Storage Class Memory (SCM) Explained – The Next Leap in Memory Technology

The Missing Link in Modern Computing:

For decades, computer architects have lived by a simple rule: Memory is fast but temporary (volatile), and storage is slow but permanent (persistent). We have relied on a rigid memory hierarchy that offers trade-offs between access speed, data storage capacity, power consumption and cost.  That hierarchy ranges from fast, expensive DRAM for active processing to slower, less-expensive NAND flash SSDs or spinning disks (HDDs) for data storage. However, in the era of artificial intelligence (AI), massive in-memory databases and real-time analytics, this traditional setup is falling short. We are experiencing performance bottlenecks, and the culprit is the significant gap—or semantic/memory wall—between DRAM and NAND flash. 

To overcome this wall, we need a next-generation memory technology that bridges the gap between performance and persistence—offering near-DRAM speeds with non-volatile, scalable, and cost-efficient characteristics. 

Why Should Companies Care?

In a world where data is the new gold, the speed at which you can "mine" it may very well define your competitive edge. While CPUs and GPUs have become incredibly fast, they are constantly held back by a data traffic jam caused by the latency gap between fast, volatile DRAM and slow, persistent SSDs. Storage class memory (SCM) is a solution that clears this jam. 

What is SCM?

SCM is a high performance, non-volatile memory technology that bridges the gap between DRAM and NAND flash storage. It combines the durability and persistence of storage with the low latency and byte addressability of host memory, allowing data to be accessed faster than traditional PCIe^® SSDs while retaining information during power loss. 

Different types of SCM and Materials/Techniques Used: 

• 3D X Point: Uses a cross-point structure that is faster than traditional NAND flash, often used as persistent memory. It uses germanium-antimony-tellurium (GST) as its phase-change data storage material.
• Resistive RAM (ReRAM/RRAM): Uses a memristor to change the resistance of a solid dielectric material, allowing for high-density and fast non-volatile storage.
• Magnetoresistive RAM (MRAM): Relies on magnetic states rather than electrical charges, offering high endurance and near-SRAM speeds.
• Spin Transfer Torque MRAM (STT-MRAM): Low-power version of MRAM that uses spin-polarized current to switch magnetic layers.
• Nanotube RAM (NRAM): Uses carbon nanotubes to form, in essence, a high-speed, non-volatile technology.
• Low Latency NAND: Uses NAND flash as a core material and stores 1-bit-per-cell (SLC), offering faster read/write speeds than standard TLC/QLC NAND flash while providing non-volatile persistence. 

KIOXIA XL-FLASH™ Memory: The New Gold Standard for Low-Latency NAND SCM:

KIOXIA has focused on low-latency NAND because it is well positioned to support today’s key applications and use cases.  XL-FLASH is KIOXIA’s high-performance solution that bridges the gap between DRAM and NAND flash storage and meets several of the market’s critical needs. Unlike standard TLC/QLC NAND flash, XL-FLASH is specifically designed to deliver significantly lower latency and higher throughput, making it ideal for the mission-critical/high-performance storage layer.

 Key Features of XL-FLASH: 

• Ultra-low latency: XL-FLASH dramatically reduces read latency (typically in the range of 5 microseconds), filling the performance gap between DRAM and standard enterprise SSDs.
• High parallelism:It uses specialized architecture with more "planes" and shorter bit lines, allowing it to handle more simultaneous data requests than standard 3D NAND.
• Byte-addressability support:While fundamentally based on flash, it is designed to work with protocols like CXL™ (Compute Express Link), enabling it to act as a system memory for the CPU/GPU.
• Extreme endurance:Designed for enterprise workloads, XL-FLASH supports high Drive Writes Per Day¹ (DWPD). Some implementations demonstrated up to 60 DWPD, making it durable enough for heavy write-logging and caching.
• Super High IOPS:Custom PCIe^® 5.0 implementations using XL-FLASH can reach staggering speeds, with some drives delivering up to 3.5 million random IOPS, roughly three times faster than typical high-end SSDs.  Higher performance is possible, up to 100 million IOPS, with future generations. 

The Evolution of Performance:

KIOXIA’s SCM solution has evolved from XL-FLASH generation 1’s high-endurance "true SLC" (1-bit) to XL-FLASH generations 2’s cost-effective pSLC (MLC-based) technology, addressing the void between DRAM and standard NAND. While XL-FLASH generation 1 focuses on maximum performance and 50k-100k cycle endurance for mission-critical systems, XL-FLASH generation 2 lowers the cost-per-gigabyte² for large-scale AI and in-memory databases by easing back on endurance (~20k-30k cycles) for increased capacity and reduced cost.

