How many watts would need to be drawn by a chip in what unit time to cause fire?
Bill Henning
Posts: 6,445
Due to a discussion with tonyp, I am now extremely curious as to how many watts drawn by a chip is needed to literally cause a fire.
Q1: How many watts before a TQFP-100 center ground/heat pad package will literally catch on fire? and how long would it take?
(Small smoking hole in the center of the package that cannot spread a fire does not count.)
Sample responses, picked out of the air, as I don't know if it would be enough / too much to cause fire:
1000W for 1ms
100W for 10ms
Q2: How many watts needed to be drawn before a TQFP-100 center ground/heat pad package will cause the PCB it is on to catch fire?
The trace current capacity tables I use do not have the "PCB WILL CATCH FIRE" data :-(
Only the heat rise vs current at different copper thickness. Of course this depends on solder mask fire resistance, pcb material fire resistance, trace width etc., but a ball park would be good.
Q3: Is it even possible?
Voltage regulators tend to have internal thermal shutdown, so it seems to me that unless there was an awfully high current spike (100A? 1000A?) in a very short time, the whole board would go into thermal shutdown.
Given I don't see how to draw enough current in such a short time to bypass the thermal shutdown, I am curious if it is even possible to cause a modern PCB to catch fire!
(dropping an aligator clip into a high wattage power supply is not a fair answer, as it is against the spirit / nature of this question)
I appreciate your responses!
Q1: How many watts before a TQFP-100 center ground/heat pad package will literally catch on fire? and how long would it take?
(Small smoking hole in the center of the package that cannot spread a fire does not count.)
Sample responses, picked out of the air, as I don't know if it would be enough / too much to cause fire:
1000W for 1ms
100W for 10ms
Q2: How many watts needed to be drawn before a TQFP-100 center ground/heat pad package will cause the PCB it is on to catch fire?
The trace current capacity tables I use do not have the "PCB WILL CATCH FIRE" data :-(
Only the heat rise vs current at different copper thickness. Of course this depends on solder mask fire resistance, pcb material fire resistance, trace width etc., but a ball park would be good.
Q3: Is it even possible?
Voltage regulators tend to have internal thermal shutdown, so it seems to me that unless there was an awfully high current spike (100A? 1000A?) in a very short time, the whole board would go into thermal shutdown.
Given I don't see how to draw enough current in such a short time to bypass the thermal shutdown, I am curious if it is even possible to cause a modern PCB to catch fire!
(dropping an aligator clip into a high wattage power supply is not a fair answer, as it is against the spirit / nature of this question)
I appreciate your responses!
Comments
Tony's concern was extremely well meant, but misplaced.
Thanks!
I don't know about "modern" PCBs but back during the '80s and '90s when I still worked as a EE it wasn't unusual to see holes burnt through PC boards. That was mostly underneath "high power" devices like power transistors, triacs, very large diodes and power resistors though.
So the FR bit did its work then.
FR additives work by preventing flame causing the material to char instead. Which is why wooden houses are normally OK in building codes.
I *think* that that configuration isn't permitted in various Standards.
Well, I've had "FR" double sided circuit boards physically in flames, and the flame spreading. I suspect "fire retardant" does not mean that it won't catch fire. I could not tell you what the wattage being dissipated was. A 24 volt 5 amp source power supply was feeding it.
Cheers,
Peter (pjv)
I'm Back from Hospital.
On Yours question.
Made it AMD compatible and everyone will BURN without cooling!!!
P1+ won't work from without a core voltage regulator.
Which would shut down first.
Using a P1+ or P2 without any voltage regulation is an incredibly bad idea, and will fail and/or fry the processor BEFORE it could catch fire.
Tony,
I have been going out of my way to nicely explain, even started to make a thread to make sure no one else knows of the failure mode you are worrying about, that I knew full well from your first post could not happen. Sorry, your concern is totally unfounded.
Don't believe me, see the other responses in this thread.
We now have another REAL data point!
8W CANNOT cause fire
120W CAN cause fire
Peter is an experienced designer, and would have had wide enough copper that was think enough to take 120W.
Sorry to hear you were in the hospital, I was wondering where you were.
I an VERY glad you are back
LOL!
I have a couple of 125W eight core AMD's... I can believe they would burn without the thermal shutdown if there was no cooling.
The power issue is not about PCBs burning up, it is about a valid measure, like performance vs cost, performance vs power.
Higher power means bigger power supplies, more complexity (ie. going to switching supplies), and more cost. And makes the P2 compare unfavorably to what is available today.
I've seen lots of parts fail, sometimes even flash (like tantalum cap in backwards), or a diode die and melt a trace. I have yet to see the PCB itself catch on fire, though I suppose it is possible, but not at 5W. In most digital electronics I've seen fail (and I've seen a lot), the part itself will fuse and the power goes away.
However, I do remember when I worked at HP eons ago, I shared a bench with the power supply designer for the mini-computer group. And he would always have a fire extinguisher handy whenever he tested out a switching power supply design for the first time.
Theirs thermal shutdown don't work as desired --- That biggest problem.
Only Intel made good thermal shutdown that works.
For the traces we use on a P1/P2 setup they would blow long before generating enough heat to catch fire. For a power supply, or for an output designed to carry 5 - 10 amps, that is another story. Those traces are large enough to generate a lot of heat before they blow. Same for the bigger power components.
I always derate my outputs and hope that some nut does not hook up a big device without an in line fuse as extra protection. In really critical apps you build in a fuse (but then they just bypass it with a piece of foil the first time they blow it LOL).
I tried - many times - in other threads to address it, with no success.
Instead of giving up, I started a thread so others would chime in and say fire with a P1+ could not happen.
