DNN加速器技术发展及航空计算系统应用展望

Development of DNN Accelerator and its Application in Avionics System

新一代航空计算系统将以高性能智能数据处理为核心,能够支撑智能化的图像/雷达目标识别、大数据分析、指控决策、故障自诊断等多种应用场景。深度神经网络硬件加速器是面向人工智能领域专用的硬件加速平台,能够为深度卷积神经网络、循环神经网络等人工智能算法提供一个灵活可配置的运行环境,与通用计算硬件相比将获得高性能、低时延、低功耗等优势,其核心技术主要包括智能计算单元设计、并行计算加速结构、存储结构优化等。重点分析国内外深度神经网络硬件加速器技术发展情况,总结关键技术解决方案,探讨其在航空计算系统的应用场景及未来需要重点突破的技术方向,为新一代智能航空计算系统研制提供理论支撑。

The high performance intelligent data processing will be the main concern in the design of next generation airborne computing systems,especially for various intelligent application scenarios,such as image/radar target recognition,big data analysis,control and decision,and fault self diagnosis.Deep neural network(DNN)hardware accelerator is a domain specific accelerating platform for artificial intelligence(AI).It can provide a flexible and reconfigurable operating environment for AI algorithms,including deep convolutional neural network and recurrent neural network,with the advantages of high performance,low latency,and low power consumption compared with the general purpose computing hardware.The key technologies of DNN accelerator mainly include:design of intelligent computing elements,parallel computing architecture of accelerator and optimization of memory structure.This article gives an overview on the development of DNN accelerators,concludes the solutions of key technologies,and discusses the probable application scenarios of DNN accelerator in avionics system.This article can provide some technique support for the design of next generation intelligent airborne computing systems.