What are the NAND flash memory types and where do they work best?

Single-level cell

Single-level cell (SLC) NAND, as the name indicates, stores one bit in each cell. There are two states SLC NAND can be in: programmed (0) or erased (1), which depends on the level of charge that is applied to the cell. Because the number of potential states is limited to two, determining the state of a cell is a quick process.

SLC NAND is the simplest of the NAND flash memory types and has a very low chance of error. SLC NAND memory cells can take about 100,000 write operations before failure, giving them the highest endurance of the NAND types. However, it is also the most expensive option; it's more than twice the price of multi-level cell NAND. Because of the high price, performance and reliability of SLC, it is most often used in commercial and industry applications.

Multi-level cell

Perhaps the most cryptically named of the bunch, multi-level cell (MLC) NAND stores two bits per cell. This means that there are four possible states (00, 01, 10, 11,) as opposed to SLC's two. Serving as a midpoint between single- and triple-level cells, MLC has a higher bit rate than SLC, which lowers the number of write-cycles it can sustain and increases the chance of error. MLC NAND can take about 10,000 write-cycles per cell.

As mentioned previously, however, MLC NAND is significantly less expensive than SLC and is the right choice for many use cases. Manufacturers of consumer-based electronic devices, such as PCs, are partial to MLC NAND thanks to its lower price tag.

However, MLC NAND is not limited to consumer-grade flash. Enterprise MLC (eMLC) is an enhanced type of MLC NAND that accommodates more write-cycles than the consumer variety. EMLC NAND can handle 20,000-30,000 write-cycles, making it better suited to more demanding use cases while avoiding steep SLC costs.

Triple-level cell

Boasting three bits per memory cell, triple-level cell (TLC) NAND is another type of NAND flash memory suited to consumer-level products. Also referred to as MLC-3, 3-bit MLC and X3, TLC NAND comes with a lower price tag than both SLC and MLC NAND. There are eight possible states for TLC NAND: 000, 001, 010, 011, 100, 101, 110 and 111.

TLC has a higher storage density than SLC and MLC and a lower cost per bit. However, these come at a cost; TLC NAND has lower performance, longevity and reliability. TLC NAND benefits from the addition of 3D NAND, expanding its potential enterprise use cases.

Quad-level cell

Quad-level cell (QLC) NAND stores four bits in each memory cell. Continuing with the previous trend, QLC NAND is cheaper than the above options, but has lower endurance for writes. QLC was developed for the additional storage capacity it provides SSDs and is capable of faster reads than other types of NAND flash.

QLC NAND is suited particularly to read-intensive applications and is used for applications supporting AI, machine learning and deep learning, where data is typically written once. It is not suited to write-intensive workloads, supporting around 100 write cycles. As with other NAND flash memory types, the addition of 3D NAND can boost the number of write cycles QLC NAND can endure. However, with QLC, this change only increases the number of write cycles to about 1,000.

3D NAND

While 2D or planar NAND has one layer of memory cells, 3D NAND stacks cells vertically in multiple layers. 3D NAND SSDs are available using MLC, TLC and QLC NAND technology, but not SLC NAND. With 3D NAND architecture, an SSD has much higher density than with planar NAND, in a smaller physical space. Higher density means that 3D NAND SSDs are lower in cost per gigabyte, require less power consumption and have a higher write performance.

However, there are some disadvantages to using 3D NAND instead of 2D. The cost of manufacturing 3D NAND is higher at the outset, requiring additional steps not necessary in the production of 2D NAND SSDs. 3D NAND can be used in all the same scenarios as planar NAND, but its use will depend on the organization's requirements and budget constraints.