Glossary
TLC flash
TLC flash (triple-level cell flash) is NAND flash that stores three bits in every memory cell by holding the cell at one of eight distinct charge levels.
It is the mainstream cell type in data-centre SSDs, sitting between MLC, which stores two bits per cell, and QLC, which stores four.
Why TLC flash matters for large storage estates
Once a storage estate reaches petabytes, the choice of flash cell type stops being a component detail and becomes a cost and lifetime decision. TLC is the default for flash that takes a steady stream of writes: metadata, small objects, indexes, databases and the hot working set of analytics and AI jobs. QLC packs more capacity into the same silicon but tolerates far fewer writes. Hard drives remain cheaper still per terabyte. Deciding which data lands on TLC, which on QLC and which on disk sets a large share of the hardware budget and the replacement schedule for the next five years.
How TLC stores three bits per cell
A NAND flash cell holds electrical charge in an insulated layer, and the amount of charge sets the voltage at which the cell conducts. Reading the cell means working out which voltage band it sits in. Three bits need 2³ = 8 bands, separated by seven boundaries, all squeezed into roughly the same voltage window that a single-bit cell divides once.
Narrow bands have three consequences. Programming takes longer, because the controller nudges each cell into place with a series of small pulses and checks after each one. Reads need stronger error correction, because a small drift in charge can push a cell across a boundary. And the cell wears out sooner, because every program and erase cycle damages the insulating layer a little and a worn TLC cell crosses a boundary earlier than a worn single-bit cell would.
TLC compared with other cell types
| Cell type | Bits per cell | Charge levels | Capacity per cell relative to TLC | Typical role today |
|---|---|---|---|---|
| SLC | 1 | 2 | 0.33× | Write caches inside drives, write-heavy logs |
| MLC | 2 | 4 | 0.67× | Largely replaced by TLC in new designs |
| TLC | 3 | 8 | 1× | Mainstream data-centre SSDs, read-intensive and mixed-use |
| QLC | 4 | 16 | 1.33× | High-capacity, read-dominated flash tiers |
Density also grows through stacking. 3D NAND builds cells in many vertical layers, and each added layer multiplies with bits per cell. A modern TLC die therefore holds far more data than the cell count of a flat chip would suggest.
Endurance and write behaviour
Drive vendors express TLC endurance as drive writes per day (DWPD) over the warranty period, or as total terabytes written. Data-centre TLC drives are commonly sold in two classes: read-intensive drives rated around one drive write per day, and mixed-use drives rated around three. A 7.68 TB drive at 1 DWPD for five years is rated for 7.68 × 365 × 5 = 14,016 TB of host writes.
The rating covers what the host sends. Inside the drive, garbage collection moves valid data around to free blocks, so the flash absorbs more writes than the host issued. The ratio is called write amplification. Large sequential writes of whole objects keep it close to 1; small random overwrites of a nearly full drive push it well above 2, which halves the useful endurance. Client drives also hide TLC's slow programming behind an SLC cache that fills during long writes; data-centre drives are specified for sustained, steady-state rates instead.
What TLC flash means for large-scale storage
The practical consequence is that endurance becomes a fleet-level budget. A 2 PB TLC tier rated at 1 DWPD can absorb roughly 2 PB of host writes a day for the warranty period. A platform ingesting 200 TB a day, stored with an erasure-coding overhead of 1.33×, writes about 266 TB a day to that tier and uses around one eighth of its endurance. The same tier holding a metadata store that rewrites small records in place can wear far faster than the ingest figure suggests, because write amplification multiplies the load.
For a team planning flash at this scale, TLC versus QLC turns on the write pattern of each data set more than on raw performance. Data that is written once and read many times, such as AI training corpora, media archives and backup copies, rarely uses the endurance TLC pays for. Data that churns, such as object metadata, indexes and small-object buckets, is where TLC's extra endurance and steadier write latency earn their cost per terabyte.
Drive size changes the failure picture as well. TLC drives of 15 TB and more put more data behind each failure, so rebuild time and the traffic it generates grow with drive capacity. The tiering design, the protection scheme and the drive size are best weighed together, since each affects what a single drive failure costs in time and bandwidth.
TLC flash in Scality RING
Scality RING is software-defined object and file storage that runs on standard x86 servers, so the flash cell type in a RING cluster is a property of the drives chosen for those servers. RING does not depend on a particular NAND type, which leaves the TLC, QLC and hard drive mix as a design choice per deployment. Hard-drive-based RING clusters commonly place metadata on flash, and the all-flash RING XP configuration uses NVMe mixed-use drives.
When a drive fails, RING rebuilds within the affected server, writing data across its remaining drives and rebuilding only data that was actually written. For a large TLC drive that is only partly filled, rebuild work tracks the data it actually holds.