Under the influence of geopolitical factors, China's AI-related industry is facing the challenge that high-end GPU chips supply is being controlled by overseas supplier. In response, the Chinese academic and industrial sectors have come up with alternative approaches, such as bypassing Nvidia AI GPUs and focusing on the development of integrated systems (IS) while reducing dependence on advanced EUV (extreme ultraviolet) systems.
According to Chinese media reports, Zhao Lidong, founder of China's AI startup Enflame Technology, believes that market requirements for AI chips mainly lie in high performance, high bandwidth, high storage, high versatility, efficient distributed computing, and cluster interconnection, which basically cannot solely depend on a single chip but rather on the concept of system clusters, emphasizing distributed computing.
To address this challenge, Zhao stressed, the Chinese chip industry needs to make breakthroughs in AI chip architecture, power consumption and clustering capabilities, and the ultimate goal is to make future computing power as easily accessible to everyone as water and electricity.
In this regard, he continued, Enflame Technology has proposed an innovative architecture and an open-source ecosystem support, which means that businesses can perform AI acceleration tasks without relying on GPU chips.
On another front, China also faces challenges in accessing EUV systems due to restrictions imposed by foreign suppliers. In response, Mao Junfa, an academician of the Chinese Academy of Sciences has indicated that integrated systems (IS) will be one of the important paths to bypass the limitations of Moore's Law.
The term IS refers to the overall design and production from a system perspective, integrating various chips, sensors, components, antennas, interconnects and more onto a single substrate. Compared to traditional ICs, the IS concept enables high density, low power and easier design and manufacturing of systems. Chiplets, which are well known in the semiconductor industry, represent the embodiment and prototype of integrated systems, according to Mao.
Mao pointed out that the development of the IS pathway presents an opportunity for a detour to surpass existing challenges. However, IS itself needs to be carried out across multiple scales, materials, process technologies, dimensions and physics, which is more difficult to plan, design and manufacture at the system level, involving a huge engineering scale, he added.
Nevertheless, these alternative development paths and solutions also raise concerns among leading European and American vendors leaders, including ASML and Nvidia, who are worried that China's rapid progress in autonomous development may dent their potential market opportunities or even replace them, according to market observers.
