Dual-wheel drive, Intellifusion launches 12TOPS edge vision SoC

At the 3rd Dishui Lake China RISC-V Forum, Zhang Fulin, Chip BD Director of Shenzhen Intellifusion Technologies Co., Ltd., introduced the company's latest Deep Edge10V, an edge AP-level SoC with 12 TOPS computing power.

In 2022, Yuntianlifei's new self-developed chip Edge10V was successfully taped out. In 2023, the company will focus on productization and is now close to quasi-mass production.

In three sentences, Yuntianlifei's chips can be summarized as follows: it has gone through four generations of neural network processors, has three generations of instruction set architectures, and has released two generations of SoC chips. Zhang Fulin said that chip development adopts a two-wheel drive concept and customizes chip design based on algorithms. On the one hand, applications are created through chips, and the data generated by applications is used to counter-complement algorithms.

Edge10V supports Chiplet and is a domestically produced AI chip. In addition to design, including process, packaging and testing, all rely on the domestic supply chain, and supports dual Gigabit network ports, multi-channel network access, and different sensors. It uses the Pingtouge C920 dual-core RISC-V and NNP 400T AI processors, and the PCle 3.0 interface can support different peripheral expansions.

Zhang Fulin introduced that one of the major advantages of Yuntian Lifei is its strong effective computing power. It has a software stack that matches the heterogeneous chip architecture and can schedule computing resources, thereby optimizing computing resources.

In addition, the chip also supports mainstream AI tool chains and models, including frameworks such as Pytorch, Onnx, and Caffee, as well as models such as CNN, RNN, and Transformer. "Edge computing is very fragmented, so we also made a MINI PCIe accelerator card that supports embedded devices and is used in edge servers," Zhang Fulin added.

Edge10V application scenarios can be flexibly applied to the end and edge, including service robots, smart robots, smart education, smart manufacturing, smart security, smart warehousing, smart industry and other applications.