[Michael Dornisch] was surprised to find that the main processor of the Raspberry Pi reaches about 56 degrees Celsius (about 133 degrees F) while streaming video over the network. He thought it might help the longevity of the device if he was able to cool things off a bit. But why stop with just the processor? He added heat sinks to the SoC, Ethernet/USB chip, and voltage regulator.
From his parts bin he grabbed a small heat sink that was probably used on a graphics card. After measuring the three chips with his digital calipers he cut out the footprint he needed, resulting in three smaller heat sinks. We didn’t realize that thermal compound has enough gripping power to hold the sinks in place without any mechanical fastener, but apparently it does. [Michael] mentions that it’s possible to use other adhesives, like JB Weld. What’s important is that you use something (ie: thermal compound or a liquid adhesive) to prevent any air gap from coming between the chip surface and the aluminum.
He measured the result as a 17.3 degree C (31 degree F) drop in temperature. We looked around and it seems there’s no internal temperature sensor on the Broadcom chip so these surface readings will have to suffice. Do you think this will prolong the life of the board if it is used regularly to play back high quality video? We already know that these temperatures are within the specifications for the hardware.
71 thoughts on “Adding Heat Sinks To A Raspberry Pi”
Therefore, a drop of 17.3°C is only equivalent to a drop of 31°F – not 63°F.
Good job on the math there. That’s a simple enough thing that it would fool most people.
I was just going to say the same… good catch. (Oh how I love offset temperature scales!)
Thanks for pointing that out. That’s a bizarre gotcha.
cute, but that temperature is insignificant to most electronics.
Hate to be an asshole here, but your temperature drop calculation is incorrect. 56C is 132.8F. 56C-17.3C = 38.7C. 38.7C = 101.66F. 132.8F-101.66F = 31.14F degree drop. You have to do the conversions for the temperatures separately, you can’t do (17.3*1.8)+32.
What he said.
Heat does not kill chips so long as you are within spec, they actually last longer when running hot as they’re not so baldly effected by hot carriers.
Reasons to cool a chip are to prevent BGA soldering from breaking or to achieve higher clock speeds, not to make the semiconductor last longer.
Reducing the voltage to a value where it remains stable would also make it last longer, but seriously, it’s £35, you can get another one.
Wow, thanks, I learned something new today. :)
It’s only 5 bu it is a 4 months wait :)
I hope the thermal compund glues thees things very well, or he might end up with a big metallic part falling on his runnig Pi, shorting everything. That woulsd kill it way more quickly than a 56°C chip temperature.
I agree, we run FPGA’s at over 60C , some applications even run as hot as 75C, case temperature. The maximum is 85C, so even if ambient rises (which stops at 30C anyway) a bit we’re under that.
If you’re thinking the chips are running hot some air convection will help a lot too. However, that’s only useful if you encase the Pi. These heatsinks will probably do the job too, but unless you live on the equator with summer days of 45C+, I don’t think there wasn’t an issue at all.
That’s not practically true. There’s diffusion and electromigration at the semiconductor junctions that only gets worse with heat.
The rule of thumb is that for every 6 degrees increase in temperature, you halve the lifetime of the components. A part that has been specified for an MTBF of 50,000 hours at 85 C will survive roughly 182 years at 56 degrees. It’s not usual for the chips themselves to break, but everything else around them.
What about thermal cycling between power on/off?
The main effects on start/stop are current surges and voltage spikes. Main failures for semiconductor junctions are electromigration and overvoltages breaking down dielectrics. On/off thermal cycles mainly do mechanical stressing.
On the RasPi, though, probably your safest bet for improving longevity (outside of reducing temperature) would be adding inrush current limiting and replacing the tantalum/electrolytics with ceramics.
Anything under 60 should not affect the lifespan of a silicon chip, at least not to the point where you would generally care. (should last a good 40 years if temperature is the only factor).
Now many components today are designed close to the limit and will break after only a few years, but often not due to temperature but due to simple bad construction or single faulty semiconductor on chip etc.
I agree with all of the above. It won’t make the chips last longer. It will make the board as a whole last longer if the temperature fluctuates less- But who cares! If you have heatsinks lying about I’d expect you to be able to fix any problems the board might have ten years from now in mere minutes.
As an aside I regularly run a high end rig within a few degrees of max spec for hours and hours at a time. It is a 1200 watt space heater. Been running just fine for four years. I expect something twice as good as the pi to come out within two.
