A specialist will know this topic more deeply - but if you want to do research yourself then focus on magnetic stability; hdd designs have magnetic domains which can remain stable over very long periods unless something external disrupts them, there are some interesting articles out there with these key terms.

i understand SSD relies on an electrical charge which slowly dissipates away

HDD spinny disks store data as a series of magnetic oscillation patterns using a methodology called RLL. (IDE and scsi disks still use RLL, even though the name evokes an earlier period in hard disk history. It uses a different, more dense version of RLL encoding, but it is still RLL.) This encodes binary data as a series of magnetic inversion pulses which are then transferred to the media, literally by magnetizing it with the read/write head.

The durability of these magnetic flux inversion patterns depends on factors such as, the health of the media surface, the temperature of the room it is stored in, and the like, but CAN undergo problems where the flux patterns flip themselves because of magnetic forces exerted by nearby magnetic domains. This phenomenon is known as "bit rot" and does occur with spinny disks. its just rather slow.

SSDs on the other hand, are a very different beast.

These are flash memory devices, which store data in a series of cells. Depending on the flash design, this could be SLC, for Single Layer Cell, or MLC for Multi-Layer Cell. SLC flash is considerably more performant than MLC, but also more expensive to manufacture, and as a consequence, nearly all modern SSDs are MLC flash.

A flash chip stores data as a "captured" electronically excited state, sort of like how a capacitor stores a small charge. This small stored charge changes how the cell conducts electricity, and thus switches a cell from a 0 to a 1, when there is a captured charge.

The process of storing this charge and releasing this charge, very slowly degrades the cell, because it slowly breaks down the doping layers used to make the flash into a semiconductor. After a certain number of writes, the cell becomes unstable, and the built in flash controller will lock it.

Because these are captured electric charged, they can self-discharge over long periods of being stored on a shelf. This is why Flash devices can lose data when stored in a drawer.

wierd_w wrote on 2025-05-16, 09:51:

HDD spinny disks store data as a series of magnetic oscillation patterns using a methodology called RLL. (IDE and scsi disks sti […]
Show full quote

HDD spinny disks store data as a series of magnetic oscillation patterns using a methodology called RLL. (IDE and scsi disks still use RLL, even though the name evokes an earlier period in hard disk history. It uses a different, more dense version of RLL encoding, but it is still RLL.) This encodes binary data as a series of magnetic inversion pulses which are then transferred to the media, literally by magnetizing it with the read/write head.

The durability of these magnetic flux inversion patterns depends on factors such as, the health of the media surface, the temperature of the room it is stored in, and the like, but CAN undergo problems where the flux patterns flip themselves because of magnetic forces exerted by nearby magnetic domains. This phenomenon is known as "bit rot" and does occur with spinny disks. its just rather slow.

SSDs on the other hand, are a very different beast.

These are flash memory devices, which store data in a series of cells. Depending on the flash design, this could be SLC, for Single Layer Cell, or MLC for Multi-Layer Cell. SLC flash is considerably more performant than MLC, but also more expensive to manufacture, and as a consequence, nearly all modern SSDs are MLC flash.

A flash chip stores data as a "captured" electronically excited state, sort of like how a capacitor stores a small charge. This small stored charge changes how the cell conducts electricity, and thus switches a cell from a 0 to a 1, when there is a captured charge.

The process of storing this charge and releasing this charge, very slowly degrades the cell, because it slowly breaks down the doping layers used to make the flash into a semiconductor. After a certain number of writes, the cell becomes unstable, and the built in flash controller will lock it.

Because these are captured electric charged, they can self-discharge over long periods of being stored on a shelf. This is why Flash devices can lose data when stored in a drawer.

Your last point is much less of an issue with newer Nand than it was with the older stuff, newer SSD Nand can have data retention times longer than Spinning Rust can, if stored in a controlled environment naturally. Still neither storage medium should be used alone and its always best to have multiple long term storage setups for critical data. Its also good to remember you do get what you pay for with Nand devices, the cheaper it is the worse its going to be at keeping your data safe.

IIRC some enterprise SSDs now have onboard long-term battery backup via a small long life lithium cell right on the PCB, it doesn't take much energy to keep the cells stable.

The issue I have with these "Long Term" data storage options is .. will we even be able to read these device in a hundred years ? or even 50, look at how difficult it is right now to read data easily and reliably from HDDs, Discs, Optical media that are at beast only 20-30 years old. Will we even have devices capable of interfacing with current SATA/NVME/USB devices or still have the knowledge required to do so 100 years down the road ? more interestingly . .will the data even if we can retrieve it be usable on future systems.

