AI 2019 22019.6 iResearch I AI Gartner 12 AI AI 2022 500 AI PPACR AI AI AI AI GPU+CUDA AI GPU FPGA ASIC SMS3AI 1AI 2AI 3AI 4 542019.6 iResearch I CPU CPU AI AI AI 1 AI GPUFPGAASICDSP AI Artificial Intelligence AI AI 2 AI 3 AI AI DL Deep Learning Deep Learning Ian Goodfellow Yoshua Bengio Aaron Courville ControlALU ALUALU ALUCacheDRAM DRAMALU AI CPU GPU FPGA/ASIC 52019.6 iResearch IAI AI AI 80/GPUFPGAASIC AI ASIC AI AI PPA POWER PERFORMANCE AREA Intel 0.00181015.745Intel-XEON Intel/Altera-Stratix 10SoC Nvidia-TESLA V100 Google-TPUv2 Performance TFLOPs CPU FPGA GPU ASIC CPU FPGA ASIC FPGA ASIC CPU GPU PPA 62019.6 iResearch IAI AI CPU CPU AI CPU GPU CPU AI CPU AI AI CPU AI PCIE CPU AI IP SoC AI PCB PCIE CPU SoC CPU CPU GPUXPU DSPModem Camera ISP4K Video Hifi AudioLPDDR 4X UFSSensor Processor Security EngineSoC72019.6 iResearch IAI AI AI AI/+SoC+IP DSP/GPU/ASIC GPU/FPGA/ASIC GPU/FPGA/ASIC/AI 82019.6 iResearch I20.2172.13.471.915.52108.192017 2022e AI AI AI AI AI 5 10 1 AI AI/2 2017 AI 39.1 AI 20.2 3.4 AI 15.5 2022 AI 352.17 CAGR55%AI 172.1 CAGR 54%71.9 CAGR 84%AI 108.2 CAGR 48%AI AI 2017-2022 AI CAGR:53.5%CAGR:84.1%CAGR:47.5%9AI 1AI 2AI 3AI 4 5102019.6 iResearch I IT 5 IaaS PaaS SaaS 1.IAAS-2.PAAS-3.SAAS-IT IT IDC IT IBM IT O/S IaaS O/S PaaS O/S SaaS O/S 112019.6 iResearch I 2019.6 iResearch I889.7 952.5 967.2 988.7 1009.1 1107.9 1110.3 1145.1 18%7%2%2%2%10%0%3%2010 2011 2012 2013 2014 2015 2016 20172010-2017%490046006700101001243168225494756-6%46%51%270%298%296%2014 2015 2016 20172014-2017 Amazon&Amazon Amazon%IT 24 54.1%5%2017 IT 2.4 IT 9.92%GDP AMAZON Wind 122019.6 iResearch I AI AI PaaS AI AI AI AWS AI Infrastructure AI Frameworks AI Platforms AI ServicesDL-aaS AI-aaS 132019.6 iResearch I AI$/h 0.056.45GPU 1.2634.27 0.136.45GPU 1.2634.27 0.076.45GPU 1.2634.27 11 Web PaaS SaaS AI AI AI CPU$DGX-2 16*Tesla V1002*intel Xeon Platinum8168$399,000DGX-1(Volta 8*Tesla V1002*intel Xeon E5-2698 v4$149,000DGX-1(Pascal)8*Tesla P1002*intel Xeon E5-2698 v3$129,000 AI AWS AI V100 GPU 10+P4 GPU 2+FPGA 45000 AI142019.6 iResearch I AI GPGPU+CUDA AI SDK 7 AI TESLA GPU CPU GPU 1/8 1/15 1/8 GPU CUDA SDK CUDA GPU+CUDA CUDA-Datacenter AI purpose GPUsLanguage Support TESLA P4AI-INFERENCE TESLA P40AI-INFERENCE/GRAPHICS TESLA V100AI-TRAINING/ACCELERATERDeep Learning Framework Support Deep Learning Math/Linear Algebra Siganl/Imge/Video Parallel AlgorithmNVgraphcuBLAS CUDA Math cuRAND cuFFT cuFFT152019.6 iResearch I 2019.6 iResearch I6.7 13.7 35.3 62.5 87.5 0.0 3.4 18.3 57.0 105.9 6.3 14.8 35.7 66.0 98.8 0.0 0.8 5.3 19.7 37.0 2016 2017 2018e 2019e 2020e2016-2020 AI AI AI AI AI AI AI GPU GPU AI