首页 > AI前沿 > Scalable fault vulnerability analysis for ReRAM-based process-in-memory CNN accelerators

Scalable fault vulnerability analysis for ReRAM-based process-in-memory CNN accelerators

arXiv机器学习 2026-11-19 23:55 6 阅读 查看原文
用户现在需要把给定的英文新闻正文排版成语义化的HTML片段,按照要求,每个自然段落用p,重要的可以加strong,要保留所有内容,不能增删改。首先,先通读正文,有没有需要小节的?看内容是一段连贯的,所以直接处理段落。首先第一句里的重点?不对,要保留原文,然后重要的句子可以用strong?不对,先理清楚结构,原文是一段?不,不对,看正文的内容,整体是一个段落?不对,再看:首先介绍ReRAM的PIM加速器,然后说问题,然后说现有故障注入的问题,然后提出本文的方法,最后对比效果。哦,不对,用户给的正文是一个连贯的段落吗?不对,等下,看内容,是不是可以分?不对,要求是每个自然段落用p包裹,原文的自然段落就是一段?不对,再看原文的内容,哦,不对,原文是一整段?不对,仔细看: 首先开头:Resistive random-access memory...leading to system unreliability in safety-critical applications. 然后下一句:Fault injection is a commonly...especially for large-scale CNNs. 然后下一句:In this paper...enabling the design of fault-tolerant accelerators for edge and safety-critical applications. 然后下一句:The comparison between...confirms the effectiveness of our proposed method. 然后下一句:Unlike exhaustive fault injection...within only a few minutes or hours. 哦,不对,原文是连续的一段?还是本来就是多个自然段?不对,用户给的正文看起来是一整段?不对,等下,用户给的正文是连续的,没有分段?不对,再看用户的输入:“正文如下:”后面的内容是连续的,没有空行,所以是不是一个自然段落?不对,不对,要按原文的自然段落?哦,原文的自然段落如果是一个的话,那用p包裹?不对,不对,再仔细看,有没有需要用h2或者h3的?用户说长文用h2/h3划分小节,这里的内容是不是可以划分?比如,背景与问题、现有方法局限、本文方法、实验验证?不对,不对,规则是不得增删改内容,所以不能加小节标题,因为原文没有,所以不能自己加h2/h3,因为没有对应的原文内容,加了就增了内容了。哦对,规则4是保留原文所有信息,不得增删改内容,所以不能自己加标题,因为原文没有标题。那接下来,处理段落:原文是一个自然段吗?不对,等下,用户给的正文有没有分段?看: 用户给的正文: Resistive random-access memory (ReRAM)-based processing-in-memory (PIM) accelerators offer a highly efficient platform for parallel matrix-vector multiplications (VMM) in neural networks, leveraging analog computation, high density, low leakage, and non-volatility. However, their susceptibility to permanent faults, such as stuck-at-high (\textit{SaH}) and stuck-at-low (\textit{SaL}) resistance states, permanently corrupts the convolutional neural networks (CNNs) weights mapped onto ReRAM cell conductance values, thereby degrading inference accuracy and leading to system unreliability in safety-critical applications. Fault injection is a commonly used technique to assess the vulnerability of neural networks by systematically injecting faults into individual parameters and measuring the resulting accuracy degradation. Although this method is accurate, it is prohibitively time-consuming and computationally expensive, especially for large-scale CNNs. In this paper, we propose a fast, scalable, and analytical framework to quantify the fault vulnerability of convolutional neural networks deployed on ReRAM-based PIM accelerators. This method quantifies vulnerability at different granularities, including weights, kernels, and layers, without the need for exhaustive simulation, enabling the design of fault-tolerant accelerators for edge and safety-critical applications. The comparison between the results of the proposed vulnerability score (VS) metrics with performance metric of CNNs captured with fault injection simulations confirms the effectiveness of our proposed method. Unlike exhaustive fault injection, which requires days or weeks, whereas the proposed approach achieves a speedup of about $10^4\times-10^5\times$, enabling fast and accurate vulnerability analysis for large-scale CNNs within only a few minutes or hours. 哦,这确实是一整段?不对,但是有没有需要用strong的地方?比如重要的句子?比如本文提出的方法相关的?不对,规则是重要句子可用strong,但是不能改内容。哦,不对,还要注意,原文里的LaTeX格式的内容,比如\textit{SaH}、$10^4\times-10^5\times$,要原样保留对吧?对,因为不能增删改内容,所以这些符号都保留。那现在,是不是整个内容用一个p包裹?不对,等下,有没有可能原文的自然段落就是多个?不对,用户给的正文没有分段,所以是一个自然段?不对,不对,再想,是不是我理解错了?哦,不对,可能用户的正文虽然写的是连续的,但其实可以按语义?不对,规则说每个自然段落用p包裹,自然段落是原文的分段,原文没有分段,所以就是一个p?不对,等下,再看规则2:每个自然段落用