I think for historical record data these long term storage requirements will be useful but I doubt storing a DOS game from 1980 wil lbe of any real use 100 years from now.

In layman's terms - information on the HDD is stored in magnetic domains, they resist any attempts to change it, you need to apply a significantly strong field (read as spend some energy and efforts) to flip the magnetic domain orientation.
The SSD store the information as a charge in a tiny capacitors, isolated by imperfect materials, and that charge wants to leak away. Nothing can stop it leaking, and the quality of materials only define how fast the charge is leaking, but not whether it will leak or not, it is not preventable.
So magnetic HDD won't lose information without significant destroying efforts applied, while SSD loses the information without any involvement. SSD requires that the information is refreshed, either as side effect of wear leveling mechanisms (data is moved around to make space for new erased blocks), or when the error detection is triggered during any access to the information, and then controller would rewrite the corrected data. This requires the SSD not only being powered, but actively used to keep the data alive. The data needs to be touched in order to be kept alive.

vstrakh wrote on 2025-05-16, 10:07:

In layman's terms - information on the HDD is stored in magnetic domains, they resist any attempts to change it, you need to apply a significantly strong field (read as spend some energy and efforts) to flip the magnetic domain orientation.
The SSD store the information as a charge in a tiny capacitors, isolated by imperfect materials, and that charge wants to leak away. Nothing can stop it leaking, and the quality of materials only define how fast the charge is leaking, but not whether it will leak or not, it is not preventable.
So magnetic HDD won't lose information without significant destroying efforts applied, while SSD loses the information without any involvement. SSD requires that the information is refreshed, either as side effect of wear leveling mechanisms (data is moved around to make space for new erased blocks), or when the error detection is triggered during any access to the information, and then controller would rewrite the corrected data. This requires the SSD not only being powered, but actively used to keep the data alive. The data needs to be touched in order to be kept alive.

All that would require is for the on board controller to read each cell periodically, not something that would require much power to do once every 10 years. Modern SSDs already have the capability to do this even when not connected to a PC so long as they have some power source.

Trashbytes wrote on 2025-05-16, 10:09:

All that would require is for the on board controller to read each cell periodically, not something that would require much power to do once every 10 years.

The question was not about achieving the long storage with minimum power required, but on how HDD is fundamentally different from SSD so they don't need any actions at all on supporting the data from degradation. It's irrelevant how modern the SSD is, the charge holding cells are inherently leaky.

Again, bitrot is a thing for spinny disks.

That is EXACTLY what RAID scrubbing is for, for enterprise storage.

(RAID is an acronym for Redundant Array of Inexpensive (Sometimes stated as Independent) Disks. There are multiple "levels" of RAID, in which simple volume mirroring is used, or more complicated forms such as data striping with parity is used, and combinations of the two, with variations on how many parity disks in the group, etc... I could write an entire article on JUST what each of the RAID options actually do, on the nitty gritty level, but that's not necessary here. Instead, the purpose of RAID is to catch when data becomes unreliable on a disk, and automatically correct it with a redundant backup. This could be something quite large, like a whole disk failure, but much more often it is a momentary blip caused by bitrot, that then gets corrected by a process called a 'Scrub'. A RAID scrub compares the data from the rest of the array, and determines if one of more disks contain incorrect information, then uses parity data (if available) to mathematically reconstruct what the data SHOULD be, then re-writes that data to the affected disk to correct it. The RAID controller (or driver) does this when it detects such a condition, after notifying the administrator, and getting approval to start.)

Is this the same as a patrol read?

My RAID controller does this once a week, which is annoying because it makes the array unusable for half the day.

Things are much more complicated than they seem...

First of all - powering SSD does nothing. It is a very persistent myth, but it is a myth.

In a simplified way each SSD cell holds charge, as a tiny capacitor. This charge can and will leak over time, with many factors affecting actual rate - storage temperature, temperature when written, previous wear, quality of specific chip, etc, etc. MLC/TLC/QLC only complicates things - in those specific voltage levels have to be measured, not just "0" and "1".

The only way to "recharge" a cell is to erase and reprogram it. And since flash can only be erased in large blocks whole block has to be read, stored somewhere temporarily, erased and rewritten.

So even when you use your PC, or your phone for that matter, stored data is slowly decaying. And if it is not rewritten by user it will not be "refreshed" since there is no evidence of such mechanism existing in modern SSD firmwares. The only way that could happen is if it was rewritten for some other reason - wear leveling, garbage collection, read disturb prevention algorithms, user actions, etc, etc.

Practically the only way to be sure it is actually happening is to copy the data off the SSD and then write it back.

