浙江大学 航空航天学院,浙江,杭州,310058
纸质出版:2020
移动端阅览
傅思扬, 陈华, 郁发新. 基于RISC-V的卷积神经网络处理器设计与实现[J]. 微电子学与计算机, 2020, 37(4): 49-54.
FU Si-yang, CHEN Hua, YU Fa-xin. Design and implementation of convolutional neural network processor based on RISC-V instruction set[J]. 2020, 37(4): 49-54.
针对卷积神经网络对于运算资源需求的不断增长,和传统的硬件卷积加速方案在功耗、面积敏感的边缘计算领域难以应用的问题,设计并实现了一个低功耗嵌入式卷积神经网络加速处理器.目标处理器基于RISC-V指令集架构,内核扩展4条自定义神经网络指令,并在硬件层面实现加速处理.该卷积神经网络处理器最大程度的复用了原RISC-V的数据通路和功能模块,减小了额外的功耗和芯片面积等资源开销.目标处理器通过RISC-V官方标准测试集验证,并对MNIST手写数据集进行识别测试,正确率达到97.23%.在TSMC 40nm标准数字工艺下,目标处理器面积仅为0.34 mm ,动态功耗仅为11.1 μw/MHz,与同期处理器相比,面积和功耗方面均具有一定优势.
A low power embedded convolutional neural network acceleration processor is designed and implemented in response to the growing demand for computational resources in convolutional neural networks and the difficulty of applying traditional hardware convolution acceleration schemes in power- and area-sensitive edge computation applications. The target processor based on the RISC-V instruction set architecture
extend four custom neural network instructions
and accelerates the processing on the hardware architecture. The convolutional neural network processor maximizes the reuse of the original RISC-V data path and functional modules
reducing additional resourceoverhead. The target processor is verified by the RISC-V official standard test set
and the MNIST handwritten data set is identified and tested
with a correct rate of 97.23%. The target processor occupies 0.34mm area and consume 11.1μw/MHz dynamic power using TSMC 40nm technology. Compared with the correlation processor
it has certain advantages in area and power consumption.
MOONS B, BANKMAN D, YANG L, et al. BinarEye: An always-on energy-accuracy-scalable binary CNN processor with all memory on chip in 28nm CMOS [C]// 2018 IEEE Custom Integrated Circuits Conference (CICC). San Francisco: IEEE, 2018: 1-4.
阎强.卷积神经网络处理器的设计与实现[D].西安: 西安理工大学, 2017.
PEEMEN M, SETIO A A A, MESMAN B, et al. Memory-centric accelerator design for convolutional neural networks[C].// 2013 IEEE 31st International Conference on Computer Design (ICCD). Asheville, IEEE, 2013: 13-19.
BANKMAN D, YANG L, MOONS B, et al. An always-on 3.8μJ/86% CIFAR-10 mixed-signal binary CNN processor with all memory on chip in 28nm CMOS [C]//2018 IEEE International Solid - State Circuits Conference - (ISSCC). San Francisco: IEEE, 2018: 222-224.
ASANOVIC K, AVIZIENIS R, BACHRACH J, et al. The rocket chip generator [R]. EECS Department, University of California, Berkeley, 2016.
杨一晨, 梁峰, 张国和, 等.一种基于可编程逻辑器件的卷积神经网络协处理器设计[J].西安交通大学学报, 2018, 52(7): 158-164. DOI: 158
吕倩茹, 王彦鹏, 曹壮, 等.面向网络报文转发的RISC-V压缩指令定制[J].计算机工程与科学, 2018, 40(3): 381-387. DOI: 381
杨维科, 贺光辉, 景乃锋.基于Rocket-Chip开源处理器的CNN加速模块的设计及实现[J].微电子学与计算机, 2018, 35(4): 17-21. DOI: 17
雷思磊. RISC-V架构的开源处理器及SOC研究综述[J].单片机与嵌入式系统应用, 2017(2): 56-60. DOI: 56
WATERMAN A, LEE Y, PATTERSON D A, et al. The RISC-V instruction set manual, volume I: Base user-level ISA [J]. EECS Department, 2011, 7(9): 475. DOI: 475
方睿, 刘加贺, 薛志辉, 等.卷积神经网络的FPGA并行加速方案设计[J].计算机工程与应用, 2015, 51(8): 32-36. DOI: 32
马珂.具有卷积神经网络扩展指令的微处理器的设计与实现[D].西安: 西安理工大学, 2018.
GUO K, SUI L, QIU J, et al. Angel-Eye: A Complete Design Flow for Mapping CNN onto Embedded FPGA [J]. IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems, 2017(99):1-1. DOI: 1
0
浏览量
237
下载量
0
CSCD
关联资源
相关文章
相关作者
相关机构
京公网安备11010602201714号