Lithium niobate chips: fast light switches on a chip

Lithium niobate is a crystal whose optical properties change when a voltage is applied to it. In thin-film form, a very thin layer of the crystal is bonded onto an insulating base wafer and carved into channels narrow enough to guide light. An electrical signal sent along the channel can then switch the light or write data onto it. This is what engineers call a modulator, and it is one of the basic parts of an optical link.

The reason this matters is that the links carrying data between data centres and across telecom networks need modulators that are fast, small and easy to drive with ordinary electronics. Thin-film lithium niobate is one of several materials being tried for that job, alongside silicon and indium phosphide. This article separates three things: what has been shown in a test, what companies and the press say, and what is still speculation.

What has been demonstrated

An early result is a 2018 set of modulators from a Harvard-led team, published in Nature. The paper's abstract reports monolithically integrated modulators with a drive voltage compatible with CMOS electronics, data rates of up to 210 Gbit/s and on-chip optical loss below 0.5 dB (Wang et al., Nature, 2018). A Laser Focus World report on the paper puts conventional lithium niobate modulators at 3 to 5 V, against about 1 V for typical CMOS circuits, which is why a low drive voltage matters (Laser Focus World). We have read the abstract and the trade report, not the full paper. Neither gives a numeric bandwidth, half-wave voltage, device length or test setup, so those conditions are not stated here.

A separate study reports a frequency comb made in a thin-film lithium niobate microring. A frequency comb is a light source that emits many evenly spaced colours at once, like marks on a ruler. The study was published in Nature in 2019 (Zhang et al., Nature, 2019). Its arXiv preprint abstract describes a comb of over 900 lines spaced about 10 GHz apart, spanning more than the whole telecom L-band (a band of wavelengths used in long-distance fibre). The abstract says the spacing can be finely controlled from 10 Hz to 100 MHz. The authors describe a comb spanning a full octave, which would need further dispersion engineering, as a future possibility. That has not been shown in this work.

A review of the platform, published in Advances in Optics and Photonics in 2021, states that some thin-film devices, such as modulators and wavelength converters, have already outperformed their bulk-crystal versions (Zhu et al., 2021; arXiv version). The abstract gives no numbers for that comparison, so it is a qualitative statement only.

Stability is a further test. A study on thin-film lithium tantalate, a close relative of the material, published in Optics Express in 2024, reports that over 46 hours at 12.1 dBm input power, tantalate modulators held their output power fluctuation at quadrature bias below 1 dB, while equivalent lithium niobate modulators fluctuated by 5 dB (Powell et al., 2024; arXiv version). The abstract does not give a cause for the difference, and this is one group's comparison of one set of devices.

What companies say

These are company claims, not independent measurements.

HyperLight, a commercial supplier that Optics.org describes as a Harvard University spin-out (Optics.org), says on its website that its packaged modulators reach up to 145 GHz bandwidth, and that its transmitter chips deliver up to 448 Gbps per lane. The page gives no test conditions. It describes the drive voltage only in words, such as "ultra-low" and "direct CMOS drive", with no figure. A news headline on the site announces a partnership with UMC and Wavetek for high-volume foundry production. The page does not print a date for that partnership.

CCRAFT, a foundry spun out of the Swiss research centre CSEM in April 2025, told Optics.org in June 2025 that it planned two multi-project wafer runs that year, with a September deadline for the first. Such runs let several designers share one wafer. The company says it aims for millions of chips a year, which the report says will likely not be reached before 2028. The wafers come from a European supplier, Soitec, and are initially 6-inch. We have not checked whether those 2025 runs went ahead as planned.

Lightium, a Zurich firm, is reported by Optics.org as putting volume wafer supply at the top of its list of challenges. The ELENA consortium, also cited in that report, described a limited supply chain reliant on a single commercial supplier outside the EU.

Limits and open problems

• Wafer supply. Volume production depends on wafers, and the sources name wafer supply as a leading challenge.
• Independent testing. Most product figures come without test conditions, and we found no independent measurement of a commercial part.
• Bias drift. The tantalate result suggests lithium niobate devices can drift under DC bias, but one comparison does not show that the problem is solved in a product.
• Comb reach. The comb result is a single chip-scale device spanning the L-band. A full-octave comb has not been demonstrated in the sources reviewed.
• Drive electronics. The 2018 devices are described as CMOS-compatible, but neither the abstract nor the trade report gives a numeric half-wave voltage, so a reader cannot tell how much the electronics would have to do.
• Cost and yield. We did not find published figures for wafer cost, fabrication yield or chip cost in the sources we checked.
• Packaging and coupling. The sources we checked give no figures for the loss when light is coupled between fibre and chip, or for packaging cost. This is a practical question that any product comparison would need to answer.
• Lab to product. The 2018 and comb results come from research devices. The commercial offers are chiplets and packaged parts whose figures remain company claims.

Speculative uses

Several uses are plausible but have not been shown in the sources reviewed. Low-voltage optical links inside computing clusters are one. Compact frequency combs for ranging and spectroscopy are another. Quantum transducers, which would convert signals between microwave and optical frequencies, are another possible use of the low-loss resonators. None of these has been demonstrated on this platform in the sources we checked.

What to watch next

• Independent measurements of a commercial chip, with stated test conditions.
• Whether CCRAFT's wafer runs produce working devices, and whether its volume timetable holds.
• Published figures for yield, cost and fibre-to-chip coupling loss.
• A full-octave electro-optic comb on the platform.
• Follow-up work on the cause of the stability difference between lithium tantalate and lithium niobate modulators.