According to Tom’s Hardware UltraQLC, rather than being a specific type of memory, combines SanDisk’s BICS 8 QLC 3D NAND, an advanced proprietary controller with 64 NAND channels, and innovative firmware.
The custom-built controller plays a crucial role by incorporating domain-specific hardware accelerators to offload storage functions from firmware, reducing latency, improving reliability, and enabling higher bandwidth for hyperscale storage environments.
Additionally, the controller dynamically scales power consumption based on workload demand, ensuring energy efficiency. A sophisticated bus multiplexer further enhances performance by managing increased data loads from high-density NAND memory stacks, using channels without compromising efficiency.
SanDisk's chief of Engineering and Product Management, Khurram Ismail, said: “UltraQLC [is] custom built based on our decades of experience and our current learnings to be deployed in the modern data infrastructure while not compromising on density, performance, and power efficiency.”
“It is really built around those three things: BICS 8 NAND technology [and future NAND too], customized controllers, and advanced system design.”
Currently, UltraQLC SSDs use 2Tb NAND memory ICs to achieve 128TB capacities. While the 64-channel controller could theoretically support even larger capacities, it would come at the cost of performance. However, with higher-capacity NAND memory devices in development, SanDisk envisions future SSDs reaching 256TB, 512TB, and eventually 1PB in the coming years.
SanDisk acknowledged the ongoing "memory wall" issue—the growing disparity between increasing AI training demands and available memory technology. Large language models (LLMs) are expanding rapidly, requiring exponential increases in memory performance. This limitation is driving the industry to explore alternatives beyond conventional DRAM scaling.
SanDisk’s memory technology chief Alper Ilkbahar said: “This work has been going on for quite some time, but the technological challenges have been pretty daunting and there is no clear line of sight to getting to 3D DRAM right now in the industry.”
SanDisk outlined three potential approaches to addressing this challenge. The first is investing more resources into DRAM scaling, though diminishing returns make this less viable. The second is 3D DRAM, which seeks to stack memory vertically like 3D NAND, though significant technological obstacles remain. The third and preferred approach involves developing new scalable memory technologies such as High Bandwidth Flash (HBF), which SanDisk is actively exploring as an alternative to traditional DRAM scaling.
