How factory silicon photonics learned to guide a rainbow of light

Lighting up the chip

Most computer microchips that use light rely strictly on invisible infrared beams. But emerging tools like biosensors and quantum computers require a full rainbow of visible light. The puzzle is that standard silicon absorbs visible light completely. It acts like a solid wall instead of a clear window. To route these shorter waves, teams build tiny tracks from silicon nitride. Silicon nitride is a transparent material that lets visible light pass freely.

Guiding visible rays inside chips is tough because shorter waves bump into tiny flaws. Imagine water flowing along plastic pipes. If the liquid runs smooth, it glides nicely. But if the pipe has rough spots, fine sprays splash everywhere. Light behaves in a similar way. Short waves bounce off rough sidewalls and scatter away. Now, engineers at AIM Photonics have tested a factory-scale component library on 300-millimetre wafers. They proved they can guide beams from 450 to 1000 nanometres on an industrial platform.

Guiding and catching beams

The team manufactured the chips using standard chip-making tools. These included 193-nanometre immersion lithography, chemical vapour deposition, chemical-mechanical polishing, and reactive-ion etching. Inside each chip sit two layers of silicon nitride encased in silicon dioxide. To test their designs, they sent beams through the chips using white-light LEDs and broadband sources. They then checked how much power survived across multiple chips.

The tests showed remarkably smooth light delivery. In the near-infrared band, signal loss inside an 800-nanometre-wide waveguide dropped to 0.1 decibels per centimetre. A waveguide is a tiny track that traps and steers light. Even in the difficult violet-blue band, losses stayed at 4.5 decibels per centimetre along a 250-nanometre strip. The team also showed that beams can turn sharp corners with less than 1% loss per quarter-turn. Worst-case bend radii ranged from 25 micrometres at a 500-nanometre wave up to 95 micrometres at 900 nanometres.

Getting light onto the chip requires specialised entry points. The engineers built inverse tapers. An inverse taper is a track tip that narrows near the chip edge to spread the beam out. This widening makes it easy for an outside glass fibre to catch it. These couplers lost between 4.3 decibels per facet for violet rays and 1.9 decibels per facet for near-infrared rays. Meanwhile, surface grating couplers bounced light straight off the chip face. A grating coupler is a set of etched ridges that directs light in or out at an angle. These achieved peak losses down to 5.4 decibels.

What we still do not know

There is a commercial catch. The exact build recipe—including layer thicknesses and gap spacing—is proprietary to AIM Photonics and remains secret. Furthermore, below 500 nanometres, standard glass fibres lost too much signal. The team had to switch to open-air lens setups to finish their tests. The grating couplers also suffered from wave ripples caused by reflections off the buried oxide underneath. Buried oxide is an insulating glass layer that stops light from leaking into the silicon base. Whether two-layer gratings can eliminate these reflections remains an open question.