How long can a computer last?

[audion_1908]

I am not talking about the computer as sold stock but if retro fitted for long term reliably, using current teck to replace the weak links, like the exploding RIFA X2/Y caps, but not gutting it and putting a new computer guts in.

I know that electrolytic and tantalum capacitors are also unreliable, but can be replaced with modern equivalent ceramic capacitors, and use suitable mains filtering and surge suppressor, to stop lightning killing it, and used in a cool dry low dust environment with no detect sun.

I know mechanical parts can be unreliable too, but hard drives can be replaced with SSDs, and CPU heatsinks with high speed fans replaced with larger heat pipe based heatsinks with low speed high quality fans instead of the other way round! or no fans.

High efficiency passive power supplys are available, one I have seen is tiny belly bigger than the connector, a Pico 160w ATX psu that take 12v dc in but at 160watts this would only be suitable for the smallest pc, or reworking the old power supplies, to run cooler, using lower resistance high capacity switching fets?

how long could a computer last? I know this would depend on the generation of computer, a BBC b computer treated as such, may go another 40+ years!
But what about a 486, Pentium 3 or more modern !3 or !3 based computer, are there any other common computer failure modes I have missed?
Not counting software incompatibility, and just being too obsolete for the job

How long can a PCB last? When will the CPU eventually wear out, what is the optimal temperature to run consumer grade computers in, and can the environment be too dry? What about pvc cable how long till it drys up or softens to jelly? Would PTFE or silicone cable last longer if the original pvc cable is suspect or failing?
I know there are computers that have been used for 40+ years and still going, the Pegasus 25 was used from 1959 to 2009 when a small electrical fire put out of action, that's 50 years! not bad for a valve computer, but apparently it was designed to last from the start, it has since been repaired, and could be used but won't.

[Craig Sawyers]

Failure of all electrical equipment follows a bathtub failure curve. There is a fast fall initially due to "infant mortality", and after that a long period in the bottom of the bath when failure rate is small, and then a slow rise in failures at end of life.

The stable period failure rate is governed by the Arrhenius equation, which is a chemical rate equation of the form exp(-Ea/kT) where Ea is an "activation energy", and kT is the usual Boltzmann constant and absolute temperature that you find in semiconductor theory.

Anyway, Ea is different for different components. But the general behaviour is that keeping things cool is critical. That is why electrolytic caps fail much quicker at higher temperatures.

See for example https://www.jedec.org/standards-docu...on-reliability

But keep things cool in a computer (or any electronics for that matter) is a good thing for longevity. Periodically open the case and blow dust and grot out (particularly processor and graphics card heatsinks), and clean fan filters is essential maintenance.

But as to long term reliability? Well it is a statistical thing so there is not a hard and fast answer. But keep it cool and clean is a good thing.

Craig

[Craig Sawyers]

Turfed out another half-way decent reference that underlines how tricky failure analysis really is, in practice:

https://energiforskmedia.blob.core.w...esentation.pdf

[Craig Sawyers]

And finally a 410 page white paper from 2006 just on semiconductor failure mechanisms from Renesas

https://www.digchip.com/datasheets/d...mber=R8A77850A

[audion_1908]

Thanks for the links but statistics not my strong point, I am looking for some practical knowledge to extrapolate off, I know we throw teck out way before it's worn out or only needing a small repair to get it going agen.

[audion_1908]

Originally Posted by Craig Sawyers

And finally a 410 page white paper from 2006 just on semiconductor failure mechanisms from Renesas

https://www.digchip.com/datasheets/d...mber=R8A77850A

Wow 410 pages of info absorb thanks this is great, I had somehow forgotten about latent ESD damage though building and maintenance, but moisture diffusion through the epoxy into the chip as a new one!
It looks like humidity has to be kept under 50%

[Craig Sawyers]

Well:

1. Failure mechanisms are statistical by nature
2. Fixing a modern computer (i.e. anything in the last 15-20 years) built using surface mount everything is non trivial.

Good luck!

Craig

[audion_1908]

But are the hard to fix parts the weak link? or for a more reliable computer is it better to go with much older teck using through hole or surface mount chips? Or go pre year 2000 teck if you want to use the same computer for 50+ years?

