具有自主产权的安诚嵌入式处理器上支持AES及GF(2^n)运算的指令扩展结构研究

In 2010, Intel launched two series of 32nm processors that support Advanced Encryption Standard New Instructions (AES-NI): Core and Xeon. This is the first time that Intel has extended the x86 instruction set for the purpose of enhancing cryptographic computations. This project is based on our leading theoretical results in GF(2^n) cryptographic computations, and will be carried out on the Ancheng microprocessor developed by the No.705 Institute of the China Shipbuilding Industry Corporation (CSIC), which is a RISC embedded microprocessor with self-owned intellectual property. We will focus on designing and implementing two sets of instruction extensions: 1、GF(2^n) arithmetic operations, especially GF(2^n) multiplication and GF(2) matrix operations； 2、AES specification, which will be based on a new isomorphic representation of GF(2^8). Similar to Intel's PCLMULQDQ（Carry-less Multiplication Instruction）introduced together with AES-NI, the first set of instruction extensions in this project is also aimed to increasing the performance of software multiplication over GF(2^n). But we will follow a totally different design approach: other than the straightforward polynomial multiplication adopted by Intel, we will adopt the matrix-vector approach jointly developed by the applicant and his coauthor. This matrix-vector approach is currently the best VLSI method to perform GF(2)[x] subquadratic multiplication for large operands, and our earlier theoretical analysis proved that its asymptotic time and space complexities are much better than the polynomial method. The second set of AES instructions extensions will be based on a new isomorphic representation of GF(2^8), which is a highlight of this project.

Intel新发布的"AES New Instructions"是其首次对x86指令集进行密码方面的扩展。本项目基于我们在GF(2^n)密码芯片算法设计领域的国际领先理论成果，将对GF(2^n)域运算算法进行深入研究，并将理论成果应用到昆明七零五所开发的具有自主产权的安诚嵌入式处理器上，主要研究内容包括： 1、研究并设计实现支持GF(2^n)乘法和GF(2)上矩阵计算的指令扩展模块； 2、研究GF(2^8)同构表示，设计支持AES密码算法的指令扩展模块。 与Intel新发布的GF(2)[x]无进位乘法指令相同，前一项研究的目的亦是为了提高基于GF（2^n）的密码系统的性能。但本项目并未沿袭Intel所采用的多项式直接实现方案，而是遴选矩阵方案，我们最近的理论分析表明该方案在门数量和门延时两方面均优于Intel的直接实现方案。而基于移位多项式基的GF(2^8)同构表示则是后一项研究的创新点。

目前主流对称密钥密码体制和公钥密码体制的基本操作均基于有限域GF(p^n)。在硬件实现方面，有限域GF(2^n)乘法器是基于GF(2^n)的密码芯片的核心模块。作为并行GF(2^n)乘法器领域大多数国际纪录的唯一保持者，近20年来我们在该领域的开创性工作得到了国际同行的认可，在项目执行期间，我们应该期刊Finite Fields and Their Applications之“20周年”特辑之邀撰写乘法器方向综述性文章，该期刊为有限域领域唯一专刊，其编委包括R. Lidl，H. Niederreiter和N. Koblitz教授等。2015年， 我们将亚二次设计方法TMVP法由三项式推广到了五项式，这使得该技术可以运用于所有的实用GF(2^n)域，该成果发表在IEEE Transactions on Computers，由项目负责人及其所指导的硕士生（第一作者）共同完成。2016年，我们打破了保持近30年的最快乘法器的门数量复杂性纪录，这是目前最快混合二次乘法器的国际最好成果。 该结果发表在IEEE Transactions on Computers，由项目负责人独立完成，是该刊创刊65年来中国大陆第8篇独立作者论文。我们将这些设计方案在安诚处理器上进行了模拟验证。