As AI data centers rapidly scale bandwidth requirements, optical interconnects are moving from 400G to 800G, 1.6T, and even 3.2T architectures. At these speeds, the limiting factor of optical transceiver performance is no longer laser sources or packaging technologies—but the optical modulator, which is responsible for encoding electrical data onto optical signals.
While indium phosphide (InP) and silicon photonics (SiPh) have long dominated modulator technologies, both are approaching performance and scalability constraints at the next generation of ultra-high-speed systems. In this context, a new material platform is emerging as a strong candidate: Thin-Film Lithium Niobate (TFLN), also known as Lithium Niobate on Insulator (LNOI).
Thin-film lithium niobate (TFLN) is a photonic integration platform based on single-crystal lithium niobate (LiNbO₃), a well-established electro-optic material widely used in modulation, nonlinear optics, and acoustic devices.
Lithium niobate has been used in optical communications for decades, but traditional devices are typically centimeter-scale bulk components. The innovation behind TFLN lies in transforming this material into a thin crystalline layer (nanometers to microns thick) integrated onto a silicon dioxide substrate.
This structure is commonly referred to as Lithium Niobate on Insulator (LNOI).
By reducing the material thickness and integrating it into a waveguide platform, TFLN enables:
Importantly, “thin-film” does not mean flexible material—it still consists of rigid single-crystal lithium niobate, only engineered into a much thinner optical layer.
In optical communication systems, digital information is transmitted by modulating a continuous-wave (CW) laser source. The optical modulator determines how efficiently and how fast electrical signals can be converted into optical signals.
At data rates beyond 400G and toward 1.6T, modulation requirements become extremely demanding:
Existing technologies face structural limitations:
InP-based modulators are highly mature and can integrate lasers, modulators, and detectors on the same chip. However, their modulation bandwidth is gradually reaching physical limits for beyond-400G single-channel systems.
Silicon photonics offers excellent scalability and CMOS compatibility. However, silicon lacks strong native electro-optic properties. Modulation relies on carrier injection or depletion effects, which introduce trade-offs between speed, power consumption, linearity, and optical loss.
TFLN is fundamentally different because it operates based on the Pockels effect (linear electro-optic effect):
An applied electric field directly changes the refractive index of the crystal.
This enables:
As a result, TFLN is increasingly viewed as a key enabling technology for next-generation ultra-high-speed optical transceivers.
Unlike silicon photonics, TFLN is not grown directly on silicon substrates. Instead, it relies on a layer-transfer engineering process combining crystal growth and wafer bonding technologies.
High-purity lithium niobate crystals are grown using the Czochralski method. The crystals are then sliced and polished into wafers.
Hydrogen or helium ions are implanted into a controlled depth inside the wafer, forming a weakened layer beneath the surface.
The lithium niobate wafer is bonded to a silicon dioxide (SiO₂) or silicon handle wafer using direct wafer bonding techniques.
Thermal or mechanical treatment is applied, causing the wafer to split along the implanted layer. A thin crystalline film is transferred onto the substrate.
Chemical mechanical polishing (CMP) is used to smooth the surface, followed by standard photolithography, etching, metallization, and packaging processes.
Despite its promising process, several technical barriers remain:
It is important to clarify that TFLN is not a light source material. It does not generate lasers.
Instead, it functions as a high-speed electro-optic modulation layer.
In a typical optical system:
Most TFLN modulators are based on the Mach-Zehnder Interferometer (MZI) structure.
This enables high-speed encoding of digital data onto optical signals.
The future of optical interconnects is not defined by a single material platform, but by a heterogeneous multi-material ecosystem.
Together, these technologies form a hybrid photonic architecture for next-generation optical transceivers.
Despite strong performance advantages, TFLN is still in an early industrial scaling phase.
Maintaining uniform thin-film thickness, low defect density, and stable bonding interfaces remains challenging.
Lithium niobate is significantly harder to etch than silicon, leading to scattering losses caused by sidewall roughness.
Impedance matching, microwave loss control, and electro-optic velocity matching are complex RF-photonic co-design problems.
Bonding yield, thermal stress management, and process standardization are still evolving.
Differences in refractive index require advanced coupling structures such as taper waveguides, edge coupling, and evanescent coupling.
As AI infrastructure continues to push the boundaries of bandwidth and energy efficiency, optical transceiver development is shifting from single-material optimization to system-level material collaboration.
Thin-film lithium niobate does not aim to replace InP or silicon photonics. Instead, its value lies in addressing a critical bottleneck in the optical chain: ultra-high-speed, low-loss electro-optic modulation
In future 1.6T, 3.2T, and co-packaged optics (CPO) architectures, TFLN is expected to become a key enabling component within hybrid photonic systems—working alongside InP and silicon photonics to support the next generation of AI-driven optical networks.
