Longevity of microcontroller devices?

[DaveC1964]

It seems many things these days uses micro controllers of some kind.  Everything from battery volt gauges, to flashlights, to blue tooth speakers.  From what I understand though they all use flash memory to store the program and flash memory is not a permanent solution and eventually the bits be subject to errors as they age and the electron charge dissipates on the floating gates.  Doesn't that mean that every device using a micro controller will become unusable with in a few years like a ticking time bomb, or is there some other technology in use?

[crossroad]

You can look in the datasheets, the vendors usually specify the lifetime/endurance of the flash. It's often a million or so cycles for over ten years. Additionally, TI makes MSP430s with FRAM which have memory that will last practically forever.

So, no, probably not. What will fail will be the other components like mechanical parts and capacitors before the micon gives, generally speaking

Also, there are techniques to mitigate data loss like error correction codes (quite an ingenious process of perfectly reconstructing corrupted data without storing three repetitive parities in most conditions) and memory conditioning (re writing the data periodically)

[ataradov]

Most MCUs guarantee 20 year retention over the whole temperature range. It would be way more under normal conditions.

If application writes the flash, then there are possible longevity issues, but hopefully application designers though of that.

I just looked up SAM D21. It guarantees > 100 years of retention on cells with < 100 write/erase cycles at 55 C ambient temperature.

[techman-001]

I have PIC mcu's in gear I installed 22 years ago. They have been running 8 hours a day, 7 days a week for that entire time and no failures other than mechanical (stepper motor bearings getting dust in them past the seals etc) and one was blown to bits by a direct lightning strike.

I think you can relax about flash memory losing its data under normal circumstances.

[woofy]

It is a legitimate concern for long life products. I have many thousands of boards running 24/7 and the earliest of those are over 25 years old.
Back then I used a PIC16C84 with a quoted data retention of >40 years. Later PICs seem to be going backwards with "min 20 years" retention. It all depends on the environment. For the SAME70 with 1k cycles on the clock its min 20 years at +85C, dropping to 10 years with 10k cycles. At 1k cycles and +105C its just 5.5 years.

[Gyro]

That might be something to do with the early PIC 16Cs being UVEPROM based (even if they were plastic packaged OTP), whereas the later parts use Flash memory.

[woofy]

PIC16C84's were Microchips first EEPROM based chips. They were the only C version chip to be electrically erased and reprogrammed. All later chips were F types such as the 16F83 and 16F84 which replaced the 16C84.
Prior UV and OTP C devices were different technology.

[Zero999]

Some high volume products still use mask programmed devices, which essentially have infinite program retention life.

[amyk]

There are quite a few threads in the repair forum where old nonvolatile memory has corrupted itself somehow... so I wouldn't say it can't happen, but it's relatively rare.

[srb1954]

Some high volume products still use mask programmed devices, which essentially have infinite program retention life.

Not if those mask programmed ROMs were made by Mostek.

These were used in some Tek and HP gear from the '70s and they suffered from data corruption over time. This was a common enough occurrence to be given a special description:  "ROM-rot"

[Ice-Tea]

If application writes the flash, then there are possible longevity issues, but hopefully application designers though of that.

Any SW programmer worth his salt should. Off course, the Tesla lot didn't seem to have gotten the memo 

[Bassman59]

If application writes the flash, then there are possible longevity issues, but hopefully application designers though of that.

Any SW programmer worth his salt should. Off course, the Tesla lot didn't seem to have gotten the memo 

They were too busy moving fast and breaking things.

[David Hess]

At least up until recently, NOR Flash could be treated as EPROM or EEPROM as far as retention and essentially ignored.  The need for NOR Flash also prevents microcontrollers from being built on processes with the smallest feature sizes.  I think they only recently achieved 28 nanometers but were previously limited to 45 nanometers?

What is a real danger is designs going back at least 10 years which rely on NAND Flash for storage of state or even code.  (1) I have run across a few of these now where the device just quit, but of course well after the warranty expired.  They might as well be designed with a self destruct.

(1) Tesla provides a current example of this.

[SiliconWizard]

As said above, refer to datasheets, most of which are pretty conservative regarding Flash data retention. Although Flash memory has certainly a limited retention, I've personally never witnessed data loss in devices up to 25 years old.

20 years minimum is common from manufacturers' specs. That's never going to be failures "within a few years", unless you constantly WRITE to it. So that being said, if you're using part of the internal Flash of some MCU as NV memory that you often write to (to replace EEPROM for instance), then of course you'll shorten the MTBF.

