Flash Memory: From NOR and NAND Experiments to Persistent Storage

Flash memory is nonvolatile semiconductor storage that can be electrically erased and rewritten. Its development created a practical alternative to memory that required power to retain data and to storage media whose moving parts constrained size and access. The technology did not replace every prior medium at once. Instead, two major cell organizations, NOR and NAND, developed for different tradeoffs in random access, density, cost, and erase behavior.

The history is often reduced to a single inventor story, but the technology’s path involved device research, patents, manufacturing, product engineering, and competing commercial uses. Fujio Masuoka and Toshiba’s research are associated with early flash work; the Semiconductor History Museum of Japan records the 1984 announcement of NOR flash and a subsequent NAND development milestone. Intel’s historical reports show how commercial NOR products entered embedded applications. These sources are complementary: one records the research lineage, another an industry history, and another a manufacturer’s product history.

Before flash: nonvolatile memory had constraints

Volatile memory loses its stored information when power is removed. It is useful for fast working data, but a system needs another mechanism to preserve firmware, settings, or user files through shutdown. Earlier nonvolatile semiconductor memories included read-only and electrically erasable technologies, but their write and erase processes imposed tradeoffs in speed, granularity, endurance, and circuit complexity.

Flash offered electrical erasure and rewriting using a cell structure related to floating-gate transistor memory. A floating gate can retain charge without power, and the charge state can be read as a difference in transistor behavior. Writing and erasing require carefully controlled electrical conditions. Flash cells are not ordinary RAM locations that can be overwritten arbitrarily at the same granularity as a processor word. Erase blocks and program rules are fundamental to the technology.

The phrase “nonvolatile” also does not mean permanent. Flash cells have finite program/erase endurance, and stored charge can degrade over time or under unsuitable conditions. Controllers and file systems need to account for wear and error correction. Modern products include sophisticated management not present in the earliest devices, so contemporary behavior should not be projected onto 1980s chips.

The NOR and NAND distinction

The early NOR arrangement connects memory cells in a way that supports relatively direct random access to stored data. That made NOR suitable for reading code and executing firmware in place, subject to device and processor requirements. It could be used for embedded programs that needed to retain instructions without a disk. NOR’s structure tended to make high-density storage more expensive compared with NAND, but random reads and code access were useful properties.

NAND organizes cells as a string and is accessed through pages and erase blocks rather than treating every cell as a directly addressable byte. That organization supports high density and lower cost per bit, but requires a controller or host software to manage data placement, bad blocks, error correction, and wear. NAND became the basis for many mass-storage products, including memory cards and solid-state drives.

The comparison is not “NOR is old and NAND is better.” Each architecture fits different constraints. NOR’s random read behavior helped firmware applications; NAND’s density and cost supported larger storage. System design determines which characteristics matter. A device booting code may value predictable reads, while a camera storing large images may prioritize capacity and price.

Toshiba’s research and the early announcement

Fujio Masuoka worked at Toshiba when the early flash research was announced. The Semiconductor History Museum of Japan dates the public presentation of NOR flash to the International Electron Devices Meeting in December 1984 and describes a cell based on polycrystalline silicon layers that could be electrically erased collectively. Toshiba’s later engineering history records subsequent developments in NAND devices. These dates are milestones in the development, not proof that a complete modern flash storage product existed in 1984.

The name “flash” is commonly explained as a reference to the speed or breadth of erase compared with earlier processes. Retrospective accounts attribute the name to a colleague’s comparison with a camera flash. The important device concept was not that every memory bit could be rewritten instantly; erase operations occurred over a defined region and needed appropriate voltage conditions. The erase granularity became a key architectural consideration.

Flash inventorship and commercial product history are sometimes argued in broad terms. A careful account separates the underlying cell concepts, patent filings, conference demonstrations, memory products, and later high-density manufacturing. Research papers and patents establish what was disclosed and by whom; product reports establish when a company began selling a particular device. One milestone does not necessarily settle every legal claim about related semiconductor structures.

