An SSD (solid-state drive) is a non-volatile storage device that stores data in NAND flash memory chips managed by a dedicated controller, with no moving mechanical parts. In hardware, and specifically in storage, the SSD has replaced the hard disk drive as the standard system drive because it accesses data about a hundred times faster for random operations.
An SSD is more than a set of flash chips. Its controller and firmware translate the simple block interface the operating system expects into the complex rules of flash memory, and this hidden layer largely decides the drive’s speed, endurance and reliability.
How an SSD stores and manages data
NAND flash stores information as electrical charge trapped in memory cells. Modern drives use 3D NAND, in which cells are stacked vertically in more than 200 layers to raise density without shrinking each cell further.
Flash has an awkward constraint: data is written in pages of typically 16 KB, but can only be erased in whole blocks of several megabytes, and a page cannot be overwritten in place. The controller therefore writes new data to free pages, marks old copies as invalid and later reclaims blocks through garbage collection.
The flash translation layer (FTL) maps logical block addresses to physical pages and performs wear leveling, spreading writes across all blocks so that none wears out early. The TRIM command lets the operating system tell the drive which data is deleted, which reduces unnecessary copying. Because garbage collection moves valid data around, the flash receives more writes than the host sends, a ratio called write amplification.
| Component | Role | Effect on the drive |
|---|---|---|
| NAND flash | Stores the data persistently | Capacity, endurance, base speed |
| Controller | Runs the FTL, error correction and the host interface | Performance consistency and reliability |
| DRAM cache or HMB | Holds the address mapping table | Fast random access, especially on large drives |
| SLC cache | Temporarily writes in one-bit mode | High burst write speed |
| Over-provisioning | Spare capacity hidden from the user | Lower write amplification, longer life |
Cell types and endurance
Each NAND cell stores a number of bits determined by how many charge levels it must distinguish. Every extra bit doubles the number of levels, which raises density and lowers cost but reduces endurance and write speed.
| Type | Bits per cell | Charge levels | Approximate P/E cycles | Typical use |
|---|---|---|---|---|
| SLC | 1 | 2 | 50,000 to 100,000 | Industrial, caching layers |
| MLC | 2 | 4 | 3,000 to 10,000 | Older enterprise drives |
| TLC | 3 | 8 | 1,000 to 3,000 | Mainstream client and enterprise |
| QLC | 4 | 16 | About 300 to 1,500 | High-capacity, read-heavy storage |
Manufacturers express endurance as TBW (terabytes written) or DWPD (drive writes per day) over the warranty period. A 1 TB drive rated at 600 TBW with a five-year warranty allows 600 / (1 x 365 x 5) = about 0.33 drive writes per day, which is ample for typical desktop use of a few tens of gigabytes per day.
Interfaces and form factors
The interface sets the ceiling for sequential throughput. Older SATA drives are limited by the 6 Gbit/s link, while NVMe drives connect directly to PCIe lanes and use a protocol designed for flash, with up to 65,535 command queues instead of the single queue of AHCI.
| Interface | Typical form factor | Practical sequential read | Notes |
|---|---|---|---|
| SATA III | 2.5-inch, M.2 | About 550 MB/s | Universal compatibility, AHCI protocol |
| NVMe over PCIe 3.0 x4 | M.2 2280 | About 3.5 GB/s | Common in older laptops |
| NVMe over PCIe 4.0 x4 | M.2 2280 | About 7 GB/s | Current mainstream |
| NVMe over PCIe 5.0 x4 | M.2 2280, E1.S, E3.S | About 14 GB/s | Higher power, often needs a heatsink |
An M.2 slot is a form factor, not a protocol: the same connector can carry SATA or NVMe, and a drive works only if the slot supports its protocol.
How SSDs compare with hard disk drives
The decisive advantage of an SSD is access latency, not sequential speed. A hard disk must move its head and wait for the platter to rotate, which takes about 5 to 10 ms, whereas an NVMe SSD answers a random read in about 50 to 100 microseconds.
| Property | SSD | HDD |
|---|---|---|
| Random 4 KB read | Hundreds of thousands of IOPS or more | About 100 to 200 IOPS |
| Sequential read | 0.5 to 14 GB/s | About 150 to 280 MB/s |
| Shock resistance | High, no moving parts | Low during operation |
| Cost per terabyte | Higher | Several times lower |
| Wear mechanism | Limited write cycles | Mechanical wear |
| Unpowered retention | Can degrade over long periods | Generally stable for years |
Limitations and common mistakes
Advertised SSD speeds describe best-case bursts. Sustained performance depends on the cell type, the SLC cache size, the controller and how full the drive is.
- Ignoring post-cache write speed: once the SLC cache is full, some QLC drives drop to a few hundred MB/s or less.
- Filling the drive completely: with little free space, garbage collection works harder and both speed and endurance suffer.
- Using SSDs as unpowered archives: flash cells slowly lose charge, so long-term cold storage needs periodic power or other media.
- Defragmenting an SSD: it adds writes without benefit; the operating system should run TRIM instead.
- Overlooking thermal throttling: fast PCIe 4.0 and 5.0 drives can slow down without adequate cooling.
The idea in brief
An SSD is the part of a computer that keeps files, apps and the operating system when the power is off. It does this with memory chips instead of a spinning disk.
Because nothing has to move, the computer finds any file almost instantly. This is why a computer with an SSD starts in seconds and opens apps quickly. The trade-off is that each memory cell can be rewritten only a limited number of times, although a normal user rarely comes close to that limit.
A practical scenario
A small accounting office runs its accounting database on a server with four hard disk drives in RAID 10. At month-end closing, many users run reports at the same time, and simple queries take up to a minute because the disks deliver only a few hundred random IOPS in total.
The office replaces the disks with two enterprise TLC NVMe SSDs in a mirror, each rated at 1 DWPD. The database fits easily on the 3.84 TB drives, and the same month-end reports now complete in a few seconds.
Daily writes average about 150 GB, far below the rated 3.84 TB per day, so the drives have a large endurance margin for their five-year service life.