akimmet wrote on 2025-04-10, 03:59:

I have had hundreds of tantalums fail short over the years. It is a big enough problem that anything over 30 years old I automa […]
Show full quote

I have had hundreds of tantalums fail short over the years. It is a big enough problem that anything over 30 years old I automatically replace them. Examples from the 1970's and earlier are especially failure prone. The failure mode is well known and its causes.
https://www.kyocera-avx.com/docs/techinfo/Fie … tallization.pdf
https://www.edn.com/what-a-cap-astrophe/

Tantalums do degrade in storage, almost always from moisture ingress. I have had to throw out entire reels of them because of improper storage in a wet environment.

Moisture is a problem when you go to solder the part, not a problem for parts already on the board. The paper you linked to points out that there are only some potential wear out mechanisms, and the one they looked at turned out not to be significant, so tantalums are even more reliable than was thought. A failed tantalum would by default be a symptom rather than a cause so you'd look into the rest of the system or design to find what's very possibly the actual cause. It could be the tantalum but probably not.

Seventh subject: Matrox G100. Seven SMD wet alus, 10 uF 16 V. Matrox typically used tantalums for their graphics cards, but accidentally put wet caps on some of the G100. These particular ones showed signs of either crustied leakage or glue remnants, capacitance was normal but ESR in the 20-30 range. I did remove them by twisting which could've influenced the readings afterward.

Replaced with 10 uF 20 V SMD tantalums.

Card works. Very good 2D quality, but probably just as good as before.

vvbee wrote on 2025-04-10, 05:49:

akimmet wrote on 2025-04-10, 03:59:

I have had hundreds of tantalums fail short over the years. It is a big enough problem that anything over 30 years old I automa […]
Show full quote

I have had hundreds of tantalums fail short over the years. It is a big enough problem that anything over 30 years old I automatically replace them. Examples from the 1970's and earlier are especially failure prone. The failure mode is well known and its causes.
https://www.kyocera-avx.com/docs/techinfo/Fie … tallization.pdf
https://www.edn.com/what-a-cap-astrophe/

Tantalums do degrade in storage, almost always from moisture ingress. I have had to throw out entire reels of them because of improper storage in a wet environment.

Moisture is a problem when you go to solder the part, not a problem for parts already on the board. The paper you linked to points out that there are only some potential wear out mechanisms, and the one they looked at turned out not to be significant, so tantalums are even more reliable than was thought. A failed tantalum would by default be a symptom rather than a cause so you'd look into the rest of the system or design to find what's very possibly the actual cause. It could be the tantalum but probably not.

You are correct that in most failure circumstances a shorted tantalum can likely be traced back to poor design.
It doesn't change that most of my repairs involve finding the shorted tantalum or multi layer ceramic capacitor.
Tantalums made since the 2000's are approaching the reliability you claim, especially AEC-Q200 qualified parts.
I still prefer polymer tantalums, since they do not exhibit thermal runaway (short out and combust).

https://www.kemet.com/en/us/capacitors/polyme … lymer-faqs.html
https://www.kemet.com/content/dam/kemet/light … Differences.pdf

If you have data about your repairs then that would be interesting, like which type of component failed where in what board under what circumstances and whether the rest of the system was measured to operate normally so that you could estimate the probability that the component failed while operating within spec. Otherwise we have failure data from manufacturers' testing, where regular tantalums look to be the most reliable out of the typical capacitor choices. I wouldn't be surprised if properly derated tantalums from the 1980s were already safe, and I'm not aware of 1990s graphics cards with burnt-out tantalums even though they were used plenty there, just typically with 16 V rated parts on 3 and 5-volt rails. Combustion isn't a likely scenario anyway since the inherent failure rate is small and it's not the only failure mode, but you also have control over it by derating. Infant failures could have worse outcomes.

Eighth subject: Diamond Fire GL 1000 Pro (Permedia 2). About ten wet SMD alus, 10 uF 16 V. Measured two of them after twisting off the board, capacitance was ok and ESR about 12. Slightly crusty on the bottom. Don't know what voltages they see in use, probably 5 or less. There's also one tantalum, might've been 4.7 uF 16 V.

The wet caps were replaced with 10 uF 20 V tantalums.

Card works. 2D image quality was slightly ghosting before and unchanged now. Still good quality, just not absolutely perfect.

