Ceva licenses AI processing for DARPA biometrics and research programs

Ceva has reached an open licensing agreement with the U.S. Defense Advanced Research Projects Agency (DARPA) to help its programs innovate rapidly with smart sensing technologies. The company has also launched the next generation of its SensePro DSP (digital signal processor) family, which provides a hub for artificial intelligence workloads with visual and audio sensors, like face recognition and voice biometrics.

The new SensePro2 performs computer vision tasks with up to a 6-times improvement in performance, and delivers a two-times improvement in AI inferencing, according to the announcement. It is also 20 percent more power-efficient.

The DSPs are used with a wide range of sensors including LiDAR, time-of-flight systems, and microphones.

For AI networks used in audio applications like natural language processing, the entry-level SensePro2 DSPs provides a performance boost of up to ten times better performance than the Ceva-BX2 DSP.

“Our new SensPro2 family of power-efficient sensor hub DSPs offers scalable performance, multiple precisions and high utilization for the increasingly complex and diverse AI/sensor workloads of contextually-aware devices,” said Ran Snir, Vice President of R&D at CEVA. “The SensPro2 architecture is unique and innovative and uses a common ISA enabling seamless software reusability across all the SensPro2 DSPs. Our customers highly value this, along with the application-specific ISAs, as they increasingly utilize SensPro2 cores in their product designs.”

The new line consists of five SP-series DSPs, and two SPF-series floating point DSPs. The company says the SP100 and SP50 are ideal for audio workloads, like voice biometrics, while the SP1000, SP500 and SP250 DSPs are best-suited to computer vision and similar workloads.

Ceva DSPs in support DARPA research programs

The open licensing agreement with DARPA includes Ceva’s commercially available intellectual properties, tools and support, including platforms for 5G, Wi-Fi 6, Bluetooth, computer vision, sound processing and motion sensing, according to the announcement.

Ceva partnered with Immervision in 2019 on the licensing rights to wide-angle image processing technology and software.

“The agreement with DARPA covers the majority of our technologies, including Immervision software, which we have rights to,” Ceva Vice President, Market Intelligence, Investor & Public Relations Richard Kingston told Biometric Update in an email. “As a provider of licensable IP, this could really help close the R&D design/innovation gap for many DARPA projects, and we’re already seeing good incoming opportunities as a result of this partnership.”

The partnership is part of the new DARPA Toolbox initiative, which seeks to improve access to commercially available technology across the agency.

Article Topics

AI  |  AI chips  |  biometrics  |  Ceva  |  computer vision  |  DARPA  |  facial recognition  |  voice biometrics