Caltech puts telecom fiber's glass on silicon chips, approaching fiber-grade light loss

Caltech's Vahala lab has put the same germano-silicate glass that gives telecom optical fiber its ultralow loss onto standard 8- and 12-inch silicon wafers, reporting waveguide loss at visible wavelengths that approaches what a drawn fiber delivers. The advance, published in Nature, translates a 50-year-old glass recipe used in spool-based fiber fabrication onto chip-compatible lithography. The result is a photonic integrated circuit platform, a class of light-based circuits that route information with photons instead of electrons, that approaches drawn fiber's loss in a wavelength range where on-chip photonics has long been limited.

The Caltech group measured resonator Q factors above 180 million from violet through telecom wavelengths, according to the Nature paper and the Caltech press materials aggregated by ScienceDaily. Q factor measures how long a resonator holds light before losing it; higher Q means lower loss. In the telecom band, the new platform delivers a 10-dB improvement in Q over prior art without thermal treatment, the team reports, so the waveguides are fabrication-ready as-deposited, a step that simplifies integration with other photonic components.

"Ultra-low loss" here means so little signal is lost that light can travel across a chip with nearly the efficiency of an optical-fiber cable. Most on-chip photonics platforms hit a wall in the visible band, the part of the light spectrum the human eye sees, distinct from the infrared used in telecom, because the materials that work for infrared absorb and scatter at shorter wavelengths. The visible band is the working range for atomic clocks, gyroscopes, certain quantum systems, and many of the most precise laser-based sensors, which is why the visible-band result is the headline. A waveguide, the photonic analog of a wire, is the nanoscale on-chip pathway that channels light. The Quantum Insider's re-report flags the visible-band unlock as the main story.

The fabrication path is what makes the result manufacturable rather than just interesting. Germano-silicate is the same glass family used in drawn optical fiber, and the Caltech team deposited it on 8- and 12-inch silicon wafers using a process the lab describes as fully CMOS-foundry-compatible, meaning the deposition can be slotted into the same commercial chip-fab lines that produce ordinary silicon electronics. The lead authors are Caltech postdoc Hao-Jing Chen and graduate student Kellan Colburn (MS '25), with the work carried out in Kerry Vahala's lab. Vahala, BS '80, PhD '85, now the Ted and Ginger Jenkins Professor of Information Science and Technology and Applied Physics, has framed the result as translating spool-based fiber fabrication onto silicon wafers while preserving the loss profile that makes fiber special.

The application list that trails these announcements is real but forward-looking, and the researchers are the source. Stated targets include more coherent on-chip lasers, miniature atomic clocks, gyroscopes, atomic sensors, quantum systems, and more energy-efficient optical communications inside AI data centers. None are shipping products; the Nature paper reports a waveguide material, a fabrication process, and loss numbers, not deployed systems. SciTechDaily's writeup carries the same caveat. Any read that this paper has just dropped the energy bill for an AI data center is reading ahead of the result.

Three open questions will decide whether the advance lands in real hardware. First, scaling: a single Nature paper demonstrates a material and a process. Running the same recipe on full 12-inch production lines at volume, with the yield and uniformity commercial photonics requires, is a separate problem. Second, full photonic-circuit integration: the waveguides are the lowest-loss part of a circuit, but every circuit also needs modulators, detectors, and coupling to fibers. Whether germano-silicate plays nicely with those active components, or whether heterogeneous integration is required, is a question the paper does not fully resolve. Third, the competitive landscape: silicon nitride is the existing workhorse for low-loss on-chip photonics and has its own foundry-compatible processes. Why germano-silicate wins, and on which axes, is a fair question for the field to answer next. HPCwire's earlier writeup treats this comparison as still open.

The Nature paper's strongest claim is the visible-band number, Q factors above 180 million in a regime where on-chip photonics has historically struggled. That is the result to watch, because it is what makes on-chip atomic clocks, gyroscopes, and visible-light quantum hardware worth re-engineering around. The rest of the application wish list can wait on the next set of papers.