An SSD stored in a drawer, an external drive forgotten on a shelf, an old system disk kept aside just in case: we assume they are safe because nothing spins or heats up anymore. That assumption is only partly true. The flash memory inside an SSD records information as electrical charge, and that charge slowly leaks away over time. While the drive is powered, its controller compensates. Once it is switched off, nothing corrects the drift. This article explains how the mechanism works, what the JEDEC standards actually require, and how to reduce the risk.
How a NAND cell holds (and loses) information
An SSD keeps its data in NAND memory chips. Each cell is a transistor in which electrons are trapped during a write operation, using a high voltage: the amount of trapped charge determines the value the controller reads back later. Depending on the generation, those electrons are held in a conductive floating gate (the historical planar design) or in an insulating layer that captures them at discrete sites (charge trap, widely used in 3D NAND).
In both cases the insulating barrier is not perfect. Over time, some electrons cross it, through quantum tunnelling or thermal agitation, and the cell’s threshold voltage drifts. Academic research on flash reliability calls these retention errors: the longer the time since the data was written, the more read errors appear.
While powered, an SSD has tools to handle this drift: error-correcting codes (ECC), read retries with adjusted voltage thresholds, and rewriting ageing data elsewhere in the memory. Dell’s documentation on enterprise SSDs points out that these retention operations cannot run while the drive is off. The charge leakage, however, carries on.
What the JEDEC standards say
The industry reference is JEDEC’s JESD218 standard, which defines endurance and retention requirements for SSDs by usage class. It does not promise that data will last indefinitely: it sets a minimum, measured under specific conditions.
| Usage class | Power-off storage temperature | Minimum retention time |
|---|---|---|
| Client (consumer) | 30 °C | 1 year |
| Enterprise (server) | 40 °C | 3 months |
One point is often misunderstood: these durations apply to a drive that has reached the endurance rating declared by its manufacturer (the TBW, or terabytes written). In other words, this is the standard’s worst case, not the lifespan of a new drive. A lightly written SSD should normally retain data longer than this minimum. On the other hand, the standard does not publish a guaranteed duration for a specific model, and manufacturers rarely give a detailed figure.
Two further points complete the picture. First, the enterprise class is deliberately less demanding on duration, because these drives are designed to stay powered around the clock, and Dell states that a system with enterprise SSDs left off for more than 3 months may show errors. Second, the figures are verified through temperature-accelerated testing with a 60% statistical confidence level, not by actually waiting a year: they are a design threshold, not an individual promise.
The factors that shorten retention
Temperature
Charge leakage is thermally activated: the hotter the storage, the faster it goes, and the acceleration is exponential (the Arrhenius model, which JEDEC applies with an activation energy of 1.1 eV for this mechanism). You will sometimes read that every 10 °C halves the retention time. That is a rule of thumb that varies between sources and chips, not a normative value. What matters is the direction of the relationship. A temperate cupboard is better than an attic, a garage or a car glovebox. Dell recommends keeping stored enterprise drives at an ambient temperature no higher than 40 °C.
Drive wear
Every write and every erase forces electrons through the insulating layer. These program/erase (P/E) cycles wear it down gradually, and a tired cell holds its charge less well. This is why the standard reasons in terms of end of endurance: a heavily used drive, pulled from a workstation or a server and put away as an archive, is the most exposed profile.
Bits per cell
The more bits a cell stores, the more charge levels must be told apart: 2 for SLC (1 bit), 4 for MLC (2 bits), 8 for TLC (3 bits) and 16 for QLC (4 bits). The margin between two levels shrinks accordingly, so the same leakage becomes a problem sooner. Most consumer SSDs today use TLC or QLC. The retention times quoted per cell type vary widely from one source to another and are not confirmed by manufacturers: treat them as orders of magnitude, not guarantees.
2D or 3D architecture
3D NAND stacks cells in several layers and mostly relies on charge-trap technology. It behaves differently from planar NAND over time and temperature: published measurements on 3D chips show, among other things, faster charge loss shortly after programming and a distinct sensitivity to temperature. It would therefore be wrong to conclude that 3D is simply “safer” or “less safe” in every situation.
What happens to a file when the charge weakens
Degradation is gradual. At first the affected cells remain readable or are corrected by ECC without the user noticing anything. Then the number of faulty bits exceeds what the correction can handle: a file will not open, a photo appears truncated, an archive refuses to extract. If the damage reaches the structures that describe the drive’s content (partition tables, file system metadata, the controller’s internal tables), the whole volume can become unreadable, appear unformatted, or no longer be detected at all.
Unlike a hard drive, which often gives warning signs (noises, slow sectors, SMART alerts), an SSD can look perfectly healthy until you try to read old data. It is therefore impossible to say exactly when an unpowered drive becomes risky: it depends on the model, the wear, the temperature and the manufacturing batch.
Good practice for keeping your data
- Never make an unpowered SSD the only copy. The 3-2-1 rule (three copies, on two types of media, one of them off-site) remains the foundation, as CISA among others points out.
- Store it cool and dry. A temperate room, away from heat sources, well below 40 °C.
- Power drives up regularly. For enterprise SSDs holding data, Dell advises powering them on every 2.5 months for at least 3 weeks, because the time needed grows with capacity. No equivalent recommendation has been published for consumer drives, and a brief connection does not guarantee that every cell is read: it is a full read of the drive that triggers the maintenance operations.
- Monitor wear. SMART health and endurance indicators help you rule out a heavily worn drive for archival use.
- Verify your copies. A checksum (hash) calculated at archive time and checked again later reveals silent alteration before it is too late.
- Choose the medium for the intended duration. For long-term archives, a mechanical hard drive or a dedicated medium, duplicated and tested regularly, is a better fit than an SSD in a drawer. A hard drive is not immune either: it also needs to be copied and checked.
These precautions matter for businesses too. An SSD pulled from a NAS or server after replacement and kept “for the record” combines several unfavourable factors: high wear, an enterprise class that is less demanding on unpowered duration, and storage that is often not climate-controlled. Today’s market, with component shortages, also encourages reusing or keeping older drives: our article on the SSD and hard drive shortage covers what this means for your data.
What to do if the SSD no longer responds or files are corrupted
The right reflex is to limit handling. Avoid running a format, an automatic file system repair or a drive reset: these operations write to the medium and can permanently ruin data that was still partly readable. If the drive is detected, copying what can be read to another medium, in read-only mode, is better than repeated attempts. In a RAID environment the same caution applies: a rebuild started on weakened drives increases the risk.
Recovering data from a degraded SSD is laboratory work: direct reading of the memory chips, analysis of residual voltages, reconstruction of the mapping between logical and physical addresses, then advanced error correction. The outcome depends on how much charge remains. When the degradation is recent and limited, a significant share of the files can often be reconstructed. When the charge has almost disappeared, no tool can recover the information, because the cell no longer holds any trace of it. Only a diagnosis can tell which situation your drive is in.
An SSD is not fragile by nature: it is designed to work powered, and that is when it protects its data best. Stored unpowered, it should be treated like any archive medium, with redundancy, careful storage and regular checks. If files are already unreadable or the drive is no longer detected, a quick opinion from a specialist limits the risk of making things worse.