storage-class memory

For decades computers have lived with a hard wall between two worlds: fast-but-forgetful memory (DRAM, accessed byte by byte, gone on power loss) and slow-but-permanent storage (disks and SSDs, accessed in big blocks, surviving shutdown). Programs must constantly shuttle data across this wall — loading files into memory, saving memory back to files. Storage-class memory is the idea of a single technology that lives in the gap: fast enough to be used like memory, yet persistent like storage, dissolving the wall.

Concretely, storage-class memory is non-volatile memory fast enough and fine-grained enough that the CPU can access it with ordinary load and store instructions — byte-addressable, like DRAM — while it keeps its contents without power, like a disk. Built from emerging persistent memories (phase-change, MRAM, ReRAM) or specialized flash, it slots into the hierarchy between DRAM and SSD: slower and denser/cheaper than DRAM, but far faster and byte-addressable compared with block storage. A program could keep a data structure 'in memory' and simply have it still be there, intact, after a reboot — no save/load step.

Why it matters and the honest reality: SCM promises to collapse the memory/storage distinction, which would reshape how software is written (no more serializing everything to files). It is a genuinely exciting direction. But be candid: real products have struggled. The best-known commercial SCM (Intel Optane persistent memory, built on 3D XPoint) was discontinued, and SCM is harder than it sounds — programming for crash-consistent persistence is subtle, and the price/performance has not yet clearly beaten the simpler combination of lots of DRAM plus fast SSDs. SCM is a powerful concept whose practical triumph is still unsettled.