
Researchers at the California Institute of Technology have developed a way to guide light across silicon wafers with extremely low signal loss, approaching the performance of optical fiber. The advance could lead to more efficient photonic integrated circuits for precision instruments, quantum technologies, and data center communications, tells Science Daily.
Optical fiber can carry information over long distances with minimal loss because its highly pure glass and smooth surfaces limit light absorption and scattering. Caltech researchers have spent years trying to reproduce these properties on silicon wafers while retaining compatibility with chip manufacturing.
Their new approach creates nanoscale waveguides from germano-silicate, a glass also used in optical fiber. The material can be incorporated into a lithography-based fabrication process for standard 8-inch and 12-inch silicon wafers. Instead of forming straight paths, the waveguides are arranged as spirals, allowing light to travel long optical distances within a compact chip.
Performance improvements are particularly significant at visible wavelengths. The researchers use a thermal reflow process to smooth waveguide surfaces to nearly atomic levels, reducing the scattering that limits conventional visible-light photonic circuits. At visible wavelengths, the platform achieves 20 times lower loss than the previous silicon nitride record. Lasers built using the technology also demonstrate more than a 100-fold improvement in the time their light remains coherent.
Such low losses are important for ring resonators, where light circulates repeatedly through a small structure. Lower losses allow light to travel greater effective distances, improving device performance.
The technology could support chip-scale optical clocks, gyroscopes, atomic sensors, ion-trap systems, lasers, quantum computing, and energy-efficient AI data center communications. Researchers have already demonstrated ring resonators, several laser designs, and nonlinear resonators using the platform, while noting that further performance improvements remain possible.
