How long does flash hold data with the power off? Data retention explained
- Endurance and retention are different axes: endurance is how many writes a cell survives; retention is how long it holds its charge once the power is off. A cell can be nowhere near worn out and still lose data if it sits hot and unpowered long enough.
- Two things drain retention — wear and heat. Every program/erase cycle shrinks the cell's voltage margin, and a higher power-off storage temperature speeds up charge leakage, roughly halving retention for every ~5 °C rise (JEDEC's Arrhenius-based figure).
- The JEDEC numbers are worst case, not everyday: JESD218 specifies retention at the drive's full rated endurance (100% worn) — client 1 year at 30 °C, enterprise 3 months at 40 °C, powered off. A fresh drive at room temperature holds data for years; the scary "loses data in days" headlines came from misreading an end-of-life, hot, enterprise case.
- It matters for export stock and archives: flash is not good unpowered long-term archival media. The real risks are parts that sit months in a 50–60 °C warehouse or container, and worn or remarked grey-market parts that have already spent their margin. Spec industrial / pSLC grade, and power drives up periodically so the controller can refresh.
Most people ask "how long does an SSD last?" and mean how many times can I write to it. That's endurance. There's a second, quieter question that catches people out: once you've written your data and switched the power off, how long does the drive remember it? That's data retention, and it's a different axis entirely. A drive can be barely used — almost all its endurance intact — and still lose data if it sits hot and unpowered for long enough.
How a flash cell forgets
A NAND cell stores bits by trapping electrons on a floating gate. The amount of trapped charge sets a voltage level, and the controller reads the level back as your data. Nothing holds those electrons in place actively — over time they leak away, and when enough have leaked, the voltage drifts far enough that a bit flips. The error-correction (ECC) catches a few; past that, the data is gone.
Two things govern how fast that happens:
- Wear. Every program/erase cycle damages the oxide a little and shrinks the voltage margin between states. As P/E cycles climb, retention drops — a fresh cell and a near-worn-out cell holding the "same" data are not equally safe [3].
- Heat. Charge leakage is a thermally driven process. The hotter the cell sits while powered off, the faster the electrons escape. This follows the Arrhenius relationship — roughly, retention halves for every ~5 °C rise in power-off storage temperature [4].
There's a counter-intuitive twist: while heat hurts powered-off retention, a little warmth actually helps the act of writing (cold cells program less cleanly). That's why the JEDEC model cross-references two temperatures — the temperature during use, and the temperature in storage.
What JEDEC actually requires
The standard is JEDEC JESD218. The key thing to understand is the scenario it specifies: a drive is written all the way to its full rated endurance (100% worn — the worst case), then powered off, and must still retain data for a set time at a set temperature [1]:
| Application class | Active use (powered on) | Power-off retention requirement |
|---|---|---|
| Client | 40 °C, 8 hrs/day | 1 year at 30 °C |
| Enterprise | 55 °C, 24 hrs/day | 3 months at 40 °C |
Read that table carefully, because almost every "SSDs lose data fast!" scare misreads it. The 1-year and 3-month figures are floors at end of life — the guarantee for a drive that is fully worn out. Enterprise looks worse than client only because enterprise drives are rated for vastly more writes, so their cells are more worn when they hit "100%."
The temperature curve — and the 2015 myth
The famous chart behind the headlines comes from a JEDEC presentation by Alvin Cox [4]. It's an Arrhenius lookup table: you cross-reference your power-off (storage) temperature against your power-on (use) temperature, and read retention in weeks.
A worked reading: a client drive at end of endurance, stored at 30 °C with a 40 °C use history, lands on 52 weeks — exactly the JEDEC client spec. Move the storage temperature and it swings hard: the same client curve runs from about 404 weeks (~7.7 years) at 25 °C down to roughly 8 weeks at 55 °C [4]. That steepness is the whole story.
So why were the 2015 "data gone in days" headlines wrong? Three reasons:
- They were worst-case, end-of-life numbers — a fully worn drive, not the lightly used one in your laptop.
- "Retention" means reliable readback, not destruction. The number marks when the first bits get risky, with engineering safety margin baked in. At room temperature, real data often stays readable for many years beyond the spec figure.
- No one guarantees an exact number — it depends on wear, design, NAND generation and temperature history [1][4].
For everyday use the situation is reassuring: a normally-used drive kept at room temperature holds your data for years. The danger is specific — a heavily worn drive stored hot.
Cell type sets the starting margin
How many bits a cell holds decides how much margin there is to lose. SLC, MLC, TLC and QLC pack 1, 2, 3 and 4 bits into the same voltage window — the more levels crammed in, the smaller the gap between them, and the less leakage it takes to flip a bit. SLC has the widest margin and typically holds ~10 years at operating temperature without help; QLC has the least. pSLC — running TLC or MLC at one bit per cell — buys back most of that margin, which is exactly why industrial cards and SSDs lean on it for harsh or long-dormant roles.
Why this matters for sourcing and export
Retention is usually invisible — until your product sits unpowered, hot, for a long time. That describes a lot of real B2B situations:
- Stock in transit and storage. A pallet of cards or SSDs can spend months in a warehouse or shipping container that hits 50–60 °C in a Gulf or Latin-American summer — powered off the entire time. Fresh, cool-rated parts are fine for years, but it's a real argument for industrial-grade or pSLC parts on long-dormant or write-heavy deployments, and for not letting inventory bake.
- Grey-market and remarked parts. A used or remarked part may have already spent much of its endurance — and therefore its retention margin — before you ever power it on. You can't see consumed retention on a label; it's one more reason to buy from a known source and write-verify a sample.
- Dormant devices. Dashcams and security cards left in a hot parked car combine both stressors — wear from constant logging and heat — so retention and endurance both have to be specced.
The good news for anything that stays in service: modern controllers run a background refresh when powered — they watch how long blocks have held data and rewrite the ones approaching their limit. A drive that gets powered on periodically retains far better than the shelf spec. A drive left in a drawer doesn't get that help — which is the one-line reason flash is not a good cold-archive medium.
Bottom line
Endurance asks how many times you can write; retention asks how long it remembers. Retention falls with wear and, steeply, with power-off heat — but the alarming JEDEC numbers are worst-case, end-of-life floors with safety margin, not everyday behaviour. Keep unpowered media cool, don't rely on flash as offline cold storage, and for stock that may sit hot for months or for write-heavy roles, spec the grade up front. Tell us the operating and storage temperatures and how long parts may sit unpowered, and we'll spec the NAND grade and retention headroom to match — with the conditions in writing.
FAQ
How long does an SSD keep data when it is unplugged?
Does heat erase SSD or flash data?
Is an SSD or USB flash drive good for long-term backup or archiving?
Which flash type retains data the longest?
References
- JEDEC — JESD218 Solid-State Drive (SSD) Requirements and Endurance Test Method (defines power-off retention)
- ATP — SSD data retention in high-temperature environments
- Macronix — Program/Erase cycling endurance and data retention of NAND flash (App Note AN0339)
- JEDEC — SSD Specifications Explained (Alvin Cox, JC-64.8): the data-retention vs temperature chart
We publish measured usable capacity and welcome trial-batch verification — automotive-grade, direct from the source factory.
Get a quote