Did you mean you expect something “twice as good as the pi” to come out within 2 years? If so, they’re already here. There’s actually been a few jumps in the dev board area in terms of size and performance. There’s power friendly 64-bit 8-core ARM development boards that have gigs of dual-channel DDR3 RAM, eMMC memory (arguably the “board’s SSD”), GPUs with half the power of a XBox 360, gigabit LAN, the list goes on.
Not that Samsung is willing to give the Exynos 5420 cheap.. There are still plenty of alternative boards out there. There’s the VoCore for smaller WiFi projects. Don’t even need a board for DIY hacking? A MK802 is impressive — I could play RuneScape Oldschool (i.e. circa 2007) on Linux with OpenJDK and retain 22 FPS. Realistically, that’s not bad for around $38.
I have a brother who used to work for a semiconductor company and apart from the traditional clerical error(s) of sending military spec components to toy manufacturers, and toy spec components to … any how that is a different story. What I was going to say is that most semiconductor companies cook components, while powered on, in ovens for several hours to in effect force the ones that would fail within the first 10 years of use, to fail in subsequent quality tests. I’m not saying that running a RaspberryPi at a cooler temperature will NOT extend the lifetime of the chips, I’m saying that in terms of years of extended lifetime it may not be all that much. But where it may extend the lifetime is by causing the crappy European lead-free solder [Restriction of Hazardous Substances Directive (RoHS)] not to grow whiskers (http://en.wikipedia.org/wiki/Whisker_%28metallurgy%29) as fast.
Keeping the device above 13.2°C may actually extend the life of the device, by slowing down the progress of Tin pest (tin disease, tin blight or tin leprosy)
The bottom line is that Tin without Lead is good for a short product lifespan (lots of replacement sold more often than is necessary $$$/€€€/元元元, but it is worse for the global environment).
Don’t forget about tin whiskers. Temp. fluctuations are said to be one cause of them.
There are very few ‘military spec’ parts anymore. The military has been using commercially or rather industrially rated parts for decades now.
And baking the parts is to remove any moisture from the parts slowly, so that they don’t “pop” when soldered or lead to internal corrosion from the moisture.
Thermal CYCLING is what breaks down electronics, not just temperature by itself (within reason of course.) If the raspberry Pi is run constantly, and that temperature remained pretty constant, it would not cause problems. Actually, as others have noted, those temperatures are not going to be a problem anyway.
Cycling temperature causes the tiny wirebonds inside the chip to expand and contract, which eventually breaks them.
Finally, with some power electronics, heat is more of an issue because a hot conductor = less resistance, which means more current, which means more heat, etc…
I would be more concerned about the extreme heat generated by a part affecting the PCB, glue in vias, glue holding traces, etc… localized heat is typically not a good thing.
I am an electrical engineer in the aerospace defense industry. We bake our parts and our boards to get rid of moisture, do our thing, then conformal coat them. Moisture is the real issue. Heat only becomes an issue if the convection temperatures could reach the point that solder breaks down, glues breakdown, or electrolytics boil.
A “burn-in” test to determine a potential failure is extreme temperature cycling over a period of time. More important than the temperatures themselves is the rate of change in temperature. And that typically finds bad solder connections, not parts that can’t handle it. Though that does ocassionally happen, but usually we find it is because moisture is trapped in the part.
“because a hot conductor = less resistance,”
There are some negative coefficent materials, but your statement is generally untrue.
The hotter things are, the more resistance.
Dax, correct. I think I just bungled up trying to explain “thermal runaway.” Obviously a hotter conductor would mean more resistance.
There are some complexities (and this is definitely not my expertise, I admit) based on whether the conductor is generating the heat, or the heat is from an outside source. But, that paragraph would have been better just left out. Too awkward. Or perhaps you can offer a better explanation to benefit other readers?
Thank you for the correction.
Maybe he meant the carrier mobility (or current gain) in bipolar junction transistors, especially secondary breakdown?
Here’s an interesting question:
Suppose a Pi, or anything else with BGA parts, is used in a car environment. Where it is subjected to ambient temperatures up to 120°F, and vibration.
Assume cracking of lead-free BGA solder joints is the primary mechanism of failure.
Would heatsinks like the ones shown in this article help, by reducing peak temperature and thermal stress? Or would they hurt more, by adding more vibrating mass?
I’d just use zip ties to fasten the heatsinks to the board…
Thermal stress? Reduced!
