Non-volatile memory

Non-volatile memory is defined by persistence rather than by a particular circuit or storage medium. A device is non-volatile when its stored state remains readable after its supply voltage falls to zero, although retention is finite and can depend on temperature, manufacturing variation, and prior use.¹ The category includes mask read-only memory, programmable read-only memory, electrically erasable programmable read-only memory (EEPROM), flash memory, and several emerging solid-state technologies.

It is distinct from the usual working memory of a computer. Dynamic random-access memory and static random-access memory provide fast access but normally lose their contents without power. Non-volatile devices are commonly used for boot code, device settings, removable storage, and persistent files; a computer may copy persistent data into volatile memory during operation.

Persistence does not mean unlimited durability. A memory can retain data for years yet tolerate only a limited number of rewrites, or it can endure many writes while requiring energy or specialized circuitry to preserve its state.

Non-volatile technologies store information by maintaining a physical state after power is removed. In floating-gate and charge-trap devices, electrons alter a transistor's threshold voltage; flash memory uses these mechanisms in dense arrays and erases blocks rather than individual bytes.² EEPROM provides finer-grained electrical updates, while EPROM is programmed electrically but traditionally erased with ultraviolet light.

Other devices encode state in different physical properties. Magnetoresistive random-access memory (MRAM) uses magnetic orientation, ferroelectric memory uses polarization, and phase-change memory uses the resistance difference between structural phases of a material. Resistive memories instead switch the conductance of a cell, often through nanoscale changes in a material or its interfaces.³⁴

These families differ in density, latency, energy consumption, endurance, retention, and manufacturing compatibility. No single technology simultaneously provides the lowest cost per bit, the fastest writes, the highest endurance, and the longest retention.

Non-volatile memory occupies several levels of the storage hierarchy. Small amounts hold boot firmware in computers, routers, vehicles, appliances, and embedded controllers; larger flash devices provide solid-state drives, memory cards, and USB storage. The operating system normally treats storage as persistent and loads active code and data into faster volatile memory.

Controllers are essential to practical flash systems. Because erase operations occur in blocks and cells wear with repeated program-and-erase cycles, controllers distribute writes through wear leveling, mark defective blocks, translate logical addresses, and use an error-correcting code to recover corrupted bits. Garbage collection can improve usable capacity but may increase write amplification and latency.

Persistent-memory designs place some non-volatile media closer to the processor and expose byte-addressable or memory-like access. They can reduce restart and recovery costs, but software must still handle ordering, atomicity, metadata consistency, and power-loss protection; persistence alone does not make an application crash-safe.⁵

The practical boundary between memory and storage is not fixed. Flash is commonly called storage because it is organized in large blocks and accessed through controllers, while some newer media can be mapped more directly into an address space. The same physical principle may therefore appear in products described as memory, storage, or persistent memory.

Retention and endurance are opposing engineering concerns in many charge-based devices. Higher-density flash stores multiple voltage levels in one cell, increasing capacity but narrowing the distinction between states and making reads, correction, and retention management more demanding.² Temperature also accelerates charge loss and other failure mechanisms, so specifications state conditions rather than promising permanence.

Non-volatile memory also appears in less visible roles: calibration constants, cryptographic keys, boot-loader recovery images, event logs, and tiny configuration regions in microcontrollers. Research devices based on resistive switching, ferroelectricity, and magnetism are notable because they seek combinations of fast writes, low standby power, high endurance, and compatibility with conventional semiconductor fabrication, but their commercial maturity varies.³⁴

Flash memory

Electrically programmable and erasable non-volatile memory generally erased in blocks.

EEPROM

Electrically erasable programmable read-only memory that can usually update smaller units than flash.

Endurance

The number of program, erase, or write cycles a memory can tolerate before reliability degrades.

Retention

The period for which a memory can preserve data under specified conditions without power.

Wear leveling

A controller technique that spreads writes across physical cells to extend device life.

Persistent memory

Non-volatile memory presented with memory-like access and intended to retain state across power loss.