[Edaily Reporter SOYEON KIM ] SamsungElectronics and SK hynix will showcase a wide range of next-generation technologies designed to resolve AI memory bottlenecks at “FMS 2026,” the world’s largest conference dedicated to memory and storage. As the AI inference market expands rapidly and the volume of data to be processed surges, competition for next-generation AI memory technologies that offer both bandwidth and capacity scalability is heating up.

SamsungElectronics and SK hynix will participate in FMS 2026, taking place from the 4th to the 6th (local time) at the Santa Clara Convention Center in California, USA, to unveil their next-generation AI memory technologies and strategies. Through keynote speeches, panel discussions, and exhibitions, the two companies plan to present a blueprint for the memory infrastructure needed in the AI era.

On the 4th, SK hynix, in collaboration with SanDisk, unveiled the first standard specifications for High-Bandwidth Flash (HBF), a next-generation storage technology. HBF is a new memory tier positioned between High-Bandwidth Memory (HBM) and solid-state drives (SSDs). It is characterized by its ability to provide data processing speeds on par with HBM while significantly expanding capacity through its NAND flash-based architecture.

The industry-standard Chiplet Interconnect (UCIe) was adopted as the interface connecting HBF to processors. This enables flexible connectivity between HBF and various processors, such as graphics processing units (GPUs) and central processing units (CPUs). These standard specifications were released through the Open Compute Project (OCP), the world’s largest open data center technology consortium. The strategy is to accelerate the expansion of the AI storage ecosystem by establishing open standards that the industry can use collectively. Currently, the HBF Consortium includes participants such as Google and Tenstorrent.

On this day, Kim Cheon-sung, Head of Solution Development at SK hynix, and Kang Wook-sung, Head of Next-Generation Product Planning, will deliver a joint keynote speech. The presentation topic is “The Era of Agent AI: Strategies for Efficiently Implementing AI Infrastructure Based on Hierarchical Memory.” The two speakers will propose “hierarchical memory systems” as a key strategy for resolving AI data bottlenecks. They plan to introduce a method for constructing a memory hierarchy by combining 3D-stacked DRAM for computing units (xPUs), HBF, and memory pooling based on Compute Express Link (CXL).

SamsungElectronics will also unveil next-generation memory technologies designed for the AI era. Senior Vice Presidents Song Taek-sang and Lee Jin-yeop will present the company’s next-generation memory and storage strategies required for the AI era during their keynote speeches. Leveraging its capabilities as an Integrated Device Manufacturer (IDM)—encompassing memory, foundry (semiconductor contract manufacturing), and advanced packaging—SamsungElectronics will present integrated solutions that reduce bottlenecks in AI infrastructure and enhance scalability.

At this FMS, SamsungElectronics will showcase zHBM, a next-generation 3D stacked memory architecture, following the introduction of zNAND. Unlike conventional methods where memory is placed next to the GPU, zHBM features a 3D structure in which memory is directly stacked on top of the GPU. Since implementing this requires the organic integration of design, foundry, DRAM, and advanced packaging technologies, SamsungElectronics plans to leverage its IDM competitiveness to lead the next-generation HBM market.

SK hynix will also unveil for the first time the wafers and products of its 10th-generation (V10) 375-layer 4D NAND, currently under development. The 375-layer 4D NAND features 2.5 times the power efficiency of the previous generation, and the company plans to begin mass production of high-performance, high-capacity enterprise SSDs (eSSDs) incorporating this technology early next year.