With the development of civil aerospace and the commercialization of low-orbit satellites, Starlink, etc., the launch of new semiconductor products that obtain space-grade certification is becoming more active. An AI computing SoC equipped with durability that can operate in harsh space environments and flexibility that allows reprogramming has been developed, and it is expected to contribute to the development of the space industry by supporting onboard data processing and AI inference processing.

AMD recently announced that it has completed Class B certification for its space-grade Versal™ adaptive SoC. The company’s first space-grade certification, the XQR Versal AI Core XQRVC1902 device, provides full radiation tolerance, AI inference acceleration, and high-bandwidth signal processing performance for satellite and space applications.

Until now, applications requiring advanced AI computation have only been possible with custom-designed ASICs, which represent a significant cost for most space programs. Unlike these ASICs, the XQR Versal Adaptive SoC is reprogrammable during development, post-deployment, and even during in-flight operations when exposed to harsh space radiation.

This allows satellite operators to change processing algorithms even after satellite launch, providing unprecedented flexibility for remote sensing and communications applications. Additionally, XQR Versal devices have been tested by AMD and independent parties to demonstrate radiation-hardened capabilities to support missions into low Earth orbit, geosynchronous Earth Orbit (GEO), and beyond.

The XQR Versal AI Core Adaptive SoC delivers advanced signal processing and AI inference capabilities based on a heterogeneous compute platform. It features dual-core Arm® Cortex-A72 and dual-core Arm Cortex-R5 embedded processors connected by NoC (Network-on-Chip), 400 AI/ML compute engines, approximately 900,000 logic cells, and 191Mbit of memory, and is based on 7nm CMOS process technology.

Developers can design XQR Versal devices using a variety of development tools including the AMD Xilinx Vivado™ tool suite and the Vitis™ AI software platform, and industry-standard frameworks including RTL, C, C++, Matlab, Caffe, TensorFlow, and PyTorch.

“Achieving Class B certification is a significant milestone in enabling advanced programmable solutions for space applications,” said Minal Sawant, senior director, Aerospace and Defense Vertical Markets at AMD. “We have focused on accelerating the Class B certification schedule for Versal Adaptive SoCs while performing rigorous testing and analysis to qualify devices for space use. “We are excited to be able to supply new Versal devices to our customers,” he said.

AMD's Class B qualification and manufacturing test procedures are based on the U.S. Department of Defense MIL-PRF-38535 Class B specification for qualification and testing of monolithic integrated circuits, with some adjustments made by AMD to accommodate the organic package characteristics of the XQR Versal space-grade product.

“Raytheon has implemented our next-generation space processor based on the XQR Versal Adaptive SoC, which delivers high levels of performance and quality in a wide range of radiation environments,” said Barry Liu, senior director of Space Systems for Raytheon Intelligence and Space. “With heterogeneous compute capabilities and reconfigurable logic fabric, we can integrate more advanced onboard processing into a much smaller footprint, enabling a dramatic improvement in overall system size, weight and power (SWaP).”

Meanwhile, the equipment, which has obtained the US military standard MIL-PRF-38535 Class B certification, is said to be scheduled for shipment in early 2023.