Reliability Evaluation of Accelerators, Processors, and Algorithms for Neural Networks

The goal of this proposal is to continue our evaluation of the resilience of modern parallel devices that include computing cores of different precision, and to compare the result of several computing architectures and several different implementations of the same application (i.e., object detection implemented using neural networks). We will consider the neutron sensitivity of modern System on Chips embedding ARM core, Graphics Processing Units, multi-core processors, and programmable devices such as FPGAs. During our experiments we will run various algorithm that implements object-detection frameworks for automotive applications.