AI BriefingHigh-Bandwidth Flash, Efficient Storage for Model Weight Storage
Key point
By stacking NAND flash using 3D packaging technology, it realizes high-bandwidth memory suitable for storing LLM weights.
Details
Amid growing importance of HBM (High-Bandwidth Memory) and DRAM driven by the explosive demand for LLMs, High-Bandwidth Flash (HBF) technology is gaining attention as a way to overcome the limitations of existing NAND flash.
NAND flash has non-volatile characteristics and high density, but has the drawback that its read/write speed is markedly slower than DRAM or HBM. HBF aims to solve this problem by applying the 3D packaging and vertical stacking technology that made HBM successful to NAND flash.
Key features and outlook are as follows:
- Performance improvement: According to Sandisk's roadmap, the 1st-generation HBF will stack up to 16 NAND flash chips to provide up to 512GB of capacity and 1.6TB/s of read bandwidth.
- Generational progress: Bandwidth is planned to expand further with the 2nd generation (2TB/s) and 3rd generation (3.2TB/s).
- Market outlook: HBF is expected to be commercialized within at least 1 year, and by offering far higher bandwidth than existing NVMe storage, it is expected to contribute to high-performance tasks such as storing LLM model weights.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.
Related
Next-Generation Memory Technologies Preparing for the Post-HBM Era: The Rise of Magnonics and Vertical FeRAMMicron Invests $10 Billion in AI Memory Research
The History of Memory Prices from 1960 to 2026
Hierarchical Memory Architecture Needed to Resolve AI Inference Bottlenecks