The US government recently announced new export controls on high-tech memory chips, specifically high bandwidth memory (HBM), that are crucial for artificial intelligence (AI) applications. These restrictions target both American-made and foreign-produced chips, with the regulations becoming effective as of December 2, 2023. This move marks the latest in a series of restrictions put in place by the Biden administration aimed at curbing China's access to advanced technology that could bolster its military capabilities.

High bandwidth memory is a sophisticated form of semiconductor that surpasses older technologies, specifically dynamic random access memory (DRAM), by offering greater data storage capacity and faster transmission speeds. HBM chips are essential for a variety of applications, including graphics cards, high-performance computing systems, and especially in the rapidly expanding field of AI, most notably generative AI models. G Dan Hutcheson, the Vice Chair of TechInsights, explained the significance of memory in AI systems, stating, "The processor and the memory are two essential components to AI. Without the memory, it’s like having a brain with logic but not having any memory."

Experts predict that these export restrictions will significantly hinder China's progress in developing its own AI chips, particularly HBM technology. Despite efforts to cultivate domestic capabilities in this area, China's production of HBM is currently lagging behind leading manufacturers in South Korea, such as SK Hynix and Samsung, as well as the American company Micron. Jeffery Chiu, CEO of Ansforce, remarked, "What the US export restrictions would do is cut China’s access to HBM of better quality in the short run. In the long run though, China will still be able to produce them independently, albeit with less advanced technologies."

The leading Chinese manufacturers, Yangtze Memory Technologies and Changxin Memory Technologies, are reportedly increasing their production capacities to meet the demand for HBM, which remains vital for the nation's goal of technological self-sufficiency.

The importance of HBM lies in its ability to manage complex computing tasks necessary for seamless AI application functionality. HBM chips not only store more data but also facilitate quicker data transfer, resulting in improved performance for AI models. Hutcheson likened the efficiency of HBM to that of a multi-lane highway, stating, "It’s like the difference between a two-lane highway and a hundred-lane highway. You just don’t get traffic jams."

In the current global market, the dominance of HBM technology is held by a few key players. As of 2022, SK Hynix commanded approximately 50% of the HBM market share, followed by Samsung with 40% and Micron with 10%. These figures are expected to remain relatively stable in the coming years. Micron has set an ambitious target of increasing its market share to between 20% and 25% by 2025, according to a source from Taiwan’s Central News Agency.

The manufacturing of HBM is a highly complex process that involves stacking multiple memory chips in a precise manner. This results in a significantly higher price compared to conventional memory chips. The intricacy of creating HBM lies in the need for extremely thin layers of chips and precise engineering to ensure that electrical connections between the chips function correctly. Chiu described the difficulty with HBM production, saying, "Each of those memory chips need to be ground to as thin as the height of half a strand of hair before they are stacked together."

In summary, the recent US export restrictions on high bandwidth memory chips are set to have profound implications for AI development in China, potentially setting back advancements while the nation attempts to forge its own path to tech self-sufficiency. The stakes are high in the race for AI dominance, and the market for memory technologies is expected to evolve as companies adapt to changing regulations and competitive pressures in the sector.

Source: Noah Wire Services