HBF secures both scalability and power efficiency for AI inference infrastructure
Two companies to form a joint workstream under OCP
SK Hynix has announced a global standardization strategy for High Bandwidth Flash (HBF), a next-generation memory solution targeting the era of artificial intelligence (AI) inference, together with SanDisk.
SK Hynix said on the 26th that on the 25th (local time) it held an event titled "HBF Spec. Standardization Consortium Kick-Off" at SanDisk's headquarters in Milpitas, California, United States.
SK Hynix explained, "Together with SanDisk, we will establish HBF as an industry standard and build a foundation on which the entire AI ecosystem can grow together," adding, "Under the world's largest open data center technology collaboration body, the Open Compute Project (OCP), we will form a dedicated workstream (a collaboration framework within OCP operated around a specific technology topic) with SanDisk to take charge of key tasks and begin full-fledged standardization work."
Recently, the AI industry has been rapidly shifting its focus from the "training" phase, in which large language models (LLMs) are created, to the "inference" phase, in which actual services are provided. As the number of users simultaneously accessing AI services surges, fast and efficient memory has become essential. However, with the existing memory architecture alone, it is difficult to meet both the large-scale data processing and power efficiency required in the inference phase.
HBF is being evaluated as an alternative that can overcome these limitations. Positioned as a new memory tier between HBM, an ultra-high-speed memory, and SSDs, large-capacity storage devices, HBF fills the gap between the outstanding performance characteristics of HBM and the high-capacity characteristics of SSDs, while securing both capacity scalability and power efficiency required in inference workloads. While conventional HBM is responsible for the highest level of bandwidth, HBF complements it in this architecture.
In particular, HBF is expected to enhance the scalability of AI systems while reducing total cost of ownership (TCO). The industry expects that demand for composite memory solutions, including HBF, will begin to expand in earnest around 2030.
SK Hynix and SanDisk plan to proactively drive rapid standardization and commercialization of HBF, leveraging their accumulated design and packaging technologies and mass production experience in the HBM and NAND fields.
Ahn Hyun, President and Chief Development Officer (CDO) of SK Hynix, said, "The core of AI infrastructure goes beyond performance competition in a single technology and lies in optimizing the entire ecosystem," adding, "Through the standardization of HBF, we will build a cooperative framework and present an optimized memory architecture for customers and partners in the AI era, thereby creating new value."
© The Asia Business Daily(www.asiae.co.kr). All rights reserved.
