17th International Forum on MPSoC for Software-defined Hardware
July 3-7, 2017
Yukoh Matsumoto, TOPS Systems, Japan
Dataflow : Manycore Architecture and It's Programming for Computer Vision and AI
Recently programming with Dataflow Processing is a trend can be seen in such as Google's TensorFlow; an Open Source Software Library for Machine Intelligence, OpenVX for Portable, Power-efficient Vision Processing, etc.
Dataflow is expected as a solution for speed-up of Computer Vision and Deep Learning Applications with exploitation of more parallelism than simple data parallel like GPU as well as reduction of required memory access by passing intermediate data from core to core. Optimization of data granurarity and merging of nodes for a target processor enables portability between diverse of computing platform, such as CPU, GPU, Specialized Manycore, DSP, FPGA, hardware accelerator, etc.
This presentation discusses the software and hardware architecture of the world fastest object detection with Dataflow based stream processing on SMYLEvideo(Gen-2) Manycore that achieved more than 150fps of Oject Detection speed with 8 cores running at only 50MHz. Amount of memory access is reduced to 1/280 compared to a conventional CPU. Furthermore, the presentation addresses a concept of Software Desined Sensor system using SMYLEvideo(Gen-2) Manycore. As well as the research work on Programming Language for dataflow based Manycore processor for Next-Generation Artificial Intelligence, the project commissioned by the New Energy and Industrial Technology Development Organization (NEDO).
Dr. Yukoh Matsumoto is Chief Architect, and President and CEO of TOPS Systems Corp. His current research project "Development of Dataflow Programming Language for making the best use of Manycore" is supported by NEDO as a Next-Generation AI technology in. Recently, he designed and implemented SMYLEvideo(Gen-2) Manycore for Computer Vision Applications. In his 30 years of carrier, he has architected and designed more than 10 advanced microprocessors such as, Embedded Application Domain Specific Manycores / Multicores for Video Codec, Real-Time Ray Tracing, WiFi Base Band, superscalar x86 processors, and DSPs for voice processing. He founded TOPS Systems Corp. in 1999 for Multicore R&D and IP business. He received the Takeda Techno-Entrepreneurship Award, Tsukuba Venture Award, ET Award, Small Enterprises Excellent Product & Technology Award, in 2001, 2010, 2011, 2016 respectively. Prior to TOPS Systems, he worked for Texas Instruments in US and Japan. He received the Dr. of Information Sciences (the Ph.D.) degree from the Graduate School of Tohoku University in 2007. He has participated in the MOT (Management of Technology) program at the Graduate School of Engineering in Tokyo University.
* If you wish to modify any information or update your photo, please contact the Publicity Chair at the following address.