{
  "id": "nvm",
  "title": "Non-volatile memory",
  "lang": "en",
  "source_lang": "en",
  "summary": "Non-volatile memory (NVM) retains stored information without continuous electrical power and includes technologies used for firmware, persistent storage, and specialized main memory.",
  "category": "computer-science",
  "subcategory": "hardware",
  "type": "concept",
  "aliases": [
    "NVM",
    "Nonvolatile memory",
    "Non-volatile storage"
  ],
  "generated_at": "2026-11-05",
  "model": "pro/gpt-6.1-sol",
  "url": "https://aiwiki.page/en/nvm/",
  "translations": {
    "zh-Hans": {
      "url": "https://aiwiki.page/zh/nvm/",
      "json": "https://aiwiki.page/zh/nvm.json"
    }
  },
  "sections": [
    {
      "heading": "",
      "text": "Non-volatile memory (NVM) is memory that retains stored information when its electrical power is removed. It contrasts with volatile memory, which requires power to preserve its contents. The term most commonly refers to electronic memory devices, especially semiconductor memories such as flash memory, but it can also be used broadly for persistent storage technologies, including magnetic and optical media. Non-volatility describes data retention, not a particular access speed, interface, or ability to rewrite data."
    },
    {
      "heading": "Scope and terminology",
      "text": "In a computer, non-volatile memory stores information that must survive shutdowns, resets, and power interruptions. Typical contents include firmware, configuration settings, application software, and user data.\n\nSeveral related terms describe different properties:\n\n- Non-volatile means that stored information survives removal of operating power under specified conditions.\n- Read-only describes restrictions on changing stored information. Some non-volatile memories are permanently programmed, while others can be rewritten repeatedly.\n- Random access means that locations can be accessed without traversing preceding data. It does not imply volatility.\n- Persistent memory usually refers to non-volatile storage exposed through a memory-like interface, often allowing processor load and store instructions rather than conventional storage input/output.\n\nOrdinary static random-access memory (SRAM) and dynamic random-access memory (DRAM) are volatile. Battery-backed SRAM can preserve information through an interruption of the main power supply, but the SRAM cells themselves still require power. Hybrid modules may instead copy volatile contents into non-volatile storage using reserve energy.\n\nA device’s non-volatility also does not guarantee that every recently issued write has reached durable storage. Data may remain temporarily in volatile buffers or processor caches."
    },
    {
      "heading": "Physical principles",
      "text": "Non-volatile memory stores information in physical states that remain distinguishable without continuous electrical power. A memory cell may encode a bit or, in some technologies, several bits.\n\nStored electrical charge\n\nEPROM, EEPROM, and flash memory store information using trapped electrical charge. In floating-gate designs, charge is held on an electrically isolated conductor within a transistor. Charge-trap designs store charge in insulating material.\n\nThe stored charge changes the transistor’s threshold voltage. Reading circuitry determines the stored value by measuring how the cell conducts under selected voltages. Programming and erasure move charge through insulating barriers using mechanisms such as tunnelling.\n\nCharge confinement is not perfect. Leakage, defects, temperature, and accumulated programming damage can eventually alter the stored state.\n\nMagnetic states\n\nMagnetic memories encode information through stable magnetization states. In magnetoresistive random-access memory (MRAM), a magnetic tunnel junction has a resistance that depends on the relative orientation of magnetic layers. Writing changes a layer’s magnetic orientation; reading senses the resulting resistance.\n\nMagnetic disks and tapes also retain information magnetically, although their recording mechanisms and access arrangements differ from semiconductor MRAM.\n\nFerroelectric polarization\n\nFerroelectric RAM (FRAM or FeRAM) stores information using remanent electric polarization. Conventional capacitor-based designs distinguish polarization states through their electrical response. Reads in many such designs are destructive and must be followed by restoration of the data.\n\nFerroelectric field-effect transistor memories use polarization to modify a transistor’s electrical characteristics.\n\nMaterial phase and resistance\n\nPhase-change memory stores data through differences in electrical resistance between material states, commonly relatively crystalline and amorphous states. Electrical pulses heat a small region to establish the desired state.\n\nResistive random-access memory (ReRAM or RRAM) uses electrically induced changes in resistance, often associated with the formation and disruption of conductive paths within an insulating material. Its behaviour depends strongly on the materials and cell structure."
    },
    {
      "heading": "Major types",
      "text": "Type | Programming or rewriting method | Principal characteristics\nMask ROM | Defined during fabrication | Fixed contents; suited to large production runs with unchanging data\nProgrammable ROM | Electrically programmed once | Permanent programming, commonly using fuse or antifuse structures\nEPROM | Electrically programmed; erased with ultraviolet light | Reusable, but erasure generally affects the whole exposed chip\nEEPROM | Electrically erased and programmed | Often supports fine-grained updates; widely used for settings and small data stores\nFlash memory | Electrically programmed and erased in blocks | High density; dominant in solid-state mass storage\nMRAM | Magnetic-state switching | Fast access and high endurance in many implementations\nFRAM | Ferroelectric polarization switching | Low write energy and high endurance in many implementations\nPhase-change memory | Electrically induced material-state changes | Offers non-volatile storage with technology-dependent access and endurance characteristics\nReRAM | Electrically induced resistance changes | Supports a range of embedded-memory and research applications\n\nROM, EPROM, and EEPROM\n\nRead-only memory (ROM) originally referred to memory whose contents could be read but not changed during normal operation. Mask ROM fixes those contents during manufacturing. Programmable ROM permits programming after fabrication, generally only once.\n\nErasable programmable read-only memory (EPROM) introduced reusable semiconductor storage that could be erased using ultraviolet light. Electrically erasable programmable read-only memory (EEPROM) made electrical erasure possible, allowing updates without removing a chip for ultraviolet exposure.\n\nEEPROM remains useful where relatively small amounts of information must be updated independently, such as calibration constants, device identifiers, and configuration records.\n\nFlash memory\n\nFlash memory provides block erasure, reducing the cell and circuit overhead associated with fine-grained erasure. Its two major organizations are NOR and NAND.\n\nNOR flash supports efficient random reads and is often used to store executable firmware. Suitable systems can execute instructions directly from it.\n\nNAND flash is organized for dense storage, with reads and programming typically performed in pages and erasure performed in larger blocks. It is used in memory cards, USB drives, and solid-state drives (SSDs).\n\nFlash cells may represent multiple bits using several distinguishable threshold-voltage ranges. Increasing the number of bits per cell improves storage density but narrows sensing margins and generally creates trade-offs involving write performance, endurance, and retention. Three-dimensional NAND increases density by arranging memory cells in vertically stacked structures.\n\nAn SSD is not simply an array of flash cells. It also contains a controller and management software that handle address translation, error correction, defective blocks, and allocation of writes."
