SK hynix has partnered with SanDisk to reveal the first open specifications for High Bandwidth Flash, a new class of storage designed to bridge the gap between memory and SSDs. This technology uses standard chiplet interfaces to allow processors to access massive pools of flash memory at speeds nearing standard system memory. While the standard is now public through the Open Compute Project, commercial hardware is still in the planning stages.
Modern artificial intelligence applications require processing vast quantities of data, creating massive bottlenecks in current hardware setups. High Bandwidth Flash aims to solve this by creating a middle tier of memory that offers high speed transfers while keeping costs low through NAND flash technology. Capacity scales up to 512GB. This is achieved using 2 stack configurations of 8 high or 16 high NAND dies. Bandwidth is divided into 3 performance grades, scaling from 0.4TB/s to 3.0TB/s to meet different infrastructure needs.
To make integration easy, the architecture adopts the Universal Chiplet Interconnect Express connection standard. This open interface lets hardware designers link the memory directly with different processors, including graphics chips and standard central processors. Rather than keeping this proprietary, the partners released the standard through the Open Compute Project to encourage industry adoption. Google and Tenstorrent are already participating in the consortium to build out this ecosystem.
The announcement coincides with the opening of a major memory and storage conference in California this week. Alongside the new standard, SK hynix showcased a wafer for its 10th generation 375 layer 4D NAND technology. This new silicon boosts power efficiency by 2.5 times compared to the prior generation. Mass production of high capacity enterprise drives using this flash will start early next year. Company executives believe this design will reshape how systems handle heavy computational loads.
During the opening sessions of the event, executive vice president Kim Chun sung addressed the growing demands of autonomous agent systems. He pointed out that traditional structures cannot keep pace with modern data workloads. Discussing the potential of the new technology, Kim Chun sung stated:
Through HBF, SK hynix will expand the boundaries between memory and storage and contribute to building new architectures that enhance overall system efficiency.
With massive data requirements showing no signs of slowing down, the open standard could prove highly influential. Success will ultimately depend on how quickly other silicon design companies adopt the chiplet interface for their upcoming processors.
