No, Hackaday hasn’t become a baking blog. We’re just here to give you a bit of advice: if [MickMake] ever offers you one of his fresh-baked Pis, proceed with caution. While we have no doubt that there will be some interesting smells wafting out of his kitchen, these aren’t the tasty pies you’re looking for. There’s no delicious home-baked treat when that timer dings, just a handful of Raspberry Pis that have had an exceptionally hard day.
To properly explain the odd sight of some Raspberry Pis laid out on a cookie sheet, we need to take a step back. [MickMake] originally set out to see how everyone’s favorite Linux SBC would handle the harsh Australian heat, and thought that setting them up on his car’s dashboard would be a suitable torture test. But as luck would have it, a storm rolled in while he was making the video which brought temperatures down to a “cool” 30 C (86 F); basically jacket weather at the bottom of the world. So naturally, he decided to put them in his oven instead.
Placed on an insulating sheet and with a thermocouple between them to get an accurate idea of the temperature they were experiencing, an original Pi, a Pi 2, and a pair of Pi 3s were sent on the ride of their lives. In addition to monitoring them over the network, he also added a “heartbeat” LED to each Pi so he’d be able to tell at a glance if any of them had given up the ghost. As if these poor little Pis didn’t have it bad enough already, [MickMake] decided to take things a step farther and run sysbench on them while they took their trip through Hades.
The Pis are actually rated for temperatures up to 85C, and all the participants of the experiment hit that point without any issues. At 87.3 C (~190 F) the original Pi dropped off the network, but its LED was seen bravely blinking on. At 105.7 C (~222 F) it finally breathed its last, followed by the pair of Pi 3s tapping out at 112 C (233 F). The Pi 2 fought on, but it fell right at the 119 C (246 F) mark.
But what about when they cooled off? Somewhat surprisingly, [MickMake] successfully powered all four back up and was unable to find any damage to the Pis, either physically or operationally. Even the SD cards survived, and the Pis popped right back onto the network and were ready for another round of Silicon Chef. Not bad considering they were subjected to temperatures three times higher than the official limit.
Testing electronics in your home oven might seem a bit suspect, and admittedly we’d probably turn down a slice of the next few frozen pizza’s [MickMake] runs through it, but it’s not really so far removed from how proper reliability testing is performed.
[Thanks to BaldPower for the tip.]
“subjected to temperatures three times higher than the official limit”
Huh? 85C is 358K, and three times that is 1074K.
Perhaps I did not read the article correctly.
I was going to ask the same thing. I sense maybe the listed operating temperature is actually something like ’30C Ambient’ whereas the chips are rated upto 85C internal. 90C is indeed 3x 30C….?
No, the point is, you can’t have a Celsius temperature that’s “three times” the limit, because the Celsius scale has an offset origin. This is why Peter A Nielson converted to Kelvin.
I believe the error is down to mixing up C and F. The rated max is 85 deg C and the temp reached 246 F
even so, 240°C is not three times 80°C. You need to work in Kelvin (absolute temperature) if you want to be able to say that sort of thing
85°C * 3 ~= 246°F
*for large °C and/ or small °F
Maybe it means that
the official limit is from temperature A to B
but it survives temperature A + B-A + B-A + B-A
In what world is 119C “three times” 85C? Is your idea of baseline temperature 74C?
Hang on a minute…
“but it fell right at the 119 C (246 F) mark”
“The Pis are actually rated for temperatures up to 85C”
246F / 85C = 2.9 temperatures
Hackaday math!
What is a temperature “3 times higher than the official limit”?
That would be (273+85) * 3 = 1074 Kelvin, and I would be surprized if the pi survived that.
I wonder if the failure is caused by the Pi’s oscillator(s) drifting at the higher temperature, to the point network timing was shafted and/or DRAM refresh was flaky? Be interesting to repeat with an external oscillator rated for the full temperature range to test the hypothesis.
