Emerging memory explained: MRAM, ReRAM, PCM

Emerging memory technologies store information in a physical property of a material, such as magnetisation, electrical resistance or atomic structure, rather than as stored electric charge. The three best-known families are MRAM (magnetoresistive RAM), ReRAM (resistive RAM) and PCM (phase-change memory). All of them are non-volatile, meaning they keep data without power, and they are being developed to close the gap between fast but volatile working memory and slower, denser flash storage.

Why new kinds of memory?

Computers use a hierarchy of memories, each with a compromise between speed, capacity, cost and energy:

Memory	Speed	Keeps data without power	Typical use
SRAM	Very fast	No	Caches inside processors
DRAM	Fast	No	Main working memory
NAND flash	Slower, especially for writing	Yes	SSDs, memory cards, phones
Embedded flash	Moderate	Yes	Program storage in microcontrollers
MRAM, ReRAM, PCM	Between DRAM and flash, depending on design	Yes	Embedded memory, specialised products, research

Two pressures drive interest in alternatives. First, embedded flash becomes harder to integrate as chip manufacturing shrinks to smaller structures. Second, moving data between memory and processor costs a large share of energy in modern computing, which motivates memories that sit closer to the logic or even compute inside the memory.

MRAM: storing bits in magnetism

MRAM stores each bit in a magnetic tunnel junction: two ferromagnetic layers separated by a very thin insulating barrier. One layer has a fixed magnetic orientation, the other can be switched. When both point in the same direction, electrical resistance is low; when they point in opposite directions, it is higher. Reading means measuring that resistance.

Modern variants such as spin-transfer torque MRAM (STT-MRAM) switch the free layer with a spin-polarised current, and spin-orbit torque MRAM (SOT-MRAM) is a research direction that separates the read and write paths.

• Strengths: fast, very high endurance, non-volatile, robust against radiation, compatible with standard chip processes.
• Limitations: more complex manufacturing, lower density than flash, and a trade-off between how long data is retained and how much energy writing takes.

ReRAM: switching resistance in an oxide

ReRAM cells consist of a thin insulating layer, often a metal oxide, between two electrodes. Applying a voltage forms or dissolves a tiny conductive path, often called a filament, which switches the cell between high and low resistance.

• Strengths: simple structure, low-voltage operation, potential for dense 3D stacking, and the ability to store intermediate resistance levels, which is interesting for analogue and neuromorphic computing.
• Limitations: variability from cell to cell and cycle to cycle, and endurance that is generally lower than MRAM.

PCM: from glass to crystal

Phase-change memory uses materials, typically chalcogenide alloys, that can switch between an amorphous (disordered, high resistance) and a crystalline (ordered, low resistance) state. A short, strong current pulse melts and quickly cools a small volume into the amorphous state; a longer, gentler pulse lets it crystallise.

The same family of materials was long used in rewritable optical discs, where the two phases reflect light differently.

• Strengths: mature physics, good scalability, multiple resistance levels per cell.
• Limitations: relatively high write energy because of the heating, and a gradual drift of resistance over time in the amorphous state.

Where the industry stands

The picture changes quickly, so treat this as an overview as of October 2026, not a market report:

• Several large semiconductor foundries offer embedded MRAM in their process portfolios, aimed at microcontrollers, automotive and low-power devices.
• Embedded ReRAM is offered or announced by foundries and specialised IP companies, with a broad ecosystem around it.
• A high-profile storage-class memory product line based on a resistive cross-point technology was discontinued in the early 2020s, which showed how hard it is to compete with DRAM and flash on cost.
• Research continues on other candidates, such as ferroelectric memories, and on using resistive memories for in-memory and neuromorphic computing.

How to follow this field

If you want to dig deeper, start with review articles and conference proceedings in device physics and electronics; a structured literature search in databases such as IEEE Xplore helps. Major device conferences publish many first results, and our guide to submitting to a conference and giving the talk explains how such meetings work. Because memory research is close to industry, patents matter here too; see patents for researchers. And if you are preparing your own results, how to publish a research paper walks through the process.