Applications for XL-FLASH: 

• AI and Machine Learning (ML):High-performance SSDs are currently being designed to hit 10 million IOPS to keep pace with powerful GPUs. This prevents storage from becoming a bottleneck during training cycles where the GPU is constantly requesting tiny bits of data.
• Next-gen high IOPS SSDs:Upcoming PCIe 6.0 and 7.0 drives are targeting 10M to 100M IOPS to support the massive data demands of future AI servers.
• Memory expansion (CXL.mem): Servers often hit their memory limits long before they run out of processing power. By using XL-FLASH with CXL technology, systems can create a high-capacity memory expansion layer. This provides an effective option to scale and increase total memory, making it an optimal solution for memory-heavy workloads like AI inference and large-scale virtualization.
• Direct GPU-to-storage connectivity for LLMs and generative AI: With CXL™, GPUs can pull data directly from XL-FLASH™ without going through the CPU first. This direct path supports the 100M+ IOPS required for real-time AI training and high-speed checkpointing.
• CXL memory pooling: In massive data centers, memory is often stranded on one server while another needs more. Using a CXL switch, a pool of XL-FLASH SCM can be shared dynamically across multiple servers. This ensures that high-performance, low-latency memory is always available where the workload is heaviest.
• Real-time analytics: In-memory databases can process millions of small, random queries every second, allowing for instant business intelligence on live data streams.
• Virtual Desktop Infrastructure (VDI) at scale:Large organizations running thousands of virtual desktops from a single storage array require millions of IOPS to handle "boot storms" (everyone logging in at once) without lag.
• High-Frequency Trading (HFT):Financial systems use million-IOPS storage to execute trades in microseconds, where even a slight queue in storage requests could lead to a failed trade or lost profit. 

Looking Forward:

While early SCM solutions faced challenges, KIOXIA XL-FLASH is positioning itself as the high-performance backbone for the next generation of data-intensive computing. 

As AI and data-intensive workloads push traditional storage to its limits, the future of XL-FLASH lies in establishing itself as the premier, high performance SCM bridging the gap between DRAM and NAND. By evolving into MLC-based solutions and leveraging CXL technology, XL-FLASH is poised to deliver ultra-low latency and impressive scalability, offering a crucial option for real-time inference and high-performance database caching. Moving forward, expect XL-FLASH to be present in the enterprise as well as in mainstream data centers, redefining performance benchmarks for GPU-accelerated storage. 

For more information, visit the KIOXIA XL-FLASH web page and read the insightful article in Electronic Design^® on the benefits of low latency NAND.   

NOTES:

¹ DWPD: Drive Write(s) Per Day: One full drive write per day means the drive can be written and re-written to full capacity once a day, every day, for the specified lifetime.  Actual results may vary due to system configuration, usage, and other factors. 

² Definition of capacity: KIOXIA Corporation defines a megabyte (MB) as 1,000,000 bytes, a gigabyte (GB) as 1,000,000,000 bytes, a terabyte (TB) as 1,000,000,000,000 bytes and a petabyte (PB) as 1,000,000,000,000,000 bytes. A computer operating system, however, reports storage capacity using powers of 2 for the definition of 1Gbit = 2³⁰ bits = 1,073,741,824 bits, 1GB = 2³⁰ bytes = 1,073,741,824 bytes, 1TB = 2⁴⁰ bytes = 1,099,511,627,776 bytes and 1PB = 2⁴⁰ bytes = 1,125,899,906,842,624 bytes and therefore shows less storage capacity. Available storage capacity (including examples of various media files) will vary based on file size, formatting, settings, software and operating system, and/or pre-installed software applications, or media content. Actual formatted capacity may vary. 

TRADEMARKS:

CXL is a trademark of Compute Express Link Consortium, Inc.  PCIe is a registered trademark of PCI-SIG.  All other company names, product names and service names may be trademarks of third-party companies. 

Electronic Design is a registered trademark of Endeavor Business Media, LLC.  All rights reserved.  

DISCLAIMERS:

KIOXIA America, Inc. may make changes to specifications and product descriptions at any time.  The information presented in this blog is for informational purposes only and may contain technical inaccuracies, omissions and typographical errors.  Any performance tests and ratings are measured using systems that reflect the approximate performance of KIOXIA America, Inc. products as measured by those tests. In no event will KIOXIA America, Inc. be liable to any person for any direct, indirect, special or other consequential damages arising from the use of any information contained herein, even if KIOXIA America, Inc. are advised of the possibility of such damages.