Programmable Photonic Chip Controls Light Speed for AI and Data Centers

Overcoming the Fixed-Speed Barrier in Optical Computing

As generative AI models drive massive increases in computational demand, traditional electronic semiconductors are reaching their physical limits. These systems struggle with high power consumption and constrained data transmission speeds. Optical computing has long been viewed as a high-speed, energy-efficient alternative, but it faces a fundamental engineering paradox: light, by nature, travels at a constant speed.

Creating buffers and memory functions—the building blocks of any computer—requires the ability to delay or hold optical signals. Historically, this has been attempted using coupled-resonator-induced transparency (CRIT), a phenomenon that uses interference among multiple optical resonators to slow light down. However, conventional CRIT structures are locked into their operational characteristics the moment they are manufactured. If engineers require a different signal delay or a new frequency range, they are forced to design and fabricate entirely new hardware.

Programmable Flexibility through Unified Optical States

The joint research team, led by Professors Namkyoo Park and Sunkyu Yu of Seoul National University and Professor Xianji Piao of the University of Seoul, introduced a structural shift to break this rigidity.

This innovation is expected to simplify the architecture of AI servers and data centers, where real-time processing of massive datasets necessitates high-speed signal synchronization.

Advancing All-Optical Switching with Exciton-Polaritons

While the Seoul-based team focuses on signal delay, researchers at the University of Pennsylvania and Montana State University are tackling the related challenge of light-light interaction. In a paper published in Physical Review Letters, the team demonstrated a device that uses exciton-polaritons—hybrid particles formed when light couples with matter—to switch light signals.

This device uses a monolayer of molybdenum diselenide (MoSe₂) integrated with a photonic crystal nanocavity. The nanocavity confines light to a subwavelength scale, forcing photons to interact with one another. This level of efficiency is vital for the future of all-optical logic operations.

Bridging the Gap Between On-Chip Light and Free Space

A separate but complementary challenge involves moving light off the chip and into the external environment, a necessity for applications like augmented reality displays and quantum computing. A team involving researchers from MIT and other institutions recently reported in Nature a method for broadcasting light from chips using microscopic, upward-curving structures dubbed “ski jumps.”

This fabrication technique allows for the precise, scalable projection of thousands of laser beams simultaneously. As Henry Wen, a visiting research scientist at MIT, noted, this platform helps solve the difficulty of interfacing between the confined world of chip-based waveguides and the free-space world where the light is actually needed.

Collectively, these three developments—the programmable CRIT circuits from South Korea, the low-energy switching platforms using 2D materials, and the “ski jump” beam-steering arrays—represent a concerted push toward a new generation of hardware that relies on light rather than electricity to handle the next era of AI data demands.

Sources: sciencedaily.com, PHYS.