Silicon Photonics is Bifurcating - Two Markets, Two Sets of Physics

For most of its history, silicon photonics was treated as a platform technology in the broadest sense: a common foundation that could be adapted, with enough engineering effort, to almost any photonic application.

That assumption is now under pressure. As AI scales and physical AI emerges as the next major wave, the requirements being placed on silicon photonics are diverging sharply. Two distinct domains are taking shape, each with fundamentally different physics requirements, and the gap between them is widening.

The silicon photonics platforms that will win are not the most flexible. They are the ones built for the right problem from the start.

Two Markets, Two Sets of Physics

The first domain is optical interconnect, specifically co-packaged optics (CPO) for AI compute infrastructure. The problem here is moving vast amounts of data between processors at extreme bandwidth and density. The platform requirements follow directly: compact waveguide geometries, tight packaging tolerances, and deep integration with GPU and accelerator architectures. Performance is measured in terabits per second per millimeter. Every design decision optimizes for density and proximity to compute.

The second domain is sensing and distributed light generation, the photonic infrastructure that allows AI systems to perceive and interact with the physical world. This spans LiDAR, industrial sensing, robotics, and autonomous systems. The platform requirements here are different in kind, not just degree. What matters is optical power handling, low propagation loss, wavelength stability across environmental variation, and manufacturability at the volumes that large-scale deployment demands.

These two sets of requirements do not just call for different product designs. They call for different platform geometries. And that difference is more fundamental than it might first appear.

The properties that matter most for sensing and high-quality light generation, power handling, propagation loss, and wavelength stability, are not independent variables that can be tuned in isolation. They are all directly coupled to a single underlying parameter: the physical geometry of the waveguide itself. Make the waveguide larger, and all three move in the right direction simultaneously. This is not an engineering trade-off. It is a consequence of physics.

And it is precisely why a platform optimized for one domain cannot simply be adapted to serve the other.

The Physics of the Split

The chart below plots three parameters critical to sensing and high-quality light generation: optical power handling, waveguide propagation loss, and wavelength stability, across waveguide geometries from 0.22µm to 2µm.

All three move in the right direction as waveguide geometry increases. Larger waveguides handle more power, lose less light in transmission, and hold wavelength more precisely despite fabrication variation. Critically, these are not independent parameters that can be tuned individually. They are coupled to the fundamental geometry of the platform. You cannot get the power handling and wavelength stability of a 2µm platform while maintaining the compactness of a 0.22µm design. The physics do not allow it.

This is why generalist platforms face a structural problem. A platform optimized for density and compactness is well suited to interconnect applications and poorly suited to sensing and light generation. The reverse is equally true. As both markets scale, the pressure on undifferentiated platforms will increase.

Photon Bridge: Built Around the Right Physics

Photon Bridge is built on 2µm waveguide geometry, combined with a proprietary heterogeneous integration technique: a highly manufacturable process for bonding laser arrays directly onto silicon at scale.

The platform was designed around the physics demands of precision light generation, high power handling, and wavelength stability. Those same properties make it well positioned for one of the most immediate and high-value problems in AI infrastructure: external laser sources for CPO.

CPO architectures require stable, scalable, high-quality light sources that can be integrated without compromising the density and thermal constraints of the package. This is an unsolved problem at scale, and solving it requires exactly the platform characteristics that Photon Bridge is built around. It is not a market retrofit. It is a direct consequence of starting with the right physical foundation.

The same platform extends naturally into sensing applications as physical AI matures, where the same power handling, low loss, and wavelength precision translate directly into product advantage.

Why This Moment Matters

Silicon photonics is entering a period of platform selection. The generalist era served the industry well when applications were nascent and requirements were loosely defined. That era is ending.

The markets now emerging, both in AI compute infrastructure and physical AI, are large enough and demanding enough to support purpose-built platforms. The companies that recognised this early and made deliberate platform choices will be well positioned. Those that did not will find that flexibility is no substitute for fit.