2
\$\begingroup\$

We have a production batch of PCBs with faulty regulators that produce 2.4 Volts instead of 1.25 Volts. The only load on the 1.25 Volt node is the VccInt input on a Xilinx Spartan 3E FPGA. The max limit is 1.32 Volts according to the datasheet. Oddly, nothing has blown up. We see random odd behavior from the FPGAs where they seem to partially lose their configuration but recover after a reboot. They run a little warm too.

Would an overvoltage on the 1.25 Volt input cause what appears to be partially corrupt configuration at random times?

Why do you suppose the overvoltage didn't smoke the FPGAs? Is there built-in overvoltage protection?

Edit: BTW, the 'bad' regulators turned out to be counterfeit parts. Everybody seems to think that an over-voltage would smoke the FPGA but we ran the voltage up to 4.8 Volts and it still didn't smoke - it would stop running occasionally but would spontaneously re-start after an apparent cool-down. The only reason we stopped at 4.8 Volts was because we melted a test lead.

\$\endgroup\$
  • 1
    \$\begingroup\$ i.stack.imgur.com/wFfAa.jpg \$\endgroup\$ – Olin Lathrop Sep 21 '11 at 22:53
  • \$\begingroup\$ So a crazy grey clown will appear? I gotta try this one out... \$\endgroup\$ – Oli Glaser Sep 21 '11 at 23:05
  • 2
    \$\begingroup\$ You might not have seen smoke, but look at the big picture: they got whacked with double their spec'd voltage, and now they're a little flaky. Don't count on the parts getting better, write the off. \$\endgroup\$ – JustJeff Sep 22 '11 at 2:48
5
\$\begingroup\$

IF the random reconfigurations and running warm happen with 2.4 Volts applied rather than 1.25V applied you MAY have been lucky.

If the above results occur when 1.25V is applied then they HAVE been "smoked". They are walking wounded and you have no certainty that they will not die or behave in any possible way whatsoever in future. If these are for anything other than non critical in house use by competent technical people they should not be used. For personal in house use by competent people you can assess the value of the time wasted against their replacement cost.

Note that they MAY appear to work perfectly in ,any cases but do bad things occasionally.

Why did they not die? They did by any reasonable definition. example only: pedestrian hit by a car will die about V^2/50 % of the time, V in kph. ie at 70 kph death is near certain. At 50 kph death occurs about 50% of the time. Above 70 kph SOME people will survive but its a fluke. At 40 kph almost 40% fewer people will die than when hit at 50 kph so even modest braking helps a lot.
Overvolatged FPGA's are obviously vastly different than this but will also have a death / voltage relationship.

Long long long ago (30+ years?) I (stupidly) transiently applied about 50 VDC to a complex multi IC assembly with several separate PCs in it. Circumstances meant I had to repair it. Replacement was not an option. Most ICs were socketed (thankfully). I found that most higher current capable driver type ICs had died and that about none of the glue logic ICs had died. An interesting lesson. I haven't managed to do anything quite so major since.


NB !!!

TESTING !!!

You already know the following, but all of us often need to know it better :-).

A major "problem" here was lack of testing.
Testing of a manufactured product needs to be carried out by somebody who has your interests paramount and not those of a supplier.

While you cannot test for everything, as this sort of problem could be caused by other types of faults (track short, via open, wrong resistor value, misinsertion, poor soldering, ...) and as the outcome is potentially fatal to the product, testing should detect such problems.

You need to trade off testing cost and complexity against product cost and run size etc, but chances are that good testing here would have saved you money.

Cost of faults can be far far more than component costs and remediation. Loss of customer business and reputation is often an issue and impact on end user profitability is likely.