Heh, I’ve used zip ties before.
But let’s assume that in this case, there’s too many obstructions to use zip ties to get anywhere near a decent clamping force. And so the heatsinks must be self-adhesive, not supported by anything else.
Yes the Thermal stress would be reduced however some one has now put a big blob of metal on the chip and increased the mechanical stresses.
Any way 55c is nothing for most electronics even cheap comercial stuff are good for 80c. Best just leave your raspberry Pi alone like I did.
BTW since when did sticking an heat sink on a chip constitute a HAD article anyway ?
Umm. How was the second temp measured or deduced?
With the heatsink on its awfully hard to get a comparable surface measurement. Measuring the sinks tells nothing.
Aaaaand…. In a world where IC’s are typically rated for a minimum of 85 degrees C (105 degrees C available for many of them)… This will accomplish… Exactly nothing!!
Congratulations. You can stick heat sinks on things.
C’mon guys, this post is pretty lame. Incorrect temperature values, crappy build, and doing nothing really more than slapping a heatsink onto IC’s that can handle stock high temperatures.
Lets see some of the older quality posts that I started browsing hackaday for.
IMHO, a temperature of 56C on the package surface is serious enough to warrant attention. The epoxy packaging material is a poor heat conductor (and also blocks IR from reaching the thermal imager), so it is reasonable to assume the chip temperature is much higher. I suppose a heatsink further attenuates the amount of IR reaching the thermal camera. At my workplace, the chip temp is measured by removing part of the package to expose the chip. A thermal modeller like the one included in the freeware AppCAD is useful for understanding where is the heat bottlenecks.
Thermal interface materials (e.g. silicon grease) have poor thermal conductivity relative to metal (e.g. heatsink). Hence the heatsink must be clamped tightly (300-600kPa) to the device to thin out the thermal compound – instead of being slapped on top of the package. Rigid metal to metal contact (e.g. package flange to heatsink) benefits from thermal compound to fill in the gaps due to the microscopic surface irregularities. With sufficient clamping force, I suspect the epoxy package can deform enough to conform to the heatsink and maybe obviate the need for the gap filler. A good tutorial is Altera’s appnote, http://www.altera.com/support/devices/power/thermal/pow-thermal.html
And that’s why they make TIM/thermal compound out of silver or even diamond, because of its higher thermal conductivity. There are plenty of epoxy like TIM/thermal compounds made for exactly this type of use, they can also be mixed with silver or diamond based compounds for better thermal conductivity… It would be relatively easy to get the necessary clamping force onto the chips temporarily until the TIM/thermal compound dries.
IMHO, you have no idea what you are talking about.
IC’s rated for 105 degrees C have much higher thermal shutdown temperatures – Generally in the range of 150 degrees C die temperature. 56 degrees C is NOTHING for this IC, nor virtually any other IC. I have seen IC’s with thermal lockout die temps of almost 200 degrees C.
Just an FYI.
The broadcom chips have a very high temperature limit and are designed to run hot, ever felt the top of a cable box or sat receiver ? As humans we see something like 50C and think, wow that is too hot, but remember we use our body temps as the reference for what is hot, 50, 60, 70C is nothing for semiconductors designed for that, it is a cool spring day.
Hey guys, those “specs” are not “everything’s just dandy” temperatures, they are maximums. The processor was breaking the temperature rule of thumb- “If you can’t hold your thumb on it continuously, it’s too hot.”
Hotter is better? Definitely NOT! Here, the rule is that you double the life of a component for every 10 degrees C you drop its temperature. Why? The devices consist of P and N doped regions which are created by putting the dopants (like arsenic or phosphorous)on the chip surface and drifting it into the silicon with HEAT. Long periods of high chip temperature can cause doped areas to drift a tiny bit more until a device on the chip moves out of spec enough to cause failure. With electronics, cooler is ALWAYS better, if you want reliability!
Michael’s work is spot on!
I don’t disagree that cooler is better (to a point), but I am not certain that was achieved here. I don’t think anyone could be.
1. The Heat Sink is applied without enough force to properly mate the heat sink to the chip, making it inefficient.
2. The Heat Sink is applied above a layer of insulation (the chips ‘cap’), also causing an inefficiency.
3. The heat sink is passive, so if the Pi is placed in a closed box will it have an effect anyway? It won’t facilitate heat transfer out of the box.
