Chip Smaller Than Fingernail Could Be Key to 6G’s Blinding Speed

The ring is barely visible to the naked eye. A loop of silicon nitride perhaps a fraction of a millimetre across, sitting on a chip the size of a shirt button, in a laboratory at Tokushima University in southwestern Japan. Pump enough light into it at the right frequency and something strange happens: the ring starts producing dozens of new light frequencies simultaneously, spaced out in a precise, comb-like pattern across the infrared spectrum.

This is a soliton microcomb, and until recently it lasted only a few minutes before thermal wobble killed the signal. Now researchers have kept one running for more than 27 hours continuously, and used it to push data through the air at 112 gigabits per second on a carrier wave far above the frequencies that electronics alone can generate. It is, in a fairly literal sense, a glimpse of what the next generation of wireless communication might look like.

The sixth generation of mobile networks, which engineers and regulators are already planning for, will need to carry data at rates that would make a 5G connection look modest. The ambition is speeds of a terabit per second or more for fixed backhaul links, the connections that ferry traffic between base stations and the internet backbone. Achieving that requires moving to higher carrier frequencies, up into the sub-terahertz range above 300 gigahertz, where there is still plenty of unallocated bandwidth and contiguous spectrum. The trouble is that conventional electronics struggle to generate clean, stable signals up there. “This result represents a major step toward practical 6G wireless systems and ultra-high-speed mobile backhaul,” said Prof. Takeshi Yasui, who led the Tokushima team.

The 350 GHz Wall

Electronics-based transmitters rely on frequency multiplication, taking a microwave signal and multiplying it up to terahertz frequencies. This works tolerably well up to about 300 GHz, which is why the spectrum around there is already allocated, already congested, and already attracting commercial interest. Above 350 GHz the approach starts to break down. Phase noise (roughly, unwanted jitter in the signal’s frequency) increases, output power drops, and the signal quality deteriorates in ways that make reliable high-order modulation difficult. Higher-order modulation formats, like the 16QAM used in this experiment, pack four bits into every transmitted symbol; they are exquisitely sensitive to phase noise. Push a 16QAM signal through a noisy carrier and the tightly clustered constellation of data points smears into indistinguishable blur.

Photonic approaches sidestep the multiplication problem by generating the terahertz signal optically instead. Take two laser beams at slightly different infrared frequencies, combine them in a specialised photodetector called a uni-traveling-carrier photodiode, and the detector produces a terahertz wave at exactly the difference frequency between the two beams. The approach is in principle far more scalable than electronic multiplication, and can reach carrier frequencies that electronics cannot. The difficulty has been keeping those two laser beams phase-coherent with each other. If the lasers drift independently, the terahertz carrier inherits all that noise. Which is where the microcomb comes in.

A soliton microcomb generates what amounts to a ruler in frequency space: dozens of evenly spaced optical teeth, all locked together by the same nonlinear physics, all sharing the same phase reference. Lock two separate lasers to adjacent teeth and you get a pair of sources separated by exactly one comb-tooth spacing, with essentially no relative phase drift between them. At 560 GHz, that spacing produces a terahertz carrier of remarkable cleanliness.

The Packaging Problem

The difficulty, until now, was that soliton microcombs required fussy free-space optical coupling to work, aligning a tiny lensed fibre precisely in front of the chip’s waveguide facet. Vibrations from nearby equipment, small temperature changes across the optical bench, even air currents from a ventilation system, could shift the alignment enough to kill the soliton state within minutes. In Tokizane et al.’s experiments, that conventional approach produced soliton lifetimes of around four minutes on average. Usable for a proof-of-concept in a quiet lab at three in the morning, perhaps, but hardly a platform for commercial deployment.

The Tokushima team’s key innovation was to eliminate the free-space coupling entirely. They spliced a standard optical fibre to a high-numerical-aperture fibre with a much smaller mode-field diameter, then permanently bonded it directly onto the chip’s input waveguide using a UV-curable adhesive. The resulting assembly is smaller by roughly two orders of magnitude than the conventional setup (5 millimetres compared with 450), handles pump powers up to at least 3 watts without degradation, and drifts by only a tiny fraction of a percentage point in coupling efficiency over ten hours, against more than a full percentage point of drift for the free-space equivalent. The soliton lifetimes improved accordingly: 27.7 continuous hours in the endurance test, with typical passive runs of between one and three hours.

