Despite the US' latest restrictions on exports of advanced-node chips and related manufacturing equipment to China, China will still be capable of developing its homegrown AI chips for edge computing applications, according to industry sources.

Under the restrictions, Chinese foundries are unable to acquire the most advanced chip manufacturing processes and their technology will stay put at the 28nm node; many HPC chip designers fail to import advanced-process wafers; and terminal device makers cannot procure high-end processors from US suppliers such as Nvidia and AMD to develop AI equipment, the sources said.

But if abandoning the development of supercomputer chips or cloud computing chips that require the support of advanced process nodes, and instead focusing on edge AI chips carrying lower process requirements, China will still have a chance to gain a solid presence in the global AI market, the sources stressed.

Actually, Chinese AI-related IC designers have not yet progressed to the development of high-end CPU and GPU chips in recent years. They still mostly focus on mature-process MPU (micro processing unit) and ASIC chips for image processing and security surveillance applications, which will surely become a major segment for the designers to tap deeper, sources continued.

Meanwhile, many AI chips have yet to fully demonstrate their actual computing power due to constraints in software designs, the sources indicated, and therefore upgrading the computing performance of the chips through software enhancement will be a key task for the AI chip design sector in China.

With software breakthroughs, theoretically, the 28nm node will be able to fabricate relatively high-level edge AI chips that can well serve AIoT, automotive electronics and other related applications, the sources emphasized.

In fact, for the vast edge computing market, what AI chips pursue is definitely not the extreme computing power but the optimal cost-effective solution. With its own robust software prowess, China will be able to achieve its self-sufficiency in edge AI chips, the sources remarked.

Jay Liu, Taipei; Willis Ke,
DIGITIMES Asia