基于GPU的密码分析技术实现和基于硬件实现的S盒构造

发布时间：2018-05-18 20:38

本文选题：CUDA + SPECK　；参考：《山东大学》2016年硕士论文

【摘要】：本文主要研究了两个方面,一是CUDA在密码分析中的加速作用,是利用塔域技术构造较小硬件面积的S盒。CUDA是英伟达公司于2006年11月推出的应用在GPU上的通用并行开发平台,主要利用GPU并行计算的特点,用来开发适合运行在GPU上的程序[1]。利用CUDA编程能够大幅加快加密算法的速度,对有些密码分析技术来说也同样适用。本文首先根据GPU运算的特点,提出了适合使用CUDA加速的算法应该满足的一些特点,并且研究了哪类加密算法更适合用于CUDA加速,哪种密码分析方法运用CUDA会有更高的效率。而后本文确定将CUDA优化技术运用到一个具体的分组密码算法——SPECK [2]算法。SPECK算法是由美国国家安全局(NSA)设计,在2013年发布的具有Feistel结构的轻量级分组加密算法,专门适用于软件端实现。在16MHz的Atmel ATmega128 8-bit微控制器上,分组长度为64比特,初始密钥长度为96比特的SPECK算法加密吞吐量可以达到888Kbps。本文在Tesla C2070 GPU下,适用CUDA技术,加密分组长度为64比特,初始密钥长度为96比特的SPECK加密算法,加密吞吐量可以达到298Gbps。GPU除了可以加快算法的加解密速度,还可以利用CUDA技术,用来寻找加密算法的平衡性质。同样对SPECK进行实验,在Tesla C2070 GPU下的速度是在Inter(R)Xeon(R)CPU E5-2620 @2.00GHz单核下的147倍。这个结果说明了CUDA对满足某些密码分析方法能够起到很好的加速作用。在AES[3]的硬件实现上,其s盒可以通过塔域技术实现,达到减少硬件资源消耗的目的。AES的S盒是由一个仿射变换和一个在有限域GF(28)上的求逆运算构成。塔域方法是通过一个同构映射及其逆映射,将有限域GF(28)上的求逆运算转换为塔域GF((24)2)或者GF(((22)2)2)上的求逆运算。之前利用已有S盒的塔域方法构建的S盒比AES的S盒硬件实现面积小,本文通过塔域方法并巧妙利用了仿射变换和塔域上的求逆运算之间的关系构建了新的S盒。新构建的S盒由两个仿射变换和一个在塔域GF(((22)2)2)上的求逆运算构成,其中巧妙选择的仿射变换可以减少S盒和逆S盒中塔域求逆运算的硬件面积,可以进一步的减少总的硬件实现面积。通过对新的S盒进行密码安全指标分析,发现新的S盒具有可靠的安全性质。
[Abstract]:This paper mainly studies two aspects. One is the accelerating function of CUDA in cryptographic analysis. The S box. CUDA, which uses tower domain technology to construct small hardware area, is a general parallel development platform which was introduced by Nvidia in November 2006 and applied on GPU. Mainly using the characteristics of GPU parallel computing, it is used to develop programs suitable for running on GPU [1]. The speed of encryption algorithm can be greatly accelerated by using CUDA programming, and it is also applicable to some cryptographic analysis techniques. According to the characteristics of GPU operation, this paper first puts forward some characteristics that should be satisfied by the algorithm suitable for using CUDA acceleration, and studies which encryption algorithm is more suitable for CUDA acceleration and which cryptographic analysis method has higher efficiency using CUDA. Then this paper determines that the CUDA optimization technology is applied to a specific block cipher algorithm.SPECK algorithm is a lightweight block encryption algorithm with Feistel structure, which was designed by National Security Agency (NSA) and published in 2013. It is suitable for software implementation. On the Atmel ATmega128 8-bit microcontroller of 16MHz, the encryption throughput of the SPECK algorithm with 64 bits of packet length and 96 bits of initial key length can reach 888Kbps. In this paper, under Tesla C2070 GPU, the SPECK encryption algorithm with 64 bits of block length and 96 bits of initial key length is suitable for CUDA technology. The throughput of 298Gbps.GPU can not only speed up the encryption and decryption speed of the algorithm, but also make use of CUDA technology. It is used to find the balance of encryption algorithm. SPECK is also tested at a speed of 147x at Tesla C2070 GPU than at 2.00GHz Inter(R)Xeon(R)CPU E5-2620. The results show that CUDA can accelerate the performance of some cryptographic analysis methods. In the hardware implementation of AES [3], its s-box can be realized by tower technology, and the purpose of reducing the consumption of hardware resources. The S-box of AES is composed of an affine transformation and an inverse operation on the finite field GFF28). The tower domain method is to convert the inversion operation on the finite field GFN28) to the inverse operation on the tower domain GFN / 24 / 2 / 2 / 2 / 2 by an isomorphism map and its inverse map. The S-box constructed by using the existing S-box tower domain method is smaller than that of AES's S-box hardware. In this paper, a new S-box is constructed by using the relationship between affine transformation and inverse operation in the tower domain by using the tower domain method and skillfully using the relation between the affine transformation and the inverse operation in the tower domain. The newly constructed S-box is composed of two affine transformations and one inverse operation on the tower domain GF22 / 2), in which the skillfully selected affine transformation can reduce the hardware area of the inverse operation in the S-box and the inverse S-box. Can further reduce the total hardware implementation area. It is found that the new S-box has reliable security property by analyzing the cryptographic security index of the new S-box.
【学位授予单位】：山东大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TN918.2

【相似文献】