A Test Methodology for Neural Computing Unit
Minho Cheong, Ingeol Lee, Sungho Kang · 2018
As convolutional neural networks (CNN) has been widely employed in deep learning applications, the accelerator for CNN has been proposed. Neural computing unit (NCU), which is an accelerator for CNN, includes thousands of identical cores named multiplier and accumulate (MAC), so testing NCU with the conventional methods are inefficient. This paper proposes a novel method to test NCU by applying test patterns for a MAC to all MACs in NCU. The experimental results indicate that the new method reduces test time to 1.38% and test data volume to 0.03%.