Tolerating Defects in Low-Power Neural Network Accelerators Via Retraining-Free Weight Approximation
Date
2021-09-23
Journal Title
Journal ISSN
Volume Title
Publisher
ACM Transactions on Embedded Computing Systems
Abstract
Hardware accelerators are essential to the accommodation of ever-increasing Deep Neural Network (DNN) workloads on the resource-constrained embedded devices. While accelerators facilitate fast and energy-efficient DNN operations, their accuracy is threatened by faults in their on-chip and off-chip memories, where millions of DNN weights are held. The use of emerging Non-Volatile Memories (NVM) further exposes DNN accelerators to a non-negligible rate of permanent defects due to immature fabrication, limited endurance, and aging. To tolerate defects in NVM-based DNN accelerators, previous work either requires extra redundancy in hardware or performs defect-aware retraining, imposing significant overhead. In comparison, this paper proposes a set of algorithms that exploit the flexibility in setting the fault-free bits in weight memory to effectively approximate weight values, so as to mitigate defect-induced accuracy drop. These algorithms can be applied as a one-step solution when loading the weights to embedded devices. They only require trivial hardware support and impose negligible run-time overhead. Experiments on popular DNN models show that the proposed techniques successfully boost inference accuracy even in the face of elevated defect rates in the weight memory.
Description
This article was originally published in ACM Transactions on Embedded Computing Systems. The version of record is available at: https://doi.org/10.1145/3477016
Keywords
Computer systems organization, Reliability, Neural networks, Embedded software, Hardware, Error detection, error correction, Neural network accelerator, defect tolerance, memory faults, approximation
Citation
Fateme S. Hosseini, Fanruo Meng, Chengmo Yang, Wujie Wen, and Rosario Cammarota. 2021. Tolerating Defects in Low-Power Neural Network Accelerators Via Retraining-Free Weight Approximation. ACM Trans. Embedd. Comput. Syst. 20, 5s, Article 85 (September 2021), 21 pages. https://doi.org/10.1145/3477016