Ninth subject: Edimax EN-9130 (Realtek RTL8139) network adapter, from 2000. Three wet through-hole alus, 22 uF 16 V. Capacitance measured ok, ESR in the 5-12 range. One had 5 volts across it, the rest I didn't measure but I don't see them getting 12 volts.

Replaced with 22 uF 16 V SMD tantalums. Would've preferred through-holes but these will do.

Works, i.e. the driver installs and the lights come on.

Voltages measured after the fact: 5, 5, and 3.

Nice! Although I'm slightly disappointed by the lack of orange or blue colored TH tantalums (they taste better than yellows).

I didn't like tantalums at all, until only after a couple of years ago. After building DIY 8088 clones, I became a fan of their long storage life. I think I still have about 400 10uf in storage. I stored them well so I am looking forward to still using the batch until I reach retirement age.

Tenth subject: Datapath VisionRGB-PRO2 capture card, from the 2000s. Lots of tantalums but also four wet SMD cans, 2 22 uF 6 V and 2 47 uF 16 V. The bigger ones have 5 volts across them and 3 for the smaller ones, at least when idling. Capacitance was ok, ESR 3 for the bigger ones and 11 for the smaller. No signs of leakage.

Replaced with 22 uF 16 V and 47 uF 16 V SMD tantalums.

Maybe works and maybe not, don't have a supported system on hand to test. The PCI device and vendor ids look ok anyway.

I was able to find public copies of some earlier articles discussing field crystallization failures.

https://www.vishay.com/docs/49268/tn0003-dcleakfailmode.pdf
https://onlinelibrary.wiley.com/doi/abs/10.1155/APEC.3.171
https://onlinelibrary.wiley.com/doi/abs/10.1155/APEC.3.233

Here are several videos that have caught tantalum failures on camera.

https://www.youtube.com/watch?v=Qoy82YBg7Pw
https://www.youtube.com/watch?v=yCQ7ncuT468
https://www.youtube.com/watch?v=A5aoIgz7pXc
https://www.youtube.com/watch?v=X15hLGrxD0s
https://www.youtube.com/watch?v=FItTR0wBywE

Here are a few examples of repairs, where I had a clear diagnosis of why a tantalum failed.

1. Poor design. 486 and earlier AT motherboards tended to have several tantalums serving bypass duty right next to the power supply connector. Almost always directly across voltage rails without any over current protection. Even worse, 16V rated parts were used for the 12V rail.
Many early AT power supplys had poor phase margins. This meant voltage overshoot or ringing would occur at power on, or abrupt load changes. These voltage transients cause small amounts of damage to the dielectric layer in all of the bypass capacitors. This damage is cumulative for dry tantalums, and eventually leads to thermal runaway. Since common 250W AT power supplies could supply 20A on the 5V rail and 9A on the 12V rail, a shorted capacitor will often not trip over current protection.

2. Physical damage to the outer epoxy or plastic encapsulation. Once moisture can get inside, it will damage the dielectric. This leads to an increase in leakage current, which in turn leads to thermal runaway.

3. Incorrect soldering techniques can damage tantalums. Excessive heat will damage the outer encapsulation, again allowing moisture ingress.
While it sounds counter-intuitive, overheated parts and lifted pads are usually caused by an under powered or non temperature controlled soldering iron.
An under powered iron will dump full heating current into the tip to get a solder joint to melt. Once the solder liquefies the power required to maintain the molten solder joint decreases. With an under powered iron, there will be more temperature overshoot compared to a temperature controlled iron with more power.

In conclusion.
Modern manufacturing techniques have minimized impurities inside the capacitor that can lead to field crystallization. Avoid old or salvaged stock.
Modern power supplies are better designed, and are less likely to oscillate or overshoot. Avoid vintage power supplies if possible.
New assembly and preheating techniques minimize exposure to high temperatures. If your budget allows, get a temperature controlled soldering station if you do not have one already. A hot air rework station is also nice to have.

Since none of this is likely to change your opinion, I will now leave you be. My apologies for hijacking your thread.

akimmet wrote on 2025-04-14, 17:22:

Since none of this is likely to change your opinion, I will now leave you be. My apologies for hijacking your thread.

Don't hold it against me if your argument doesn't convince me. Every capacitor can fail, the question is which type is least likely to fail, and you'd agree the answer is tantalum. Especially with long storage times.