With a stable microcomb in hand, the transmission experiment itself could proceed. Two distributed-feedback lasers were each locked onto adjacent comb lines via optical injection locking, transferring the comb’s low phase noise to high-power laser sources capable of driving the modulation and photomixing hardware. One laser carried the data, encoded as QPSK or 16QAM symbols; the other served as an unmodulated reference. Their combined light fed the photodiode, which produced the 560 GHz carrier. At the receiver, a sub-harmonic mixer and a real-time oscilloscope completed the link. At 28 gigabaud with 16QAM, the system delivered 112 gigabits per second, passing the hard-decision forward error correction threshold that certifies a link as commercially viable. That figure is more than fifty times higher than any previous transmission demonstrated at 560 GHz.

Why 560 GHz Is Both Right and Wrong

There is an awkward irony in the choice of frequency. The 560 GHz band happens to sit near an atmospheric water vapour absorption line, which means the air attenuates the signal at roughly 7 decibels per metre. The experiment covered ten millimetres of free-space propagation, so that was not a problem in the lab; in the real world, you would not get far. The researchers are candid about this. Their simulations suggest that shifting to 500 GHz, where atmospheric absorption falls to a tiny fraction of a decibel per metre, would extend the reach to perhaps 55 millimetres with the same hardware, and potentially to tens of metres with improved antennas and higher-power photodiodes already in development. The 560 GHz demonstration is not the destination; it is a proof that the physics and the packaging work.

What matters practically is that the soliton microcomb approach is inherently scalable to other frequencies. The comb-tooth spacing can be engineered by choosing the microresonator’s dimensions, and silicon nitride microresonators are fabricated using processes compatible with standard semiconductor manufacturing. A commercially deployable 6G backhaul radio built around photonic generation is not a fantasy, and the gap between proof-of-concept and field unit now looks rather smaller than it did before this paper appeared.

The next steps Yasui’s group has identified are roughly what you would expect: reducing the residual phase noise further, pushing to higher-order modulation formats (which would squeeze more bits per hertz from the available bandwidth), and improving the terahertz output power to extend transmission distances. None of those are easy, but none of them looks intractable given the pace at which photonic integration technology has been moving. The silicon nitride ring sitting in its bonded fibre assembly in Tokushima may not look like the future of wireless communications. But it is surprisingly difficult, once you know what it does, to think of it as anything else.

Frequently Asked Questions

What is a soliton microcomb and why does it matter for wireless communications?

A soliton microcomb is a tiny optical device, typically a microscale ring of material like silicon nitride, that uses nonlinear light-matter interactions to generate a precise series of evenly spaced laser frequencies. Because all those frequencies share the same phase reference, two of them can be combined to produce an extremely stable terahertz signal. That stability is critical for the high-order modulation formats needed to push wireless data rates above 100 gigabits per second.

Why do 6G networks need terahertz frequencies rather than the millimetre-wave bands used for 5G?

The sub-28 GHz and millimetre-wave bands used in 5G are already congested, and the available bandwidth in those bands is limited. The terahertz region above 300 GHz contains large stretches of unallocated spectrum, which could support data rates an order of magnitude higher than current 5G links. The challenge is building transmitters that can generate stable signals in that range at useful power levels.

What was the main engineering breakthrough in this research?

The key advance was replacing fragile, alignment-sensitive free-space optical coupling with a compact, bonded fibre-to-chip interface. Earlier microcombs would maintain their soliton state for roughly four minutes before thermal or mechanical drift killed the signal. The new packaging kept the soliton running for more than 27 hours, which is what made continuous wireless transmission feasible.

If the 560 GHz band has severe atmospheric absorption, is this system actually practical?

The 560 GHz demonstration is primarily a proof of concept for the photonic approach and the chip packaging. The researchers’ own simulations show that operating at 500 GHz instead, where atmospheric absorption is around 170 times lower, could extend the transmission range from millimetres to potentially tens of metres with improved antennas and more powerful photodiodes. The frequency choice in the experiment was driven by the microresonator’s comb spacing, not by what would eventually be deployed.

How far away is a commercial 6G product based on this technology?

The paper is an early-stage demonstration, and the path from a 10-millimetre laboratory link to a deployable backhaul radio involves significant engineering, including reducing device size, increasing output power, and validating performance under real-world conditions. That said, silicon nitride microresonators are already manufactured using semiconductor-compatible processes, which removes one major barrier to scale-up. Most researchers in the field would likely consider a decade a reasonable timescale for early commercial deployment, though 6G standardisation timelines will play a large role.

https://doi.org/10.1038/s44172-026-00659-8