计算体系架构研究综述与思考

Review and thoughts on the development of computing architec-ture

随着摩尔定律(Moore's law)与迪纳德(Dennard)缩放定律逐步走向终结,依靠集成电路制程工艺的进步提升计算系统性能与效能越来越困难,计算体系架构的演进成为了未来计算系统发展的重要技术途径.本文首先从应用适应性、计算驱动方式、系统重心变化、计算核心构成,以及计算逻辑使用等不同的角度回顾了体系架构的发展历程,总结了不同体系架构的优缺点;然后着重分析了在人工智能、大数据等应用飞速发展的条件下未来计算系统的能力需求特征;最后提出了软件定义计算体系架构,并梳理了其重点研究内容与关键技术,为未来计算体系架构的发展提供了一条可行的技术途径.

With the end of Moore’s law and Dennard’s scaling law,it is becoming increasingly difficult to improve the performance and efficiency of computing systems by relying on the progress of the integrated circuit(IC)manufacturing process.The evolution of computing architecture has become an important technical approach for the development of computing systems in the future.In this article,the evolution of the architecture is first reviewed from different perspectives,including the application adaptability,computation-driven mode,changes in the center of the computing system,the composition of the computing cores,and computing logic usage.The advantages and disadvantages of different architectures are summarized,and then the ability to computing systems in the future under the conditions of rapid development of manual intelligence and big data is analyzed.Finally,a software-defined computing architecture is proposed,and its key research contents and key technologies are summarized,which provides a feasible technical approach for the future development of computing architectures.