NAND flash is a type of non-volatile semiconductor memory that stores data as electric charge trapped in memory cells connected in series strings, keeping that data without power. In storage hardware it is the medium inside SSDs, smartphones, USB drives and memory cards, valued for its high density, low cost per bit and fast random access compared with spinning disks.
The name comes from the cell arrangement, which resembles a NAND logic gate. NAND flash cannot be used like ordinary RAM: it is read and written in pages but erased only in much larger blocks, and each cell survives a limited number of program and erase cycles. Every flash-based device therefore depends on a controller that hides these constraints.
How NAND flash stores and reads data
Each NAND cell is a transistor with an extra layer that can hold charge, either a floating gate made of polysilicon or, in most current 3D designs, a charge trap layer of silicon nitride. Programming pushes electrons into that layer through a thin oxide using high voltage; the stored charge shifts the cell’s threshold voltage, and reading means detecting which voltage range the cell falls into.
Cells are grouped into pages of typically 4 to 16 KB, which are the units of reading and programming, and pages are grouped into blocks of several megabytes, which are the units of erasing. A page cannot be overwritten in place. New data goes to an empty page, and the old copy is marked invalid until its whole block is erased.
Modern chips are built as 3D NAND, stacking cells vertically in more than 200 layers. Stacking increases density without shrinking cells further, which also improves endurance compared with the last generations of planar NAND.
Bits per cell: SLC, MLC, TLC, QLC and PLC
Storing more bits per cell requires distinguishing more threshold voltage levels: two for one bit, sixteen for four bits. Each step raises density and lowers cost, but the narrower voltage windows reduce speed and endurance.
| Cell type | Bits per cell | Voltage states | Typical endurance (P/E cycles) | Typical use |
|---|---|---|---|---|
| SLC | 1 | 2 | About 50,000-100,000 | Industrial devices and SSD write caches |
| MLC | 2 | 4 | About 3,000-10,000 | Older and enterprise SSDs |
| TLC | 3 | 8 | About 1,000-3,000 | Mainstream consumer and data center SSDs |
| QLC | 4 | 16 | About 100-1,000 | High-capacity, read-heavy storage |
| PLC | 5 | 32 | Lower still, emerging | Experimental and archival use |
Endurance figures vary by manufacturer and process, so they are best read as orders of magnitude. Drive makers translate them into a TBW rating, the total terabytes that may be written over the warranty period.
Operation timing and the role of the controller
The three basic NAND operations differ by orders of magnitude in duration, which shapes how SSDs are designed. Values below are typical ranges for current TLC devices.
| Operation | Unit | Typical duration | Consequence |
|---|---|---|---|
| Read | Page | Tens of microseconds | Fast random reads compared with HDDs |
| Program (write) | Page | Hundreds of microseconds | Writes are slower than reads |
| Erase | Block | Several milliseconds | Erasing is done in the background |
The SSD controller runs a flash translation layer (FTL) that maps logical addresses to physical pages. It performs wear leveling to spread writes evenly, garbage collection to consolidate valid pages and free whole blocks, and strong LDPC error correction to handle the raw bit errors that dense NAND produces. Many TLC and QLC drives also use part of the flash in SLC mode as a fast write cache.
NAND flash compared with related storage technologies
NAND flash is the medium, while terms such as SSD, NVMe and SATA describe the device and its interface. The comparison places NAND among its neighbors.
| Concept | What it is | Relation to NAND flash |
|---|---|---|
| NOR flash | Flash with cells in parallel | Byte-addressable and fast to read, but far less dense; used for firmware |
| SSD | Storage device | Combines NAND chips, a controller and often DRAM cache |
| HDD | Magnetic spinning disk | Cheaper per terabyte, much slower random access |
| NVMe | Protocol over PCIe | Exposes NAND parallelism with low latency |
| SATA | Older storage interface | Limits NAND SSDs to about 550-600 MB/s |
| DRAM | Volatile working memory | Much faster and unlimited writes, but loses data without power |
Endurance, retention and reliability limits
Every program and erase cycle damages the tunnel oxide slightly, so cells gradually leak charge and become harder to read. This wear defines both endurance and data retention, the time a cell holds data without power.
- Write amplification: internal copying during garbage collection means the NAND absorbs more writes than the host sends.
- Retention loss: worn cells, especially QLC, hold data for a shorter time when stored unpowered and warm.
- Read disturb: repeated reads of one area slowly shift neighboring cells, requiring periodic refresh.
- Performance drop when full: little free space leaves garbage collection struggling and writes slow down.
- SLC cache exhaustion: long sequential writes on QLC drives can fall to a fraction of their advertised speed.
Controllers counter these effects with over-provisioning, spare area hidden from the user, and with the TRIM command, which tells the drive which pages the file system no longer needs.
Choosing NAND-based storage for a workload
The right NAND type depends mainly on write intensity and capacity needs. For most users the difference matters less than the controller quality and the drive’s free space.
TLC is the balanced default for operating systems, applications and general work. QLC suits large, read-mostly libraries such as media or game collections, while write-heavy databases and logging systems call for enterprise TLC or SLC-based drives with high TBW ratings. Keeping 10 to 20 percent of capacity free helps any NAND drive maintain steady performance.
Stripped to the essentials
NAND flash is the memory chip that keeps photos on a phone and files on an SSD, even with the power off. It works by trapping tiny amounts of electric charge inside billions of cells. The amount of charge in each cell represents data.
Squeezing more data into each cell makes storage cheaper. The trade-off is that the cells wear out sooner and work more slowly. The key idea is that NAND is fast and compact but wears out with writing, so a small built-in computer manages it carefully.
Where it is applied
A video production studio buys storage for two jobs. For active editing, where several terabytes are written each week, it chooses 4 TB NVMe drives built on TLC NAND with a rating of 2,400 TBW. At that pace the drives would take years to reach their endurance limit, and sustained write speed stays high during long exports.
For the archive of finished projects, which are written once and read occasionally, the studio picks 8 TB QLC drives. They cost noticeably less per terabyte, and their lower endurance does not matter for read-mostly data. The studio keeps about 15 percent of every drive free, which preserves performance and gives the controller room for garbage collection.