Modern SSD controllers include a few pretty clover algorithms to combat the leakage itself though, including, for example, applying offsets to measured voltage levels when read fails.

Typical consumer devices are rated for minimum of 1 year data retention when fully worn (all rated p/e cycles used up) at specific temperature (IIRC typically 25C). Realistically it tends to be a lot better. Like old SSDs, thumb drives, memory cards, BIOS chips, etc still hold data even decades later.

But if bad conditions are created, like worn (possibly beyond rated number of cycles) or low quality flash, primitive controller (thumb drives, sd cards), high storage temperature (left in a hot car for days) etc, data can be lost very fast.

Then... "HDDs can be stored unpowered for decades and still retain data 100%" is a myth too.

HDD platters get slowly demagnetized too and it depends on various things like temperature too. Some HDDs survive decades, just as some SSDs do. Some do not. I've seen plenty of cases when data can not be read from old HDD, but after rewriting it works perfectly - so there was no actual damage to the surface, just lost data. This often goes unnoticed because it is just assumed that "hdd died" (appropriate issues appear in smart and everything) without trying to erase it.

And that's not even taking possible mechanical issues like degrading lubricants, degrading plastics, seals, etc which can kill old drive instantly before reading data can be even attempted.

Archer57 wrote on 2025-05-16, 11:21:

And if it is not rewritten by user it will not be "refreshed" since there is no evidence of such mechanism existing in modern SSD firmwares.

There isn't? That's odd, I've seen semi-random latency spikes in SD cards that are cyclic in nature vs time, not amout of data written. In fact these cards are written once and then only read pretty much. And that latency issue can only be explained by the Flash controller on the card doing some housekeeping, like LRU lists for sector/block rewrites for example. Nothing else would bog it down so much.

Somebody mentioned RAID scrubbing - this is actually not that great idea on modern very high capacity HDDs. There is such a thing as error rate for the data channel, which is never 100% perfectly covered by CRC/ECC and such. Coupled with random events like power noise and cosmic radiation any scrubbing operations on large datasets have a non-zero chance of introducing bit glitches that are not caught at the time it happens. This is not limited to HDDs or any other particular storage system, just mentioning that to point out that long-term data storage is much more complicated then most people realize.

Deunan wrote on 2025-05-16, 12:46:

There isn't? That's odd, I've seen semi-random latency spikes in SD cards that are cyclic in nature vs time, not amout of data written. In fact these cards are written once and then only read pretty much. And that latency issue can only be explained by the Flash controller on the card doing some housekeeping, like LRU lists for sector/block rewrites for example. Nothing else would bog it down so much.

Some housekeeping yes. May be related to read disturb, may be something else. SD card controllers have so little resources that any small task would cause latency issues.

In case of some SSD controllers, however, pretty interesting telemetry can be extracted. Some also show pretty interesting stuff in smart. So far i've seen no indication of significant amount of writes (p/e cycles) initiated by controller itself.

If controller did have such algorithms simply being on would cause wear over time, which manufacturers want to avoid.

Worse - only some controllers would take action if a block requires multiple retries to read, some will just continue doing it resulting in noticeable read speed degradation for old data over time.

Hmm, interesting topic. I use CF cards for all my vintage systems. So it sounds like I should take full backups of each card and periodically write them back to the card to avoid degradation of the data, and that simply using my vintage systems periodically isn't enough to retain the full data on the card. Is there a rough estimate for how often I should do the backup and re-write process? Like can I expect a CF to hold the data well enough for a year or two, or I must do backups every 3-6 months?

Does the same principle apply to disk-on-modules and USB thumb drives? I've personally never had a USB thumb drive go bad on me. I've used some of them for over 15 years with no issue. For DOMs, I use a DOM in one of my vintage systems, it is more cumbersome to remove from the system to take a full backup like the CF cards, but if that's needed, I guess I'll put in the effort. You'd think DOMs would have pretty good stability though since they are often marketed for use as HDD replacements for industrial use.

Archer57 wrote on 2025-05-16, 13:11:

Some housekeeping yes. May be related to read disturb, may be something else. SD card controllers have so little resources that […]
Show full quote

Deunan wrote on 2025-05-16, 12:46:

There isn't? That's odd, I've seen semi-random latency spikes in SD cards that are cyclic in nature vs time, not amout of data written. In fact these cards are written once and then only read pretty much. And that latency issue can only be explained by the Flash controller on the card doing some housekeeping, like LRU lists for sector/block rewrites for example. Nothing else would bog it down so much.

Some housekeeping yes. May be related to read disturb, may be something else. SD card controllers have so little resources that any small task would cause latency issues.

