Sandisk is significantly accelerating the commercialization of High Bandwidth Flash (HBF). This new technology, regarded by the industry as a critical storage tier for the AI inference era, is attracting full-scale participation from Sandisk, SK Hynix, and Samsung Electronics, marking the official start of the next wave of structural competition in the memory industry following HBM. According to a report from The Bell cited by Korean media ETNews on April 13, Sandisk has begun engaging with materials, component, and equipment partners to build an ecosystem for an HBF prototype production line, with plans to introduce a prototype product in the second half of this year. The pilot production line is expected to be completed and operational in the second half of 2026, with commercialization targeted for 2027. The report cites informed sources stating that, with sample production accelerating, Sandisk may move its entire HBF development schedule forward by approximately six months, indicating a strong intent to capture first-mover advantage in the market. The advancement of HBF has profound implications for the AI hardware supply chain. By incorporating Through-Silicon Via (TSV) stacking packaging into NAND flash, the technology can provide approximately ten times the storage capacity of HBM while maintaining high bandwidth, specifically designed to bridge the storage tier gap between HBM and SSDs. As AI workloads increasingly shift towards the inference phase, the strategic value of addressing this technological gap is becoming more pronounced. Sandisk possesses design, packaging, and mass production experience in both HBM and NAND, providing a technical foundation for its push into HBF. According to ETNews, because HBF and HBM share highly similar process flows, the equipment, materials, and component ecosystem developed for HBM production lines is expected to maintain a technological lead in the HBF domain. Specifically, TSV process equipment for signal transmission between stacked NAND chips, bonding materials for die attach, and related equipment are anticipated to remain dominated by companies that have already established strong competitiveness in the HBM market. This continuity in the technological pathway suggests that the potential market for existing HBM equipment suppliers will directly expand with the progression of HBF. New developments are also emerging on the materials front. JK Materials, a subsidiary of South Korea's Hanul Materials Science, recently announced it has completed development of high-performance polymers for HBF and has begun supplying major customers. The report indicates that the company's high-performance KrF polymer is a key chemical material required for stacking hundreds of layers of NAND flash. In terms of the competitive landscape, SK Hynix and Sandisk are promoting standardization through an OCP working group, attempting to establish a first-mover advantage at the specification level. Industry observers previously anticipated that Samsung Electronics and Sandisk plan to integrate HBF into products from NVIDIA, AMD, and Google by late 2027 or early 2028. Regarding Samsung, according to The Chosun Ilbo, Samsung Electronics began HBF research in the early 2020s and has recently intensified its acquisition of a series of HBF-related patents, actively expanding its technology portfolio. Although it has not made a public announcement similar to SK Hynix, its patent activities indicate that Samsung is steadily advancing its布局 in the HBF arena. Kim Joungho, a professor at the Korea Advanced Institute of Science and Technology (KAIST) renowned as the "father of HBM," recently stated that the implementation timeline for HBF is significantly ahead of previous expectations. He predicts HBF will see widespread adoption around the time HBM6 is introduced and is expected to surpass HBM in market size by approximately 2038. He also noted that, thanks to the process and design experience accumulated from HBM, the commercialization cycle for HBF will be considerably shorter than the development journey of HBM was. The core value of HBF lies in creating a new storage tier positioned between ultra-high-speed HBM and high-capacity SSDs. SK Hynix points out that in AI inference scenarios, as user bases expand rapidly, existing storage architectures face a structural dilemma: balancing high-capacity data processing with power efficiency. HBM offers excellent bandwidth but limited capacity, while SSDs provide ample capacity but insufficient read/write speeds. HBF is designed specifically to bridge this gap. By vertically stacking NAND flash memory, it maintains high bandwidth while offering roughly ten times the storage capacity of HBM. In a system architecture, HBM handles high-bandwidth demands, while HBF acts as a supporting layer for capacity expansion, working together to meet the dual requirements of massive data processing and power efficiency for AI inference. " The key to AI infrastructure lies in optimizing the entire ecosystem, moving beyond competition based solely on the performance of individual technologies," said Ahn Hyun, President and Chief Development Officer at SK Hynix. SK Hynix also stated that HBF becoming an industry standard would lay the foundation for the collective growth of the entire AI ecosystem.
Comments