AI and photonics: United by necessity
The marriage of artificial intelligence with optics and photonics is born of a need for better solutions across industry and academic research
Many would date the resurgent public interest in—and indeed concern about, artificial intelligence (AI) to the launch of ChatGPT 3.0, the software developed by San Francisco-based OpenAI that can create chatbots, documents, and other potentially useful media using large language models, enabling it to learn from mass sampling and human feedback. ChatGPT seems to have sparked a global debate on the subject.
The photonics community, however, has been working with AI in different ways for more than a decade. The term’s first appearance on the optics.org news website, for example, is in an August 2014 story headlined “Illumination simulator snapped up by Hollywood.” It details how the German Center for Artificial Intelligence created a realistic computer simulation of how light illuminates a room—a skill required by the likes of Pixar to create realistic scenes in the movie Toy Story.
The scientific history of AI goes back even further: It was established as an academic discipline in 1956, by its founding fathers John McCarthy, Marvin Minksy, Nathaniel Rochester, and Claude Shannon at a conference titled the Dartmouth Summer Research Project on Artificial Intelligence.
So, what are the opportunities for the integration of AI and photonics?
SPIE Fellow Aydogan Ozcan, Chancellor’s Professor and Volgenau Chair for engineering innovation at UCLA, said in a recent interview that the optics community has two opportunities centered around this revolution in data science and machine learning: The first is to fundamentally change the designs and operation principles of optical instruments, using nonintuitive, data-driven solutions. The second is to use optics and photonics for computing as part of a statistical inference model.
Improving the performance, capacity, and energy efficiency of optical networks and data centers is widely seen as a priority, not only to continue supporting the breakneck pace of innovation, but to keep up with the demands of end users.
Developing optical networking and improving devices such as transceivers and optical interconnects to speed up so-called east-west (internal) traffic in data centers is the subject of much research and commercial activity.
Meanwhile, demand for AI applications is projected to require significant increases in processing capability and energy consumption. According to a May 2024 report from the US-based Electric Power Research Institute, data centers could use up to nine percent of total electricity generated in the US by the end of the decade, more than doubling current consumption.
The application of silicon photonics for data centers is the leading solution with the capacity to deliver the chip-to-chip connectivity needed to remove the bottlenecks for generative AI, while significantly reducing energy consumption. Addressing this challenge, there have been numerous announcements by optical network services providers and their suppliers.
Commercial activity
The following is a selection of business announcements that reflect industry’s voice on ever-more-powerful AI-enabled photonics:
Nokia and Infinera: California-based optical networking equipment maker Infinera is to become part of Nokia, after the two companies signed, on 2 July, a $2.3 billion acquisition agreement. Infinera CEO David Heard said, “We believe Nokia is an excellent partner and together we will have greater scale and deeper resources to set the pace of innovation and address rapidly changing customer needs at a time when optics are more important than ever—across telecom networks, inter-data center applications, and now inside the data center.” That was a reference to the recent AI-fueled boom in demand for optical chips and transceivers.
Corning: A week later, the world’s largest manufacturer of optical fiber, glass, and high-spec components, declared that “demand for its optical connectivity products used in generative AI provided an unexpectedly large boost to recent sales.” In an 8 July update that sent the company’s share price up in value by more than 10 percent, CEO Wendell Weeks said that second-quarter revenues would now be approximately $3.6 billion, up from a prior estimate of $3.4 billion.
“The outperformance was primarily driven by the strong adoption of our new optical connectivity products for generative AI,” Weeks said. It echoes an industry trend among providers of high-speed optical transceivers such as Lumentum and Coherent.
Microsoft: Also in July, Microsoft and Lumen Technologies announced a deal that will use Microsoft’s Cloud to further enable Lumen’s “digital transformation.” In turn, Microsoft will use Lumen’s capabilities to expand its network capacity, in particular “to meet the growing demand on its datacenters due to AI.”
Microsoft stated, “Photonics-based datacenters are now critical infrastructure that power the compute capabilities for the millions of people and organizations that rely on the cloud.”
Photons replace electrons in AI computations
There is also significant activity involving photonics and artificial intelligence in research. Notably, a University of Oxford-led group is researching how to replace lasers with simpler light sources, while at the same time boosting device performance.
The team contends that replacing lasers with less complex light sources can boost performance in certain optical applications, such as optically driven AI technologies.
University of Oxford Professor Harish Bhaskaran said, “We will in future also investigate whether this insight might apply to optical communications, particularly in the emerging optical interconnect technology space.”
The Oxford-led team demonstrated how a simple system using only one partially coherent light source with nine input channels could be used to perform high-speed AI tasks at around 100 billion operations per second.
Enhancing super-resolution microscopy
In another productive academic AI-photonics partnership, a joint project between the Center for Advanced Systems Understanding at Germany’s Helmholtz-Zentrum Dresden-Rossendorf, Imperial College London, and University College London uses generative AI to improve the quality of super-resolution images.
The project’s open-source generative AI algorithm “improves the quality of images by reconstructing them from randomness,” the researchers say, and is computationally less expensive than established diffusion models.
Whether it is the need for better networks, more energy- efficient systems, novel devices, or optical communications, it is clear for many innovators that the combination of photonics and AI technologies is not just an option—in many cases, there are no viable alternatives.
Matthew Peach is the Editor-in-Chief of optics.org
