SK hynix, together with SanDisk, unveiled the first standard specification for HBF (High Bandwidth Flash), a next-generation AI memory technology. As the performance gap between HBM (high bandwidth memory) and SSD widens with the spread of generative artificial intelligence (AI), the companies proposed a new storage technology that bridges the two memory tiers.
SK hynix said on the 4th (local time) that it unveiled the HBF standard specification co-developed with SanDisk at FMS 2026, the world's largest flash memory event, which opened in Santa Clara, California.
HBF is a new memory tier positioned between HBM and SSD. It combines HBM-level high data processing performance with NAND flash–based large-capacity storage to reduce memory bottlenecks that arise in AI inference and large-scale data processing environments.
The standard released this time supports up to 512GB of capacity based on a structure that stacks NAND dies in 8 and 16 layers. It defines three levels of data transfer speed from about 0.4TB/s up to 3.0TB/s, and adopts the industry-standard UCIe (Universal Chiplet Interconnect Express) as the interface for consolidation with the processor.
SK hynix has been promoting the standardization of HBF with SanDisk, and Google and Tenstorrent are also participating in the related consortium. The company plans to expand the AI storage ecosystem based on an open standard.
Along with this, SK hynix also revealed a tiered memory strategy at FMS 2026. At its booth, it showcased for the first time a 10th-generation 375-layer 4D NAND wafer and products applying it, and said it plans to begin mass production of high-performance, high-capacity enterprise SSDs (eSSD) early next year based on this.
Kim Cheonseong, an institutional sector head at SK hynix, said, "HBF is a new architecture that expands the boundary between memory and storage to improve efficiency across systems," adding, "We will continue to broaden memory solutions optimized for the AI era."
Earlier, in Apr., SK hynix also disclosed at the TSMC Technology Symposium that it would expand its next-generation AI memory lineup beyond HBM to include HBF and three-dimensional (3D) stacked DRAM.