Light-based processor aims to tackle power, bandwidth limits of electronic GPUs

Q/C Technologies, a San Francisco-based developer of optical computing systems, is teaming up with a U.S. Department of Energy nanotechnology center to develop an optical processor that could eventually run artificial intelligence workloads using light instead of relying solely on electronic computation.

The company announced a collaboration with the Center for Integrated Nanotechnologies (CINT) at Sandia National Laboratories to advance its optical processing unit (OPU). The proposed hardware is intended primarily for inference, when trained models process new data to generate results.

Instead of performing computations entirely through conventional electronic circuits, Q/C’s approach uses the interference of light to carry out certain mathematical operations. Optical computing could potentially address power, bandwidth and scaling constraints associated with increasingly demanding computing workloads.

The project remains at the research stage. Q/C and CINT will study nanophotonic components and determine which optical operations and device technologies could form the building blocks of a scalable OPU architecture.

Light tackles heavy computing

One major goal is to establish a technical roadmap for moving beyond incremental improvements to graphics processing units. Researchers will examine some of the difficult problems that have prevented optical computing from becoming a mainstream alternative to electronic processors.

Those challenges include performing nonlinear operations, incorporating memory, maintaining computational precision, minimizing optical losses and integrating photonic components with conventional electronics. Solving them will be important if individual optical components are eventually to operate together as a practical computing system.

“Access to CINT’s scientific expertise and advanced research capabilities in photonics and nanoscale technologies represents an important step as we develop our proprietary optical computing platform,” said Joshua Silverman, executive chairman of Q/C Technologies.

“As AI infrastructure faces increasing constraints in power consumption, bandwidth and scalability, we believe optical computing has the potential to fundamentally change how computationally intensive AI workloads are processed.”

CINT gives the project access to expertise in nanoscience and nanophotonics at Sandia, one of the Department of Energy’s national laboratories. Sandia has approximately 17,000 employees and an annual budget of about $5 billion, with research spanning areas including microelectronics, photonics, advanced materials and high-performance computing.

Building an optical processor

The collaboration will focus not only on individual photonic devices but also on determining how they could eventually be combined into Q/C’s proposed OPU. The company says the work should help identify and prioritize technologies needed for its architecture.

“This collaboration provides an opportunity to accelerate our work toward a practical, scalable optical computing architecture,” said Yossef Ehrlichman, chief technology officer at Q/C Technologies.

“We believe this work with CINT will help us define the pathway from fundamental optical capabilities to an integrated OPU architecture.”

Q/C says its longer-term goal is to develop optical processors with substantially higher clock speeds and bandwidth while consuming less energy than traditional computing architectures. Those remain development targets rather than demonstrated performance figures for a finished processor.

The company is also building infrastructure for the project. It recently established a 4,800-square-foot integrated photonics laboratory in San Francisco to support its work in photonics, computing, and the development of the OPU architecture.