I have watched videos on YouTube on ball grid array chip replacement in laptops and it looks like almost a black art! and one place an x-ray could be useful! (Watch louis rossmann on YouTube) but it looks like most of the failure are due to water damage or damaged by something plugged in, so maybe for more reliability the bga chips need to be sealed better to the PCB maybe with a high temp wax so it can still be repaired, but better still not exposed to damp, and ESD protection adapters used on the sensitive ports? Or retrofitted, is bga teck unreliable or just very hard to fix?

To dry an environments could exacerbate ESD problem, so better maintained climate control.

Also some manufacturers deliberately make it hard to get replacement chips, but that's another issue

[SiriusHardware]

I didn't have time to read the document but one factor which uniquely affects computers and computer controlled devices is the longevity of the firmware code - in older devices for example the programmed firmware code may be held in EPROM where the initial programming is performed by forcing a charge (or no charge) into each memory cell.

Over time this charge gradually fades away and it is a matter of when, not if, the firmware will begin to fail as bits gradually return to their unprogrammed state. So one essential bit of maintenance which needs to be performed on equipment with firmware in EPROM is to 'refresh' the EPROM code by over-programming it every few decades. Every time you do that, you reset the 'data fade' clock back to zero. If you have equipment with EPROMs which were programmed forty or more years ago, they may be getting very close to the onset of bit-rot now. You'll put them away working, and then the next time you get them out they'll be dead or going haywire. The internet is a terrific resource here, though - wherever it is possible to read and archive firmware you will generally find that someone, somewhere, has already done it. With the rise of emulation, firmware for home computers of the 8 and 16 bit eras has been especially widely preserved.

Systems with their firmware programmed in MASK ROM or one-time programmable PROM should not be subject to these problems, although there are no doubt other interesting ways in which they can fail.

More modern computers / computer controlled devices will typically have their firmware code stored in 'flash' memory and this, too, may be subject to gradual degeneration of the firmware code over time although with flash still being a relatively recent technology, that will be for the next generation to worry about.

Whereas with discrete EPROMs it was a simple matter to read and back-up the firmware code while it was still working, most microcontrollers are designed to make their internal code difficult for third parties to read out so it would be much harder to refresh the code in those cases.

[Craig Sawyers]

I have experience in space hardware, in which there has to be 100% inspection of solder joints in flight hardware. For that reason ball grid array devices are not allowed.

Reliability in high radiation environments over may years is of course a given.

[GMB]

I know mechanical parts can be unreliable too, but hard drives can be replaced with SSDs,

SSDs big feature is that they are fast and generally survive being dropped while powered, unlike their mechanical cousins. But all Flash technology intrinsically has a finite life so I am not sure they last longer than hard drives under ideal conditions.

The biggest worry for any microprocessor-based thing is loss of firmware - because apart from the oldest examples, they use EPROM and Flash which have a finite life so are certain to die one day, although the best examples seem to have outlived their specified life expectancy. So make sure you have copies - but that is hard because there is a lot of sneaky firmware like the programming of gate arrays, and some peripherals have their own processors (e.g. modern hard drives).

Very high density chips will enevitably fail due to solid state diffusion - so that's another vote for keeping cool and using older technology. In fact, looking back at my pile of dead PCs I think that is the thing that did for most of them.

A number of very early computers still work because their main worry was tin-whiskers in germanium transistors - but easily replaced, so maintainable. And of course the very early computers came with absolutely everything documented. Wouldn't that be nice?

[jjl]

Originally Posted by SiriusHardware

I didn't have time to read the document but one factor which uniquely affects computers and computer controlled devices is the longevity of the firmware code - in older devices for example the programmed firmware code may be held in EPROM where the initial programming is performed by forcing a charge (or no charge) into each memory cell.

Over time this charge gradually fades away and it is a matter of when, not if, the firmware will begin to fail as bits gradually return to their unprogrammed state. So one essential bit of maintenance which needs to be performed on equipment with firmware in EPROM is to 'refresh' the EPROM code by over-programming it every few decades. Every time you do that, you reset the 'data fade' clock back to zero. If you have equipment with EPROMs which were programmed forty or more years ago, they may be getting very close to the onset of bit-rot now. You'll put them away working, and then the next time you get them out they'll be dead or going haywire. The internet is a terrific resource here, though - wherever it is possible to read and archive firmware you will generally find that someone, somewhere, has already done it. With the rise of emulation, firmware for home computers of the 8 and 16 bit eras has been especially widely preserved.

