Why Has Thin-Film Lithium Niobate Suddenly Become So Popular?

Not long ago, NVIDIA CEO Jensen Huang stated that next-generation AI infrastructure will require a massive amount of optical interconnects, as copper cables can no longer meet the demand.

This is not alarmist talk.

We Are Entering the World of Light

With the rapid development of information technology, global data traffic is growing exponentially, and the demand for information capacity and processing capability continues to rise. Driven by emerging technologies such as 5G communications, the Internet of Things, cloud computing, big data, and artificial intelligence, traditional electronic communication systems are increasingly facing bandwidth bottlenecks and high power consumption challenges.

Optical communication technology, with its advantages of high bandwidth, low loss, and immunity to electromagnetic interference, has become a key solution to these challenges.

The fundamental reason why next-generation AI infrastructure must rely heavily on optical interconnects is that the “interconnect wall” has replaced computing power as the biggest bottleneck. As GPU clusters scale to tens of thousands or even hundreds of thousands of cards, single-channel data rates are moving toward 224G. At the physical layer, copper cables are hitting the limits imposed by skin effect and dielectric loss, compressing their effective transmission distance to less than 2 meters. This makes them unable to meet the scale-out requirements across server racks.

At the same time, all-optical interconnects can reduce power consumption per unit bandwidth by more than 40%, making them one of the most promising paths to solving the energy-efficiency crisis in AI factories.

Lithium Niobate: A Material That Waited Decades for Its Moment

Electro-optical modulators, or EOMs, are key components in optical communication systems. Their main function is to convert and modulate electrical signals into optical signals. Their performance directly affects the transmission speed, power consumption, signal quality, and stability of the entire communication system.

Overall structure of an optical fiber communication system

Lithium niobate, or LiNbO₃, is a critical electro-optic material. With its excellent electro-optic effect, relatively high refractive index of around 2.2, wide transparency window from approximately 350 nm to 5 μm, and strong chemical stability, it has long been regarded in the photonics community as “optical silicon.” Since the 1960s, lithium niobate has been widely used in electro-optical modulators.

However, although lithium niobate has been indispensable at the system level, it was largely left behind for nearly three decades during the wave of chip-level integration.

The reason lies in the limitations of traditional bulk lithium niobate modulators. These devices modulate optical signals by using an electric field to control the phase or intensity of light. However, due to the physical properties of the material and the limitations of conventional processing technologies, bulk lithium niobate waveguides are typically on the millimeter-to-centimeter scale. This limits the interaction efficiency between the optical field and the electric field, meaning that effective modulation often requires high driving voltages ranging from several volts to tens of volts.

In addition, the large device size makes it difficult to integrate with silicon photonics platforms, restricting its application in chip-level optoelectronic systems. Traditional processing methods also result in relatively high waveguide transmission loss, further limiting device efficiency and long-distance transmission performance.

As a result, platforms such as silicon photonics, InP, and SiN began to rise rapidly, while lithium niobate was once viewed as a material with excellent performance but poor scalability and low integration density.

Thin-Film Technology Arrived Just as the Industry Needed It

The turning point came with the maturation of thin-film lithium niobate, or TFLN, technology.

Thin-film lithium niobate is based on a “lithium niobate–insulator–substrate” heterogeneous structure. Through advanced fabrication techniques such as crystal ion slicing and chemical mechanical polishing, single-crystal lithium niobate thin films can be separated from bulk material and transferred onto substrates such as silicon, sapphire, or silicon dioxide.

Compared with bulk lithium niobate, thin-film lithium niobate enables submicron-scale waveguide structures with much stronger optical confinement. This greatly enhances the interaction efficiency between optical and electrical fields, often by tens of times, thereby significantly reducing driving voltage and shrinking device size.

In addition, the low transmission loss of thin-film lithium niobate gives it unique advantages in long-distance photonic integrated circuits. Its compatibility with silicon-based platforms also provides a new route for heterogeneous integrated photonics.

Lithium niobate single-crystal thin film, source: Jinan Jingzheng Electronics Co., Ltd.

Whether a technology becomes popular depends partly on how good it is, and partly on whether the era gives it the right application demand.

Looking at several key performance indicators, it becomes clear why TFLN is being aggressively adopted in the 1.6T and 3.2T era:

1. Bandwidth: Easily exceeds 100 GHz and is moving toward 200 GHz.
2. Power consumption: Only tens of femtojoules per bit.
3. Signal quality: Low insertion loss, extremely low chirp, and excellent linearity.
4. Versatility: A single platform can support electro-optic, nonlinear, and quantum photonic functions.

From the perspective of industry demand, AI computing power is growing explosively. Data center optical interconnects are rapidly upgrading from 400G to 800G, 1.6T, and even 3.2T. This is exactly the kind of era that thin-film lithium niobate was made for.

Take today’s highly discussed co-packaged optics, or CPO, as an example. CPO moves the optical engine from the front-panel pluggable module directly onto the same package substrate as the switch chip or ASIC. After NVIDIA took the lead in mass-producing CPO solutions for its Spectrum-X and Quantum series, the measured results were striking: insertion loss dropped from around 22 dB to approximately 4 dB, signal integrity improved by about 63 times, and system optical power efficiency increased by up to 5 times.

However, CPO is not simply a matter of “moving” existing optical modules to a new location. The package volume is dramatically reduced, the power budget is cut to the bone, thermal conditions become harsher, and the electrical environment becomes extremely demanding. Every component inside the optical engine is pushed toward its physical limits.

It is under these new constraints that thin-film lithium niobate arrived at exactly the right time. It has evolved from a “performance benchmark” into an “engineering necessity.”

In other words, thin-film lithium niobate has become popular not merely because it has become thinner, but because the AI computing infrastructure has finally reached the level where it needs TFLN as a structural load-bearing technology.

This is why we are seeing NVIDIA invest USD 4 billion in companies such as Coherent and Lumentum, two companies that together account for around 80% of the global high-end thin-film lithium niobate modulator market.