Research Insight: US export curbs on chips to impact China's data center infrastructure

Eric Chen, Taipei; Jingyue Hsiao, DIGITIMES Asia 0

Credit: AFP

The US has announced additional chip rules against China. DIGITIMES Research posits that these measures will heighten the challenges associated with China's acquisition of AI and high-performance HPC chips, subsequently complicating their efforts to establish AI and supercomputer systems, potentially impacting China's data center development.

To close loopholes left by the export curbs unveiled in 2022, the US Commerce Department announced updated rules to prevent China from acquiring advanced chips and developing supercomputers on October 17.

In October 2022, the US announced control measures against China based on chip processing power and transmission bandwidth to stop China from using US chips or chips using US technologies to enhance China's performance in AI and supercomputer systems, with potential military applications, to reduce the threat to US national security. On October 17, the US also announced the Interim Final Rule (IFR) as last year, allowing the implementation of certain rules without the procedure of collecting public comments, which are used in situations that require swift action, especially in national security issues.

DIGITIMES Research observes that although the US has removed the interconnect bandwidth parameter in the updated measures, the chip processing power threshold remains at 4,800 TOPS and more. The US also added the performance density threshold parameter to limit China from purchasing many AI chips to build computing power of 4,800 TOPS and more for AI systems, supercomputers, or data centers. Besides, the IFR creates a new License Exception Notified Advanced Computing for consumer-grade chips with AI capabilities below the restricted threshold of 4800 TOPS, which means the US export curbs have reached beyond total processing power and into application areas.

The curbs will restrict HPC chip suppliers, such as Nvidia, AMD, and Intel, from shipping lower-end chips compared to their most premium products to China-based customers, which may impact Nvidia's A800 and H800 and Intel's Gaudi 2.

In addition to restricting the export of high-performance chip technology to China, the US has also added emerging China-based GPU suppliers, including Biren Technology and Moore Threads and their subsidiaries, to the entity list, further constraining these companies from developing high-end GPUs using US technology, thereby curtailing China's development of high-performance computing chips and heading off the possibility of these companies using overseas foundries for advanced processes to produce high-performance chips.

IFR also emphasizes that foundries should strengthen due diligence to understand their products' ultimate use and end customers. Additionally, the chips' total processing power must be summed if they are packaged together. Furthermore, the US closed a loophole by preventing companies from exporting US-controlled technology and chips to China through regions like Macau.

The IFR definition of advanced processes remains at 16/14nm and below for logical ICs, 18nm and below for DRAM, and 128-layer and more for NAND Flash. Adjustments have been made for semiconductor equipment exports, and restrictions on US individuals or US companies supporting the production of advanced chips in China continue.

The curbs on semiconductor equipment are similar to the ones proposed by Japan in July 2023, which include restrictions on equipment designed for silicon or silicon germanium (SiGe) epitaxial growth, equipment designed for dry etching, wet chemical processing and having the largest silicon germanium to silicon etch selectivity ratio of greater than or equal to 100:1, and spatial ALD equipment, etc.

Furthermore, the additional measures also prevent US companies from supporting advanced semiconductor production in China, especially during the chip development or early production stages.

The US not only seeks to control high-performance chips that previously fell into the gray area (i.e., did not meet control thresholds) but also increases restrictions on sales and shipments to certain countries and regions, an attempt to address the issue of companies circumventing US control policies to supply products or technology to China. Additionally, Chinese GPU designers have been added to the entity list to cut off their chip design and production capabilities.