South Korean chipmakers Samsung Electronics and SK Hynix are stuck in a perpetual game of memory cat-and-mouse with their global rivals, all vying for dominance in the next frontier of AI: high-bandwidth memory (HBM) is dead, and 3D XPoint (3DXP) is dying.
Chris Miller, the author of Chip War: The Fight for the World’s Most Critical Technology, recently spoke with Kyunghyang Daily News about the state of the industry. Miller stresses that both companies need to move quickly to the next big thing: 3D stacked DRAM.
The End of HBM
High-bandwidth memory, or HBM, was touted as the future of AI computing when it emerged a few years ago. By stacking memory on top of a microprocessor, HBM offered incredibly fast data transfer speeds, up to 256 GB/s. But NVIDIA’s decision to axe its HBM2-based graphics cards in favor of more practical and affordable technologies should be seen as a harbinger of HBM’s impending doom.
The Rise of 3D XPoint and the Fall of 3DXP</hassistant
3D XPoint, also known as Intel-Micron Flash Technologies (IMFT) Optane, has been a significant player in high-performance computing for several years. However, IMFT’s inability to increase 3DXP’s capacity and the relatively high cost of the technology have led to its slow decline.
“Intel’s Optane P5800X is the last gasp of 3DXP technology,” says Miller. “It won’t be long before 3DXP becomes obsolete, just like HBM, making way for the next big thing.”
The Next Frontier: 3D Stacked DRAM
3D stacked DRAM, also known as 3D XStack, is the next major development in AI memory technology. By stacking DRAM layers on top of each other, companies can create high-capacity memory chips with incredibly fast transfer speeds, rivaling those of HBM.
Miller emphasizes that Samsung and SK Hynix need to invest heavily in 3D stacked DRAM to stay competitive in the AI market. “If they don’t move quickly, they risk falling behind their rivals, including Chinese companies like Huawei and Xiaomi.”
What this means for consumers and AI developers is that we can expect faster, more efficient computing experiences in the future. As companies continue to push the boundaries of AI memory technology, we can expect to see even more incredible advancements in fields like deep learning, natural language processing, and computer vision.