AI GPU AI V100 10 DGX AI AI GPU GPU AI IDC Gartner IDC 90.0%90.0%80.0%75.0%70.0%65.0%60.0%60.0%50.0%40.0%35.0%30.0%30.0%2016 2017 2018e 2019e 2020e 2021e 2022e2016-2022 AI GPU GPU%GPU%162019.6 iResearch I AI AI ASIC AI GPU CPU GPU GPU GPU Google TPU MLU AI XPU forCloud SRAM Cache AI MLNNs 172019.6 iResearch I/1 2 3/4 5/SaaS PaaS IaaS EC-SaaS EC-PaaS EC-IaaS*vFWvLB/182019.6 iResearch I AI PPACR AI IP in SoC/1 2 3 4 1-2-3 AR 4 1 2 1 2 3 Slam 4 Performance1TOPs 18TOPs 420TOPs 204000TOPs L3L5 Power 2,0005,000mAh 200,000500,000mAh Area SoC SoC SoC Server PCIE contains multiple SoC Chips Cost IPC NVR Reliability/AI PPACR192019.6 iResearch I 2019.6 iResearch I289215891811366113 4 1-25%42%-24%172%112%6%481%-8%1349%-82%2018%55 276684123018883458110001380017100205002360026600402%148%80%53%83%25%24%20%15%13%2017 2018 2019e 2020e 2021e 2022e2017-2022 AI AI IP AI%IP%1 G AI AI 2020 AI IDC 202019.6 iResearch I AI AI 20GB AI AI DSP ASIC GPU ASIC ISP DVR SoC A/D/IPC SoC CPU ISP NVR SoC IP IPC SoC+AI-IP/AI-IPC-SoC AI DL GPU/ASIC GPU ASIC AI212019.6 iResearch I 78%22%2016 IPC DVR/NVR TI AI AI Hi3559 IPC 7 CPS IHS 62%8%7%3%20%2016 IPC Ambarella TI NXP JETSON TX1-GPU1TOPs Hi3559AV100 4TOPs T01-ASIC 1TOPs TESLA P4 5.5TFlops BM1682 3TFlops AI222019.6 iResearch I 2 AI 13.0 14.3 14.7 14.7 14.9 15.0 15.2 15.2 15.2 0.4 1.5 3.0 4.6 6.1 6.8 10%3%0%1%1%1%0%0%237%102%51%33%13%2014 2015 2016 2017 2018 2019e 2020e 2021e 2022e2014-2022 AI CAGR%AI CAGR%10 0 AI SoC AI AI AI AI/VR/AR AI IDC 232019.6 iResearch I AI SoC AI IP AI AI CPU AI SoC AI AI ASIC DSP AI 20005000mAh AP CP CPU GPU ISP/AI SoC AI-A/VRAI IP CPUGPUDSPISPAI-IP SoC/SoC242019.6 iResearch I 2019.6 iResearch I SoC AI SoC 845 Hexagon 685 DSP+CPU+GPU 855 Hexagon 690 DSP+CPU+GPU7TOPs 3 A11 Neural Engine 0.6TOPsA12 Neural Engine 5TOPsMTK P60 APU 0.56TOPsP90 APU2.0 2.25TOPs Kirin970 NPU 1A 0.512TOPsKirin980 NPU 1H 5TOPsIPARM N/A ML/OD ProcessorCadence N/A P5/P6/Q6/C5Synopsys N/A EV5x/6xCEVA N/A XM6/XM4 1 AI 2 AI SoC IP AI-IP SoC SoC 3 SoC IP AI IP IP AI-IP AI CounterPoint AI 2017 SoC 41%21%18%8%6%5%1%Qualcomm Apple MediaTek SamsungHisilicon Spreadtrum Others252019.6 iResearch I 3 510 L45 SAE L0L5 L0 L3 L5 Audi A8 Tesla L2 3 ADAS 2020 L3 L4 5 2025 2030 IHS,9581134662001030219410084832054330595236359566620687347056573278755727867610.0%14.0%21.0%30.0%40.0%45.0%48.0%49.0%48.0%47.0%43.0%39.0%35.0%31.0%30.0%0.2%0.4%0.6%1.0%1.8%2.8%4.4%7.0%10.0%12.0%14.5%16.0%18.0%20.0%21.0%0.0%0.3%0.6%1.2%2.3%4.0%7.2%10.5%13.0%15.0%10.0%14.0%21.0%31.0%41.0%48.0%53.0%57.0%59.0%61.0%62.0%62.0%63.0%64.0%66.0%2016 2017 2018e 2019e 2020e 2021e 2022e 2023e 2024e 2025e 2026e 2027e 2028e 2029e 2030e2016-2030 L1-L5 L1/L2%L3%L4/L5%L1-L5%262019.6 iResearch I ECU ECU ECU 100 L3 DCU MDC ADAS/CPU+XPU ECU+HAL/Lv12 Lv3 Lv45 13 46 610 1 24 68 N/A 01 13 50TOPS 272019.6 iResearch I AI AI+AI CPU+XPU XPU DSP/GPU/FPGA/ASIC DSP GPU/FPGA/ASIC CPU DRIVE PX Xavier Pegasus ZF Bosch Audi Benz Tesla Tier1s OEMs DRIVE OS NV Media CUDA TensorRT Developer ToolsDRIVEWORKS DRIVE AV DRIVE IX DRIVE Core DRIVE Calibration DRIVE NetworksDRIVE PerceptionDRIVE MappingDRIVE PlanningVisualizationAI CoPilotAIAssistanceDRIVE PX Parker AutoChaufferDRIVE PX Parker AutoCruiseDRIVE XavierDRIVE Pegasus-SelfDriving Hardware+GPU+Software Stack282019.6 iResearch I 1 L3 612 312 5 L5 4000GB 2 3 ASIL-D ASIL-D Intel Mobileye Altera OEMs Tier1 AI+Mobileye SoC AutoCruise 2*Denver+4*Cortex A57+2*Pascal GPU 256 CUDA coresAutoChauffeur 4*Denver+8*Cortex A57+4*Pascal GPU 512 CUDA cores+2*dedicated M*M modules20TFLOPS FP16/250WXavier 8*custom ARM64+Volta GPGPU 512 CUDA cores 20 TOPS INT8/30WPegasus 16*Custom ARM64+2*Volta GPGPU CUDA cores+2*Volta GPU 320TOPS(INT8/500WMobileye Intel EyeQ3 0.256TFLOPS/2.5WEyeQ4 MIPS Warrior CPU+6*VMP Vector Microcode Processors+MPC Multithreaded Processing Cluster 2.5TOPS/3WEyeQ5 8*CPU+18*Computer Vision Processors 17TOPS/5W292019.6 iResearch I AI AI AI AI AI SoC SoC+XPU forEdge/AI SoC ASIC AI/AI MAC Clock-gating AI30AI 1AI 2AI 3AI 4 5312019.6 iResearch IAI EDA IC Fabless IDM IDM Fabless IDM Fabless-Foundry Foundry Fabless IDM IC Fabless/Fab-lite EDA FablessFoundryOSATIDM AI322019.6 iResearch IAI IP IP IP IP CPU GPU DSP NPU IC 970 980 NPU IP IP 1%3%IP IP IP IP IP 8:20=8 20 60%60%40%IP 8:20IP Fabless SoCFoundryIP+1%3%SoC IP 332019.6 iResearch IAI EDA 1 IC 2 Foundry 3 OSAT 4 EDA IP 50100 EDA EDA CR3 70%EDA IC 20 EDA 100/EDA LPDDR IP Xavier 2000 20 Xlinix ACAP 1500 10 EDA/IP IC Foundry Tape Out Chip Test Final Test OSAT 58 90 342019.6 iResearch IAI+=12 300mm 4500/100 50/1.5 50%16nm 60 6 3 60 6 3 4,800,000 20%48.0 6.00 2.00 8.2 0.8 0.4 120.2 112.8 112.4 4.8 6.00 0.20 8.2 0.8 0.4 30.2 22.8 22.4 0.5 6.00 0.02 8.2 0.8 0.4 21.2 13.8 13.4 0.0 6.00 0.00 8.2 0.8 0.4 20.3 12.9 12.5 IP IP 16nm*成本（掩膜 封装）片成品率 晶片成本；晶片成本 晶 成本（每片 晶 晶片数）成品率；每片晶 晶片 数 晶 面 晶片面 晶 周 晶片面积35AI 1AI 2AI 3AI 4 5362019.6 iResearch I&2018.12 2018 12 5G 2018.10 2017.7 1 2 3 4 5 6 1.2020 1500 1 2.2025 4000 5 3.2030 1 10.2017.3 2017.2 2016.5+AI 2016.3 2015.7+2015.6 2025 372019.6 iResearch I CPU&2017.4 2016.12 5G 32/28nm 16/14nm 10/7nm 2016.7 2015.05 2025 2014.06 2015 3500 2020 20%2012.07 2012.02 2011.01 382019.6 iResearch I AI CPU AI AI AI CPU AI 12 AI AI,1-2 5G AI 5G AI AI392019.6 iResearch I76.0 66.8 1666.0 487.5 62.0 93.0 109.0 144.0 2015 2016 2017 2018 AI AI AI 1517 18 17 32%18 IT 2013 10 18 8 4.4 B+146.66 2015 7 19 1 6 30 2016 3 18 6 25 2016 5 18 10 2 A 5.8 2012 6 18 7 6 C+4.57 19 4 D 2007 10 18 6 5 5 2016 3 3 Xlinix 2018 3 18 8 3.4 Pre-A 1.75 2017 2 17 12 4.5 A 1.55 2017 11 18 4 TCL 3 A 1.05 2015-2018 AI AI402019.6 iResearch I AI AI 5G AI AI AI AI AI AI AI CPU AI IoT+AI 10 AI IOT 5G AI+ASIC AI 1 2 AI AI 41AI 1AI 2AI 3AI 4 5422019.6 iResearch IAI AI 12 GPU AI AI 15 17 AI AI AI Intel Altera Nervana Movidius Mobileye AI AI Intel VerisiliconGRAPHCOREWaveComputing GoogleAWS 2015 2018 2017 20162015-2019Q1 AI AI432019.6 iResearch I 1993 1999 GPU 80%10 GPU VR HPC AI AI AI 16 17 18H1 10 OEM&IP,AI 18H2 GPU Wind 2015-2018 36%37%44%13%59%7%145%52%1%-6%11%21%-13%-41%-11%1%85%75%52%15%2015 2016 2017 2018 OEM&IP 2015-2018442019.6 iResearch I AI AI AI Tesla GPU Volta Tesla V100 15.7TFLOPS CUDA 10 50 AI&Big Data ResearchEdge Computing AI TESLA GPUFermi Kepler Maxwell Pascal Volta2010 2018 2016 2014 2012-CUDADeep Learning Math/Linear Algebra Siganl/Imge/Video Parallel AlgorithmNVgraphcuBLAS CUDA MathcuRAND cuFFT cuFFT452019.6 iResearch I AI DianNao DianNao PuDianNao DianNaoYu Cambricon ISA NN ASIC Google TPU DianNao TPU DianNao DianNao DianNao 2014 ASPLOS2014 CNNDNNRNN 0.98GHz 0.45TOPs 65nm 0.485W CPU100 1/10 GPGPU DaDianNao 2014 MICR2O14 DianNao 16 NFU 606MHz 28nm 16W GPU21 1/330 training ShiDianNao 2015 ISCA2015 DRAM ShiDianNao CNN SRAM DRAM PuDianNao 2015 ASPLOS2015 k-k-7 1GHz 1.06TOPs 3.51 0.596W 65nm PuDianNao GPGPU GPGPU DianNaoYu 2016 ISCA2016 Cambricon GPUX86MIPS 6.41 9.86 13.38 X86 GPU 91.72 3.09 GPU Cambricon 130.53462019.6 iResearch I AI AI AI AI SDK ISA Cambricon-1A 970SoC Cambricon-1H 980CoC MLU Edge Cambricon-1A/1H/1MCloud MLU100/270 CN-ML CN-GenCN-CCCN-Perf CN-GDB DianNao AICambricon DianNaoYu ISACambricon NeuWare SDKDeep Learning Framework SupportingDeployment472019.6 iResearch I AI 7nm 7nm 74.5%2017 AI BM1680 2018 3 BM1682 2018 9 AI BM1880 2019 AI BM1684-2013 2013 10 28 55nm 2014 28nm 2015 BTC 2017 AI BM1680 252016 16nm 2018 7nm AI BM1682 AI BM1880 50 2019 7nm AI 482019.6 iResearch I AI BM1680 2TFlops 32MBytes 25W AI BM1682 3TFlops 16MBytes 30W BM1680 50%/BM1880 1TOPSINT8 2MBytes 2.5W BMNNSDK BMNet bmodel BMRuntime bmodel BMNet BModel DevelopeBMBMRuntime Trained BM1682 BM1682 AI AI PCIE3.0 X86 CNN/RNN/DNN BM1682 BM1880SA3 SE3 SC3 BM1880 USB AI BM1880 USB BM1880 ARM CORTEX A53 492019.6 iResearch I 2019.6 iResearch I AI AI 618 98 30 3D 6 150 AI Ubitus Ubitus AI Wind AI 1.4 2.8 25.2 28.4 0.7 1.5 12.1 10.3 2015 2016 2017 2018H12015-2018H1 50 Keep Learning 400-026-2099 51