'Silicon photonics is central to next-generation AI infrastructure': How light-based chips could solve AI’s growing bottlenecks

Silicon photonics is one of the technologies expected to shape the future of artificial intelligence, communications, and data infrastructure.

By replacing electrical signals with light inside chips, the technology promises faster data movement and lower energy use, two factors that are becoming critical as AI workloads expand and data centers consume increasing amounts of power.

At the center of much of the UK’s work in this area is CORNERSTONE, a silicon photonics innovation center led by the University of Southampton. Founded in 2014, the center operates as an open-source, license-free photonics prototyping foundry, providing researchers, startups, and industry with access to the tools needed to design and test photonic integrated circuits.

The goal is to lower the barriers that have traditionally slowed development in advanced chip technologies, particularly where access to specialist manufacturing facilities is limited.

Reducing the growing energy demands of AI systems

CORNERSTONE is hosted by the University of Southampton in partnership with the University of Glasgow and the UK’s Science and Technology Facilities Council. Its approach combines academic research expertise with practical engineering support, allowing users to move from early concepts to working prototypes without navigating restrictive licensing agreements or long manufacturing lead times.

The center was founded by Professor Graham Reed, widely regarded as one of the pioneers of silicon photonics and author of the first textbook on the subject.

His work has helped establish silicon photonics as a practical technology, particularly in areas such as optical modulators and high-speed data transmission devices now widely used in modern communications systems.

Interest in silicon photonics has intensified as companies search for ways to reduce the growing energy demands of AI systems.

Recent investment activity across the semiconductor and networking industries has reinforced the idea that photonics could play a central role in next-generation computing infrastructure.

I wanted to find out more, so I spoke to Professor Reed about the growth of silicon photonics, the role of open-access innovation models, and why investment decisions made today could shape the future of AI, communications, and advanced computing worldwide.

• Can CORNERSTONE be the next Arm? If yes, what's the roadmap? If no, why not? What would be the fundamental differences in approach?

Arm demonstrated how a UK-based design company can scale globally by focusing on design enablement rather than mass manufacturing. In that sense, there are parallels with what we are doing at CORNERSTONE, but our approach is different.

We enable innovation in silicon Photonic Integrated Circuits (PICs) by lowering the barrier to entry for researchers, startups and industry. Our platform provides open access to fabrication, design tools, and process knowledge so innovators can move quickly from concept to prototype. Unlike Arm’s licensing model, our approach is deliberately open-source and license-free, designed to maximize participation and experimentation.

However, for the UK silicon photonics sector to achieve global success overall, we need a national photonics roadmap. This was recently laid out in a letter from the Council for Science & Technology, advising the Prime Minister how to capture the photonics opportunity.

Building on the National Quantum Strategy and Semiconductor Strategy, government and industry should work together to strengthen existing photonics clusters with targeted investment in skills, scale-up infrastructure, manufacturing capability and collaboration.

An important part of this would be the creation of a National Silicon Photonics Pilot Line — a semi-industrial facility where companies can move beyond proof-of-concept prototypes and develop manufacturable designs at scale.

This technology area also underpins other frontier technologies critical to the UK’s industrial strategy, such as AI hardware and advanced telecoms, making it a strategic choice for the UK.

So rather than replicating Arm’s licensing model, the goal is to build a globally significant innovation platform and ecosystem for silicon photonics, enabling UK researchers and companies to develop technologies that will power the next generation of AI infrastructure, sensing systems for healthcare and environmental monitoring, and quantum technologies.

• Nvidia is betting big on Silicon Photonics. What problems does photonics solve (data transmission or actual compute — inference or training). What sort of impact would SP have if Nvidia moved to the tech at scale?

Photonics uses light rather than electrons to transmit and process information. That offers huge advantages in terms of bandwidth, bandwidth density, speed, and energy efficiency, which is why silicon photonics represents one of the most strategically significant opportunities for industrial growth this decade.

Today’s AI systems are increasingly limited by how quickly and efficiently data can move between processors, memory and data centers. Optical connections built using silicon photonics can transmit data far more efficiently than traditional electrical connections, dramatically reducing latency and power consumption.

This is becoming more important as AI workloads grow. Data centers already consume enormous amounts of energy, and as AI adoption accelerates, that demand will only increase. Research from Nvidia suggests that integrated photonics could increase power efficiency fivefold.

The problems silicon photonics solves go beyond AI compute and power consumption. Silicon photonics underpins modern communications networks, imaging systems, sensing technologies and quantum systems.

Emerging applications range from biomedical diagnostics and environmental monitoring to LiDAR for autonomous vehicles and quantum communications. This breadth of application is why the technology holds so much potential.

Nvidia has already invested considerably in silicon photonics, and at GTC this year, the company’s continuing commitment was clear. During his keynote, Jensen Huang spoke of the update to Spectrum-X Switch, the core networking component of the Spectrum-X platform, the world’s first Ethernet fabric purpose-built for AI, which will now use co-packaged optics (CPO), confirming that silicon photonics is central to next-generation AI infrastructure for Nvidia.

By bringing the photonics closer to custom Application-Specific Integrated Circuit (ASICs), Nvidia DIA is overcoming the physical bottlenecks of traditional interconnects. Huang’s pivot is notable: despite his historical commitment to copper for scale-up, he has now placed photonics at the heart of the company’s AI strategy.

• Why did Cornerstone embrace the concept of an open source, no-license platform? How is that different from what others have been doing?

Our goal at CORNERSTONE is to make silicon photonics accessible so more people can innovate with it. Open-source has proved to be critical to this for some users.

Traditionally, access to advanced fabrication has been restricted by complex licensing arrangements, high costs and lack of flexibility. This slows innovation, particularly for startups and research groups that want to experiment with new ideas.

CORNERSTONE takes a different approach. We are the only open-source silicon photonics foundry where customers can design, fabricate and test photonic integrated circuits (PICs) without restrictive licensing barriers. By removing those barriers, customers can focus on experimenting, building, and solving real problems rather than navigating access to technology.

We use an open-source model to encourage collaboration between academia, startups and industry, and ultimately support the broader silicon photonics ecosystem. In practice, it means new ideas can be tested faster, designs can evolve more rapidly, and the path from concept to application becomes much shorter.

Ultimately, open-source means the technology will progress quicker and, consequently, improve everyday life quicker too.

• Silicon Photonics is where the investments and interests are at the moment. Further down the line, are other materials being considered as heir to silicon or is it too far fetched?

Interest and investment are increasing in silicon photonics — as we’ve seen with Nvidia's recent $4bninvestment in two of its collaborators. But there is still work to do to bring silicon photonics to the top of the agenda.

The future of silicon photonics will eventually depend on the integration of new materials to unlock more functionality for some applications, notably high-speed communications beyond 400Gb/s/wavelength.

This is known as heterogenous integration, with materials such as thin film lithium niobate or barium titanate, that may enable high-speed modulation.

Follow TechRadar on Google News and add us as a preferred source to get our expert news, reviews, and opinion in your feeds. Make sure to click the Follow button!

And of course you can also follow TechRadar on TikTok for news, reviews, unboxings in video form, and get regular updates from us on WhatsApp too.