4. Only the temperature of the OUTSIDE of the chip was taken, and presumably after the heat sink was installed the temperature of the outside of the heat sink. So for all we know the temperature inside the chip climbed. He could have just placed a thermal blanket over the chip, (Due to 1 and 2)
5. He added mass to the chip, and thus stress, to the solder joints, board… This was done with no ability to prevent or mitigate the extra stresses.
So if you take those 5 points and add them all together, I am betting this will lead to faster failure, not more longevity.
The real questions are even if you assume that that the heat sink is properly installed; did lower the temperature of the chips; do not add stress to the solder joints: What is the expected lifetime change? 10 years to 12? Do I care for a $35 item? Do I expect to be using it in 15 years?
If you thought the chip was too hot I would think, given the above points, you could get better cooling of the chip by adding a small fan blowing across the board. No heat sink required, no additional stress to the board and solder joints.
Of course I could be wrong. But when you are talking about these things it is important to consider things as they are and not talk about one point (lowering the temperature of a chip) when it has not even been shown that was actually accomplished. That’s why HAD has comments. ;^)
I dont think a 5 gram heatsink adds stress to the solder joints. As solder spot is metal to metal (60/40 Sn/Pb touching copper), it takes lot more weight to brake a solder spot.
For the other stuff, you are right. That chip will last enough time – more then we need it to. I’m sure the next version of PI’s is already in development. If the engineers thought that would need a heatsink, they would have installed it (thus raising the price for .50C) lol. If I plan to use mine 24/7 and if it runs hot – I’ll install a heatsink on it. Otherwise NO.
I think with the new overclocking and overvoltage on demand as recently released, heatsinks do make sense as the overclocking is reduced when the die gets hot ( There actually ARE a couple of temperature sensors on the die ). As for the thermal resistance there’s little daubt in my mind that it’s reduced this way a lot. Thermal resistance from die to package and from processor package to ram package probably is of no importance compared to thermal resistance to air. Thats the one reduces when adding heatsinks. I found some heatsinks designed for TO220 package and used 2 component glue to attach it to the SoC and ethernet chips. I agree, for the ethernet chip it probably makes no sense. Adding a heatsink to the regulator is imho useless
Bob, an IC rated for 85 degrees C can generally withstand a die temperature of at least twice that temperature, for a significant amount of time, without any damage. I know this from experience.
The “if it’s too hot to touch, it’s too hot” sentiment is nice, but incorrect. Most IC’s should operate near their maximum rated temperature for long past their useful lifetime.
While the temperature might not affect the chip itself, a high temperature in the case will affect the lesser rated components inside. CHEAP CAPACITORS for one. And overclocking capabilities.
The 220 uF electrolytic capacitor is located near the hot voltage regulator. The life span of that is a lot shorter at high temperatures. The broadcom chip is not likely to fail due to high temperatures in 40 years or so. Some people have reported unstable raspberry’s but a problem in the psu is far more likely in that case ( tried some really cheap psu’s myself ….. learning the hard way .. )
Thermal compound is generally not a suitable adhesive. Sure, it might stay on with surface tension – for a little while.
Use a real thermally conductive epoxy if you want the heatsinks to stay attached.
was it paste or adhesive?
Regular thermal paste may not stick very long on its own.
There are thermal adhesives out there, from aluminum oxide/boron nitride adhesives to thermal adhesive tape/pads.
I wonder if it wasn’t one of these.
HUR DUR I PUT HEATSINKS ON MAH PI, NOW IT RUNS %200 FASTER !!11!1!
HURRR DURRRR, no one said it ran faster, just cooler… Although, it IS possible to overclock the PI, and this mod would help tremendously if heat was a limiting factor (it usually is)… But you’re probably too ignorant to realize this, hence your idiotic comment… Don’t worry, you’re not alone, there are plenty of other like minded idiots like yourself who are quick to pass judgment on something they know absolutely nothing about.
ummm… since you have a thing for articulating [your] ignorance: to what extent does the clamping force play a role in the heat transference mechanism? In your answer please make reference to how this unclamped configuration will make the chip temperature hotter not cooler. Please also point out at what thickness thermal compound becomes ineffective at transferring heat to the sink. Extra points for pointing out the experimental error(s) demonstrated in the final test.
seriously tho, this is a great article, very useful and well written. keep up the great work guys !!
oh did i mention i put heatsinks on my calculator expect the 16 page write up soon.
i try not to be too critical about stuff but come on, putting heatsinks on a board counts as a hack?
i would rather have no new content than fluff like this.
