Axera Showcases at 2024 WAIC: Advancing Edge and Terminal Intelligence for Universal AI Access
Abstract
On July 4, 2024, the World Artificial Intelligence Conference (2024 WAIC), one of the most awaited events in the tech community, officially opened its doors at the Shanghai World Expo Exhibition and Convention Center. Running from July 4 to July 7, this year’s conference features prominent industry players, including Axera Semiconductor Co., Ltd. Axera took center stage at booth C1525 in Hall H2, where it unveiled its latest innovations under the theme “Mimicking Human Perception with large language models: Advancing Edge and Terminal Intelligence for Universal AI Access.” The company presented an array of significant advancements in AI technology, including advanced SoCs (System-on-Chips), pioneering solutions, and sophisticated AI models. These innovations are poised to offer conference attendees an insightful look into the future of smart technology.
Small Chips,large language models: The Core Technology Transforming Edge and Terminal Intelligence
At this year’s WAIC, Axera unveiled multiple cutting-edge technological achievements for the first time, continuously capturing the spotlight. In the dedicated chip exhibition area, Axera showcased a range of AI chips, including those for intelligent applications, TX and RX chips for coaxial HD cameras, hybrid analog-digital chips, and MCU chips. This varied showcase demonstrated Axera’s thorough product planning and robust market presence across various segments, from high to low-end. The Axera chip family is celebrated for its low power consumption, high efficiency, safety, and reliability, positioning it as the backbone of intelligent edge and terminal devices. By enhancing the cost-efficiency, effectiveness, and environmental friendliness of AI technology, Axera’s chips are pivotal in propelling us into a new era where AI is universally accessible.
Notably, in the dedicated large model exhibition area, the AX630C chip for terminal applications, powered by Alibaba Cloud’s Tongyi Qwen 2.0 language model, has successfully demonstrated seamless human-computer interaction with an impressive computing power of 3.2 teraflops. Its processing speed is remarkable, handling over 10 information units per second while maintaining a power consumption of less than 1.5 watts. This “small” chip, with its controllable cost, is suitable for devices priced at around a hundred yuan, marking the advent of an era where large AI model applications are accessible to all.
Additionally, Axera has unveiled the AX650N and MiniCPM V2.0 combination, a potent AI duo that provides enhanced performance with smaller models and reduced costs. This breakthrough allows terminal devices to effortlessly support AI models, generating text from images, recognizing fine details in high-definition pictures, and capturing key information from vast data sets.
Cutting-Edge Innovations: More Small Chips Unleash Infinite Creativity with Ultra-Low Power Consumption and Higher Efficiency
Axera’s suite of small chips for terminal applications, including the AX620Q, AX650N, and Axera-Pi Zero, when integrated with the Depth Anything model, showcases exceptional deep learning capabilities, and pioneers the transformation of standard 2D photos into 3D visual experiences. This breakthrough enhances extended reality (XR) technologies and paves the way for novel AI-driven applications. The AX650N chip excels in both edge and terminal applications by smoothly processing and analyzing 32 channels of HD video in real-time, all under 8 watts. This showcases a perfect balance between ultra-low power consumption and high performance. Additionally, the AX650N, when integrated with the CLIP (Contrastive Language-Image Pre-Training) model, simplifies image searches, allowing users to find matching images from extensive databases using simple text inputs. The integration of the AX650N chip with the Segment Anything Model (SAM) enables automatic photo editing without the need for manual labeling. Axera’s image segmentation technology allows for the instant categorization of complex images, dramatically simplifying the processes of image editing and analysis. Furthermore, when paired with the OWL-ViT (Vision Transformer for Open-World Localization), a multi-modal multi-dimensional perception model, the AX650N significantly boosts its capabilities in detecting unknown objects and automating image labeling.
Mini Darkroom and Ecosystem Wall: Witnessing the Future Impact of AI Chip Applications
Axera’s classic AXProton AI-ISP Mini Darkroom Lab, powered by the AX650A chip-driven night vision full-color camera, demonstrates exceptional image quality in extremely dark environments. It restores colors while preserving details, creating a visual miracle that makes night view appear as bright as day.
The booth also showcased a rich lineup of Axera’s AI products, including the developer-favored AXera-Pi series, computing power cards, AI boxes, smart medical devices, and more. This display highlighted the infinite possibilities of Axera AI chips in practical applications.
In the burgeoning realm of smart technology, Axera champions the belief that edge computing and edge and terminal intelligence will emerge as pivotal forces driving innovative applications. With its robust foundation in proprietary AI technology, coupled with its rapid iteration and mass production capabilities, Axera is poised to create world-class perception and computing platforms, and work with industry partners to further its mission of “AI for All, AI for a Better World.”