Starting from the new paradigm of AI, the AX8850 opens a new era of high-efficiency AI processing.
In the current global wave of artificial intelligence, the consensus that “AI is everywhere” has emerged. However, how to popularize AI in edge scenarios with lower costs and higher energy efficiency has become a key competition among chip companies. Aixinyuan Intelligence is an active promoter in this embodied intelligence revolution. Its latest AI processor, the AX8850, not only reflects a profound understanding of AI development trends but also brings practical and efficient solutions to the industry through architectural innovation.
On May 13, the 15th Songshan Lake China IC Innovation Summit Forum, the most well-known local IC promotion platform in China, was held in Dongguan Songshan Lake. During the chip promotion session, Liu Jianwei, co-founder of Aixinyuan Intelligence, pointed out that AI has evolved from an abstract technical concept into a social-level “intelligent output infrastructure”—similar to water and electricity, it will exist stably and universally across various industries.
He pointed out that the fundamental change lies in the migration of programming paradigms: traditional programs rely on human-defined rules, while AI programs depend on data to train models, automatically correcting and optimizing. Furthermore, AI programs are fundamentally centered around tensor computation, which presents new challenges for hardware in terms of computational scheduling and energy efficiency due to the high-dimensional and large-volume data streams.
Architectural Innovation: Data Flow Microarchitecture + Programmable Operators = High-Efficiency AI Chips
Based on the characteristics of the aforementioned AI computation, Aixinyuan Intelligence proposed an instruction set design based on operators, combined with a programmable data flow microarchitecture, achieving energy efficiency performance that is orders of magnitude higher than traditional GPUs.
Core Highlights of the Architecture:
The microarchitecture design optimized for tensors ensures that energy consumption is truly used for high-density computation rather than data movement. It supports heterogeneous multi-core and hardware-level task scheduling, enhancing parallel execution efficiency under data dependencies. The hardware structure is better suited for AI tasks, especially excelling in environmental perception, image generation, and other tasks.
According to experimental data provided by Aixinyuan Intelligence, compared to traditional GPU architectures, its AI chip can achieve a tenfold improvement in energy efficiency, which is precisely the breakthrough needed in edge scenarios.
Flagship Product: AX8850, Bringing Edge AI to Embodied Intelligent Scenarios
As an important outcome of Aixinyuan Intelligence’s AI strategy, the AX8850 processor is positioned for “lower costs and more practical scenarios,” performing excellently in embodied intelligent terminals such as robotics, security, and intelligent driving.
Typical application cases include intelligent visual perception—capable of semantic detection of “people,” “objects,” and “clothing color” based on single frames or continuous images. Additionally, in terms of text generation from images, it can generate descriptive text or interactive images in real-time, possessing a certain degree of “emotional interaction” expression.
Moreover, it can achieve intent recognition and command generation, understanding user intent based on dialogue content and outputting control commands for human-machine collaboration. In multi-modal perception fusion, it can integrate signals from multiple cameras and millimeter-wave radar to compensate for blind spots in purely visual solutions. In continuous video event perception, it can achieve event-level recognition and judgment for scenarios such as “fall detection.”
Liu Jianwei pointed out that Aixinyuan Intelligence has invested years in building a software development platform that provides full-process support for AI program deployment, model mapping, and optimization, lowering the entry barrier for developers and accelerating the conversion speed of AI applications from algorithms to products. In the future, Aixinyuan Intelligence will continue to advance in the exploration and practice of AI hardware architecture, contributing to the realization of an intelligent world that is “accessible to everyone.”
Recent Popular WeChat Articles1、The dual “giants” of foundry officially start the battle! Who will be the final winner?2、A glimpse into the progress of China’s semiconductor industry from a small detail of Kirin 9020!3、Chip Origin opens a new era of edge AI computing4、The Chinese semiconductor industry enters the Warring States era of mergers and acquisitions!5、How to view Huada Jiutian’s acquisition of Xinheng Semiconductor?6、Mergers for growth or innovation alone? Eight industry leaders discuss the future of local EDA.7、ST CEO to be dismissed!8、Surpassing TI C2000! Local RISC-V core DSP strongly released!If you find this valuable, please follow my WeChat account.About Zhang Guobin’s WeChat Official AccountThis account is personally operated by Zhang Guobin, a senior media person in the semiconductor field, with followers mostly being executives in the semiconductor industry. This account focuses on the latest technological trends and emerging technology applications in the global semiconductor field, interpreting the latest dynamics of the semiconductor industry. Follow along, and let Lao Zhang take you to explore the infinite wonders of the semiconductor world!Welcome to add my personal WeChat number 18676786761