深度神经网络FPGA设计进展、实现与展望

Development,Implementation and Prospect of FPGA-Based Deep Neural Networks

近年来,随着人工智能与大数据技术的发展,深度神经网络在语音识别、自然语言处理、图像理解、视频分析等应用领域取得了突破性进展.深度神经网络的模型层数多、参数量大且计算复杂,对硬件的计算能力、内存带宽及数据存储等有较高的要求.FPGA作为一种可编程逻辑器件,具有可编程、高性能、低能耗、高稳定、可并行和安全性的特点.FPGA与深度神经网络的结合成为推动人工智能产业应用的研究热点.本文首先简述了人工神经网络坎坷的七十年发展历程与目前主流的深度神经网络模型,并介绍了支持深度神经网络发展与应用的主流硬件;接下来,在介绍FPGA的发展历程、开发方式、开发流程及型号选取的基础上,从六个方向分析了FPGA与深度神经网络结合的产业应用研究热点;然后,基于FPGA的硬件结构与深度神经网络的模型特点,总结了基于FPGA的深度神经网络的设计思路、优化方向和学习策略;接下来,归纳了FPGA型号选择以及相关研究的评价指标与度量分析原则;最后,我们总结了影响FPGA应用于深度神经网络的五个主要因素并进行了概要分析.

In recent years,with the development of artificial intelligence and big data technology,the deep neural network has made a breakthrough in many fields,such as speech recognition,natural language processing,image understanding,video analysis and so on.However,along with the increasing of neural network layers,a large number of parameters and complex calculations aggravate the requirements of hardware in computing power,memory bandwidth and data storage.FPGA,a programmable logic device,is of programmability,high performance,low energy consumption,high stability,parallelizability and security.The combination of FPGA and the deep neural network becomes a research hotspot to promote the industrial application of artificial intelligence.This paper introduces the development of the deep neural network in the past 70 years,the mainstream deep learning model and the fundamental hardware that support the development and application of deep neural network.Secondly,the research hotspots of FPGA combined with the deep neural network on the industrial applications are analyzed in six respects on the basis of introducing the development process,development mode,development process and type specification of FPGA.And then,the design idea,optimization direction and learning strategy of deep neural network based on FPGA are summarized according to the hardware structure of FPGA and the model characteristics of deep neural network.In addition,the model selection of FPGA together with the evaluation index and measurement analysis principle of related works are listed.Finally,we summarized the five main factors that affect the application of FPGA to deep neural networks and conducted a summary analysis.