Recently, at FMS 2024, storage manufacturers such as Western Digital and Solidigm showcased enterprise-grade SSDs for data centers with capacities up to 128TB. It appears that the AI era has significantly pushed SSD demand to a higher level.

With the rapid rise of artificial intelligence, machine learning, and large language models (LLMs), enterprises are facing challenges from two fronts. On one hand, the generation and usage of data are surging; on the other hand, there is an urgent need for enterprises to quickly extract valuable insights from massive amounts of data. As storage demands continue to climb, the performance, scalability, and efficiency of storage solutions have become crucial for the AI technology stack. Western Digital offers a diverse storage product portfolio aimed at meeting the varied needs at each stage of the AI data lifecycle.

Western Digital demonstrated the ultra-high-capacity 128TB BiCS8 QLC eSSD at the event, highlighting its technical applications in constructing high-speed AI data lakes and massive data storage performance scenarios.

The flash memory of this product utilizes the eighth-generation BiCS8 NAND jointly developed by Western Digital and Kioxia. Not long ago, Western Digital shared the 2Tb BiCS8 (218-layer) QLC NAND chip at an investor event, which can provide a 4TB capacity die through 16-layer stacking, laying the foundation for 128TB and 256TB class enterprise-grade solid-state drives. Moreover, this storage chip is specifically designed to meet the storage needs of data centers and AI, further reducing the cost of high-capacity enterprise-grade SSDs and meeting the increasingly vast data storage demands.

Advertisement

It is reported that Western Digital and Kioxia have adopted a CBA (Control Bus Architecture) technology route similar to Yangtze Memory's Xtacking, which involves manufacturing the storage stack and CMOS control circuit separately and then hybrid bonding the two parts.

According to the introduction, the 128TB SSD is expected to be officially released in 2025, with a larger capacity 256TB SSD likely to be achieved by 2027, followed by a 1PB version.

Also at the FMS 2024 summit, Solidigm demonstrated a 122TB QLC enterprise-grade SSD product with a PCIe 4.0 interface and U.2 form factor, which was running in real-time in a 2U server. It is expected to be released in 2025 and shipped in bulk in the first half of the year.

Previously, the D5-P5336 released by Solidigm, which is the world's largest capacity PCIe 4.0 SSD, is part of Solidigm's fourth-generation data center QLC SSD, offering an industry-leading high capacity of up to 61.44TB, with read-optimized performance and high value for read and data-intensive workloads. The D5-P5336 can accelerate data in widely adopted read and data-intensive workloads; offer large-scale scalability for high-density storage environments; and significantly reduce total cost of ownership and improve sustainability in hyperscale environments.

Additionally, it has been reported that Samsung will add a 128TB QLC SSD to its product line in Q4. According to Samsung's previous presentations, a 256TB SSD is planned for release between 2024 and 2026, a 512TB SSD between 2027 and 2029, exclusively for EDSFF-E3.L data centers. The 1PB SSD, however, is not expected until 2035.

Recently, SK Hynix's financial report for the second quarter of 2024 showed strong demand for AI-appropriate memory such as HBM and eSSD, with HBM sales increasing by over 80% quarter-on-quarter and by more than 250% year-on-year, and eSSD sales increasing by about 50% quarter-on-quarter. The market demand for high-capacity enterprise-grade solid-state drives is growing rapidly, and SK Hynix expects its sales to quadruple year-on-year in 2024. SK Hynix revealed plans to launch a 60TB eSSD product in 2024 and a 128TB eSSD in early 2025 to maintain market competitiveness.When QLC SSDs are applied to Machine Learning (ML) and Artificial Intelligence (AI), the hard drives primarily capture data and extract data from the drives, while analyzing and processing it across different workloads. These applications require rapid, high-bandwidth data access as well as low latency and read-optimized performance. QLC NAND SSDs are well-suited for read-intensive workloads, offering both high-density storage and cost-effectiveness. The anticipation for ultra-high capacity QLC SSDs in data centers and AI applications is promising.