace packaging technology as the core of competition, and TFLN, with its inherent Pockels-effect advantage, is poised for a leap in market share in the 3.2T era.

Of course, the transition to TFLN does not mean completely replacing other solutions; silicon photonics remains the mainstay of 1.6T shipments today. Moreover, history tells us that leadership in a single technology does not necessarily translate into success for the overall system — choosing the right solution for the scenario is what matters most.

An Independent Supply Chain Takes Shape

At the same time, industry insiders broadly agree that TFLN is the domestic optical chip segment with the best chance of competing on equal footing with global players, or even taking the lead. In other optical chip areas, such as silicon photonics and InP, Chinese players are typically playing catch-up. A complete independent chain now spans upstream materials, midstream high-speed modulator chips, and downstream packaging.

On the upstream materials side, TianTong Advanced Materials has moved both its 6-inch and 8-inch lithium niobate wafer materials into volume production, with its 12-inch materials completing R&D validation. Jinan Jingzheng holds a dominant global share of thin-film substrates, Fujian Crystech can supply high-quality lithium niobate crystals, Shanghai Xingui Polymerization Semiconductor offers lithium niobate and lithium tantalate single-crystal thin-film wafer products with multiple substrate options, and Nanjing Nanzhi Core Materials has independently developed 12-inch optical-grade lithium niobate crystals.

Midstream modulators are the most value-dense segment. In November 2019, Advanced Fiber Resources announced a $17 million cash acquisition of Lumentum's lithium niobate-related assets, including production equipment, patented technology, inventory, and the core team. Six-Cent Photonics has completed a thin-film lithium niobate photonic integrated chip based on 8-inch wafers, Anpei Photonics has built the world's first dedicated 8-inch thin-film lithium niobate production line, Xili Photonics positions itself as a "thin-film lithium niobate system-on-chip photonic engine provider," Nanzhi Photonics has established China's first "thin-film lithium niobate + X" optoelectronic chip production line, and DK Telecom subsidiary Niao Photonics focuses on the R&D and manufacturing of thin-film lithium niobate modulator chips.

Among downstream module makers, Chinese optical module vendors lead the world in market share, with Chinese companies claiming seven of the top ten global optical module suppliers, demonstrating commanding technical strength. Innolight, Eoptolink, Accelink Technologies, and Linktel Technologies, among others, have said on various occasions that they are advancing R&D on TFLN-based optical module solutions.

Overseas players are equally active. HyperLight, a startup spun out of Harvard University's Nano-Optics Laboratory, this year partnered with UMC to accelerate volume production of its TFLN Chiplet platform. Fujitsu has achieved volume production of modulators on 8-inch LNOI wafers, Sumitomo Electric and NGK hold key capacity for optical-grade lithium niobate single crystals and thin-film substrates, and France's SOITEC is advancing engineered LNOI substrates.

Challenges to Large-Scale Commercialization

As AI advances further, demand for optical interconnects will keep climbing. QYResearch data show that the global thin-film lithium niobate chip market was worth about $239 million in 2025 and is expected to reach $367 million by 2032, a compound annual growth rate of 6.4%. At the device level, the global thin-film lithium niobate modulator market is growing far faster: it was worth about $35.09 million in 2025 and is projected to surge to $740 million by 2032, a CAGR of 55.4%.

That said, frankly, TFLN is still some way from true large-scale commercialization. In its latest financial report, Advanced Fiber Resources said it is actively advancing the R&D and production of lithium niobate high-speed modulator chips and devices, pushing product validation with leading domestic and overseas customers across multiple application areas. Customer sampling has increased this year, small-volume shipments continue, and the company is working closely with end customers and module makers to develop next-generation data center products.

Even leading domestic TFLN players are still shipping only in small volumes. At the 8th Silicon Photonics Industry Conference, Cao Quan, chief expert at Feisiling Microelectronics, said that although 2026 has been dubbed the first year of large-scale TFLN commercialization, the industrialization of TFLN has "only just entered the second half."

He said thin-film lithium niobate currently faces seven engineering challenges: excessive size, mismatched drive schemes, lopsided functionality, immature packaging integration, reliability risks, unproven mass-production capability, and unclear cost expectations. Academia pursues ultimate bandwidth and ultra-low Vπ, but what industry really needs is small-size chips, differential drive, large-mode-field coupling, system-level margin, multi-wavelength consistency, and thermal stability.

Hao Ting, chip design and product manager at Advanced Fiber Resources, said at a forum co-hosted by CIOE and C114 that thin-film lithium niobate still needs continuous improvement. On the one hand, heterogeneous integration must be strengthened, efficiently combining TFLN devices with Si, SiN, III-V and other material systems to fully exploit the strengths of each material. On the other hand, advanced packaging should be promoted, including 3D packaging, thermal management, high-density array packaging, and die to wafer.

As Cao Quan put it, academia has already shown how good thin-film lithium niobate can be in the first half; the second half must be answered by industry.