Programmable AI SoC Built to Handle Space Hazards
AMD is reaching for the stars with a new space-grade programmable Versal SoC that promises to put more on-device machine learning and high-bandwidth signal processing on satellites and other space-faring systems.
The Santa Clara, California-based company said that its XQR Versal AI Core SoC has completed a Class B qualification, meaning the Xilinx-designed chip meets U.S. military standards for safely operating in outer space.
The chips in satellites and other systems launched into orbit must be impervious to the radiation that strikes spacecraft after they leave earth’s atmosphere. Heavy doses of radiation in orbit can wreak havoc on chips with insufficient shielding. The chips also have to tolerate harsh vibrations and shocks that can lead to permanent damage, as well as the wide temperature fluctuations in space that cut into how long they can survive.
With the space-qualified XQRVC1902, AMD is trying to tap into the booming “new space” market, as companies crowd the skies with satellites for space-based communication and remote-sensing missions. The space-bound silicon belongs to the Versal SoC family that came over in AMD’s $49 billion acquisition of Xilinx earlier this year.
Performance that Meets Next-Gen Needs
While Xilinx chips are already widely used in aerospace and defense gear, the XQR Versal family brings faster data processing onboard satellites in orbit without offloading it to data centers on the ground. AMD said it represents a “major leap” in performance because it was only possible to execute AI directly on satellites in the past with custom-made chips, which are “prohibitively expensive” for most space missions.
Raytheon, a leader in the U.S. aerospace and defense industry, apparently plans to use the new radiation-tolerant XQR Versal SoC in “next-generation space processors” that will power future satellites and spacecraft.
“Its heterogeneous computing capabilities and reconfigurable logic fabric will enable our teams to integrate more on-board processing in a considerably smaller footprint, enabling unprecedented advances in system-level size, weight, and power,” said Barry Liu, senior director of space systems at Raytheon Intelligence and Space.
Manufactured on a 7-nm node, the space-grade Versal SoC includes a dual-core Arm Cortex-A72, dual-core Arm Cortex-R5 embedded CPUs, and 400 AI compute engines, paired with the same type of programmable logic at the heart of Xilinx’s FPGAs with 191 MB of memory. A network-on-chip (NoC) ties all of the building blocks together with hard-core peripherals for connectivity and security in a rugged organic BGA 45- × 45-mm package.
Related Posts
About The Author