In case of some SSD controllers, however, pretty interesting telemetry can be extracted. Some also show pretty interesting stuff in smart. So far i've seen no indication of significant amount of writes (p/e cycles) initiated by controller itself.

If controller did have such algorithms simply being on would cause wear over time, which manufacturers want to avoid.

Worse - only some controllers would take action if a block requires multiple retries to read, some will just continue doing it resulting in noticeable read speed degradation for old data over time.

I'm not talking about cheap SD/USB/SATA SSD/NVME controllers we all know they hardly do any maintenance or managing of the stored data and have data retention lifetimes that make floppy discs look good ....my comment was in regards to high end NVME/SSD Sata controllers in modern drives, they are pretty much full computers in their own right these days and do far more than is exposed to the host PC/OS.

Some high end Enterprise drives even have on board batteries to help with data loss prevention and data retention with the on board controller able to manage the drive even with the host PC offline up to an including reading and writing data to the drive on power loss. Initially they were using capacitors to provide enough power to the drive/controller to prevent data loss by allowing the controller finish cache data writes.

If it did have algorithms to do this then it would also have the ability to decide when they need to be used and more than likely also have a schedule for long term maintenance for the drive, it would make little sense to provide one and not the other.

Modern Nand controllers are not dumb controllers ...they have their own internal OS and operate similarly to the ME/PSP on the CPU this is also why the new NVME drives are so fast, the host PC is doing exactly none of the drive management or data handling . .the controller is doing all that, its really the only way you can get 14GBs in both directions.

I don't believe for one second the data retention time manufacturers quote for modern SSD are inaccurate, I would go so far as to say they are rather conservative just to be safe and in controlled storage they should last just as long as spinning rust drives.

Again this is for full featured drives and not SD/CF cards or cheap USB drives these drives have cost/space factors full drives dont.

Trashbytes wrote on 2025-05-16, 13:36:

I'm not talking about cheap SD/USB/SATA SSD/NVME controllers we all know they hardly do any maintenance or managing of the store […]
Show full quote

Archer57 wrote on 2025-05-16, 13:11:

Some housekeeping yes. May be related to read disturb, may be something else. SD card controllers have so little resources that […]
Show full quote

Deunan wrote on 2025-05-16, 12:46:

There isn't? That's odd, I've seen semi-random latency spikes in SD cards that are cyclic in nature vs time, not amout of data written. In fact these cards are written once and then only read pretty much. And that latency issue can only be explained by the Flash controller on the card doing some housekeeping, like LRU lists for sector/block rewrites for example. Nothing else would bog it down so much.

Some housekeeping yes. May be related to read disturb, may be something else. SD card controllers have so little resources that any small task would cause latency issues.

In case of some SSD controllers, however, pretty interesting telemetry can be extracted. Some also show pretty interesting stuff in smart. So far i've seen no indication of significant amount of writes (p/e cycles) initiated by controller itself.

If controller did have such algorithms simply being on would cause wear over time, which manufacturers want to avoid.

Worse - only some controllers would take action if a block requires multiple retries to read, some will just continue doing it resulting in noticeable read speed degradation for old data over time.

I'm not talking about cheap SD/USB/SATA SSD/NVME controllers we all know they hardly do any maintenance or managing of the stored data and have data retention lifetimes that make floppy discs look good ....my comment was in regards to high end NVME/SSD Sata controllers in modern drives, they are pretty much full computers in their own right these days and do far more than is exposed to the host PC/OS.

Some high end Enterprise drives even have on board batteries to help with data loss prevention and data retention with the on board controller able to manage the drive even with the host PC offline up to an including reading and writing data to the drive on power loss. Initially they were using capacitors to provide enough power to the drive/controller to prevent data loss by allowing the controller finish cache data writes.

If it did have algorithms to do this then it would also have the ability to decide when they need to be used and more than likely also have a schedule for long term maintenance for the drive, it would make little sense to provide one and not the other.

Modern Nand controllers are not dumb controllers ...they have their own internal OS and operate similarly to the ME/PSP on the CPU this is also why the new NVME drives are so fast, the host PC is doing exactly none of the drive management or data handling . .the controller is doing all that, its really the only way you can get 14GBs in both directions.

I don't believe for one second the data retention time manufacturers quote for modern SSD are inaccurate, I would go so far as to say they are rather conservative just to be safe and in controlled storage they should last just as long as spinning rust drives.

Again this is for full featured drives and not SD/CF cards or cheap USB drives these drives have cost/space factors full drives dont.

Yeah, i understand that SSD controllers are more powerful than many computers discussed here. And that they could include such algorithms and may be some do. All i am saying is that so far i have not seen any evidence of this happening.

