基于木马的方式增强RRAM计算系统的安全性

基于新型存储器件RRAM的计算系统,因为能够在内存中执行矩阵点乘向量运算而受到广泛的关注.然而RRAM计算系统的安全性却未受到足够的重视.攻击者通过访问未授权的RRAM计算系统,进而以黑盒攻击的方式来获取存储于RRAM计算系统中的神经网络模型.本文的目标是阻止此种攻击.本文提出的防御方法是基于良性木马,也就是当RRAM计算系统未授权时,系统中的木马极容易被激活,进而影响系统的输出预测准确性,从而保证系统不能够正常运行;当RRAM计算系统被授权时,系统中的木马极难被误激活,从而系统能够正常运行.我们通过实验表明,该方法能够使得未授权的RRAM计算系统的输出预测准确性降低至15%以下,并且硬件开销小于系统中RRAM硬件的4.5%.     

Scan-Based Attack on Stream Ciphers: A Case Study on eSTREAM Finalists

Scan-based design for test （DFT） is a powerful and the most popular testing technique. However, while scan-based DFT improves test efficiency, it also leaves a side channel to the privacy information stored in the chip. This paper investigates the side channel and proposes a simple but powerful scan-based attack that can reveal the key and/or state stored in the chips that implement the state-of-the-art stream ciphers with less than 85 scan-out vectors.