They did.
it is still not believed by tony, now he shifted to worrying about killing the chip
But AMD's worked well for me last summer, when the poorly designed fan clip broke.
I noticed my system was getting really slow... and after one PM it shut down.
Looked inside... my 6 core AMD FX had been running without a fan!!!!
I know that... you know that... a lot of people know that.
I was trying to address a non-realistic fear in a nice way without criticizing.
Besides, given how worried tony was, I thought it would be fun to find out what is needed to actually cause a fire on a PCB!
Peter gave a data point of 124W
The P2 has more headroom than we thought
I try to design for at least 50% more capacity than I expect to need, more if it does not cost too much (component, board space etc)
One controller I ran into a long time ago had rather large board for the simple dry contact relay outputs that were on it. Looking more closely when I had a board out one day I saw that they had used 6 small traces to route from the screw terminals to the relay and back out. Later I was talking to the guy who designed the board. Apparently the first run of I/O boards they did had one large trace as is typical. He was at a job startup for the first new system and was standing there when they powered it up and the board caught fire. The contractor had wired two loads onto one relay to 'save $ and I/O' instead of running pilot relays without even checking their amp draw. Needless to say all heck broke lose and the designer got so scared of <choice words here> contractors that he redid the board with 5 medium sized traces and one smaller one. His hope was they would cascade fail without getting hot enough to have a repeat fireworks show.
I am not 'quite' that paranoid
I'm not worried, but saving people from having to implement nationwide recall is always good.
But if a 32cog P1+ thermal envelope and well being was based on that active cycles is 50% on average, first thing I would do is run full blast on all 32cogs to see what happens to it.
I always want to see:
- quiescent current
- minimum something ridiculously low like 1% capacity
- typical (with typical being defined, I am happy with 50% load you propose)
- maximum (maximum attainable utilization that can actually occur ie - 100% - most stuff the cog can do, but if it can only do one of two things at once, pick the more power hungry)
Then there is simulated worst case, which runs all possible gates, even if several blocks cannot run at once (due to processor/assembler limitations)
Unfortunately, before "real" silicon, the best that can normally be done is the simulated worst case (unless there is a handy super computer to run actual code on transistor level simulations) as the characteristics of the process can change slightly, and there is variations on the same wafer.
The more characterized a chip is, ie all of the above at different MHZ, the happier I am
3) It's certainly possible for "modern pcbs" to burst into flames. Protective devices generally work nicely when your fault condition is just outside normal operating area. Circuit breakers won't interrupt a really big nasty current. Polyfuses are notorious for bursting into flames.
2) I would say all bets are off once you hit glass transition of the PCB, because then you can short out internal power planes, or generally end up with a molten mess where there are all sorts of flammable possibilities (direct and indirect).
1) I did some tests last year using WS2812s as a programmable heat bed for a 3d printer. They can dissipate up to about 1/4 watt each, and were on a 9x9mm pitch on the standard Adafruit board, and in that test the pcb was hitting 108C after a few minutes. So roughly, that's dissipating 1 watt in 18x18mm area. I'd expect real trouble (meltdown) at double that, say 9 ws2812s in a 20x20mm area, with a temp rise heading towards 200 C.
As kids we used to light fuses by pumping a couple of watts into something that could't dissipate it. I reckon your "trouble spot" begins at around 2 watts, without proper management. It's interesting that compact Intel board heater linked to the other day claims to run at 70 C ambient ("heatsink not shown"). They will have quite a challenge with that.
But finally... why speculate? Grab a competitors IC in said package, run it in reverse polarity using the body diode as your heater element, and see how long it takes.
How do I arrive at that?
When I was 10 years old, the family rented an A-frame during the summer at King's Beach, near Lake Tahoe.
I was in the habit of reading in bed. Woke up early one morning and left the lamp on with a 25watt bulb attached to a clip on lamp shade. It fell on the bed after I left, still on.
Result, smoke everywhere. A visit from the local fire department to pull the smoldering kapok mattress outside as it was feared it might burst into flames at any moment.
In sum, there were good reasons that old-timers required a metal chassis enclose any electronic project. I suspect 5 watts of sparks or even sub-watt amounts of sparks could start a fire under the right conditions (mouse nest in your chassis?).
With enough propane fumes, watts might not be so much an issue as voltage. But excuse me, as this has all gotten absurd.... useless, unless you consider Erco's flame throwing BOEbot.
What about that smouldering cigarette that burns the house down? Cigarettes are not putting out much energy has heat now are they.
So a nice hot chip in a dusty environment and there you have your fire.
http://en.wikipedia.org/wiki/Minimum_ignition_energy
Edit: Yea OK, I re-read the title it does say "..watts..in..time.." i.e. Joules. Still, main point is it does not take much sometimes.
You can design circuits designated 'simple apparatus' that can go into hazardous areas safely, as long as you can get by without storage elements (caps, inductors).
The limits are 1.5 volts and 25mW, which is quite generous. This thread has some figures for operating down around that area. Unfortunately the P1 xtal oscillator doesn't work well below 1.7v or so.
Basically, I wanted to reassure tony that it cannot happen without deliberately trying insane circumstances.
This of course assumed modern fire retardant pcb's and decent pcb design.
*BUT*
I think we have established - or at least I have not seen any data to the contrary - that it would be almost impossible to make a modern reasonably designed FR4 pcb catch fire with a Pxx processor, at least without taking extreme measures to deliberately cause it to catch fire.
Intel had to distract from the performance issues the Pentium4 had, and they succeeded very well for many years, such was the strength of Chipzilla that no-one would dare say no. That was way back in the Pentium4 vs Athlon days, something like 1999/2000 era. It would be silly to argue that AMD is still there.