Evolution of SSD Storage Interfaces and the Emerging Role of Storage Class Memory (SCM)

SSD storage media and interface technologies have continuously evolved, starting with SATA SSDs (including SATA/SAS) that primarily used SLC and eMLC, followed by the emergence of PCIe SSDs, which suffered from a lack of standardization and split into Host‑based and Device‑based implementations.

With the advent of the NVMe era, a unified interface and protocol standard were established, offering three main product forms: U.2 (compatible with SATA/SAS), PCIe SSD cards, and the consumer‑grade M.2 form factor.

Although NVMe dramatically improves interface standards and data transfer efficiency, NAND Flash remains the dominant storage medium, prompting the question of what the next generation of storage media will look like.

Intel's Optane series demonstrates that pairing NVMe with Storage Class Memory (SCM) yields significant storage advantages, suggesting that the future of high‑performance storage lies in SCM rather than proprietary NVMe solutions.

SCM technology is already widely adopted across industries; a detailed overview of SCM/NVM status and research directions is available in the referenced original article.

Intel and Micron introduced 3DX (PRAM), the first product to combine NVMe and SCM, claiming up to seven‑fold performance gains over competing NVMe SSDs, with sequential read speeds of 2500 MB/s and write speeds of 2000 MB/s.

SCM offers persistent memory characteristics and non‑volatile memory benefits, eliminating NAND Flash’s write‑before‑erase constraints, simplifying operation, and providing superior lifespan and data retention, positioning it as a disruptive next‑generation storage medium for high‑performance and reliability‑critical scenarios.

Current mainstream SCM technologies include PCM, ReRAM, MRAM, and NRAM, each with distinct material and operational principles.

PRAM (Phase‑Change RAM) uses alloy materials that change conductivity between crystalline and amorphous states to represent bits, offering simple structure, large capacity, and low cost, and is typically employed for cache acceleration, exemplified by Intel’s 3D XPoint.

ReRAM (Resistive RAM) stores data by forming or breaking conductive filaments within a cell under different voltages, with notable implementations from HPE and Crossbar.

MRAM (Magnetic RAM) changes electron spin via magnetic fields to represent data, suitable for CPU caches (e.g., L2), with major vendors such as Toshiba and Everspin.

NRAM (Nantero’s CNT RAM) utilizes carbon nanotubes as switches, offering high density, long lifespan, and low theoretical power consumption.

Intel Optane products illustrate current SCM applications: the Optane P4800X (Clodstream SSD) uses NVMe Block interface, avoids garbage‑collection‑induced performance decay, and serves as metadata cache, data cache, and primary data storage for high‑performance workloads.

SCM DIMM products like Apache Pass DIMM provide memory‑semantic access with low latency, larger capacity than DRAM, and persistence, making them suitable for persistent memory, in‑memory databases, and system log volumes (e.g., checkpointing in HPC).

In summary, the transition from SATA/SAS to PCIe and finally to NVMe SSDs reflects two major technical revolutions—standard proliferation and unification—while leveraging SCM’s advantages introduces new challenges for hardware architecture, data structures, transaction mechanisms, and software stacks.

Network limitations currently restrict cross‑CPU memory access, but the emerging Gen‑Z standard enables SCM to connect via a dedicated Gen‑Z bus, allowing nanosecond‑level latency access across CPUs.

For a deeper dive into the gap between traditional SSDs and SCM and proposed solutions, readers are encouraged to consult the original "Detailed Analysis of SCM/NVM Technology Status and Research Directions" article.