Question Object:seabet casino chip and system design for edge terminals
Report time:October 30, 2019 3:30 pm
Reporting location: Central South seabet casino Computer BuildingConference Room 313
Reporter: Dr. Chen Xiaobai
ReportAbstract:
With seabet casino technology in mobile communications，Finance，Security，Wide range of applications in autonomous driving and other fields，Brings new computing requirements for seabet casino technology。TraditionalCPU，It is difficult for instruction-driven computing platforms such as GPU to support seabet casino technology for intensive computing and ultra-large data communication，And in mobile terminal devices with limited energy storage，It is difficult to support the application of seabet casino technology for a long time even with power consumption of hundreds of watts。
Computing power for seabet casino computing in edge terminal devices，Bottleneck of power consumption and flexibility，The reporter designed the computing power to be sufficient，Low power consumption，Highly flexible seabet casino computing solution。Pass pair algorithm，Instruction set，Compilation tools，Hardware (chip) software and hardware co-design，To gain the high flexibility of supporting all major neural network algorithms，and inThe seabet casino under the 55nm process is up to 2.2 seabet casino energy efficiency of Tops/W。At the software level，The speaker proposed a special instruction set for neural networks，This instruction set carries parallel parameters and data multiplexing information，At the same time, design the corresponding instruction generator to analyze the neural network algorithm，Comprehensive generation of instructions for optimized execution by obtaining hardware seabet casino and storage resource information。At the hardware level，Based on neural network calculation form and data characteristics，Designed lookup table multiplication array and reconfigurable calculation engine，Achieve 79% energy efficiency advantage over conventional multiplier arrays in low bit width calculations。Structural changes based on data storage requirements，Designed multi-layer convolutional layer pipeline seabet casino technology，Significantly reduces the need for high-power off-chip communication during neural network calculations or even reduces it to 0，Thus greatly reducing the overall power consumption of the chip。
About the speaker:
Chen Xiaobai，Male，Sun Yat-sen seabet casino-Carnegie Mellon seabet casino Joint School of Engineering jointly trains doctoral students，Main research direction is computer architecture，Artificial Intelligence ASIC Design。As the first author, he published 4 SCI/EI academic papers (including 2 trans journals，Respectively, TCSVT, the top journal in the field of circuits and systems，TVLSI, the top journal in the field of very large scale integrated circuits)，And in Kuala Lumpur in 2017，Being a human being at the two APSIPA ASC academic conferences held in Hawaii, USA in 2018seabet casino chipseabet casino’s oral presentation。The energy efficiency of the low-power neural network seabet casino designed and taped out2.2 Tops/W, for2018The best design of the year，Leading at home and abroadWorking together. Applied3itemInvention patent and1Integrated circuit layout and responsible forIEEE TVLSI，TCAS，TVLSI.Once inMicrosoft Asia R&D Group, Lenovo seabet casino InstituteEngineers and researchers.AndYu2017YearCreate a well-knownAIseabet casino company Zhenwei Technology, serves as president and chief scientist. Led the company to launch image-oriented video streamingALSeries, wearable device fieldBLseries of ultra-low power seabet casino chip products and has become a technical partner of Huawei, Lenovo and other system manufacturers，Successfully promoted the development and commercialization of seabet casino chip technology。led the company to obtain2018National Innovation and Entrepreneurship Champion of the Ministry of Science and seabet casino，One of the scholars in the industry with cutting-edge academic research capabilities and seabet casino industrialization。
