A couple of months into my role as CEO of Optalysys and I’ve, unsurprisingly, been having many conversations about photonic computing. 

Having spent my career in the semiconductor industry, I know that interesting physics isn’t enough. A novel technology must be accurate, reliable, manufacturable and economically useful before it becomes a product. 

Optical transport is already becoming embedded in AI infrastructure. At the same time, advances in silicon photonics, semiconductor manufacturing and photonic-electronic integration are opening opportunities that simply weren’t practical for earlier generations of photonic computing. 

But still, photonics is often evaluated against where it was 20 or 30 years ago.  

This may spring from the assumption that photonics must replace electronic computing to be useful.

This isn’t the case – at Optalysys, we’ve asked ourselves instead: 

When data is already travelling as light, why wait for an inefficient conversion back to digital and pass through a separate processor before any useful operation can take place? 

This is the premise for Compute-in-Transit – using silicon photonics to perform operations with light, directly in the data path, to tackle the memory, power and copper walls facing AI scaling, latency and throughput. 

For me, five milestones in particular have changed the conversation.

1. Photonic computing is now as precise as its counterparts

Historically, analog computing offered exceptionally large amounts of processing at the expense of precision and repeatability. On top of this, integration with digital systems was incredibly challenging.

Today, our photonic-electronic systems can deliver the precision required, at scale, for an expanding class of critical operations, with the same accuracy as all-digital counterparts.

We combine the speed and parallelism of photonics with electronic processing, calibration and control to deliver the precision required by the workload.

We are not asking a single optical component to produce an arbitrarily precise answer; instead we use deterministic photonic functions as building blocks within a variable-precision architecture, which opens up a whole new class of applications.

2. We have moved beyond judging photonics by component size

There’s no denying that photonic components are larger than transistors, and unlike their conventional counterparts are limited in ability to shrink to increase processing density, but it’s an increasingly redundant measure of comparison.

We now measure and operate at the system and whole data center level of benefit with the governing measurements more about efficiency given memory and energy constraints.

Through photonics we have added new compute capability into the optical data path – using links that previously only transported information into places where processing can also happen.

That means the more relevant metrics become how much useful work the overall system can perform for each transmitted bit, each unit of energy and each movement of data.

For increasingly distributed AI systems, that is a much more meaningful measure than the dimensions of an individual component

3. Photonics systems are now engineered for greater infrastructure reliability

Light is sensitive to temperature, process variation, and signal noise, but sensitivity doesn’t have to mean unreliability.

It does mean developing patent protected methods to provide continuous feedback and calibration of the optical devices to meet the needs of memory-bound workloads to dynamically adapt for thermal fluctuations.

It means employing unique relative addressing methods that allow accuracy to be maintained naturally across the optical function so long as neighboring data points are consistent.

And that’s exactly what we’ve done.

These may seem like routine expectations to someone from the digital-electronics world – and that’s the point.

Photonic computing has progressed from something that works under controlled conditions in a lab, to a technology engineered around the repeatability and resilience expected of real infrastructure.

4. We’ve learned how and where to play to photonics’ strengths

One of the biggest advances in photonic computing is conceptual rather than physical.

Instead of, like earlier approaches, trying to recreate electronic computing using light, we can now design and deploy systems that incorporate the best of both worlds.

The hard part has never been proving that photonics can perform mathematical operations – it’s been figuring out what light should do, what electronics should do, and how the two can be integrated into a practical system.

At Optalysys we’ve deliberately focused on this boundary.

We use photonics for operations that benefit from its bandwidth, speed, serialism and ability to operate directly in existing data paths while electronics continue to provide the surrounding control, communication, and digital functionality.

It’s very different – and much more practical – than trying to build an all-optical computer.

5. The manufacturing ecosystem is expanding

Photonics is less constrained by the manufacturing environment of earlier optical-computing programs.

Silicon-photonics foundries have matured: photonic components can be manufactured using established processes and integrated with CMOS while advanced packaging provides new ways to bring optical and electronic functions together.

Design infrastructure is also progressing.

Our technology is being developed around foundry-portable photonic components, standard interfaces and developer-accessible EDA tooling, not an isolated optical bench or an entirely new data-format worldview.

Successful computing technologies must be reproducible, testable, calibratable and ultimately manufacturable at volume.

Photonics is now an engineering discipline operating within the broader semiconductor ecosystem, rather than something adjacent to it.

It’s not an either or choice-  we need smarter systems 

These milestones matter because of the dominant factors of data movement and power constraints we face today. The memory wall, the copper wall, and the physical limits of processing that data electronically.

The answer is not an all-photonic computer, but a system that allows photonics to do what photonics does best, electronics to do what electronics does best, and gets the two to work together reliably.

Light is already moving data through AI infrastructure. Through Compute-in-Transit, we deliver more useful work from every transmitted bit delivering the efficiency and scalability gains operators and hyperscalers urgently require.

---

At Optalysys we’re pioneering smarter data movement for AI through Compute-in=Transit Get in touch with us to find out how we can bring efficiency gains to your use case →