    },
    {
      "heading": "Historical development",
      "text": "Early computers used several storage technologies that retained information without power. Magnetic-core memory was especially important: it served as working memory while preserving its contents through power loss. Its role illustrates that non-volatile memory need not be secondary storage.\n\nSemiconductor ROM provided a compact way to store fixed programs and control information. EPROM and EEPROM subsequently made stored programs and settings easier to revise. Flash memory extended electrically erasable storage toward higher densities and larger capacities, eventually becoming central to portable electronics and solid-state storage.\n\nOther technologies, including MRAM, FRAM, phase-change memory, and ReRAM, address different combinations of speed, energy consumption, endurance, density, and manufacturing compatibility. No single technology dominates every position in the memory hierarchy."
    },
    {
      "heading": "Performance and reliability",
      "text": "Non-volatile memories are evaluated along several independent dimensions.\n\nRetention is the time for which information remains readable under specified conditions. It depends on factors including temperature, material properties, and prior use. A retention rating is not a guarantee of indefinite archival preservation.\n\nEndurance describes the amount of rewriting a device can tolerate. Flash programming and erasure progressively stress insulating materials. Other technologies have different wear mechanisms and endurance limits.\n\nLatency and bandwidth describe access delay and sustained transfer rate. Reads and writes can have very different costs, and erasure may be a separate, relatively expensive operation.\n\nGranularity is the smallest unit that can be read, programmed, or erased. These units need not be the same size. In NAND flash, a small logical update can require substantial internal movement of data.\n\nEnergy consumption includes reading, writing, erasing, and supporting circuitry. Eliminating continuous refresh does not mean that a device consumes no power during operation.\n\nControllers improve practical reliability using error-correcting codes, spare capacity, defective-block management, and background maintenance. Flash devices commonly use wear leveling to distribute writes across physical storage locations.\n\nThese measures introduce overhead. Internal copying and reclamation can cause more physical data to be written than the host requested, a phenomenon called write amplification."
    },
    {
      "heading": "System integration and persistence",
      "text": "Non-volatile storage usually appears through either a storage interface or a memory interface.\n\nStorage devices accept commands to read and write blocks. An operating system and file system organize those blocks into files and directories.\n\nMemory-mapped persistent devices can make durable data directly accessible to processor instructions. This can reduce some software and data-copying overhead, but it creates additional requirements for ordering and recovery.\n\nA write may first reach a volatile processor cache rather than the non-volatile medium. Depending on the architecture, software may need cache write-back operations and ordering instructions to establish persistence. The relevant hardware guarantee is the persistence domain: the components whose contents are guaranteed to survive, or be safely transferred to durable storage, during a power failure.\n\nNon-volatility alone does not provide atomic updates or crash consistency. A power interruption can occur between related writes, leaving a partially updated data structure. Systems use techniques such as logging, copy-on-write, and recovery protocols to handle this problem. For example, a database may use write-ahead logging to make interrupted operations recoverable."
    },
    {
      "heading": "Applications and limitations",
      "text": "Non-volatile memory supports:\n\n- Boot code and firmware in computers and embedded systems.\n- Persistent settings, calibration values, and device identities.\n- Mass storage in phones, computers, and data centres.\n- Event records and state retention in industrial equipment.\n- Specialized persistent-memory applications requiring durable, relatively low-latency access.\n\nIts persistence also has security implications. Sensitive information can remain after shutdown, and logical deletion does not necessarily remove every physical copy. Flash address remapping and spare blocks complicate direct overwriting. Encryption and device-supported sanitization mechanisms address different parts of this problem, but their effectiveness depends on implementation and key handling.\n\nNon-volatile memory is not a substitute for backup. Hardware failure, controller faults, software errors, accidental deletion, and physical damage can destroy access to retained data. For long-term preservation, the relevant issue is not merely whether individual cells retain their states, but whether the complete device and its recorded information remain readable."
    }
  ],
  "links": [
    "semiconductor",
    "flash-memory",
    "computer",
    "static-random-access-memory",
    "bit",
    "transistor",
    "memory-hierarchy",
    "error-correcting-code",
    "operating-system",
    "file-system",
    "cpu-cache",
    "database",
    "write-ahead-logging",
    "encryption",
    "data-backup"
  ],
  "sources": [],
  "license": "https://creativecommons.org/licenses/by/4.0/"
}