Long Live TIME: Improving Lifetime and Security for NVM-Based Training-in-Memory Systems

Nonvolatile memory (NVM)-based training-in-memory (TIME) systems have emerged that can process the neural network (NN) training in an energy-efficient manner. However, the endurance of NVM cells is disappointing, rendering concerns about the lifetime of TIME systems, because the weights of NN models...

Descrizione completa

Dettagli Bibliografici
Autori principali: Lin, Yujun, Han, Song
Altri autori: Massachusetts Institute of Technology. Department of Electrical Engineering and Computer Science
Natura: Articolo
Pubblicazione: IEEE 2021
Accesso online:https://hdl.handle.net/1721.1/129440
_version_ 1826188022397272064
author Lin, Yujun
Han, Song
author2 Massachusetts Institute of Technology. Department of Electrical Engineering and Computer Science
author_facet Massachusetts Institute of Technology. Department of Electrical Engineering and Computer Science
Lin, Yujun
Han, Song
author_sort Lin, Yujun
collection MIT
description Nonvolatile memory (NVM)-based training-in-memory (TIME) systems have emerged that can process the neural network (NN) training in an energy-efficient manner. However, the endurance of NVM cells is disappointing, rendering concerns about the lifetime of TIME systems, because the weights of NN models always need to be updated for thousands to millions of times during training. Gradient sparsification (GS) can alleviate this problem by preserving only a small portion of the gradients to update the weights. However, conventional GS will introduce nonuniform writes on different cells across the whole NVM crossbars, which significantly reduces the excepted available lifetime. Moreover, an adversary can easily launch malicious training tasks to exactly wear-out the target cells and fast break down the system. In this article, we propose an efficient and effective framework, referred as SGS-ARS, to improve the lifetime and security of TIME systems. The framework mainly contains a structured GS (SGS) scheme for reducing the write frequency, and an aging-aware row swapping (ARS) scheme to make the writes uniform. Meanwhile, we show that the back-propagation mechanism allows the attacker to localize and update fixed memory locations and wear them out. Therefore, we introduce Random-ARS and Refresh techniques to thwart adversarial training attacks, preventing the systems from being fast broken in an extremely short time. Our experiments show that when TIME is programmed to train ResNet-50 on ImageNet dataset, 356× lifetime extension can be achieved without sacrificing the accuracy much or incurring much hardware overhead. Under the adversarial environment, the available lifetime of TIME systems can still be improved by 84× .
first_indexed 2024-09-23T07:53:24Z
format Article
id mit-1721.1/129440
institution Massachusetts Institute of Technology
last_indexed 2024-09-23T07:53:24Z
publishDate 2021
publisher IEEE
record_format dspace
spelling mit-1721.1/1294402022-09-30T00:49:04Z Long Live TIME: Improving Lifetime and Security for NVM-Based Training-in-Memory Systems Lin, Yujun Han, Song Massachusetts Institute of Technology. Department of Electrical Engineering and Computer Science Nonvolatile memory (NVM)-based training-in-memory (TIME) systems have emerged that can process the neural network (NN) training in an energy-efficient manner. However, the endurance of NVM cells is disappointing, rendering concerns about the lifetime of TIME systems, because the weights of NN models always need to be updated for thousands to millions of times during training. Gradient sparsification (GS) can alleviate this problem by preserving only a small portion of the gradients to update the weights. However, conventional GS will introduce nonuniform writes on different cells across the whole NVM crossbars, which significantly reduces the excepted available lifetime. Moreover, an adversary can easily launch malicious training tasks to exactly wear-out the target cells and fast break down the system. In this article, we propose an efficient and effective framework, referred as SGS-ARS, to improve the lifetime and security of TIME systems. The framework mainly contains a structured GS (SGS) scheme for reducing the write frequency, and an aging-aware row swapping (ARS) scheme to make the writes uniform. Meanwhile, we show that the back-propagation mechanism allows the attacker to localize and update fixed memory locations and wear them out. Therefore, we introduce Random-ARS and Refresh techniques to thwart adversarial training attacks, preventing the systems from being fast broken in an extremely short time. Our experiments show that when TIME is programmed to train ResNet-50 on ImageNet dataset, 356× lifetime extension can be achieved without sacrificing the accuracy much or incurring much hardware overhead. Under the adversarial environment, the available lifetime of TIME systems can still be improved by 84× . National Key Basic Research Program of China (Grant 2017YFA0207600) National Natural Science Foundation of China (Grants 61832007, 61622403, 61621091) 2021-01-19T15:30:41Z 2021-01-19T15:30:41Z 2020-12 2019-12 Article http://purl.org/eprint/type/JournalArticle 0278-0070 https://hdl.handle.net/1721.1/129440 Cai, Yi et al. “Long Live TIME: Improving Lifetime and Security for NVM-Based Training-in-Memory Systems.” IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems, 39, 12 (December 2020): 4707 - 4720 © 2020 The Author(s) 10.1109/TCAD.2020.2977079 IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems Creative Commons Attribution-Noncommercial-Share Alike http://creativecommons.org/licenses/by-nc-sa/4.0/ application/pdf IEEE other univ website
spellingShingle Lin, Yujun
Han, Song
Long Live TIME: Improving Lifetime and Security for NVM-Based Training-in-Memory Systems
title Long Live TIME: Improving Lifetime and Security for NVM-Based Training-in-Memory Systems
title_full Long Live TIME: Improving Lifetime and Security for NVM-Based Training-in-Memory Systems
title_fullStr Long Live TIME: Improving Lifetime and Security for NVM-Based Training-in-Memory Systems
title_full_unstemmed Long Live TIME: Improving Lifetime and Security for NVM-Based Training-in-Memory Systems
title_short Long Live TIME: Improving Lifetime and Security for NVM-Based Training-in-Memory Systems
title_sort long live time improving lifetime and security for nvm based training in memory systems
url https://hdl.handle.net/1721.1/129440
work_keys_str_mv AT linyujun longlivetimeimprovinglifetimeandsecurityfornvmbasedtraininginmemorysystems
AT hansong longlivetimeimprovinglifetimeandsecurityfornvmbasedtraininginmemorysystems