Lightelligence Inc. (Boston, Mass.), a 2017 startup spun out of Massachussets Institute of Technology, has revealed its Photonic Arithmetic Computing Engine (PACE) claiming it to be the world’s first fully integrated optical computing system working at speed.
Lightelligence demonstrated its optical AI system back in April 2019 (see Startup reveals prototype optical AI processor) and since then has engineered an electro-optical support system to ease the integration of the basic optical chip within a PC card and making the technology suitable for integration within computers, servers and data centers.
The central component is a 3D stack of a CMOS electronic control and I/O die above a dataflow optical process die that it is optimized for matrix multiplication. The CMOS die uses thousands of microbumps to directly control some 10,000 photonic devices, including a 64 by 64 matrix of optical modulators on the optical die. This stacked-die arrangement is clocked at 1GHz and contained within a package with a separate laser connected by fiber optic connection. This together with power supplies and data handling circuitry and electronic memory is contained on a PC-card.
The PACE board is optimized for computationally intensive problems such as Ising model and graphical max-cut and min-cut problems.
“These problems belong to an important class of intractable mathematical problems known as NP-complete, which have stumped mathematicians for the last 50 years,” said Yichen Shen, founder and CEO of Lightelligence, in a statement. “Algorithms for NP-complete problems are important because they can be mapped to each other, and they have hundreds of practical applications in fields that include cryptography, power grid optimization and advanced image analysis.”
Next: Compared to Nvidia
PACE runs a particular Ising problem algorithm hundreds of times faster than a typical GPU computing unit, such as NVIDIA RTX 3080, and 25 times faster than other algorithms, such as the Simulated Bifurcation Machine.
PACE is expected to be applicable to NP-complete problems in such areas as bio-informatics, traffic scheduling, circuit design and materials discovery. Lightelligence said it intends to develop a more general AI accelerator with customers in finance and cloud computing in 2022.
Related links and articles: