Slide 1: Fifth-Generation CEVA Imaging & Vision Silicon IP
Fifth-Generation CEVA Imaging & Vision Silicon IP Sep, 2016
Slide 2: The Demand for Vision Processor IP
The Demand for Vision Processor IP Smartphones Automotive Consumer Electronics Smartphone OEMs looking for Automotive industry focus is Drones, Surveillance, Action ways to ‘stand out from the completely dominated by Cameras, Auto After-market etc crowd’ ADAS/Autonomous Large volume opportunities th
Slide 3: CEVA-XM4 – "Best Processor IP 2015"
CEVA-XM4 – “Best Processor IP 2015” CEVA’s first imaging and vision processor supporting deep learning Widely licensed and silicon available Customer target applications include: Smartphone APs Smartphone Vision ‘co-processors’ Smart surveillance systems Drone ‘collision avoidance’ systems Automotiv
Slide 4: Introducing CEVA's 5th Generation Imaging & Vision Technology
Introducing CEVA’s 5th Generation Imaging & Vision Technology Comprehensive vision platform Centered on New CEVA-XM6 Vision DSP Enables embedded neural networks for mass market intelligent vision applications Simplifies delivery of powerful deep learning solutions on low-power embedded devices © Cop
Slide 5: Comprehensive and Scalable Vision Solution
Introducing CEVA’s 5th-Generation Imaging & Vision Technology Comprehensive and Scalable Vision Solution © Copyright by CEVA, Inc. All Rights Reserved 5
Slide 6: The CEVA-XM6 Vision DSP
The CEVA-XM6 Vision DSP 5th generation imaging and vision processor IP Major non-linear operations enhancements Up to Major performance increase in scatter-gather and sliding 3X window mechanisms Performance Gain* Significant performance gain* Up to 3x performance gain for vector heavy kernels 2x av
Slide 7: Hype Cycle for Emerging Technologies
Hype Cycle for Emerging Technologies 2016: Machine Learning at the hype peak Source: Gartner's Aug 2016 Hype Cycle for Emerging Technologies © Copyright by CEVA, Inc. All Rights Reserved 7
Slide 8: Neural Network Embedded Challenges
Neural Network Embedded Challenges Implementing a deep neural network in an embedded systems is an extremely challenging task! Very high bandwidth Porting and consuming and optimization computing bottleneck capabilities Stringent power budget and system cost Long “Time-To-Market” © Copyright by CEVA
Slide 9: CEVA Deep Neural Network (CDNN2)
CEVA Deep Neural Network (CDNN2) 2nd gen SW framework support Caffe and TensorFlow Frameworks Various networks* All network topologies All the leading layers Variable ROI “Push-button” conversion from pre-trained networks to optimized real-time Accelerates machine learning deployment for embedded sy
Slide 10: CDNN2 Usage Flow
CDNN2 Usage Flow OEM / Partner (offline) CEVA (offline + real-time) Live CDNN2 demo © Copyright by CEVA, Inc. All Rights Reserved 10
Slide 11: CDNN HW Accelerator
CDNN HW Accelerator Motivation Dedicated HW engine for executing the convolutions layers in CNN Convolutions are the major and most cycles consuming layers Provides the flexibility to cope with future Neural Network development >8X >95% Performance Gain* MAC utilization Compatibility: CEVA-XM vision
Slide 12: Flexible Embedded CNN Solution
Flexible Embedded CNN Solution Flexible embedded solution and 16bit support are required to cope with the evolving and leading neural networks © Copyright by CEVA, Inc. All Rights Reserved 12
Slide 13: CEVA-XM6 Platform vs. NVidia TX1 GPU
CEVA-XM6 Platform vs. NVidia TX1 GPU for Implementing Deep Learning Single CEVA-XM6 based platform is Power Faster Efficiency >25X >4X Processing** Factor* Assumptions: Based on the implementations of AlexNet and GoogleNet (single batch) TSMC 20nm technology and core @690MHz ROI single batch; (*) RO
Slide 14: CEVA's 5th Generation Imaging & Vision Technology
CEVA’s 5th Generation Imaging & Vision Technology Comprehensive vision platform Power efficient and scalable platform Targeting DL, AI and advanced CV functionality Significant time-to-market advantage Higher performance © Copyright by CEVA, Inc. All Rights Reserved 14
Slide 15: © Copyright by CEVA, Inc. All Rights Reserved
© Copyright by CEVA, Inc. All Rights Reserved 15
Slide 16: Thank You
Thank You