Systems with their firmware programmed in MASK ROM or one-time programmable PROM should not be subject to these problems, although there are no doubt other interesting ways in which they can fail.

More modern computers / computer controlled devices will typically have their firmware code stored in 'flash' memory and this, too, may be subject to gradual degeneration of the firmware code over time although with flash still being a relatively recent technology, that will be for the next generation to worry about.

Whereas with discrete EPROMs it was a simple matter to read and back-up the firmware code while it was still working, most microcontrollers are designed to make their internal code difficult for third parties to read out so it would be much harder to refresh the code in those cases.

EPROMs haven't been used much in domestic PCs since the early '90s nor in any system I've been involved with professionally after 1991, but you are correct about them "forgetting" the data programmed into them.
Flash memory is quite similar in construction to EPROM, but there are more recent innovations.
Single bit per cell Flash devices have specified retention times of decades to centuries. This Flash device has a minimum retention time of 20 years and this PIC microcontroller has a typical retention time of 40 years.
Multi bit per cell NAND Flash devices that are used in SSDs, SD cards, USB sticks etc. are likely to be less reliable.

John

[SiriusHardware]

As you may have gathered, this site deals mainly with technology from well before the nineties although computers are almost uniquely ancient before their time. There are now people actively collecting and preserving Pentium-era PCs. Computer hardware can be many decades younger than your average wooden cased wireless and still be considered vintage within its own category.

If PIC flash memory has a retention time of 'only' 40 years I consider that to be dangerously low, in line with typical EPROMs which are already starting to get to the age where they give problems.

[McMurdo]

Systems with their firmware programmed in MASK ROM...

the commodore 64 has mask roms and these tend to fade, though not sure why.
The solution here is to replace them with an EPROM, ironically.

[Vintage_RC]

I've had failures in PCB vias with multi layer boards in PC's. All PC's internal parts are subjected to regular thermal cycles in normal use and vias which are a very small thin form of inter layer connector are particularly vulnerable to thermal stress cycles. In my experience the weak point is actually the connection between the via and an internal track. The area of the connection is extremely small. The failure mode usually starts as an intermittent fault which can often be made to come and go by flexing the PCB.

Voyager (the planet exploring vehicle) is still going, I think, the data rate is very slow and it has been reprogrammed a few times over the years, 43 since launch!

[Craig Sawyers]

Originally Posted by merlinmaxwell

Voyager (the planet exploring vehicle) is still going, I think, the data rate is very slow and it has been reprogrammed a few times over the years, 43 since launch!

Yup - they are still alive, and still doing science, now at around 20 light hours distant.

https://voyager.jpl.nasa.gov/mission/status/

Like all spacecraft they use microwave links to transmit data. The Voyagers are of course powered by a nuclear reactor, but the typical transmitter power (from a travelling wave tube) is 60W. Via a high gain antenna, but nevertheless that is not much power at 13 billion miles.

They manage a data rate of 160 bits per second at that distance.

But the radioactive power generator is getting to the end of its life, and will likely conk out sometime this year or next.

They are limping it along by using a power saving strategy https://voyager.jpl.nasa.gov/mission...rty-year-plan/

Craig

[Scimitar]

I doubt that this will help a lot, but I have a RISC PC in use as a server for CID that runs 10 hours a day every day and has done for the last 20 odd years. I have replaced the power supply once in that time.

[audion_1908]

Originally Posted by Scimitar

I doubt that this will help a lot, but I have a RISC PC in use as a server for CID that runs 10 hours a day every day and has done for the last 20 odd years. I have replaced the power supply once in that time.

CID? Why did the PSU die?

[stevehertz]

Surely with computers the issue is not just hardware, but their inability to keep up with up to date apps and stuff? These all require a never ending increase in computing power that old computers just don't have. They become 'outdated', slow and unusable.