An SLC cache is a portion of an SSD’s multi-level NAND flash that the controller programs in single-level cell (SLC) mode, storing one bit per cell, so that incoming writes land quickly before being folded into denser TLC or QLC storage later. In storage hardware, it is the main reason consumer drives can advertise sequential write speeds of several gigabytes per second while their native TLC or QLC write speed is far lower.
The cache is not a separate chip. It is a dynamic or static allocation of ordinary flash blocks, managed by the firmware, and its size, behavior when full and recovery time define how an SSD performs under sustained load.
How an SLC cache accelerates writes
An SLC cache works because programming one bit per cell is much faster than programming three or four. A TLC cell must be placed into one of eight voltage states and a QLC cell into one of sixteen, which requires many fine program-and-verify steps. In SLC mode the controller only distinguishes two states, so each page programs in a fraction of the time and with less stress on the oxide layer.
The process runs in three stages. First, host writes are accepted into SLC-mode blocks at high speed. Second, during idle time the controller performs folding, also called migration or compaction, reading the cached data and rewriting it into blocks programmed in native TLC or QLC mode. Third, the freed blocks are erased and returned to the cache pool, ready for the next burst.
The following table compares the cell modes that matter for caching, with typical figures for current consumer NAND flash. Exact values depend on the NAND generation and the controller.
| Mode | Bits per cell | Voltage states | Relative program speed | Typical use in a drive |
|---|---|---|---|---|
| SLC mode | 1 | 2 | Fastest | Write cache, metadata |
| MLC | 2 | 4 | Fast | Rare in new consumer drives |
| TLC | 3 | 8 | Moderate | Main storage in most SSDs |
| QLC | 4 | 16 | Slowest | High-capacity, low-cost SSDs |
Static, dynamic and hybrid SLC cache designs
Manufacturers implement the SLC cache in three main ways, and the choice determines how much cache is available at a given fill level. A static cache reserves a fixed set of blocks, often taken from over-provisioned space, that always run in SLC mode. A dynamic cache borrows free user capacity and converts it to SLC mode on demand, so it can be very large on an empty drive and shrinks as the drive fills.
Most modern drives use a hybrid design that combines a small static portion with a large dynamic portion. Because an SLC-mode block holds only one third of what it holds in TLC mode, a dynamic cache consumes three or four times its own size in raw capacity, which is why it must shrink when user data occupies the space.
| Design | Size behavior | Main advantage | Main drawback |
|---|---|---|---|
| Static | Fixed, typically small | Predictable, always available | Limited burst length |
| Dynamic | Shrinks as the drive fills | Very large when the drive is empty | Collapses on a nearly full drive |
| Hybrid | Fixed floor plus variable part | Balances size and consistency | More complex firmware behavior |
Dynamic cache sizes vary widely. On a 1 TB TLC drive, an empty-drive cache can range from tens of gigabytes to several hundred gigabytes, while a nearly full drive may retain only the static floor of a few gigabytes. Vendors rarely publish these numbers, so independent sustained-write testing is the usual way to measure them.
What happens when the cache fills
When a sustained write exceeds the cache, the drive must write directly to native TLC or QLC and often fold cached data at the same time. The result is a sharp post-cache speed drop, sometimes called the write cliff, which is the most visible performance limitation of SLC caching.
The severity of the drop depends on cell type, die count and controller design. Drives with more NAND dies working in parallel sustain higher post-cache speeds, which is why larger capacities usually degrade less. QLC drives are most affected, and some fall below the speed of a hard disk in the worst phase, when the controller must simultaneously fold old data and accept new data.
The table below shows the typical write phases of a consumer NVMe drive during a long sequential copy. The ranges are indicative and differ by model.
| Phase | What the controller does | Indicative speed on TLC | Indicative speed on QLC |
|---|---|---|---|
| In cache | Writes in SLC mode | 3 to 7 GB/s | 2 to 5 GB/s |
| Direct write | Writes in native mode | 1 to 2 GB/s | 0.3 to 0.8 GB/s |
| Folding under load | Folds and writes together | 0.5 to 1.5 GB/s | 0.08 to 0.2 GB/s |
| Idle recovery | Folds, frees blocks | No host writes | No host writes |
Recovery time after a burst ranges from seconds to several minutes, depending on how much data must be folded. Workloads with natural pauses, such as everyday desktop use, rarely exhaust the cache, while video capture, large backups and disk imaging regularly do.
SLC cache, DRAM cache and host memory buffer
The SLC cache is often confused with the DRAM cache, but the two serve different purposes. DRAM on an SSD mainly stores the logical-to-physical mapping table, so the controller can locate data quickly; it does not hold large amounts of user data. The SLC cache, by contrast, is non-volatile flash that holds actual written data.
A DRAM-less SSD usually relies on the Host Memory Buffer (HMB) feature of NVMe, borrowing a small amount of system RAM for mapping data. Such drives still use an SLC cache for writes, so the absence of DRAM and the size of the SLC cache are independent properties that should be evaluated separately.
- SLC cache: absorbs write bursts in fast flash and survives power loss.
- DRAM cache: speeds up address translation and is volatile.
- HMB: gives DRAM-less NVMe drives part of the DRAM benefit through system memory.
Endurance, wear leveling and drive fill level
SLC caching affects endurance in two opposing ways. SLC-mode programming is gentler on cells, and blocks used in SLC mode tolerate many more program/erase cycles. However, every byte written to the cache is written again during folding, which adds to write amplification and consumes part of the drive’s TBW rating.
Firmware integrates the cache with wear leveling, rotating which blocks serve as cache so that no region wears out early. TRIM also matters: when the operating system reports deleted data, the controller gains free blocks for a dynamic cache, so regular TRIM keeps the cache larger and recovery faster.
Practitioners commonly recommend keeping 10 to 20 percent of capacity free on drives with dynamic caches. The following table summarizes frequent problems and their usual causes.
| Symptom | Likely cause | Practical remedy |
|---|---|---|
| Copy speed collapses after some gigabytes | Cache exhausted during a sustained write | Choose a drive with higher post-cache speed |
| Drive slower than when new | High fill level shrinks the dynamic cache | Free space and ensure TRIM is active |
| Benchmarks look excellent, real copies do not | Test file fits entirely in the cache | Test with transfers larger than the cache |
| Slow writes right after a large transfer | Folding still in progress | Allow idle time before the next burst |
The simple explanation
Think of an SSD as a warehouse where storing each item carefully on a high shelf takes time. The SLC cache is a loading dock near the entrance. New deliveries are dropped there quickly, and workers move them to the proper shelves later when things are quiet.
As long as deliveries come in batches, the dock keeps everything fast. If a truck keeps unloading for a long time, the dock fills up and every item must go straight to the slow shelves. That is why an SSD can be very fast for ordinary files but slow down during a huge copy.
How it is used in practice
A video editor records 8K footage from a camera and copies about 300 GB per shooting day onto a 2 TB QLC NVMe drive that is already 85 percent full. The first 25 GB copy at more than 3 GB/s, then the transfer drops to about 150 MB/s, and the full copy takes over half an hour instead of the expected two minutes.
After measuring the behavior with a long sequential write test, the editor moves active projects to a 2 TB TLC drive with a large static SLC portion and a post-cache speed of about 1.5 GB/s, while keeping the QLC drive for archived material. The same daily copy now finishes in under four minutes, and the archive drive, which mostly receives occasional small writes, no longer limits the workflow.