As I said in the first post, some people are afraid that tantalums will randomly explode, but this isn't supported by data. Don't put 20% tolerance 16 V tantalums on 12-volt rails and you likely eliminate a glut of historic failures.

No offense taken.

I can attest that tantalum capacitors do indeed explode...
I had my Amiga 600 recapped after a random failure. Then one day I was playing me some Sensible World of Soccer and all of a sudden KA-FOOKIN'-BOOM, sparks, smoke. As it turned out, one of the replaced caps located in the area of the motherboard directly behind the power socket, a tantalum cap, died with quite a bang, almost taking me with it, as I almost had a heart attack :]

One of the good points about them exploding is at least its easy to diagnose what you need to replace. I might be wrong but on the PC/XT/AT boards that had detonations, the hardware didn't become fully broken, they just needed a capacitor replacement.

Although that might be because they were through-hole tantalums, so the explosion is at least a small space away from the PCB as opposed to SMD tantalums.

Statistically it's more likely that a wet capacitor or the like had failed and destabilized the system, and this is probably what happens when people report exploding tantalums on first power-on in old machines. In any case you'd want to examine the rest of the system to find out what actually failed, then replace the tantalum with a higher rated part.

11th subject: S3 Trio 3D/2X. Three wet through-holes, 47 uF 16 V. Capacitance about 53, ESR 1.4. None had more than five volts across.

Replaced with 47 uF 16 V through-hole tantalums.

Works.

Polymer electrolytics do explode. In fact the lack of any deliberate structural weak spots (the grooves on top of typical electrolytics) causes them to build up gas and then launch the outer shell with potentialy eye-taking velocity, while the guts are rapidly spread all over the PCB. I know from my own experience, I've had that happen 3 times due to a faulty PSU that would randomly overshoot 12V rail up to 20V or so, for a fraction of a second, before returning to correct regulation. Took me days to figure it out.

As for tantalums, just because you have two in your hand and both say 10uF/16V doesn't mean they are identical. Crappy no-names are not made to the same standard as good quality ones. These days proper tantalums come fully tested, from factory, for dV/dt spikes and it's in the datasheets. The old ones... it's a bit of a lottery. So while I have nothing against tantalum caps, reusing them is perhaps not a great idea, at least not anywhere close to low impendance power rails. PC cards would be a grey area, perhpas OK, but mobos (esp. the more modern VRM sections) would not be. There you'd need to use new, good caps and maybe even derate as well to be on the safe side.

And then there are wet tantalum caps, but these are not really meant to be bulk filters for high power rails so it's a different story. Just mentioning them because unlike dry tantalums these pretty much don't fail. Don't really age either. But it's gotten to the point where, because of some crappy parts made in '90, the "internet wisdom" is to recap anything tantalum to wet electrolytics. Which will degrade and die eventually.

Every capacitor type is a compromise. Surge voltage, parasitic inductance, leakage current, size, and even price. Wet tantalums are the most reliable electrolytic type, but on average cost 100 times more than the alternatives. I have witnessed polymer electrolytics violently fail as well as almost every other type including film capacitors.

I'm personally not too worried about reusing tantalums in non-critical areas, and not that bothered even in relatively critical retro use if new parts aren't available. The extra heating cycles won't help them, but they've survived infant mortality and to the extent the failure rate decreases with age you wouldn't be worried about age as such. If tantalum failure were a known issue in the Voodoo community I'd be more concerned.

12th subject: Matrox G450, LP version. About a dozen surface-mount wet alus, 10 uF 16 V. Measured one of them, capacitance ok and ESR 11. No obvious signs of leakage, so a good time to get rid of them. A few tantalums on there as well, 10 V or less. I know on the G550 no capacitor has more than five volts across it, so let's assume that's the case here.

The wet alus were replaced with 10 uF 20 V surface-mount tantalums.

Works, although I forgot to replace one of the removed wet alus.

13th subject: Nvidia Vanta-16. Two wet surface-mount alus: 470 uF 6 V and 330 uF 6 V. Measured ok, ESR in the 0.2 range. Didn't bother measuring voltage across them.

Replaced with 10 V surface-mount tantalums of the same capacitance. The pads are on the thin side for these. There's also pads for through-hole capacitors.

Works. Needs a heatsink.