The last couple of years have seen the flash memory market stagnate, but a surge in demand for memory chips in the fourth quarter of last year triggered a price increase, marking a potential reversal for the sector. As we venture into 2024, questions loom over whether the Chinese flash memory market will achieve full recovery or take a turn for the worse. The burgeoning AI wave in China is particularly pivotal—what significant impacts will it have on the flash memory sector? Can we expect QLC SSDs to gain widespread adoption this year? Additionally, what new trends will emerge in related technologies such as PCI-E 5.0 and CXL? These were some of the queries raised at a recent summit focused on the Chinese flash memory market, where key players across the globe gathered to discuss market dynamics and technological advancements.

Artificial Intelligence (AI) has undeniably become a focal point of discourse in the tech industry, particularly regarding computational power. As the development of AI technologies such as large models advances, several challenges at the data storage level have come to light. Key stakeholders such as Solidigm, SK Hynix, and Samsung discussed these issues during the summit, emphasizing that, while AI significantly propels demand for data storage, it also introduces complexities that stakeholders must navigate.

The influence of AI on storage thus encompasses several dimensions. First, the rise of multimodal large models has led to an exponential increase in data generation, with Gartner forecasting that by 2025, AI-generated data will account for 10% of total data. Second, storage solutions must meet the performance demands of GPUs, aiming to enhance their utilization rates. Third, as memory capacity continues to increase, considerations around energy usage, spatial efficiency, and costs become paramount. Solidigm’s Vice President for Sales in the Asia-Pacific region, Ni Jinfeng, noted that reports indicate in certain scenarios, storage energy consumption can account for as much as 35% of total server power usage. This challenge is one that firms working with AI must confront head-on.

Indeed, the rise of large AI models signifies a massive engineering challenge focused on processing high-quality data at scale. This evolution will trigger cascading effects across related sectors, including data centers and edge computing. In light of rapidly evolving AI technologies, understanding how best to respond to the new data storage requirements is crucial.

Currently, there are two primary technological pathways to meet the burgeoning storage demands instigated by AI: one involves traditional hard drives, where increased platter density and additional platters enhance capacity; the other centers on flash memory, capitalizing on its high capacity and low power consumption to achieve a balance among cost, energy usage, and physical space.

According to Ni, the swift development of AI technology represents a revolutionary shift in storage requirements, highlighting the advantages of flash memory in terms of performance, capacity, and total cost of ownership. Market observations reveal that substantial investment in AI technologies, particularly in North America, has catalyzed an uptick in demand for high-capacity SSDs, a trend that Ni forecasts will soon penetrate the domestic market in China as well.

The recovery of the storage market is most clearly evidenced by the increasing demand for capacity. In the flash memory domain, 2023 has solidified a robust foundation for capacity density. Companies like Solidigm, SK Hynix, Yangtze Memory Technologies, and Samsung have elevated NAND technology to new heights. For instance, Solidigm has been actively promoting its QLC SSD products and recently launched a 61.44TB QLC SSD, representing significant strides in the field.

Industry experts are anticipating that the QLC SSD will gain traction in 2024, becoming a mainstream choice as its quality, reliability, and longevity have been substantiated across various applications. Specifically, in the consumer and mobile markets, the QLC SSD is gaining ground in sectors such as PC and mobile devices. With the increasing prevalence of AI applications, including AI smartphones and AI PCs facilitated by large model technology, the application of QLC SSDs is likely to accelerate. Ni emphasizes that due to current capacity reduction, potential supply-demand constraints may arise in the QLC SSD market this year; however, persistent long-term demand will inevitably drive the sector forward.

Likewise, in the enterprise space, China's data centers have a relatively modest representation of SSDs. But recent advances in technologies like ZNS, PCI-E, and CXL—as well as the rise of applications driven by AI and big data—position QLC SSDs as a compelling option for enterprise use. Notable players in storage array systems, such as DELL EMC, NetApp, and Inspur, have introduced products utilizing QLC SSDs in 2023. As storage array manufacturers fully embrace this technology, QLC SSDs are expected to proliferate within enterprise data centers in 2024.

Furthermore, existing implementations of PCI-E 4.0 combined with QLC SSDs have already shown significant promise in data centers for read-intensive and hybrid workloads. As PCI-E 5.0 products reach commercial availability, a continued spread of QLC SSD adoption seems imminent.

"Looking at the current landscape, both the consumer and enterprise markets are ready for a transition,” Ni states, “In the consumer space, QLC is poised to accelerate its replacement of TLC, while in the enterprise domain, it will replace HDDs."

However, the rapid development of the flash memory market demands sustained attention to eco-system building. The rapid acceleration of flash memory hardware in recent years highlights the necessity for software innovation and robust ecosystem infrastructure, particularly accounting for the unique characteristics of the Chinese market—whether it’s the operating systems, processors, or applications in use.

Ni articulates that the complexity of data center applications in China necessitates a collaborative effort among all stakeholders to promote the widespread adoption of flash technology. For instance, QLC SSD-related acceleration and application innovation will require synergy from cloud computing clients, enterprise clients, OEMs/ODMs, and operating system providers.

A prime illustration of this collaborative potential can be seen in the prevalent use of various service architectures by data center users, including Intel, AMD, ARM, IPU/DPU, and GPU. These architectures often encounter challenges related to optimizing mixed workloads, capitalizing on storage capacity and performance, and managing write amplification. To address these issues, Solidigm has been employing its CSAL software to establish a caching layer designed to optimize these complex data center workloads, converting random write workloads into sequential writes, which significantly enhances storage efficiency.

"CSAL is now open-sourced, allowing participants across the entire ecosystem to adopt it, thereby facilitating further adoption and application of flash memory technology," Ni explains.

The rise of AI also provides a timely opportunity to overhaul data center infrastructure. Take Kingsoft Cloud as an example; the integration of an AI functionality within its WPS platform has prompted a comprehensive upgrade from HDDs to SSDs, addressing various elements including accelerated AI workloads, infrastructure optimization, and GPU efficiency. This effort, in collaboration with eco-partners like Solidigm, has resulted in dramatically improving AI data processing times—from nine hours to just eleven minutes.

"Advancements in hardware will lead to simplifications in software. The development of hardware and software is a stepping-stone process that relies heavily on the collaborative impetus of the entire ecosystem," Ni concludes.