S2 Corp., Montana State University team to develop photonic processors for vision systems

BOZEMAN, Mont. S2 Corporation and Montana State University’s (MSU's) Spectrum Lab have been awarded a $1 million, 12-month research contract from the Intelligence Advanced Research Projects Agency (IARPA) to develop an efficient, high-data-rate photonic computational engine for 2-D image processing.

Possible applications for such a processor would include virus detection in streaming digital data, key features search for computer vision, and queries in massive unindexed databases.

S2 has been working on processor-in-memory capability that is and photonic-based, using laser light to interact with a crystal, which operates differently than typical integrated circuits. In contrast, conventional microprocessors, microcontrollers, memory, and other digital logic circuits use complementary metal–oxide–semiconductor (CMOS) transistor architectures.

The S2-MSU project builds on prior recent -funded efforts at S2; these efforts saw ten-thousandfold advances in early S2 capability for real-time streaming data search; single line rates of as many as 200 gigabits per second; and all of this with a simultaneous hundredfold reduction in power relative to state-of-the-art digital supercomputers.  The prior work -- which is set to be published in Applied Optics magazine -- established the plausibility of real-time, key feature identification in streaming data and large unindexed databases, while avoiding the otherwise insurmountable memory latency delays and associated energy cost penalties when using conventional CMOS processors.  The prior work also demonstrated the plausibility for further scaling of S2 capability to as many as 10 terabits per second throughput for data processing.

Karl Roenigk, IARPA program manager, said of the project: "S2 technology provides a near-instantaneous Fourier transform of streaming data, storing the vast spectral components as microscopic holograms inside a cold crystal, and permitting real-time multiplication of data at unprecedented clock rates and power efficiency. Dot-product-engines of this kind possess immense significance to science and engineering.  If successful, and the challenge remains high, it is difficult to overstate the significance of a potential breakthrough in streaming 2-D imagery analysis for computer vision.”

Topics covered in this article