The U.S. government has introduced new export controls targeting high-tech memory chips used in artificial intelligence (AI) applications, specifically high bandwidth memory (HBM). These rules apply not only to U.S.-made HBM but also to foreign-produced chips, as the global demand for AI technology continues to grow.

What is High Bandwidth Memory (HBM)?

High Bandwidth Memory (HBM) is an advanced memory technology that stacks multiple memory chips to enable faster data transmission and higher storage capacity compared to older memory technologies like DRAM (Dynamic Random Access Memory).

• Usage: HBM is critical in AI applications, graphic cards, data centers, autonomous vehicles, and high-performance computing systems.
• AI Dependence: HBM chips are essential for running advanced AI models, especially those involving large language models (LLMs) used in AI processors like Nvidia’s and AMD’s graphics processing units (GPUs).

As G. Dan Hutcheson, vice chair of TechInsights, explains, “Without memory, it’s like having a brain with logic but not having any memory.” This highlights how crucial HBM is to AI operations.

Why is the U.S. Blocking China’s Access to HBM?

The export restrictions imposed on December 2 are part of an ongoing effort by the U.S. government to prevent China from accessing critical technologies that could provide a military advantage. China’s growing capabilities in semiconductor technology, particularly HBM, have triggered these measures.

• Impact on China’s AI Development: While China currently lags behind countries like South Korea and the U.S. in HBM production, it is rapidly developing its own capabilities. These export restrictions are expected to delay China’s advancements in AI chip technology but are unlikely to halt them entirely in the long run. Chinese companies like Yangtze Memory Technologies and Changxin Memory Technologies are ramping up HBM production to reduce their dependency on foreign suppliers.
• Retaliation by China: In response, China has imposed restrictions on exports of germanium and gallium, materials essential for semiconductor production, signaling ongoing trade tensions.

Why is HBM So Important for AI?

HBM’s enhanced storage and speed make it indispensable for AI systems, which require the processing of large volumes of data quickly and without delays.

• Storage Capacity and Speed: HBM offers much more storage and faster data transfer compared to traditional memory chips, crucial for smooth AI performance.
• Better Performance: The large bandwidth in HBM chips can be compared to a multi-lane highway, where more lanes mean less congestion and faster data transmission, enhancing AI model performance.

Top Manufacturers of HBM

Three major companies dominate the global HBM market:

• SK Hynix: Holds the largest market share (50% as of 2022).
• Samsung: Follows closely with a 40% market share.
• Micron: A smaller player with 10% market share, though it plans to increase its HBM market share to 20-25% by 2025.

Together, SK Hynix and Samsung control around 95% of the HBM market, with Micron working to expand its influence in the coming years.

The Complex Manufacturing of HBM

HBM is made by stacking multiple memory chips, which requires highly specialized manufacturing techniques.

• Advanced Packaging: Each memory chip must be thinned down to a fraction of a millimeter and precisely aligned for stacking, making the process incredibly complex and expensive.
• Precision and Failure Risks: Creating these chips involves significant risks, as any misalignment or imperfection could cause failure. Advanced packaging ensures that data can pass through these stacked chips effectively.

HBM technology is at the heart of the AI boom, powering everything from AI models to autonomous vehicles. As the U.S. seeks to limit China’s access to this critical technology, it has imposed strict export controls that may temporarily slow China’s progress in AI chip development. However, China’s ability to manufacture HBM in the long term remains a key concern for the U.S. as the global race for AI dominance intensifies.