As the race to eliminate memory bottlenecks in AI infrastructure intensifies, Kioxia and SanDisk have pushed the boundaries of storage density with their next-generation 10th Gen Quad-Level-Cell (QLC) 3D flash memory. Unveiled at the Future of Memory and Storage (FMS) 2026 conference, the BiCS10 technology delivers a massive leap over the 8th generation, increasing bit density by up to 60% and surpassing 37 Gb/mm².
CBA Technology and Power Efficiency
The breakthrough is driven by the CMOS directly Bonded to Array (CBA) architecture. By fabricating CMOS logic and memory arrays on separate wafers before bonding them with high-precision alignment, the companies have significantly improved read/write bandwidth. Furthermore, the adoption of a Toggle DDR6.0 interface enables speeds up to 4.8 Gb/s, while simultaneously lowering I/O power consumption.
KIOXIA and Western Digital Announce Newest 3D Flash Memory ... — https://international.electronica-azi.ro/kioxia-and-western-digital-announce-newest-3d-flash-memory/
Bridging the Gap: The HBF Standard
Beyond raw density, SanDisk and SK hynix have introduced the first standard specifications for High Bandwidth Flash (HBF). This new memory layer is designed to sit between HBM and traditional SSDs, combining the high-speed data transfer of HBM with the vast capacity of NAND. This hybrid approach is critical for overcoming memory walls in large-scale AI training and inference.
PCIe 6.0 and Liquid Cooling Integration
The innovation extends to Kioxia's CM10 Series enterprise SSDs, the first to utilize PCIe 6.0 and support NVIDIA CMX architecture. These drives are specifically optimized for AI context caching and feature direct-cold-plate liquid cooling to handle the thermal demands of modern data centers. Performance gains are stark, with sequential read speeds increasing by up to 92% over previous iterations.
These advancements arrive as the industry pivots toward high-efficiency architectures, mirroring recent moves by Micron and Microchip to deploy PCIe Gen 6 storage to mitigate the rising costs and energy constraints of AI hardware.
No comments yet. Be the first!