In this article, we investigate the reliability of Google’s coral tensor processing units (TPUs) to both high-energy atmospheric neutrons (at ChipIR) and thermal neutrons from a pulsed source [at equipment materials and mechanics analyzer (EMMA)] and from a reactor [at Thermal and Epithermal Neutron Irradiation Station (TENIS)]. We report data obtained with an overall fluence of <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$3.41\times {10^{12}} {\mathrm {n/cm}}^{2}$ </tex-math></inline-formula> for atmospheric neutrons (equivalent to more than 30 million years of natural irradiation) and of <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$7.55\times {10^{12}} {\mathrm {n/cm}}^{2}$ </tex-math></inline-formula> for thermal neutrons. We evaluate the behavior of TPUs executing elementary operations with increasing input sizes (standard convolutions or depthwise convolutions) as well as eight convolutional neural networks (CNNs) configurations (single-shot multibox detection (SSD) MobileNet v2 and SSD MobileDet, trained with COCO dataset, and Inception v4 and ResNet-50, with ILSVRC2012 dataset). We found that, despite the high error rate, most neutron-induced errors only slightly modify the convolution output and do not change the detection or classification of CNNs. By reporting details about the error model, we provide valuable information on how to design the CNNs to avoid neutron-induced events to lead to misdetections or classifications.
Read full abstract