FWIW, I’ve mixed JB Qwik Weld and Arctic Silver 3, and 5, with excellent results. The heatsinks stuck to the chips fairly well (couldn’t budge ’em, even with twisting), and they got REALLY hot! However, the router onto which they were attached operated flawlessly afterward. Before the heatsinks, after about ten minutes of operation, it would just…die. Now, I haven’t had a single dropped connection from it in the past year or so, and even though I’ve no temperature data to fully-support my ‘hack’, it works for me.
The ratio of epoxy to heatsink compound was about 3:7, IIRC, maybe less (less epoxy that is). I just fudged it, and I might have tried to ‘balance’ the mix by specific gravity (it was a while ago, I can’t remember lol).
I think what RPi needs at the first place is proper power input. I added a 7805 regulator on top of the board to prevent voltage drops which cause problems with many usb devices.
Well, that gets hot if you feed it with typical 12V supply and absolutely requires a heatsink as RPi draws around 700mA current. I had only a light heatsink but at least it doesn’t now burn my fingers if I touch it…
You talk about proper power and then you add an 7805? If you would investigate a tiny bit more you’ll find replacements that don’t burn power like crazy.
Hi, there are these things called buck regulators, you should look in to it.
Yes the 7805 would need a fair sized heat sink to dissipate the 700ma or so failing that get so hot as to melt whatever case you put it in. The Switch mode regulators are available as pre done modules for £2 ish from ebay …Much easier solution
Wake me up when properly mounted thermocouples are involved.
“From his parts bin he grabbed a small heat sink..”
Heat sinks are made to last! The components they protect against overheating however aren’t.
My applauds on HaD using proper scientific measurements.
The heatsinks are poorly mounted, however since almost any type of paste is going to be magnitudes better at heat transfer than air, given the same factors (no moving air aside from convection from the heated air moving away from the chip), the chip will be cooler.
The one problem I’ve seen with chips that run hot but still within spec, is that the board around the area begins to discolor, almost as if the pcb and the conformal coating are burning/melting (observed in various cheap routers).
The main benefit to heatsinks here would be reducing stress from abrupt temperature changes and localized heat.
I just did this to a router this morning, and it gave me peace of mind, even if it won’t make a difference in the life of the router.
I personally would be more interested in getting the RasPi to run cooler, not run hot and get rid of the heat. Heat sinks are great for getting rid of heat as long as that heat isn’t being dumped on to /you/ (a la a wearable computer). Maybe the model A will run cooler w/o the Ethernet (which I don’t need).
You can’t put a heatsink on the SOC, because the dram is on top of it in a POP configuration, and there’s a slight gap between the packages. (I wonder if putting a heatsink on the dram might actually create some unintended thermal patterns in the POP stack…)
Undervolting the processor is usual way to make it run cooler. Evidently there is facility for undervolting, but I haven’t found any reports of successful undervolts. (As usual, underclocking may be needed in addition.)
With all the coverage here of temperatures on the Pi, would be nice if somebody tried it and checked results with thermometer, IR camera, whatever.
(Undervolting may void the warranty, as overvolting does, the comments I have seen have not made it clear whether any adjustment to the cpu voltage voids warranty, or if it is only adjustments above a certain level that do so. So approach this with caution.)
I am using RPi as a webserver.
Raspberry Pi inside a box definitely overheats and
hangs after a few hours of operation.
Power has to be switched off for some minute before it can be restarted.
Finger can not be held on the chip near the network connector.
Crash also happens after a minute or so if many files are transfered using FTP.
A cooling fan helps to avoid hanging.
Fan is somewhat noisy and used 0.5 W, but board is not intact compared to heat sinks, if want to sell it later.
Me I don’t mind my Raspi breaking in more than 5 years or so, I’m very sure at that time, they’ll release a new version which I would probably buy, and throw the old one. or use it till it breaks.
You can see my findings here for something very similar. http://r1b4z01d.com/blog/1032/projects/raspberry-pi/pirobot-build-2-adding-heat-sinks-temp-tests/
some cooling is always good. check out this article about overclovking and cooling. The results are impressive :)
Pretty good temperature difference, more than with just heatsinks and thermal tape results like this http://www.rootusers.com/raspberry-pi-heat-sinks-temperature-before-and-after/
am not say any thing but its great great
What about the chip under the board?
Please be kind and respectful to help make the comments section excellent. (Comment Policy)