\$\endgroup\$
  • \$\begingroup\$ Good point about testing. ICT clearly missed this problem - something that needs to be fixed. \$\endgroup\$ – JimFred Sep 22 '11 at 4:33
  • 1
    \$\begingroup\$ @JimFred - If this is a design that you manufacture internally then it's more than ICT - it's also production testing that is inadequate. This sort of result can occur for many reasons - all parts may pass ICT OK. If it's made externally then their production testing is inadequate. \$\endgroup\$ – Russell McMahon Sep 22 '11 at 5:09
  • \$\begingroup\$ Good point - better testing is needed. The challenge is that the units function quite well and only exhibit problems after 24 hour burn-in. No catastrophic failures and, unfortunately, no Bozo-shaped mushroom clouds. \$\endgroup\$ – JimFred Sep 22 '11 at 5:49
  • \$\begingroup\$ JimFred - I'm involved with having (somebody else's) products made in China. At 5 different factories so far. The receptiveness to testing properly varies considerably. It's always interesting to see the creative faults which manage to get through - more mechanicalish than electronic, but electronic faults have had their occasions. From what I've seen, in an ideal factory environment I'd like to test far more rigorously than seems to be the norm. \$\endgroup\$ – Russell McMahon Sep 22 '11 at 10:46
2
\$\begingroup\$

There wouldn't be built-in overvoltage protection in the FPGAs. Other than saying the parts are damaged, there isn't anything you can do to isolate or fix the problem. Replace them.

\$\endgroup\$
1
\$\begingroup\$

Parts which have been subjected to out-of-spec conditions will often kinda sorta work, but as others have noted they may not be reliable. One common effect of over-voltage conditions is that some portions of the insulators between the parts of transistors may partially fail, causing the transistors to become "leaky". This can substantially increase the amount of current used by localized areas of the chip when voltage is applied between parts of the circuit that were supposed to be insulated from each other. The effects of such localized extra current draw can be unpredictable. It's entirely possible for a part to seem to behave normally (aside from drawing excess current) most of the time, but fail in weird and bizarre fashion when certain combinations of signals switch simultaneously and momentarily overload parts of the chip's internal power-distribution network.

My favorite stories of accidental over-stress are:

  1. I accidentally applied about 9 volts to a PIC16C84; the chip actually worked at 9 volts, but would no longer work below about 6. The chip couldn't be programmed any more at 5 volts, attempting to program it erased it, and I didn't feel like adapting my programmer to power the chip with 9 volts, so the chip went in the trash. Too bad I tried to reprogram it, since it would have been interesting to know how long the chip would have run at 9 volts.
  2. I accidentally connected an ADC input of a 16C373 to AC120 via 1K 1/4-watt resistor. The 1K resistor was completely annihilated (leaving just the lead wires) but after the resistor was replaced the PIC still worked.
I doubt Microchip still makes PICs like they used to; some of those things were pretty tough.

\$\endgroup\$
  • \$\begingroup\$ I once ran a PIC24F on 5V instead of 3.3V for about a minute without noticing. When I did notice it was very hot, I thought it was fried and that was my only 24F left. But somehow it kept working long enough to get my prototype finished. :) \$\endgroup\$ – Thomas O Oct 7 '11 at 23:38
  • \$\begingroup\$ 1k resistors as fuses. Hmm. For my own overvoltage horror story, I discovered that opamps don't like being reverse biased when I burned my hand trying to unsocket one. \$\endgroup\$ – Nick Johnson Oct 9 '11 at 1:50
  • \$\begingroup\$ @Nick Johnson: That's not as interesting as a motor driver chip I saw which demonstrated that "thermal overload protection", does not imply "output short-circuit protection". When used with a supply that can source 100 amps without blinking and being asked to drive a dead short, the thing lit up like a road flare. \$\endgroup\$ – supercat Oct 9 '11 at 3:34
1
\$\begingroup\$

Linear regulators usually have internal current limiting. So it triggered when FPGA tried to eat too much power @2.5V, and this limited power draw & overheat.

2.4V might not immidiately kill FPGA, but it will very quickly degrade gate dielectric, so even without overheat it's lifetime would be minutes or hours.

Any kinds of glitches & data losses are possible during that.

\$\endgroup\$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.