As AI data centers rapidly scale bandwidth requirements, optical interconnects are moving from 400G to 800G, 1.6T, and even 3.2T architectures. At these speeds, the limiting factor of optical transceiver performance is no longer laser sources or packaging technologies—but the optical modulator, which is responsible for encoding electrical data onto optical signals.
While indium phosphide (InP) and silicon photonics (SiPh) have long dominated modulator technologies, both are approaching performance and scalability constraints at the next generation of ultra-high-speed systems. In this context, a new material platform is emerging as a strong candidate: Thin-Film Lithium Niobate (TFLN), also known as Lithium Niobate on Insulator (LNOI).
Thin-film lithium niobate (TFLN) is a photonic integration platform based on single-crystal lithium niobate (LiNbO₃), a well-established electro-optic material widely used in modulation, nonlinear optics, and acoustic devices.
Lithium niobate has been used in optical communications for decades, but traditional devices are typically centimeter-scale bulk components. The innovation behind TFLN lies in transforming this material into a thin crystalline layer (nanometers to microns thick) integrated onto a silicon dioxide substrate.
This structure is commonly referred to as Lithium Niobate on Insulator (LNOI).
By reducing the material thickness and integrating it into a waveguide platform, TFLN enables:
Importantly, “thin-film” does not mean flexible material—it still consists of rigid single-crystal lithium niobate, only engineered into a much thinner optical layer.
In optical communication systems, digital information is transmitted by modulating a continuous-wave (CW) laser source. The optical modulator determines how efficiently and how fast electrical signals can be converted into optical signals.
At data rates beyond 400G and toward 1.6T, modulation requirements become extremely demanding:
Existing technologies face structural limitations:
InP-based modulators are highly mature and can integrate lasers, modulators, and detectors on the same chip. However, their modulation bandwidth is gradually reaching physical limits for beyond-400G single-channel systems.
Silicon photonics offers excellent scalability and CMOS compatibility. However, silicon lacks strong native electro-optic properties. Modulation relies on carrier injection or depletion effects, which introduce trade-offs between speed, power consumption, linearity, and optical loss.
TFLN is fundamentally different because it operates based on the Pockels effect (linear electro-optic effect):
An applied electric field directly changes the refractive index of the crystal.
This enables:
As a result, TFLN is increasingly viewed as a key enabling technology for next-generation ultra-high-speed optical transceivers.
Unlike silicon photonics, TFLN is not grown directly on silicon substrates. Instead, it relies on a layer-transfer engineering process combining crystal growth and wafer bonding technologies.
High-purity lithium niobate crystals are grown using the Czochralski method. The crystals are then sliced and polished into wafers.
Hydrogen or helium ions are implanted into a controlled depth inside the wafer, forming a weakened layer beneath the surface.
The lithium niobate wafer is bonded to a silicon dioxide (SiO₂) or silicon handle wafer using direct wafer bonding techniques.
Thermal or mechanical treatment is applied, causing the wafer to split along the implanted layer. A thin crystalline film is transferred onto the substrate.
Chemical mechanical polishing (CMP) is used to smooth the surface, followed by standard photolithography, etching, metallization, and packaging processes.
Despite its promising process, several technical barriers remain:
It is important to clarify that TFLN is not a light source material. It does not generate lasers.
Instead, it functions as a high-speed electro-optic modulation layer.
In a typical optical system:
Most TFLN modulators are based on the Mach-Zehnder Interferometer (MZI) structure.
This enables high-speed encoding of digital data onto optical signals.
The future of optical interconnects is not defined by a single material platform, but by a heterogeneous multi-material ecosystem.
Together, these technologies form a hybrid photonic architecture for next-generation optical transceivers.
Despite strong performance advantages, TFLN is still in an early industrial scaling phase.
Maintaining uniform thin-film thickness, low defect density, and stable bonding interfaces remains challenging.
Lithium niobate is significantly harder to etch than silicon, leading to scattering losses caused by sidewall roughness.
Impedance matching, microwave loss control, and electro-optic velocity matching are complex RF-photonic co-design problems.
Bonding yield, thermal stress management, and process standardization are still evolving.
Differences in refractive index require advanced coupling structures such as taper waveguides, edge coupling, and evanescent coupling.
As AI infrastructure continues to push the boundaries of bandwidth and energy efficiency, optical transceiver development is shifting from single-material optimization to system-level material collaboration.
Thin-film lithium niobate does not aim to replace InP or silicon photonics. Instead, its value lies in addressing a critical bottleneck in the optical chain: ultra-high-speed, low-loss electro-optic modulation
In future 1.6T, 3.2T, and co-packaged optics (CPO) architectures, TFLN is expected to become a key enabling component within hybrid photonic systems—working alongside InP and silicon photonics to support the next generation of AI-driven optical networks.
