Lithium Niobate EO Modulator
A crystal road for light, with a 110-billion-times-per-second traffic light governed by voltage.
A lithium niobate electro-optic modulator (LN-EOM) encodes electrical signals onto optical light by exploiting the Pockels effect: an applied electric field changes the crystal's extraordinary-axis refractive index by Δn = −½ n_e³ r₃₃ E, with r₃₃ ≈ 30 pm/V. In a Mach-Zehnder interferometer, a voltage on one arm phase-shifts the light there; recombination yields constructive (bright) or destructive (dark) interference. Modern thin-film lithium niobate (TFLN, ~300 nm LiNbO₃ on insulator) confines optical mode area to <1 µm², achieves V_π·L ≈ 2.2 V·cm and >110 GHz bandwidth, and supports >512 Gb/s single-lane PAM-4 transmission. The same platform enables direct microwave-to-optical photon transduction for connecting superconducting qubits to fiber-optic quantum networks.
In bulk LiNbO₃, the optical group index n_g ≈ 2.21 and the microwave index n_mw ≈ 4.2 differ severely, limiting traveling-wave bandwidth to ~30 GHz. Thin-film LiNbO₃ on insulator (LNOI) uses a 300 nm crystal layer on 2 µm SiO₂ on Si: tight optical confinement pulls n_g to ~2.1–2.2; capacitively-loaded coplanar electrodes with SiO₂ undercladding pull n_mw down to ~2.1–2.3 — enabling near-perfect velocity matching across the 1.3–110 GHz band on a 15 mm device. The half-wave voltage V_π drops from ~5 V (bulk) to 1.5–3 V (TFLN) for the same length, lowering driver power. Because LiNbO₃ is also piezoelectric, the same chip can transduce microwave photons (~5–10 GHz from a superconducting qubit) into optical sidebands at 1550 nm via the Pockels effect — converting quantum information from the microwave to the optical domain for long-distance quantum networking.
In plain terms
Imagine writing data onto a beam of light by squeezing a special crystal with a voltage. When you push, the light slows down a tiny bit; when you let go, it speeds up. By splitting a beam into two crystal lanes and squeezing one of them, the two beams come back together either in step (bright = 1) or out of step (dark = 0). Lithium niobate lets you do this 110 billion times every second on a fingernail-sized chip — encoding more information per fiber than the entire global library of texts every few minutes.
Also called: LiNbO₃ EOM, TFLN modulator, Mach-Zehnder EO modulator.
Why it matters
- TFLN modulators with >110 GHz bandwidth and <3 V V_π are the backbone of 800 Gb/s and 1.6 Tb/s coherent optical transceivers in AI data centers and submarine fiber.
- Microwave-to-optical quantum transduction connects superconducting quantum computers via standard telecom fiber.
- TFLN integrated frequency combs produce >1,000 spectral lines for massively parallel WDM and arbitrary optical waveform synthesis.
- Compatible with silicon photonics back-end-of-line, enabling monolithic laser/modulator/detector platforms.
Where it lives
- HyperLight / NTT / Coherent TFLN coherent transceiver chips
- POET Technologies / Coherent 400G ZR+ optical engines
- Caltech / Loncar lab Harvard — TFLN integrated photonics
- NIST / Princeton TFLN microwave-to-optical transduction experiments
Key numbers
- EO 3 dB bandwidth (TFLN): >110 GHz (Volume-manufactured 2025)
- Half-wave voltage V_π: 1.5–3 V (10 mm device length)
- V_π·L (TFLN): ~2.2 V·cm (vs. ~10 V·cm bulk)
- Single-lane data rate: 512 Gb/s (Dual-pol PAM-4, 2025)
- r₃₃ Pockels coefficient: ~30 pm/V
- Waveguide propagation loss: <0.4 dB/cm
Surprising facts
- A TFLN modulator the size of a fingernail encodes 512 Gb/s on a single optical lane — at that rate, the modulator could transmit a million years of continuous typing in about 3 milliseconds.
- TFLN devices achieve 110 GHz bandwidth with electrodes spaced just 3–10 µm from a 300-nm crystal — closer than the rise of one DNA helix turn — yet handle enough microwave power to modulate a trillion optical bits per second.
- Because LiNbO₃ is simultaneously electro-optic, piezoelectric, and acousto-optic, a single TFLN chip can transduce between microwave photons, optical photons, and acoustic phonons — a candidate quantum-network node.
Open questions
- Acoustic resonances in the TFLN slab: LiNbO₃'s piezoelectricity couples the RF drive into mechanical resonances in the thin-film slab, producing sharp dips in the EO frequency response at 1–10 GHz acoustic resonances. Mitigation by acoustic damping and electrode loading is partial; no universal fix exists across all designs.
- DC bias drift: Bulk LiNbO₃ MZIs exhibit DC bias drift over hours due to charge migration in the crystal; TFLN devices, with thinner crystal, show reduced drift but long-term stability at production scale is still being characterized.
A short history
Bulk LiNbO₃ EO modulators have been telecom workhorses since the 1980s but plateaued at ~30 GHz bandwidth. The TFLN platform took off when Loncar's Harvard group (2018, Nature) demonstrated low-loss etched waveguides in commercially available LNOI wafers from NanoLN. Within a few years, multiple groups (Loncar, Lipson, Kahn, Mookherjea, Lin, Cheng) demonstrated >70, >100, then >110 GHz EO bandwidths and sub-1 V V_π devices. By 2025, four-inch volume TFLN wafers were producing 110 GHz devices with ~50 % yield (Light: Advanced Manufacturing). Coherent telecom companies (Coherent Corp., POET, NEL) are deploying TFLN in 400G/800G modules.
Sources
- Wang et al., Nature 562, 101 (2018) — TFLN EOM with 100 GHz bandwidth (https://www.nature.com/articles/s41586-018-0551-y)
- Light: Advanced Manufacturing 6 (2025) — Volume manufacturing 110 GHz TFLN (https://www.light-am.com/article/doi/10.37188/lam.2025.047)
- Nature Communications (2026) — TFLN 800 nm bandwidth modulator (https://www.nature.com/articles/s41467-025-67902-2)
- Zhu et al., Adv. Opt. Photonics 13, 242 (2021) — Integrated TFLN photonics (https://opg.optica.org/aop/abstract.cfm?uri=aop-13-2-242)