Here’s the thing about ovens, though: They don’t have proportional control over the heating elements, it’s strictly bang-bang. (At least old ones. I’ve never used an oven made this century.) And when you consider how slow and massive the temperature sensor is, that means there’s SUBSTANTIAL overshoot in the air temperature, especially at first warm-up.
This is why recipes always say to “pre-heat” the oven, by the way! The setpoint is an average, not a limit.
I neglected this step, while baking out a tray of desiccant packs. I set the oven on “warm”, which means it should idle around 200F, but of course by the time the temperature sensor reached 200F, the air temp had already gone high enough to crisp and burn the edges of some of the paper packs.
burnt edges are more likely caused by all that IR radiation from the heating elements rather then ambient temperature inside the oven…
My attempt to bake the water out of a laptop shell at 200F did not go well at all.
This is why the writeup said he put a thermocouple in between the pi’s. In order to get a measurement of the temperature they were ACTUALLY experiencing.
The modern ovens I have worked on are bang bang too.
Hi Nate,
I did actually take that into account. I would gradually ramp up the temperature by adjusting the temp knob up until I heard the oven kick in, and then I backed it off to avoid too much overshoot.
That way I managed to keep the temperature constantly rising.
So more “gnome gnome” than “nom nom” then? :)
Heat born death in semiconductors is due to losing control of the transistors inside. Semiconductors have a native set of charge carriers (i.e. electrons or holes) based on the temperature. These are then supplemented with doping to give carrier concentrations MUCH higher than the native and of only one type (either electrons or holes). Transistor operation is based on the junctions or interfaces between regions that are dominated by electrons and areas that are dominated by holes. As the device heats up, the native carriers increase and once they become comparable to the level of doping, we lose control of the transistors because there are no longer areas with only a single carrier type. At this point it’s like trying to turn off the flow of current in a metal (the conductivity of the semiconductors also skyrockets).
This process is absolutely reversible (so the device would still function) once cooled unless one of a few things happen. First, the packaging can fail at certain temperatures. Second, the device can get so hot that the dopants redistribute and physically change the device. However, typically these cases occur at VERY high temps, even above 300C-500C. Instead, the most common permadeath cased by overheating is usually caused by losing control of the transistors, which then create a short circuit which continues to locally heat the device, far beyond the ambient temperature. This then causes one of the above things (or a few other cases of heat based death such as metal diffusion or even cracking of the semiconductor) to occur. Even if only a couple transistors fail, the entire chip will fail.
Would it be possible to install transistor-level cooling rods to distribute the heat when the transistors short? Or possibly design the circuit so that if it were to short it wouldn’t heat up?
I guess that depends on how efficiently the rods conduct the heat away, but if they can’t move it far or fast enough it’ll probably be too slow and just heat up their neighbours.
One group has experimented with cutting water-cooling channels directly into the silicon, which might be a better approach if it was ever made practical: https://www.pcgamer.com/uk/researchers-have-moved-liquid-cooling-directly-onto-the-chip/
This is actually how it is setup, although probably not intentionally. There is a complex set of metal interconnects which connect all transistors as necessary and these are responsible for pulling most of the heat out. In terms of any sort of active cooling, I don’t think there is space and/or the added complexity would reduce the yeild too far.
Maybe it were the chargers/ power banks just failing
I’m not surprised at the things coming “back to life”. The Broadcom SoC (like most ICs of significant complexity) has an on-chip thermal sensor, and I would expect it to cut the power to the chip at some point above the maximum rated temperature. (The actual on-chip temperature was possibly well above what the presenter measured with the external thermocouple, depending on CPU load.) Note that the original Pi lost Ethernet before it died, indicating that the ethernet controller (the original Pi is the only one under test which has a separate Ethernet controller, rather than one in the SoC) had a slightly lower effective cutoff temperature than the SoC did.
That’s what I thought too – the chips shut down on their own for their own protection, so everything works as designed!
