SK Hynix and SanDisk have jointly unveiled the first standard specification for HBF, a next-generation storage technology, through the Open Compute Project. [SK Hynix]
SK Hynix and SanDisk have jointly unveiled the first standard specification for HBF, a next-generation storage technology, through the Open Compute Project. [SK Hynix]

SK Hynix announced Tuesday that it has jointly unveiled with SanDisk the first standard specification for HBF (High Bandwidth Flash), a next-generation storage technology based on NAND flash memory.

Like HBM (High Bandwidth Memory), which stacks DRAM vertically, HBF stacks NAND chips in the same fashion and is gaining importance as AI inference workloads proliferate. The move signals SK Hynix's strategy to establish an early lead in HBF standards — and carry its dominance in HBM into the emerging HBF market.

The specification developed jointly with SanDisk defines capacity, bandwidth and interconnect interface. On capacity, the standard covers two stack configurations — 8-die and 16-die — supporting up to 512GB. Bandwidth is divided into three grades (Grade 1 through 3), spanning roughly 0.4TB/s to 3.0TB/s.

The companies also chose an open interconnect approach to lower the barrier for broad industry participation. By adopting UCIe (Universal Chiplet Interconnect Express), an open-standard interface, HBF can connect flexibly to different processor types, including GPUs and CPUs.

The specification was released through the Open Compute Project (OCP), the world's largest open data center technology consortium, making it available for the entire industry to use.

SK Hynix has been actively developing HBF as a candidate for a new memory tier. The company began standardization collaboration with SanDisk in August last year, then formally established an HBF spec standardization consortium in February, launching concrete development work through a kickoff event.

The need for HBF is growing as AI shifts from the training phase to inference. The commercialization of agentic AI is driving a sharp increase in data volumes to be processed, along with rising demands for speed and efficiency.

HBF occupies a new memory tier between HBM and SSDs. It combines the fast data transfer of HBM with the power efficiency of NAND — which retains data even when powered off — and can deliver far greater capacity than DRAM at a more competitive price.

SK Hynix plans to use the specification release as a springboard to expand HBF adoption in the AI storage market and build out the broader ecosystem. Google and Tenstorrent are currently members of the HBF consortium. The company intends to widen participation through open collaboration while simultaneously advancing the technology's maturity and market readiness.

Against this backdrop, SK Hynix's Kim Cheon-seong, head of solution development, and Kang Uk-seong, head of next-generation product planning, are delivering a joint keynote address Tuesday (local time) at FMS (Future of Memory and Storage) 2026, held at the Santa Clara Convention Center in California.

The two executives will present on the theme of "Efficient AI Infrastructure in the Age of AI Agents: A Tiered Memory Approach," with HBF set to be introduced at the session. They will outline a next-generation architecture based on tiered memory — integrating DRAM-based HBM and NAND-based HBF into a single optimized framework — as a new direction for AI infrastructure.

On Thursday, Im Eui-cheol, SK Hynix's solution AT lead, Google DeepMind researcher Xiaoyu Ma, and SanDisk Vice President Rajeev Nagaviraba will hold a panel discussion titled "Breaking the Memory Wall with HBF." The three speakers will examine fundamental shifts in memory architecture across AI infrastructure and explore HBF's role and potential as a solution to those challenges.

SK Hynix will also make the world premiere at FMS 2026 of the wafer and product for its 10th-generation (V10) 375-layer 4D NAND, currently under development. The chip delivers 2.5 times better performance per watt than the previous generation and is optimized for data center environments. The company plans to begin mass production of high-performance, high-capacity eSSD products based on the new NAND early next year.

"The rapid spread of AI is demanding a fundamental redesign of data processing architecture," Kim said. "Through HBF, we will contribute to building a new architecture that extends the boundary between memory and storage and improves overall system efficiency."


jeongwan@heraldcorp.com