SK hynix Unveils Next-Generation Memory Roadmap to Address AI Bottlenecks

SK hynix is advancing next-generation memory technologies to tackle rising AI infrastructure bottlenecks, according to a technology roadmap unveiled at SEDEX 2026. The company aims to bolster its HBM leadership while pioneering 3D Stacked DRAM and High Bandwidth Flash (HBF) to boost performance and energy efficiency. SK hynix's Chief Development Officer, Ahn Hyun, emphasized the necessity of improving the entire memory hierarchy during his keynote address.

He pointed out that AI inference demand is skyrocketing in three areas: task complexity, context length, and inference request volume, creating a steep rise in AI computing infrastructure needs. By 2030, SK hynix projects AI infrastructure performance requirements could expand about 20 times compared to 2026. As AI models process larger datasets, bottlenecks emerge in data storage and transfer, particularly with key-value caches needing more memory capacity as context lengths increase.

Ahn explained that the conventional "memory wall" is shifting in the AI era due to disparities in memory tier performance and escalating energy costs for data movement. To address these challenges, SK hynix plans to develop an integrated memory architecture for seamless interconnectivity across various memory tiers, enhancing AI infrastructure scalability.

The firm will continue refining HBM technology, focusing on HBM1 to HBM4E generations, and explore custom HBM with integrated memory controllers and customer-specific logic. Apart from HBM, SK hynix is developing 3D Stacked DRAM for faster data transfer and energy efficiency by minimizing the physical distance between computing components and memory.

HBF, a potential solution for AI inference's growing capacity demands, stacks NAND flash memory using through-silicon via (TSV) and high-speed input/output technologies. HBF aims to deliver high bandwidth with significantly larger storage capacity, potentially reducing reliance on DRAM and SSD access. SK hynix is also investigating CXL and Storage-Next to diversify memory and storage architectures.

At SEDEX 2026, the company showcased a wide range of memory and storage solutions for AI infrastructure, personal computers, and mobile devices, including the 256GB 3DS RDIMM, 128GB MRDIMM, and 64GB RDIMM. The exhibition also featured HBM and HBF architecture models, HBM development displays, and an NVIDIA Vera Rubin superchip.