Flashbacks to the infamous “removing a heatsink from a Pentium 4 vs AMD Athlon during Quake 3” demo video…
1. Per individual component and materials ratings found within these R-pi models and per margin requirements, the supposed 85C operational rating has no efficacy.
2. HALT is not necessarily an indicator of MTTF per most reliability standards. Although SR322 does allow some considerations for a product that has been subject to a HALT series. Reliability calculations are based on the margin of each component for its electrical and environmental ratings.
3. HALT is not intended as a test for environmental ‘screening’.
4. Thermal control of tests to determine or verify temperature rating has no meaning for electrical equipment without well defined and controlled loads.
5. Thermal testing technique indicated in the video will have poor repeatability. Mounting and exposure surfaces were inconsistent between test units. Any air motion over any components will disallow acceptance of test data for use as a rating. The IECEE and UL web sites have published several methods for these tests.
1) You need to look up the word “efficacy”.
2) He was making no attempt to measure MTTF, just the point of failure.
3) He made no mention of HALT.
4) These are computers, not “electrical equipment”, and “well defined and controlled loads” doesn’t really have any meaning.
5) It looks like that is a convection oven he’s using, and the sound of the fan while the test is being done indicates that it is being used in convection mode, so the temperatures should be consistent enough for this non-critical test.
6) Lighten up, Francis.
what kind of masochistic sob is going to wear a jacket at 86f? i dont wear one unless its under 60.
One who has adapted to Australia. I was in Melbourne a couple years ago in July, and while I was walking around in a light jacket, which means it couldn’t have been much under 65degF, all of the locals were bundled up like it was the dead of Winter. Which I guess it was.
having lived in phoenix for some years i know what you are talking about. seems if its not tripple digits people are cold. and the only time ive been cold in az is when people had the ac up too high. having moved back to alaska (i cant seem to get out of this place) we kind of laugh at tourists who wear windbreakers during the best part of the year.
It’s a relative thing. What’s “cold” here in the fall, is suitable for shorts in the spring, an expectation of better.
But still, I don’t think Down Under is anywhere near “the bottom of the world”.
Michael
yep, and we all think the rest of the world is crazy when they complain about a nice day – say 30C (86F) which is when I start to think about putting shorts on and taking my sweater off.. We don’t consider it warm until about 35C (95F) and hot is 40C (104) or over. It was 43C (109F) where I am the other day and I quite liked it! Though I agree I was one of the few people driving around with all my windows down…
Back to the dashboard in a car, the general temperature in a car can easily get to 75C, and I’ve measured mine over 90C – and that is not on the dash in the sun! When I tried on the dash I killed my electronic thermometer – it came back when it cooled down – I’ve got a much better one now so I might try it again on a hot day…
Isn’t the point at which chips die when the temperature goes above the point where temperature spirals out of control, as in they cut off at 95 because that’s below the point where it starts to spike exponentially and once it gets to that point it inevitably gets much hotter and breaks. Once the chips had cut out this couldn’t possibly happen. There’s that graph of intel chips showing why they changed from the p4 way of doing things because the density of heat was greater than a nuclear reactor
“everyone’s favorite Linux SBC”. Lies! (Joining the Call-out Culture). Liars!
Interestingly, used for GPS displined Stratum 1 time servers, there’s some code that will run spare cores to maintain an elevated temperature for consistant timing results. And advice on insulating them for this.
https://blog.ntpsec.org/2017/02/01/heat-it-up.html
Heat pipe possibly. The problem as I see it is more a heat soak issue. I can not imagine a heat sink would do much. I guess a silver or copper heat sink might be needed. It would be interesting to see this done again. I would highly recommend a liquid metal themal compound. It could be the reason for the order in which they failed.
As to the 3x thing, I’m not a stickler. I seriously consider things like this to be invaluable information. Living where it hits 110 in the summer months. Any car projects need to survive 140+ hmu.
Maybe a milled heat spreader with heat pipes to remote heat sink. Then insulate everything but the heat sink.