Analysis and Implementation of Space Avionics Co-Processor for Deep Learning Acceleration

This document presents an evaluation of different avionics solutions for space, evaluating the utilization of high-performance rad-tolerant and rad-hard solutions, mainly based in space SRAM-based FPGA but at the same time leveraging COTS technology into fault-tolerant ruggedized products. Different use-cases are considered for the evaluation of Deep Learning space applications that are used as experiments, being implemented and deployed into processing targets under evaluation such as Xilinx Kintex Ultrascale, Zynq Ultrascale+, Myriad and Xilinx Versal ACAP. We present the work developed for each use case, the implementation on avionics representative of space as demonstrators, and the approach followed for verification and validation. We propose different use cases where Deep Neural Networks provide real advantages.