On December 17, 2021, at the first “Driphu China RISC-V Industry Forum”, Beijing Jingshi Intelligent Technology Co., Ltd. introduced its RISC-V-based AI vision chip CR182x.
Jingshi Smart COO Huang Qunhui
Facing the pan-security market, Beijing Jingshi Intelligent launched the CR182x series of light intelligent AI vision chips using the flat-head RISC-V C906 core. According to Huang Qunhui, COO of Jingshi Intelligent, CR182x series chips have 0.5TOPs computing power, support 2 channels of video input, maximum resolution 2880*1620@30fps, integrated second-generation fully self-developed deep learning tensor processor (TPU), smart Image processing engine (Smart ISP), hardware-level highly secure data protection architecture (Security), voice processing engine and H.264/265 smart codec technology are one of the most integrated products in the industry.
At the same time, through the deep integration of the self-developed AI tool chain (Too1Chain) and the excellent computing power of the CR182x series, a terminal AI light computing power platform with data security and ecological openness will be created. The CR182x series will be widely used in security, access control and attendance, after-installation of intelligent vehicles and other fields.
It is understood that the CR182x series adopts self-developed professional security-level image processing (ISP) technology and can support 5M high-definition dual sensor input. Each input supports image processing functions such as starlight-level noise reduction, wide dynamic, dehazing, and distortion correction to provide high-quality image output for camera applications. The H.264/265 codec can also be combined with ISP/TPU technology to enhance or retain the key information in the video picture, and provide a high-definition, high-reduced low-bit stream intelligent coding solution.
The CR182x series uses the self-developed AI tool chain (Toolchain) and the deep integration of excellent computing power to create a terminal AI light computing power platform with data security and ecological openness. CR182x series TPU supports BFloat16 (BF16 for short), combined with Jingshi’s intelligent mixed precision technology, can reduce precision loss and support terminal applications with high computing power and low power consumption.
The CR182x series will also adopt low-power and fast-start technologies to focus on the AIOT and battery camera markets, and provide partners with cost-effective and low-power product solutions. The launch of this series of chips has greatly lowered the AI threshold in the civilian consumer market and helped AI technology to enter the lives of ordinary people.
According to Huang Qunhui, the goal of Jingshi Intelligent is to create an artificial intelligence chip platform with the safest data and the most open ecological chain. Artificial intelligence is divided into three areas, from the server + algorithm in the first stage to the GPU + algorithm in the second stage, all facing the problems of high cost and difficult deployment. The third-stage dedicated chip + built-in algorithm can promote the realization of large-scale data structure.
According to reports, consumer-grade products have just been released at the beginning of this year, and the CR182x series chips to be released next year will further help it achieve a comprehensive product layout.
Jingshi Intelligent Technology Co., Ltd. is a chip design company focusing on the research and development of video surveillance and edge computing technology. It has the leading AI TPU computing core and SoC chip integration technology in China. The team elites come from top universities and chip design companies at home and abroad. Over 80% of the overall human resources have a doctoral or master’s degree. The team members have rich practical experience in product design, research and development, production, and assisting customers in product implementation. Chip design The team has an average practical experience of more than 16 years, and has complete R&D integration experience such as chip research and development, platform architecture design, algorithm development, and software and hardware reference design.
The Links: LQ074V3DC01 6MBP50RA060-01
