Skip to main content
ExplainerSilicon PhotonicsExplainerAug 21, 2026, 9:00 PM· 4 min read· in science

Caltech Breakthrough Achieves Fiber-Optic Efficiency on Silicon Chip

Researchers have engineered a new photonic integrated circuit that guides visible light with the ultralow signal loss of optical fiber, paving the way for chip-scale quantum sensors and highly efficient AI data centers.

By Viktoria Sokolova

Photonic Engineers 40%Quantum & Metrology Researchers 40%Computing & AI Analysts 20%
Photonic Engineers
Focus on the manufacturing scalability of printing germano-silicate on standard CMOS wafers.
Quantum & Metrology Researchers
Emphasize the coherence improvements enabling chip-scale atomic clocks and quantum sensors.
Computing & AI Analysts
Highlight the potential to solve bandwidth bottlenecks and energy loss in AI data centers.
20x
Lower optical loss at visible wavelengths
100x
Improvement in laser coherence
180 million
Resonator Q factor achieved
8- and 12-inch
Compatible standard silicon wafer sizes

Silicon photonics has a visible-light problem. For decades, the technology that uses light to shuttle data between server racks has been confined to the infrared spectrum, where silicon is naturally transparent. When engineers try to push shorter, visible wavelengths through these chips, the light scatters and absorbs catastrophically. The rough edges of microscopic pathways act like physical speed bumps, dissipating the optical signal as heat and severely limiting the performance of precision devices.[1][2]

That wall has now crumbled. Researchers at the California Institute of Technology have engineered a new type of photonic integrated circuit that guides visible light with the same ultralow signal loss as a spool of optical fiber. By printing germano-silicate glass directly onto standard silicon wafers, the team has achieved a 20-fold reduction in optical loss compared to previous state-of-the-art materials.[2][3]

Optical fiber achieves its near-perfect transmission because it is made of exceptionally pure glass that is engineered to be extraordinarily smooth. To replicate this environment on a flat chip, the Caltech team utilized the exact same germano-silicate glass, but adapted it for a lithography-based manufacturing process that is fully compatible with standard 8-inch and 12-inch silicon wafers.[2]

The germano-silicate platform dramatically outperforms previous silicon nitride records.

The critical step in achieving this efficiency is a "thermal reflow" process. Because the germano-silicate material has a relatively low melting temperature, the researchers place the fabricated chips into a furnace. This gently melts the surface of the waveguides, smoothing out microscopic manufacturing imperfections down to the level of individual atoms. This atomic-level smoothness is what suppresses the severe scattering loss that previously plagued visible-light circuits.[1][2]

Space is another major constraint on a microchip. Instead of running the optical pathways in straight lines, the engineers arranged the waveguides in tight spirals. This geometry allows light to travel a much longer optical path—analogous to winding a long optical fiber around a spool—while remaining confined within a footprint smaller than a human fingernail.[2]

Instead of running the optical pathways in straight lines, the engineers arranged the waveguides in tight spirals.

The data shows a dramatic leap in performance. At near-infrared wavelengths, the new platform matches the best existing devices made from silicon nitride. But at visible wavelengths, the germano-silicate waveguides outperform the previous silicon nitride record by a factor of 20. The platform achieves resonator Q factors—a measure of how long a device can store optical energy—surpassing 180 million across a spectrum ranging from violet to telecom wavelengths.[1][3]

Thermal reflow smoothing reduces scattering loss by a factor of 20 at visible wavelengths.

This reduction in loss translates directly to device performance. For lasers built using these new ring resonators, the ability of light to circulate longer without scattering results in a more than 100-fold improvement in optical coherence. Highly coherent light is essential for precision measurements, as it maintains a stable frequency and phase over extended periods of time.[2][3]

The implications for quantum technology and metrology are profound. Precision instruments like optical clocks, which keep time based on the high-frequency ticking of atoms, and optical gyroscopes, which measure rotation for navigation, rely heavily on visible light. Previously, these systems required bulky, tabletop-sized optical setups. The Caltech breakthrough provides a pathway to shrink these laboratory-grade instruments onto a single, mass-producible chip.[1][3]

Beyond precision sensors, the technology addresses a looming crisis in artificial intelligence. The data centers powering large language models are increasingly bottlenecked not by raw computational power, but by the energy and heat required to transfer data between nodes using traditional copper wires. Replacing electrical interconnects with highly efficient, low-loss photonic circuits could drastically reduce the power consumption and latency of AI infrastructure.[3][4]

Ultralow-loss photonic chips could drastically reduce the energy required to transfer data in AI server farms.

While the proof-of-concept is robust, the transition from a university laboratory to commercial foundries remains the next hurdle. The process relies on standard CMOS-compatible wafers, which is a significant advantage, but integrating the thermal reflow step into existing high-volume semiconductor manufacturing lines without damaging adjacent electronic components will require further engineering and thermal management.[1][4]

The researchers view this as a foundational shift rather than a final product. By marrying the performance benchmark of optical fibers with the mass-manufacturing scalability of silicon chips, integrated photonics is evolving from a specialized tool for telecommunications into a universal platform for next-generation computing and sensing.[2][4]

What we don’t know

  • How quickly commercial semiconductor foundries will adopt and scale the germano-silicate thermal reflow process for mass production.
  • Whether the technology can be seamlessly integrated with existing electronic components on the exact same chip without thermal interference.
  • The long-term reliability and degradation rates of these ultra-smooth waveguides under continuous high-power laser operation.

Key points

  • Caltech researchers have engineered silicon-based photonic circuits that guide visible light with the ultralow signal loss of optical fiber.
  • The team used germano-silicate glass and a thermal reflow process to smooth waveguide surfaces to near-atomic levels.
  • The platform achieves 20 times lower loss than conventional silicon nitride at visible wavelengths.
  • Lasers built using the technology demonstrate a 100-fold improvement in optical coherence.
  • The breakthrough paves the way for chip-scale atomic clocks, quantum sensors, and highly efficient AI data center communications.

How we got here

  1. Pre-2026

    Silicon photonics is largely confined to infrared wavelengths due to catastrophic scattering losses when attempting to guide visible light.

  2. February 2026

    Caltech researchers publish their foundational method for printing germano-silicate optical circuits on silicon wafers in the journal Nature.

  3. August 2026

    The breakthrough gains wider recognition as its implications for AI data centers and quantum computing become clear, demonstrating 20x lower loss than silicon nitride.

Sources

Source coverage

4 outlets

3 viewpoints surfaced

Photonic Engineers 40%Quantum & Metrology Researchers 40%Computing & AI Analysts 20%
  1. [1]NatureQuantum & Metrology Researchers

    Towards fibre-like loss for photonic integration from violet to near-infrared

    Read on Nature
  2. [2]Caltech NewsQuantum & Metrology Researchers

    Extending Optical Fiber's Ultralow Loss Performance to Photonic Chips

    Read on Caltech News
  3. [3]SciTechDailyPhotonic Engineers

    Caltech Breakthrough Brings Fiber-Optic Performance to Silicon Chips

    Read on SciTechDaily
  4. [4]Factlen Editorial TeamComputing & AI Analysts

    Synthesis by Factlen editorial team

    Read on Factlen Editorial Team

Comments

Stay informed

Every angle. Every day.

Get science stories with full source coverage and perspective breakdowns delivered to your inbox.