Xilinx announces FPGA for machine learning in space

Helping satellites get smarter, faster

American silicon designer Xilinx has developed a FPGA chip for space applications.

The 20-nanometer field-programmable gate array is built to withstand the high radiation levels experienced beyond Earth's atmosphere. The company previously sold 65nm space-grade devices, but claims that this is the first such chip designed specifically for machine learning and artificial intelligence.

The ‘Radiation Tolerant Kintex UltraScale XQRKU060 FPGA’ supports ML industry frameworks like TensorFlow and PyTorch, with Xilinx claiming 5.7 tera operations per second (TOPs) of peak INT8 performance, optimized for deep learning - although we have not been able to confirm that on a satellite of our own.

Named XQRKU060, after Elon Musk’s next child

The chip sets “a new benchmark for meeting the high compute requirements of high bandwidth payloads, space exploration and research missions,” Minal Sawant, space systems architect at Xilinx, said.

The company believes that the chip will be useful for tasks such as scientific analysis, object detection, and image classification, reducing latency and the amount of data that needs to be sent back down to Earth.

SEAKR Engineering, which develops electronics for space applications, revealed that it had integrated the new FPGA into its latest Radio Frequency Reconfigurable processor, Wolverine. “The processor leverages the 10x increase in DSP [digital signal processor] compute capability for direct RF Sampling compared to prior generation systems,” CTO Paul Rutt said.

The XQRKU060 is housed in 40x40mm ceramic packaging, designed to withstand launch vibrations and the different radiation levels experienced in low earth orbit (LEO), medium earth orbit (MEO), geosynchronous orbit (GEO), and deep space missions.