I wonder if HBF will feed Hynix like a second HBM in a few years, or if it will just be a fleeting technology.
For now, Nvidia, which holds the sword, is also using similar technology, and the industry as a whole seems lukewarm.
I even imagine this technology unexpectedly becoming an essential material for quantum computers.
https://www.mk.co.kr/news/business/12116500
SK Hynix has released the first standard specification for HBF (High Bandwidth Flash), a next-generation storage technology. With Google and TensTorrent, which could become important customers in the future, joining the consortium, the commercial viability of HBF has increased. SK Hynix announced on the 4th that it will release the HBF standard specification created with US SanDisk at the 'FMS (Future of Memory and Storage) 2026' to be held at the Santa Clara Convention Center in California, USA until the 6th (local time). SK Hynix plans to introduce HBF as a core technology that will solve memory bottlenecks in the AI era through keynote speeches and panel discussions.
HBF is a new memory layer located between high bandwidth memory (HBM) and solid-state drives (SSD). Like HBM, it can quickly transfer data, but it can also significantly increase capacity based on NAND. As AI inference spreads and the amount of data to be handled increases rapidly, it is attracting attention as a technology that can expand both bandwidth and capacity simultaneously. HBF is not intended to replace HBM, but rather to be used together to enhance the overall efficiency of AI inference.
[Omitted]
Unlike SK Hynix, Samsung Electronics is relatively passive in adopting HBF. Although Samsung Electronics has also started developing HBF, it is expected to take some time for commercialization. The industry predicts that commercialization will be achieved around 2030.
One industry insider explained, "The fact that Nvidia, which leads the industry, has not expressed its intention to adopt it seems to be because Nvidia believes that 'Inference Context Memory Storage (ICMS)' led by them can still meet the demand for memory semiconductors."
ICMS is also a storage product based on NAND, and it is also located between HBM and SSD. It is already used in Nvidia's Vera Rubin platform and is cited as one of the factors driving up NAND flash prices.