Intel’s role in commercial NOR products

Intel’s 1988 annual report says the company introduced a flash memory device aimed at customers who wanted to update embedded programs. That is evidence of a commercial product direction, not a claim that Intel invented every part of flash memory. The product connected nonvolatile storage to microcontroller and embedded use cases: firmware could be updated without removing a programmed ROM chip or maintaining a separate disk drive.

NOR products fit code storage because processors could read instructions through a memory-like interface. This enabled systems to boot from flash or execute firmware in place, subject to the bus and processor design. Product engineers still had to consider erase cycles, write protection, voltage requirements, and interrupted updates. A firmware update that loses power during erase can leave a device unable to boot unless the system has recovery or redundant-image support.

Over time, NOR densities increased, and multi-level cell designs stored more than one bit per physical cell by distinguishing multiple charge levels. This increased density but tightened margins between states and increased requirements for error management. Intel’s product announcements from later decades describe such density expansion. These later developments belong to the technology’s evolution, but do not define the first flash devices.

NAND and the rise of mass storage

NAND’s denser organization enabled storage devices that could replace or supplement magnetic media. But NAND requires a management layer. Flash Translation Layers (FTLs) map logical block addresses from an operating system onto physical pages and erase blocks. Since data cannot usually be overwritten in place, the controller writes new pages and later reclaims blocks through garbage collection. Wear leveling distributes writes, while error-correcting codes handle increasing bit-error rates as cells shrink or store more bits.

These responsibilities explain why a NAND storage device is more than a memory chip. A card, thumb drive, or SSD includes controller firmware, error correction, bad-block management, buffering, and host interfaces. The operating system typically sees a block device, not raw NAND cells. Directly accessing raw NAND is a different engineering task that requires a device-specific protocol and knowledge of the media characteristics.

NAND also changed manufacturing economics and media form factors. Solid-state storage had no spinning disk or moving head, which enabled smaller devices and lower seek latency. But early products faced high cost, limited capacity, and endurance concerns. As process technology and controller software improved, flash moved from niche removable media into phones, cameras, laptops, and enterprise systems. The transition was incremental, with hard disks remaining advantageous for some cost-per-capacity workloads.

The limits that shaped system design

Erase blocks create write amplification: changing a small logical region can trigger internal movement and rewriting of a larger amount of data. Controllers attempt to minimize that cost, but workload matters. Random writes, frequent metadata updates, and nearly full devices can produce more internal work than a simple sequential write. This explains why file systems and storage controllers have evolved to manage flash characteristics more effectively.

Endurance is workload-dependent. A specification may rate a device for a certain program/erase count or terabytes written, but those numbers depend on the product and test conditions. A historian should not quote a modern endurance value for an early chip. The safe lesson is architectural: flash has finite write endurance, and wear management became a necessary part of larger storage systems.

Data retention is also affected by temperature, wear, process geometry, and time. A flash device sitting unused is not guaranteed to preserve data indefinitely. Long-term preservation workflows should verify media periodically, maintain multiple copies, migrate data to current storage, and preserve checksums. “Solid state” describes the storage mechanism, not an archival guarantee.

A layered technology, not a single invention date

Flash memory’s history contains several distinct milestones: research on floating-gate devices, early public disclosures, NOR and NAND structures, commercial product launches, manufacturing scale-up, and controller innovations. Different sources may give different “invention” dates because they refer to different milestones. A reliable account names the milestone and cites the source rather than compressing all development into one year.

The division between NOR and NAND is also a reminder that an architectural fork can persist. NOR and NAND shared a family resemblance but enabled different system designs. One favored random read access for code and firmware; the other supported high-density data storage with more complex management. Neither technology eliminated volatile memory or magnetic storage. Instead, flash filled a new region of the design space and gradually changed the balance among performance, persistence, size, and cost.

Today, flash is often invisible because users see a file system and block device rather than cells and erase commands. Historical study brings those layers back into view. The cell architecture determines electrical constraints; the controller translates logical operations; the operating system manages files and blocks; and manufacturers balance density, reliability, and cost. Flash became foundational not because it was an ideal memory, but because engineers built systems around its limitations and advantages.

Related:

Sources: