一种可扩展的区块链专用协处理器架构

A scalable architecture of blockchain-specific coprocessor

随着区块链技术的发展,区块链系统对区块链底层的密码学算法等复杂运算的处理速度要求越来越高。双重哈希、默克尔树中哈希函数的连续计算、混合加密算法的组合运算等是区块链系统中经常出现的底层运算。为了处理这些运算,处理器频繁地访问寄存器和存储器,会使外设对总线的竞争加剧,随之引起延时变长和性能降低。为了解决这一问题,提出了一种可扩展的区块链专用协处理器架构:(1)内置的指令缓存和数据缓存使获得数据的路径变短,进而加快了处理速度;同时,外设对总线的竞争也有所缓解。(2)取指路径和取数路径都采取流水线方式,使数据处理可以不间断地工作,也会提升处理速度。(3)哈希计算单元支持双重哈希和多重哈希的计算,避免了计算的中间数据访问存储器和缓存,也使双重哈希和多重哈希的计算速度加快。还对区块链专用协处理器架构扩展到其他密码学算法、图像处理算法、神经网络算法等的可能性做了探讨。

With the development of blockchain technology,blockchain system is increasingly demanding faster processing of the underlying cryptographic algorithms.Double hashing,continuous calculation of hash functions in Merkel trees,and combination operations of hybrid encryption algorithms are the underlying operations that often occur in blockchain systems.In order to handle these operations,the processor frequently accesses registers and memory,which will increase the competition of peripherals for the bus,leading to longer delays and compromised performance.In order to solve this problem,the paper proposes a scalable architecture of blockchain-specific coprocessor:(1)The built-in instruction cache and data cache shorten the path to obtain data,thereby speeding up processing and mitigating competition for the bus from peripherals;(2)Both instruction-fetching path and datafetching path adopt a pipeline method so that the data processing can work without interruption,which improved processing speed;(3)The hash calculation unit supports the calculation of double hashing and multiple hashing,avoiding the access of memory and cache for intermediate data of the calculation and speeding up the calculation of double hashing and multiple hashing.The paper also discusses the possibility of extending the blockchain-specific coprocessor architecture to other cryptographic algorithms,image processing algorithms,and neural network algorithms.