Thursday, August 6, 2026
DarkSubscribe
AI Infrastructure · News & Analysis
HomeChips & HardwareReport
Chips & Hardware · Report

SK hynix and SanDisk announced the first High-Bandwidth Flash (HBF) storage standard with support from Google and Tenstorrent, designed to address memory bandwidth bottlenecks in large-scale AI inference.

New memory standard removes flash bandwidth constraints and enables terabyte-scale storage architectures required for frontier-model inference and agentic AI workloads.
Trade pressSlicast · August 4, 2026 · US · Source: Google News
importance 92

Six months after SK hynix and SanDisk partnered to drive High Bandwidth Flash (HBF) standardization, the consortium unveiled the first HBF standard specifications at the Future of Memory and Storage (FMS) 2026 conference in Santa Clara, California (August 4–6). The inaugural specification supports capacities up to 512GB using 8-high and 16-high NAND stack configurations, with three bandwidth tiers (Grade 1–3) delivering approximately 0.4TB/s to 3.0TB/s.

SK hynix is accelerating ecosystem adoption by rallying major backers including Google DeepMind and Tenstorrent. The specification, disclosed through the Open Compute Project (OCP), adopts the industry-standard UCIe interconnect, enabling HBF integration with a broad range of processors—GPUs and CPUs alike. It further defines interface and electrical characteristics, HBF die-stacking reliability, packaging guidelines, and software I/O requirements.

Unlike HBM, which stacks DRAM dies to maximize bandwidth, HBF stacks NAND flash to create a new memory tier positioned between HBM and SSDs. HBF emerges as a complementary technology to HBM rather than a competitor, with potential to address HBM's high cost and limited capacity. While NVIDIA has not yet announced adoption plans, industry sentiment remains positive. The interest shown by leaders such as Google and Tenstorrent suggests willingness to explore HBF, though specific use cases remain undetermined.

The long-term architecture is expected to combine HBM for ultra-high-speed computation with HBF for high-density data storage, creating a hybrid memory hierarchy positioned as a key enabler of large-scale AI commercialization. HBF technology is expected to reduce total cost of ownership (TCO) while increasing AI system scalability. Industry forecasts anticipate demand for complex memory solutions, including HBF, will accelerate around 2030.

SanDisk plans to introduce HBF prototypes in the second half of 2026, with Japan as a leading candidate for production. Pilot production is expected to complete in the second half of the year and begin operation around year-end, with commercialization targeted for 2027.

SK hynix also unveiled the world's first 375-layer, 10th-generation (V10) 4D NAND, with mass production scheduled for early 2027. The chip delivers a 2.5x improvement in performance per watt, directly addressing one of data centers' biggest challenges: power consumption. With 375-layer NAND-based enterprise SSDs set to enter production early next year, SK hynix will complete what it describes as the industry's only full-stack AI memory portfolio spanning DRAM (HBM), the intermediate HBF tier, and storage (eSSD).

Read the original
SK hynix and SanDisk announced the first… · Slicast