July 6 to 8, the 2023 World Artificial Intelligence Conference (WAIC) was held in Shanghai. With the theme of "Intelligent Connection to the World, Generating the Future", this conference focused on in-depth discussions on ten major topics including large models, AI for Science, general intelligent agents and computing power, with the total number of participating enterprises hitting a new record high in history. As one of the exhibitors, Axera showcased its AI chip series products, as well as the landing applications and solutions in the fields of AIoT and intelligent driving, and created a themed exhibition area of "Inclusive AI Creates a Better Life".

Implementation of Two Core Technologies and Comprehensive Layout of Three Industries, Axera Enters the 2.0 Era

At present, technological innovation has become the core competitive advantage of major technology companies. Since its establishment in 2019, Axera has focused on the R&D of artificial intelligence visual perception chips, and independently developed two core technologies: Aixin Zhimou®AI-ISP and Hybrid Precision NPU.

With the implementation of the two core technologies, Axera has become the first chip company in the industry to implement and mass-produce AI-ISP and Hybrid Precision NPU on the full range of products. All chip products have industry-leading advantages such as high computing power, low power consumption and excellent image processing capabilities. A series of Axera's chip products were also displayed collectively at this exhibition. Up to now, Axera has completed the R&D and mass production of four generations of multiple chip products, and applied them to three major industrial fields including smart cities, intelligent driving and AIoT, marking the official entry of the company's strategic layout into the 2.0 era.

At this WAIC, Axera focused on demonstrating its capability to deploy large Transformer models on the end and edge sides. As the third-generation end-side chip released by Axera, AX650N features high performance, high precision, easy deployment and low power consumption in deploying Swin Transformer on the end side, making it a second-to-none Transformer deployment platform in the industry.

Axera Pi Pro exhibited at the booth is a development kit for ecological communities and industrial applications based on AX650N, which can help developers explore richer product applications and build an open end-edge intelligent ecosystem. The kit ran DINOv2 in real time and conducted PCA visualization on site, attracting many visitors to stop and watch. DINOv2 is the most advanced computer vision self-supervised model, which can achieve SOTA-level performance in tasks such as depth estimation, semantic segmentation and image similarity comparison.

AI Large Models Booming, Axera's Liu Jianwei Talks About Opportunities and Challenges

During the conference, the "WAIC Financial Afternoon Tea" event hosted by Cls.cn and East Hope Lansheng, and co-organized by Science and Technology Innovation Board Daily, was held. Liu Jianwei, Co-founder and Vice President of Axera, attended the event together with distinguished guests from Ant Group, Baidu, Microsoft and Tezign, and conducted an in-depth dialogue on the topic of "Opportunities and Challenges of Generative AI After the Emergence of ChatGPT".

Second from the left: Liu Jianwei, Co-founder and Vice President of Axera

Talking about the impact of this round of AI upsurge on the industry, Liu Jianwei shared his views as a semiconductor industry practitioner. He stated that the influence of large models on the semiconductor industry can be summarized as traction, reshaping and inclusiveness. Traction means that the semiconductor industry needs upper-layer applications for driving force; from the early PC industry to the subsequent mobile Internet, and to today's large models, all have a traction effect on semiconductors. Reshaping means that different applications may have different requirements for semiconductors, and a new computing paradigm may emerge in the AI era, thus reshaping the entire industry. Inclusiveness means that large model technology can greatly reduce the cost of intelligent implementation and make the application of intelligent technology easier in some conventional scenarios. "In the future, large models will inevitably exist on the cloud, edge and end sides."

Liu Jianwei further shared his understanding of large models, "A large model does not have to be extremely large; it can be called a large model as long as it can reduce the cost of intelligent implementation."

Amid the upsurge of AI innovation and entrepreneurship, the ethics and security of AI have also sparked heated discussions. From the perspective of a semiconductor manufacturer, Liu Jianwei holds a relatively optimistic attitude. In his view, semiconductors are an infrastructure tool, and the large models running on this infrastructure are also a kind of tool to a certain extent. "There is no good or bad in the tool itself; it all depends on how it is used."

2023WAIC has come to an end, yet the development of AI large models is still in the ascendant. Axera stated that focusing on perception and computing, it will strive to build world-leading AI chips in the future, accelerate the landing and application of AI on the end and edge sides, and create a better life with inclusive AI.