NewsCompany newsIndustry News



Breakthrough the memory wall world difficult problem, the AI team of CAS QNPU chips end of the year or usher in streaming chips

MetInfo enterprise content manager system | MetInfo CMS

Recently, at the World Intelligence Congress, the first QNPU (Quantized Neural ProcessUnit) prototype of AI chips designed independently by Nanjing AI Chip Innovation Research Institute, Institute of Automation, Chinese Academy of Sciences, and the four-way vehicle recognition, on-board assisted driving ADAS system, terminal AI function accelerator Qine and AI The interactive system for real-time human skeleton recognition was exhibited.

According to the official information of the Institute of Automation of the Chinese Academy of Sciences, QNPU, an artificial intelligence chip, adopts the leading quantization model compression processing technology in the world, realizes the design of DDRFree, breaks through the world difficult problem of memory wall, not only meets the demands of low power consumption, low latency, small volume and low cost of edge devices, but also provides the high computing capability required for performing AI tasks. Cheng Jian, executive vice president of AiRiA Research Institute, previously disclosed that the QNPU designed by the team is expected to be released by the end of this year, named "WattA1". This will be the first AI chip in the world to focus on low bit quantization technology.

Its official publicity shows that WattA1 will adopt the TSMC 28nm process with a peak power of 24 Tops and support 1080P four-way real-time monitoring. AiRiA, Nanjing Institute of Artificial Intelligence Chip Innovation, Institute of Automation of Chinese Academy of Sciences, was established in September 2017 by the Institute of Automation of Chinese Academy of Sciences and Nanjing Qilin High-tech Zone. It is the result transformation base of AI and chip research and development in local area under the Institute of Automation of Chinese Academy of Sciences. AiRiA Institute, relying on the national platform of AI and chip design, has established its advantages in the fields of AI chips and core intelligent algorithms, and is positioned for the advancement of soft-hard double-wheel drive intelligent technology.