SK Hynix and SanDisk Introduce a New Type of Intermediate Memory

cryptonews.ru2026-08-04 tarihinde yayınlandı2026-08-04 tarihinde güncellendi

Özet

On August 4, 2026, at the FMS conference in California, SK Hynix and SanDisk unveiled the first specifications for HBF, a new memory technology designed to bridge the gap between high-speed HBM memory and SSDs. Key HBF specs include a capacity of up to 512GB, bandwidth ranging from ~0.4 TB/s to 3.0 TB/s, and connectivity via the open UCIe interface for compatibility with various CPUs/GPUs. The specifications are released through the Open Compute Project, positioning HBF as an open industry standard rather than a proprietary one. This introduction suggests future AI servers will likely adopt a four-tier memory hierarchy: 1) HBM for the hottest data, 2) CXL for accelerating classic RAM, 3) HBF as a high-bandwidth, high-capacity layer using cost-effective NAND, and 4) NVMe SSDs for cold storage. In LLM infrastructure, HBF is intended to store large AI model context and weights that don't fit into the more expensive HBM and CXL memory. The HBF standardization partnership began in August 2025, with a consortium including Google and Tenstorrent launching in February 2026. Separately, SK Hynix also previewed its developing 375-layer 4D NAND (V10) at FMS, claiming a 2.5x performance-per-watt improvement, with mass production of eSSDs planned for early next year.

On August 4th at the opening of the FMS 2026 conference in California, SK Hynix, together with SanDisk, published the first specifications for HBF—a technology aiming to serve as an intermediate layer between high-speed HBM memory and SSD storage.

Key technical characteristics of HBF:

  • Capacity — up to 512 GB (in 8- and 16-die NAND stack configurations);
  • Bandwidth — from ~0.4 TB/s to 3.0 TB/s;
  • Connection interface — UCIe, an open standard ensuring compatibility with GPUs and CPUs from different manufacturers.

The specifications are open-sourced via the Open Compute Project—a community for data center technologies. This positions HBF as an industry-wide open standard rather than a proprietary development.

Following the announcement of HBF, modern AI servers will likely have a four-tier memory hierarchy:

  1. HBM — processes "hot" data directly in the compute core (highest speed, minimal capacity, extremely high cost).
  2. CXL — accelerates traditional RAM with latencies on the order of tens to hundreds of nanoseconds, while maintaining full compatibility with CPUs/GPUs.
  3. HBF — fills the gap between RAM and SSDs. By integrating via the UCIe bus directly with the processor, these devices will provide speeds close to HBM but with large capacity based on inexpensive NAND architecture.
  4. NVMe SSDs — store "cold" data.

In LLM infrastructure, a key challenge is storing the vast context and model weights during computation (inference):

  • CXL handles data requiring instant access and frequent rewriting with ultra-minimal latency;
  • HBF takes on storing gigabytes of context caches and AI model weights that could not fit into HBM and CXL due to their high cost and capacity limitations.
Silicon Curtain

The partnership between SK Hynix and SanDisk to standardize HBF began in August 2025. The consortium was launched in February 2026 and also includes Google and Tenstorrent.

Also at the FMS 2026 exhibition booth, SK Hynix announced a demonstration of 375-layer 4D NAND (V10), which is under development. According to the company, the memory's performance has increased 2.5 times per watt of energy consumed compared to the previous generation. Mass production of eSSDs is planned for early next year.

According to Yahoo Finance, SK Hynix shares have risen more than 330% since the beginning of 2026, but have lost about half their value since reaching a peak in June.

SK Hynix stock chart since the beginning of 2026. Source: Yahoo Finance.

Recall that in July, the Ashenbrenner fund lost the majority of its stock portfolio.

end-content

İlgili Sorular

QWhat is HBF memory and who announced it?

AHBF (Hybrid Bonding Flash) is a new type of intermediate memory technology announced by SK Hynix and SanDisk at the Flash Memory Summit (FMS) 2026. It aims to fill the gap between high-speed HBM memory and SSD storage.

QWhat are the key technical specifications of the announced HBF technology?

AThe key specifications include: a capacity of up to 512GB (with 8 or 16 NAND dies per stack), a bandwidth ranging from ~0.4 TB/s to 3.0 TB/s, and it uses the UCIe (Universal Chiplet Interconnect Express) interface for connectivity to ensure compatibility with GPUs and CPUs from different manufacturers.

QHow does HBF fit into the proposed four-tier memory hierarchy for AI servers?

AIn the proposed four-tier AI server memory hierarchy: 1) HBM handles 'hot' data in the compute core. 2) CXL accelerates classic DRAM. 3) HBF fills the gap between DRAM and SSD, offering HBM-like speeds with large, cost-effective NAND capacity. 4) NVMe SSDs store 'cold' data.

QWhat is the role of the Open Compute Project (OCP) regarding HBF?

AThe HBF specifications were opened through the Open Compute Project (OCP), a data center technology community. This positions HBF as an open, industry-wide standard rather than a proprietary development.

QWhat other memory technology did SK Hynix announce at FMS 2026?

AAt FMS 2026, SK Hynix also announced a demonstration of its 375-layer 4D NAND (V10) under development. The company claims it offers 2.5 times the performance per watt compared to the previous generation, with mass production of eSSDs planned for early next year.

İlgili Okumalar

İşlemler

Spot
活动图片