Peer-reviewed measurements confirm exceptionally low-loss germano-silicate waveguides fabricated on silicon wafers, but the August story recycles research published in January and “silicon optical pathways” inaccurately implies a silicon-core waveguide.
Caltech shrinks microchip optical loss to 0.07 dB/m
Adapting fiber glass to silicon wafers curbs signal fade across visible and infrared light.
In a nutshell
Caltech researchers have adapted optical-fiber glass onto silicon chips, using furnace smoothing to cut light scatter down to a record 0.07 dB/m. The platform improves short-wavelength performance twentyfold, though matching long-haul fiber performance requires cutting remaining signal loss several-hundredfold while bypassing damaging 1360-degree heat cycles.
Highlights
- Caltech's annealed waveguides reached an optical loss of 0.08 dB/m in January and 0.07 dB/m in June.
- Furnace reflow treatment yielded a twentyfold, 13 dB reduction in light loss at 458 nanometers.
- True fiber parity demands dropping signal loss from 80 dB/km to the material limit of 0.2 dB/km.
- Flame hydrolysis doubled germanium content to 50 mol%, shrinking the calculated bend radius to 130 micrometers.
- Processing temperatures reach up to 1360 degrees Celsius, creating compatibility hurdles for active on-chip lasers.
Measured Annealed Waveguide Loss Across Light Spectra
| Wavelength (nm) | Propagation Loss (dB/m) |
|---|---|
| 458 | 0.49 |
| 532 | 0.32 |
| 685 | 0.32 |
| 780 | 0.19 |
| 965 | 0.14 |
| 1064 | 0.08 |
| 1550 | 0.09 |
- Measured propagation loss across seven sampled wavelengths reported in the January Nature paper following high-temperature furnace annealing.
- Shows how signal loss decreases toward longer infrared wavelengths, serving as the benchmark for future low-temperature integration.
From the Editor’s Diary
Demonstrating clean optical physics on a wafer is only half the battle; real commercial adoption depends entirely on whether those glass pathways can survive without extreme heat cycles that destroy the surrounding electronics.
Who's involved
Hao-Jing Chen
Caltech applied-physics researcher and co-first author
goal → adapt optical fiber glass into wafer-scale microchips with minimal optical loss
Kellan Colburn
Caltech researcher and co-first author
goal → push silicon chip waveguides toward optical fiber clarity while testing integrated lasers and resonators
Kerry Vahala
Caltech professor and senior corresponding author
goal → build a scalable microchip photonics platform matching the signal coherence of long-haul glass fiber
Henry Blauvelt
Caltech visiting associate and Emcore chief technology officer
goal → integrate low-loss glass circuits with semiconductor lasers and optical fibers for field applications
University of Southampton Optoelectronics Research Centre
British photonics research group
goal → provide flame-hydrolysis glass expertise to translate fiber manufacturing methods onto silicon wafers
ScienceDaily
research-news website
goal → redistribute university press releases to general audiences
In short
TL;DR: Caltech researchers adapted optical-fiber glass onto silicon wafers, curbing light loss to record lows across violet to infrared spans.
Q: What did Caltech achieve, and what stands between this chip glass and true fiber performance?
- Researchers guided light through germano-silicate glass on silicon wafers to hit a record low loss of 0.08 dB/m in January and 0.07 dB/m in June.
How it unfolded
Caltech unveils fiber-like chip platform at CLEO
Caltech unveiled its germano-silicate waveguide platform at the CLEO 2024 conference in Charlotte, North Carolina, demonstrating that optical-fiber glass could run directly on silicon wafers. By depositing germanium-doped silica over a silica buffer, the team recorded optical quality factors above 180 million from 532 to 1550 nanometers. The platform also sustained optical frequency combs, acoustic laser scattering, and laser locking, demonstrating that the technology was functioning long before its journal publication.
Caltech develops low-temperature glass fabrication
Furnace heating at 1000 degrees Celsius smooths glass edges but risks damaging sensitive electronic components on the same chip. Caltech submitted its comprehensive manuscript to Nature on April 15, 2025, while continuing to investigate low-temperature alternatives. At CLEO 2025 in Long Beach, Hao-Jing Chen demonstrated an unheated process running below 300 degrees Celsius that reached an optical loss of 0.15 dB/m at 1550 nanometers, beating prior unheated records by more than tenfold. Nature accepted the primary research paper on November 10, 2025.
Nature publishes benchmark records across visible light
Nature published Caltech's paper online on January 7, 2026, revealing an optical propagation loss of 0.08 dB/m at 1064 nanometers and 0.49 dB/m at 458 nanometers. The 458-nanometer result marked a roughly twentyfold, 13-decibel improvement over previous integrated platforms, achieved by melting etched sidewalls in a 1000-degree furnace. While trade outlets such as PIC Magazine and HPCwire highlighted applications in data centers and quantum networks, the paper cited photonic artificial intelligence only as a prospective use, without demonstrating live computing workloads.
Flame hydrolysis cuts bend radius and deepens loss record
Because germano-silicate bends leak light when turned sharply, the team collaborated with the University of Southampton to double germanium content to roughly 50 mol% using flame hydrolysis deposition. Published in Light: Science & Applications on June 4, 2026, the higher refractive index contrast shrank the calculated minimum bend radius from 840 to 130 micrometers and cut optical loss to 0.07 dB/m at 1064 nanometers. However, the flame method requires a consolidation step at 1360 degrees Celsius, leaving high thermal processing as an ongoing integration challenge.
Republished findings recirculate as a new breakthrough
On August 17, 2026, ScienceDaily republished Caltech's February release under the headline "Caltech breakthrough brings fiber-optic performance to silicon chips," renewing interest across social feeds. The syndicated release pointed back to the January Nature paper rather than a new discovery, omitting the June flame-hydrolysis advance that had already improved loss and bend radius. The coverage also characterized data-center and artificial-intelligence uses as immediate realities, even though the underlying scientific papers demonstrated passive light transmission rather than active computing engines.
Where things stand
Caltech's low-loss numbers are backed by two peer-reviewed papers spanning plasma vapor deposition in January and flame hydrolysis in June. The waveguides achieved light loss down to 0.07 dB/m, but they guide light through germanium-doped silica rather than silicon, and they have not demonstrated live photonic artificial-intelligence computing.
The central hurdle remains closing the distance between 80 dB/km on chips and 0.2 dB/km in standalone optical fiber. Engineers must eliminate remaining surface absorption and integrate active lasers without exposing delicate chip components to the platform's 1000 to 1360 degree furnace steps.