包裹,长文用

/

划分小节。但是这里原文没有小节标题,所以不能加,因为加了就是增加内容了,违反规则4。那接下来,有没有需要用其他标签的?比如没有列表,没有引用,没有代码,所以不需要ul、blockquote、pre这些。然后重要的句子可以用strong?比如哪些是重要的?比如“In this paper, we propose a fast, scalable, and analytical framework to quantify the fault vulnerability of convolutional neural networks deployed on ReRAM-based PIM accelerators.”这句是核心贡献,可以加strong?还有后面的“Unlike exhaustive fault injection, which requires days or weeks, whereas the proposed approach achieves a speedup of about $10^4\times-10^5\times$, enabling fast and accurate vulnerability analysis for large-scale CNNs within only a few minutes or hours.”这句是效果,也可以?不对,但是要注意,不能改内容,只是加strong标签包裹。哦对,strong是用来强调,不改变内容。那现在,整理一下: 首先,整个原文是一个自然段落吗?不对,等下,再仔细看,有没有可能我漏看了分段?不,用户给的正文是连续的,没有换行,所以是一个自然段,用p包裹。然后把重要的句子用strong包起来?等下,再确认规则:重要句子可用strong,所以可以加,但是不能改内容。那比如:

Resistive random-access memory (ReRAM)-based processing-in-memory (PIM) accelerators offer a highly efficient platform for parallel matrix-vector multiplications (VMM) in neural networks, leveraging analog computation, high density, low leakage, and non-volatility. However, their susceptibility to permanent faults, such as stuck-at-high (\textit{SaH}) and stuck-at-low (\textit{SaL}) resistance states, permanently corrupts the convolutional neural networks (CNNs) weights mapped onto ReRAM cell conductance values, thereby degrading inference accuracy and leading to system unreliability in safety-critical applications. Fault injection is a commonly used technique to assess the vulnerability of neural networks by systematically injecting faults into individual parameters and measuring the resulting accuracy degradation. Although this method is accurate, it is prohibitively time-consuming and computationally expensive, especially for large-scale CNNs. In this paper, we propose a fast, scalable, and analytical framework to quantify the fault vulnerability of convolutional neural networks deployed on ReRAM-based PIM accelerators. This method quantifies vulnerability at different granularities, including weights, kernels, and layers, without the need for exhaustive simulation, enabling the design of fault-tolerant accelerators for edge and safety-critical applications. The comparison between the results of the proposed vulnerability score (VS) metrics with performance metric of CNNs captured with fault injection simulations confirms the effectiveness of our proposed method. Unlike exhaustive fault injection, which requires days or weeks, whereas the proposed approach achieves a speedup of about $10^4\times-10^5\times$, enabling fast and accurate vulnerability analysis for large-scale CNNs within only a few minutes or hours.

不对,等下,有没有问题?比如,原文里的内容有没有被修改?没有,只是加了strong标签,内容都一样。那有没有需要用h2/h3的?原文没有小节,所以不需要,因为加了的话就是增加了原文没有的内容,违反规则4。对,因为规则说不得增删改内容,所以不能自己加标题。那有没有其他需要处理的?比如LaTeX的内容,原样保留,对吧?对,因为不能改内容。那现在,是不是这样就可以了?等下,再检查