Now if you want long-term retention (several decades), do not constantly write to Flash... and use mask ROM if you want really long-term.

For devices using Flash memory as code memory that is read only for most of its lifetime, the probability of something else failing first is pretty high IME (such as power supplies, dried capacitors, solder cracks, overheating, etc.)

[Berni]

The flash used in MCUs is better low density stuff, not the high density TLC and QLC garbage that is used in modern SSDs. The logic optimised fab process on them is not capable of the high density that dedicated flash proceses can do. For this reson some MCUs are now moving to external non volatile storage chips, usually QSPI.

As people said they last plenty long as long flash is not being constantly written to. But i suppose you could make an app read and write back blocks of flash at random long intervals to freshen up cells and do ECC checks. One needs to handle power loss in that case tho

[nctnico]

Some high volume products still use mask programmed devices, which essentially have infinite program retention life.

Very likely these are standard EPROMS without the (expensive) lid.

[DaveC1964]

The flash used in MCUs is better low density stuff, not the high density TLC and QLC garbage that is used in modern SSDs. The logic optimised fab process on them is not capable of the high density that dedicated flash proceses can do. For this reson some MCUs are now moving to external non volatile storage chips, usually QSPI.

As people said they last plenty long as long flash is not being constantly written to. But i suppose you could make an app read and write back blocks of flash at random long intervals to freshen up cells and do ECC checks. One needs to handle power loss in that case tho

The concern is embedded devices that use MCU. Since I have no control over them it will just be what the creator made.  They are in things you don't even expect.  These devices will just quit one day.  I am not even referring to write cycles, just the tendency of flash to degrade over time just by sitting.  People love using MCU because they are cheap and versatile but I wonder how many will still be working in 5 or 10 years.

[SilverSolder]

[...] I wonder how many will still be working in 5 or 10 years.

As others have said, the answer is:  "The vast majority of them"

[crossroad]

The concern is embedded devices that use MCU. Since I have no control over them it will just be what the creator made.  They are in things you don't even expect.

I mean, your favorite wooden chair might just snap one day. I think you might be placing an irrational amount of focus on microcontrollers specifically. Quite often the capacitors and mechanical parts get worn out and fail first

[Zero999]

Some high volume products still use mask programmed devices, which essentially have infinite program retention life.

Very likely these are standard EPROMS without the (expensive) lid.

Mask ROM isn't as common as it used to be but devices still exist and are used in applications such as calculators and watches which are made in huge numbers. You'll find more of it in older equipment though.

[bd139]

Only major risk in a 20 year window on MCU failure is thermal cycling and poor software destroying by the NVRAM in the device by writing too heavily. That has between 10k and 1M writes depending on which device. So if you have something which writes the same cell every second it’s not going to last the day.

I tried this with a 16F84 years ago when they first came out and it lasted about two weeks which was way above the specification.

[tmadness]

What about GPIO?
I would imagine electron migration and other phenomenon would cause increased degradation if the pin was pushing high current. eg a gpio terminated directly to a BJT, LED etc
I remember seeing a few failures at the 10 year mark where MCU GPIO had failed on pins that were connected to BJT pulldowns.

[ataradov]

That is no different than any other chip. If you don't stress the device beyond maximum ratings, you really should not worry about it. If pins that are not loaded beyond their ratings fail, something else is going on.

[james_s]

Some high volume products still use mask programmed devices, which essentially have infinite program retention life.

Not if those mask programmed ROMs were made by Mostek.

These were used in some Tek and HP gear from the '70s and they suffered from data corruption over time. This was a common enough occurrence to be given a special description:  "ROM-rot"

I don't recall who made them but a couple years ago I fixed two different early 90s arcade games that had defective mask ROMs.

[Zero999]

Some high volume products still use mask programmed devices, which essentially have infinite program retention life.

Not if those mask programmed ROMs were made by Mostek.

These were used in some Tek and HP gear from the '70s and they suffered from data corruption over time. This was a common enough occurrence to be given a special description:  "ROM-rot"

I don't recall who made them but a couple years ago I fixed two different early 90s arcade games that had defective mask ROMs.

I didn't know about reliability issues with some manufacturers of mask ROMs.

Just to clarify: my comment regarding mask ROM, was about microcontrollers, with internal ROM, rather than discrete ROM ICs. I'm not aware of any issue with mask ROM microcontrollers, but it would be interesting to know if it's a common problem, with some brands/manufacturers.