And it is much, much more complicated than people assume anyway even if it was the case. A "refresh" of whole drive would mean whole drive has to be read, erased and written while shuffling data around in whatever limited ram the controller has (or even writing into flash temporarily since not all controllers have enough ram to handle whole erase block). With a risk of losing data if power is unexpectedly lost in case of consumer SSDs. It'd take significant time to happen and when people assume that turning a drive on for a few minutes is sufficient it is wrong.

Manufacturers have no incentive to implement this - it'll cause extra wear, complexity and industry standard data retention time is 1y/30C for consumer devices and 3M/40C for enterprise anyway...

As long as there is no specific documentation on this the safest assumption - it does not happen. Even if it does in some drives there is no guaranteed way to know it so it would be... unwise to rely on it.

dukeofurl wrote on 2025-05-16, 13:35:

Hmm, interesting topic. I use CF cards for all my vintage systems. So it sounds like I should take full backups of each card and periodically write them back to the card to avoid degradation of the data, and that simply using my vintage systems periodically isn't enough to retain the full data on the card. Is there a rough estimate for how often I should do the backup and re-write process? Like can I expect a CF to hold the data well enough for a year or two, or I must do backups every 3-6 months?

Does the same principle apply to disk-on-modules and USB thumb drives? I've personally never had a USB thumb drive go bad on me. I've used some of them for over 15 years with no issue. For DOMs, I use a DOM in one of my vintage systems, it is more cumbersome to remove from the system to take a full backup like the CF cards, but if that's needed, I guess I'll put in the effort. You'd think DOMs would have pretty good stability though since they are often marketed for use as HDD replacements for industrial use.

Yes, it applies to anything based on flash. But honestly my opinion is - just have a backup of stuff you'd want to keep. As always.

Yes, data retention time is limited, but it is basically impossible to predict because there are so many variables. It depends on temperature during write in relation to temperature during storage and quite dramatically so. For example.

As you've noticed practically it lasts for quite a while so no reason to go crazy about it...

One thing that is quite bad about cards though, CF of SD, is that they have extremely poor error detection. Data can easily be corrupted without any errors reported. That's where all those damaged photos and whatnot come from, which are so common...

Industrial stuff or proper SSDs are much, much better. And that's why i personally prefer to use SSDs, not CFs, for old systems when practical.

Archer57 wrote on 2025-05-16, 14:23:

One thing that is quite bad about cards though, CF of SD, is that they have extremely poor error detection. Data can easily be corrupted without any errors reported. That's where all those damaged photos and whatnot come from, which are so common...

Industrial stuff or proper SSDs are much, much better. And that's why i personally prefer to use SSDs, not CFs, for old systems when practical.

The write-caching setting in windows is responsible for a lot of SD / USB thumb stick data corruption. Removing the drive while it's still being written to.

On SSD data retention time. Manufacturers matter. I once bought a cheap no-name brand SSD, installed an OS then left it for a year. The drive was totally empty but still functional like it just forgot everything... So I only use big name 'pro' SSDs now.

user33331 wrote on 2025-05-16, 09:31:

Why precisely HDDs can be stored unpowered for decades and still retain data 100% compared to SSDs (and CF cards) which are said to lose data and become empty when left unpowered for decades.

At low level, HDDs store data the same way as good old magnetic tape drives - the head(s) magnetize a pattern on a moving
magnetic material - This is read back by passing that magnetic media under a head which detect the magnetic field level.
This does not affect the magnetic domains on the media.

Small magnetic fields moving near the media can affect it (which is why you should never put magnets near a floppy disk),
and technically very low level magnetic fields can affect them at extremely low amount - this is negligible during several lifetimes
of the people using the media and the service life of the device containing the media, but over 1000s of years, it can degrade.
(HDDs are also magnetically shielded to minimize the effects of external magnetic fields on their media)

SSDs on the other hand store electrical charges (exact details vary with type of SSD) - this can degrade in two ways:
- Electric charges can leak away, SSDs are very well insulated to prevent this - but nothing's perfect.
- Reading an electrical charge requires you to "sample" it - this will consume a small bit of the charge.

Some devices have ability to "refill" the charge after reading ... but due to the technology used to store them, writing
slightly degrades the cell (This is referred to as "wear") ...

This also means that cells in sectors that are accessed more frequently than others "wear out" faster.

Most devices do "wear leveling" which moves sectors around so that even though some are logically accessed more than
others, physically they are accessed at similar rates. This also means that data you put on "once" and never referenced
again can still "wear out".

These were problem enough to be concerning back in the early days of SSD (I've had a few early ones fail), but improvement
in the technologies have made them MUCH more reliable now!