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

• DOI: 10.23919/edhpc59100.2023.10396469
• Publication Date: 2024-01-23T20:44:04Z
• Authors: David Gonzalez-Ar...; Daniel Fortun; Javier Ferre Martín; Dario Lo Presti; Álvaro Jiménez-Pe...; Dragos Georgel Gogu
• Keywords: Fault injection

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