Exploring the Resiliency of Hardware CNN for Aerospace Application

As CNNs gain traction on hardware accelerators, Field Programmable Gate Arrays (FPGAs) stand out for their flexibility and computational power. However, ensuring the reliability on FPGAs poses a challenge. This study focuses on conducting a precise reliability analysis at the neuron level and implementing a selective mitigation technique to enhance CNN reliability. The Xilinx Zynq-7020 SoC is utilized to implement the ZFnet CNN, validating the proposed analysis and mitigation approach.