From July 13th to 14th, the 3rd China Integrated Circuit Design Innovation Conference and Wuxi IC Application Expo (ICDIA 2023) was held. At the special forum on "AIoT and ChatGPT", Liu Jianwei, Co-founder and Vice President of Axera, was invited to give a keynote speech entitled "Axera AX650N Transformer End-Edge Deployment Platform". He officially announced the Chinese name of the company's core technology, the Hybrid Precision NPU, "Aixin Tongyuan", and introduced the leading advantages of the third-generation SoC chip AX650N in deploying Transformer on the end and edge sides. Axera Officially Announces the Chinese Name of Its Hybrid Precision NPU: "Aixin Tongyuan" In recent years, the scale of the AIoT industry has grown rapidly. It is not only the best channel for the intelligent upgrading of various traditional industries, but also an important direction for the future development of the Internet of Things. At the same time, since its release at the end of 2022, large models represented by ChatGPT have triggered the layout of major domestic and foreign technology giants, and the development of artificial intelligence has also entered a brand-new era. It can be seen that with the rapid development of AIoT and large AI models, the underlying hardware is facing extremely high requirements in terms of data storage, computing performance, and graphics processing capabilities. In Axera's view, large models have general cognitive capabilities, which can reduce the cost of artificial intelligence landing in scenarios. In the future, everyone may have an intelligent assistant on the terminal. In the process of artificial intelligence development, Axera provides chip infrastructure, that is, providing basic capabilities of perception and computing through artificial intelligence chips to help the landing of AI on the end and edge sides. At the perception level, Axera is committed to making cameras see more clearly, thereby creating an entrance for the digitalization of the physical world; at the computing level, Axera is committed to making cameras understand what they see, and will provide a higher computing power foundation for the landing of various AI models on the end and edge sides. At the ICDIA 2023 site, Axera also officially announced the official Chinese name of the Hybrid Precision NPU, "Aixin Tongyuan", which is intended to provide basic computing power support for various intelligent algorithms, enable intelligence to have a deeper understanding of the world on the end and edge sides, and contribute to a better life. Positioned as a basic computing power platform for artificial intelligence perception and edge computing, Axera has focused on the basic capabilities of perception and computing since its establishment in 2019. It has independently developed two core technologies: Aixin Zhimou AI-ISP and Aixin Tongyuan Hybrid Precision NPU. The latter reduces the obstacles of the memory wall and power consumption wall, and supports more intelligent algorithms with higher effective computing power under the condition of limited area and power consumption on the end and edge sides, reducing the cost of AI deployment. At present, Axera has completed the R&D and mass production of four generations of multiple chip products, and has gradually landed in three major markets: smart cities, intelligent driving, and AIoT. "All these markets take perception and computing as basic capabilities, which is also the logic for Axera to choose different markets," said Liu Jianwei. High Performance, High Precision, Easy Deployment: Axera AX650N Becomes the Best Deployment Platform for Transformer In the design and development of AI chips, Axera attaches great importance to the in-depth integration of applications, algorithms and NPU. In terms of applications, Axera achieves higher performance through data flow optimization and front and rear stage processing acceleration; in terms of algorithms, Axera will improve hardware utilization from the algorithm perspective through designs such as operator acceleration, network microstructure acceleration, and memory optimization; at the same time, the Aixin Tongyuan Hybrid Precision NPU is a heterogeneous multi-core system with a built-in multi-core hardware scheduling mechanism to reduce CPU occupancy and make system applications run faster. Based on the design concept of joint optimization of applications, algorithms and NPU, the edge and end-side artificial intelligence processor chips launched by Axera all have the advantages of high performance and low power consumption. Its third-generation high computing power and high energy efficiency ratio SoC chip AX650N has become the best deployment platform for Transformer. Test data shows that Axera AX650N not only has excellent performance on traditional CNN networks, but also performs well in deploying the Transformer network SwinT on the end side, achieving high performance of 361 FPS, high precision of 80.45%, low power consumption of 199 FPS/W, and extremely easy deployment capability with the original model and PTQ quantization. At the same time, AX650N also supports low-bit hybrid precision. If users adopt INT4, they can greatly reduce memory and bandwidth occupancy, thereby effectively controlling the cost of deployment on the end and edge sides. At present, Axera AX650N has been adapted to Transformer models such as ViT/DeiT, Swin/SwinV2, and DETR. The most advanced computer vision self-supervised model DINOv2 has also achieved an operation result of more than 30 frames on AX650N. To help developers make better landing attempts on Transformer, Axera has also launched a development kit "Axera Pi Pro" based on AX650N for the ecological community and industry applications, helping to explore richer product applications. Facing the new round of AI upsurge set off by large models, Axera AX650N will also further optimize the Transformer structure and explore multi-modal Transformer models. Axera hopes to accelerate the landing of intelligence on the end and edge sides through core technologies of perception and computing, truly realize inclusive AI, and make people's lives better.