Sandisk has announced advancements in memory technology with the introduction of its High Bandwidth Flash (HBF) concept, designed to challenge existing High Bandwidth Memory (HBM) solutions. This new NAND-based alternative aims to provide significant improvements in capacity and cost efficiency for AI workloads.
The HBF technology seeks to match HBM bandwidth while offering 8 to 16 times the storage capacity at a comparable price. Sandisk’s proprietary stacking technology allows for efficient high-density stacking, making it feasible to achieve up to 4TB of memory in a single setup.
HBF combines BiCS technology with CBA wafer bonding, enabling ultra-low die warpage. This architecture, developed over the past year, incorporates feedback from key players in the AI sector. The design utilizes multiple HBF core dies linked through TSV (Through-Silicon Via) and micro bumps, interfacing with various processing units including GPUs and CPUs.
While HBF is not a direct replacement for HBM, it shares the same electrical interface, requiring only minor protocol adjustments. A comparison shows that a system using HBF can increase memory capacity significantly, with optimized setups reaching up to 4TB.
Sandisk’s roadmap indicates a focus on enhancing capacity, read bandwidth, and energy efficiency across multiple generations of HBF technology. The first generation establishes a baseline, while subsequent generations aim for improvements in capacity and read bandwidth, albeit with slight declines in energy efficiency.
As Sandisk enters the competitive landscape dominated by companies like Samsung and SK Hynix, it plans to establish an open standard ecosystem and a technical advisory board to facilitate collaboration within the industry.
For more details, visit the original article on TechRadar.