Storage class memory (SCM) is intended to occupy the space between DRAM and NAND in the memory hierarchy. For industrial automation and other data-intensive computing environments, that position is significant: it points to a way of combining some memory-like performance characteristics with storage-like persistence and capacity. But SCM is not simply a new device replacing existing memory. Its viability depends on balancing latency, endurance, volatility, density, cost and the system architecture around it.
Why storage class memory matters
DRAM and NAND serve different roles. DRAM provides fast access and high endurance, but it is volatile: data is not retained when power is removed. NAND flash is non-volatile and offers greater storage density, but its access latency and endurance characteristics differ from DRAM. The gap between these technologies creates an opportunity for an intermediate tier.
SCM describes memory intended to bridge that gap. Depending on its implementation and how a system uses it, it may be mapped as memory or as storage. That distinction affects how software and hardware access the device, as well as the performance and endurance expectations placed on it. A useful design decision is therefore not only which memory technology to select, but where it belongs in the system’s hierarchy.
Growing demands from cloud computing, mobile devices, networking, enterprise workloads and connected devices have increased interest in memory and storage architectures tailored to different workloads. The underlying requirement is not uniform: some applications prioritize low latency, while others emphasize capacity, persistence or cost. SCM is one possible response to those competing priorities.
Storage class memory and the DRAM-NAND gap
Emerging memory technologies have often been limited to niche uses, in part because DRAM and NAND have continued to scale. Directly replacing either incumbent technology is difficult while established options remain competitive in their respective roles. SCM offers another route: a new memory may be useful as an additional tier, even if it does not displace DRAM or NAND outright.
The trade-offs are central. The source material compares DRAM and NAND across latency, endurance, non-volatility and areal density, and positions emerging memory between them. In broad terms, DRAM is fast and highly enduring but volatile; NAND is non-volatile and dense, with different latency and endurance characteristics. An SCM device must offer a combination that is valuable enough to justify its cost and integration requirements.
Phase change memory (PCM) is identified as one candidate for near-term SCM applications. PCM changes the state of a material to represent data, and different cell architectures have been explored, including self-heating and heater-based designs. These approaches aim to reduce both power consumption and latency. The presented architectures also rely on silicon-based selectors, a design factor that can constrain density and cost.
Why 3D architecture is important
As conventional memory technologies encounter limits to two-dimensional scaling, three-dimensional structures become increasingly important. The same principle applies to emerging memory: competing in mainstream markets requires a credible path to density and cost, and 3D integration is presented as a key part of that path.
Cross-point memory is one approach. Its simple cell structure can support a small cell footprint and is suitable for stacking, while placing the array above underlying circuitry can improve array efficiency. In principle, adding decks can increase density. However, stacking also introduces engineering challenges. More decks require additional wiring and vias, increasing complexity and potentially cost.
A cross-point array also needs a selector device. Because the architecture has two terminals, current can take unwanted paths through neighboring cells. These parasitic paths can affect reading and programming, creating a risk of disturbance in the array. Selector requirements include high current when a cell is selected, low leakage when it is not, compatibility with the memory material, and a process temperature compatible with fabrication. Selector reliability must also align with the reliability requirements of the memory cells.
From device potential to system adoption
Memory-cell performance alone does not determine whether an SCM technology succeeds. The device must be integrated into a practical architecture, supported by suitable controllers and software, and used in ways that take advantage of its characteristics. Memory-mapped and storage-mapped approaches can lead to different expectations for latency, endurance and persistence.
The historical examples in the source illustrate both potential and limits. PCM was applied to mobile-phone use cases, including NOR replacement and execution in place, and a PCM SSD prototype was described as offering improved performance versus a NAND SSD while carrying much higher cost. These examples show why a technology’s capabilities need to be assessed against a specific application and its economics, rather than treated as proof of broad market suitability.
3D XPoint was presented as a cross-point architecture combining a storage element with a switching element, with circuitry beneath the array. The 2017 material described two 64-gigabit decks and highlighted low latency and fast writes as intended advantages. It also cited a 2019 industry forecast for media revenue in 2023; that forecast should be understood as a period-specific projection, not a current market result.
What engineers and decision-makers should consider
When assessing an emerging memory tier, technical teams should look beyond headline speed. Key questions include:
- Workload fit: Does the application need lower latency, persistence, endurance or capacity, and which of these matters most?
- Architecture: Will the device be memory-mapped or storage-mapped, and what changes are needed in the system to support that role?
- Scaling and cost: Can the technology use 3D structures to increase density without making interconnect complexity uneconomical?
- Reliability: Can the memory cell and selector meet the demands of repeated reads and writes, including the effects of leakage and potential array disturbance?
- Ecosystem readiness: Are the surrounding hardware and software capable of using the device effectively?
Conclusion
Storage class memory represents an opportunity to add a tier between DRAM and NAND, not an automatic replacement for either. PCM and cross-point designs illustrate possible approaches, while also exposing the challenges of selector integration, 3D stacking, reliability and cost. For industrial automation leaders and system designers, the practical lesson is to evaluate SCM in the context of workload requirements and the complete system. Its value depends as much on a viable architecture and supporting ecosystem as on the capabilities of the memory material itself.