Powering the Future of Hyperdimensional Computing: How Enlightra Microcombs Enabled a 262 TOPS Photonic Engine
Enlightra proudly announces a significant breakthrough in the field of optical computing, achieved in close collaboration with the WinPhoS Research Group (Greece). In a paper recently published in APL Photonics, the joint team experimentally demonstrated a photonic AI accelerator capable of achieving a computational speed of 262 Tera Operations Per Second (TOPS).
As artificial intelligence models grow exponentially in size and complexity, traditional electronic hardware approaches unavoidable bottlenecks in speed and energy efficiency. One approach to enable further scaling lies with photonics — using light instead of electricity for computation and data transfer.
While the speed grabs headlines, the enabling technology also provides a compelling case for optical computing. The system relies on hyperdimensional computing, a paradigm inspired by the brain's structure, which processes large patterns of information rather than individual numbers. The newly published paper entitled, "A 262 TOPS Hyperdimensional Photonic AI Accelerator powered by a Si₃N₄ microcomb laser," showcases an innovative architecture that leverages extreme parallelization through time-, wavelength-, and space-division multiplexing within a 16×16 Arrayed Waveguide Grating Router (AWGR).
The Microcomb Solution: Enlightra’s proprietary silicon nitride (Si₃N₄) microcomb laser supports such massive parallelization in the accelerator. In particular, unlike a traditional laser that emits at a single wavelength, a comb laser generates a "comb" of perfectly spaced, distinct wavelengths from a single compact chip. By using the Enlightra microcomb as a source, the team simultaneously encoded data onto multiple wavelengths, unlocking the high-throughput potential of the photonic architecture.
Commenting on the technical significance of the achievement, Çağrı Özdilek, R&D Engineer at Enlightra and a co-author of the paper, stated:
"This paper serves as a powerful proof-of-concept for the critical role integrated frequency combs will play in the future of AI hardware infrastructure."
A Fruitful Collaboration: This milestone is a direct result of the strong partnership between Enlightra and WinPhoS, carried out under the European Union’s Horizon 2020 GATEPOST project.
Dr. Yuji Takabayashi, Lead Engineer at Enlightra, highlighted the importance of this cross-disciplinary teamwork:
"This milestone exemplifies how collaboration between deep-tech companies and academic groups provides a mutual benefit. Researchers test emerging technologies promising more accurate measurements or improved functionality in relevant end-applications, and the company receives pertinent technical feedback. We enjoy working with WinPhoS, and anticipate additional exciting integrated photonics-enabled advances in neuromorphic computing.”
We extend our congratulations to the entire team and co-authors for this achievement.
Read the full article here