可公开版本—人工智能芯片技术发展与应用 寒武纪

1.

2. 5~6 Cambricon = Cambrian + Silicon

3. AI 2008 2013 2006 1986 1957 1982 1956 AI 1970 BP ARTIFICIAL INTELLIGENCE 1950 1990 Hopfield 2006 1982 1956 AI MACHINE LEARNING DEEP LEARNING

4. CPU GPU FPGA A ASIC

5. MLUGPUCPU- IP VS

6. VR … AR … …

7. PRIME Fused CNN Eyeriss DianNao, 2014 EIE Bit-Pragmatic Stripes Pipelayer RedEye FlexFlow Cnvlutin ScaleDeep DianNao ShiDianNao Cambricon SCNN DaDianNao PuDianNao Cambricon-X TPU 2014 2015 2016 2017 * DianNao from ISCA HPCA ASPLOS MICRO 2010~2017 Inria

8. 5G AI CPU FPGA IoT GPU 5G ASIC + + AI

9. AI .

10. AI Speed Up 20 1 Servers >100 pcs 250 5000 <50 CPU only HPC AI

11. 5G+ h
$PQZSJHIU

$BNCSJDPO

12. 5G+ + / / AR/VR <1 TOPS 1-8 TOPS 4-20 TOPS 20-200 TOPS POPS-EOPS

13.

14. ASIC vs vs vs

15.

16.

17. vs Tianshi Chen, Zidong Du, Ninghui Sun, Jia Wang, Chengyong Wu, Yunji Chen, and Olivier Temam, "DianNao: A SmallFootprint High-Throughput Accelerator for Ubiquitous Machine-Learning," In Proceedings of 19th International Conference on Architectural Support for Programming Languages and Operating Systems (ASPLOS'14), 2014. (Best Paper Award)

18. vs Reorder Buffer Vector Scratchpad Memory Matrix Func. Unit (Matrix DMAs) Matrix Scratchpad Memory IO Interface Vector Func. Unit (Vector DMAs) IO DMA Issue Queue Decode Scalar Register File ry o em M U G A c n Fu r la ca it S n U Fetch . L1 Cache Shaoli Liu, Zidong Du, Jinhua Tao, Dong Han, Tao Luo, Yuan Xie, Yunji Chen, and Tianshi Chen, "Cambricon: An Instruction Set Architecture u, for Neural Networks," In Proceedings of the 43rd ACM/IEEE International Symposium on Computer Architecture (ISCA'16), 2016. (Highest Score in Peer Review)

19. Daofu Liu, Tianshi Chen, Shaoli Liu, Jinhong Zhou, Shengyuan Zhou, Olivier Temam, Xiaobing Feng, Xuehai Zhou, and Yunji Chen, "PuDianNao: A Polyvalent Machine Learning Accelerator," In Proceedings of 20th International Conference on Architectural Support for Programming Languages and Operating Systems (ASPLOS'15), 2015.

20. vs 0 0 index index Shijin Zhang, Zidong Du, Lei Zhang, Huiying Lan, Shaoli Liu, Ling Li, Qi Guo, Tianshi Chen, and Yunji Chen, "Cambricon-X: An Accelerator for Sparse Neural Networks," In Proceedings of 49th IEEE/ACM International Symposium on Microarchitecture (MICRO'16), 2016.

21. / / & / / & / / IP IP SoC / +

22. IP 1H16 IP 1H8 IP 1A 1M IP

23. MLU100 MLU100 2018 5 3 MLU200 2019

24. Cambricon NeuWare • TensorFlow • Caffe • MXNet • CN-ML • CN-CC • CN-Gen • CN-GDB • CN-Dmp • CN-Cmp • CNPERF • CN-Adv • CNMON

25.

26.

27.

28.

29.

30.