EEVblog Electronics Community Forum
Electronics => Projects, Designs, and Technical Stuff => Topic started by: Christopher on February 16, 2015, 06:22:18 pm
-
Hi All
I am working on a project involving a lot of power dissipated from 10xTO247 packages into a heatsink.
I have done some trials with various gap-pads and am now looking for the best idea from you guys.
I've setup a test with 100W spread evenly through two devices (in a heatsink cut down to 1/5 of the regular size) so I can test various materials overnight. The device is rated to a maximum Tj of 150C. Both devices were torqued to 35cNm with spring washers.
So far I have used:
- Cheapo sil-pad sheet. Junction reached 200C and obviously short circuited after a few hours. Heatsink didn't get too hot
- 1mm thick Laird Tflex 640 sil-pad sheet. Junction reached 150C and the heatsink got nice and warm
- Thermal compound (not electrically isolated) Junction reached 70C and the heatsink got really really warm (best case!)
I am looking for a solution which is as good as the thermal compound but electrically isolated (Very modest isolation voltage--up to about 200V). Can anyone suggest anything?
-
Mica?
-
I'm not sure about industrial or high volume production solutions...but I remember using some ceramic based thermal compounds in PCs.
The first one that came to mind was Arctic Silver Ceramique
http://www.arcticsilver.com/cmq2.html (http://www.arcticsilver.com/cmq2.html)
-
If you need the best possible heat transfer avoid any additional pads, but then the complete heatsink is connected to the voltage.
Silicone pads are cheap but not the best. Mica pads are better but need thermally conductive paste.
Pads made out of alumina may be even better than mica.
-
I thought that most thermal grease has fairly high resistance. The advantage of the pads was less mess which was offset by poorer thermal conductivity. metal tabs of the package will push through grease though, and make electrical contact that way. Did you try a pad type with a little thermal grease to improve the heat conductivity? Mica with thermal grease should also be good.
-
I thought that most thermal grease has fairly high resistance.
Grease/compound, etc may very well be "insulating" to the extend that you need it.
However, I can't see how you can have any reasonable confidence in ONLY a viscous fluid to maintain a critical insulation point. That just sounds like planned disaster to me.
-
T global red pads are very good; they squish
-
That's why I pointed out that the metal tabs would push through the grease and make electrical contact and recommended a pad or mica along with the grease.
I thought that most thermal grease has fairly high resistance.
Grease/compound, etc may very well be "insulating" to the extend that you need it.
However, I can't see how you can have any reasonable confidence in ONLY a viscous fluid to maintain a critical insulation point. That just sounds like planned disaster to me.
-
The problem is the drain tab on the back of the FET is connected to 50V. Obviously a liquid won't do any good so I am looking for some better options with minimal heat transfer tradeoffs!
T global red pads are very good; they squish
It seems like the thermal pad (I have used Tflex 640 at 1mm thick, may go to 0.5mm and it is very good) will be the best but most expensive solution. Any other ideas?
-
I can confirm that a mica insulator is not sufficient with just any thermal paste. I had a half bridge of to-247 IGBT's mounted with mica pads and the gold thermal paste from ebay ~ 30-40v on the bus i heard pops and cracks and saw smoke but the circuit continued to work. I looked closely at the edges of the mica pads and i could see it arcing over the edges through the paste.
-
Kapton is pretty good- its not as thermally conductive as mica but its strength makes generally available in thinner forms, making it a better option.
-
How about taking a more engineering approach and look in datasheets for thermal resistance of various materials?
-
I have tried various gap pads the company use on other products (We deal with a lot less power normally) after looking at the datasheets.
I am wondering now if there are any other better options! As a junior engineer with not a lot of experience it really helps having the opinions of experts.
Kapton is pretty good- its not as thermally conductive as mica but its strength makes generally available in thinner forms, making it a better option.
I thought Kapton tape was fairly thermally insulated? Do you mean a layer of tape then a squirt of thermal grease as per normal? Sounds like it won't work to me.
-
I have tried various gap pads the company use on other products (We deal with a lot less power normally) after looking at the datasheets.
I am wondering now if there are any other better options! As a junior engineer with not a lot of experience it really helps having the opinions of experts.
Calculation (http://en.wikipedia.org/wiki/Thermal_resistance) various temperatures in a system is very much like ohm's law. Better to calculate what you need, then use trial and error.
Determine how many watts you need to dissipate, and then add all the termal resistances. The hightest ambient temperature and the temperature you allow on the chip will fill in the last blanks ;)
-
Of course I have done basic thermal analysis on the ideas I have had and tried out.
My next step is to gather more ideas of how we in the real world get heat out of devices into heatsinks!
-
The idea is to minimize or eliminate any air trapped between the tab and the heat sink. Those tiny bubbles of air are thermal insulators. Try a pad or mica type along with a tiny amount of grease to displace any air and form a good seal. Make sure the pad (or mica) is a little larger than the tab all around. Don't use any of those greases that are filled with silver or gold or anything else conductive. Places like Mouser or Digi-Key sell plenty of suitable types of thermal grease and thermal pads.
-
Laird Tgard K52. It's a kapton film coated on both sides with a ceramic-filled phase change material. Available with 1-3mil thick kapton depending on how much cut-through resistance you need. Drawback is that the PCM is only 1mil thick, (0.5mil on each side) so it's far from a gap filler, but we use it between machined surfaces and find it works pretty well. As a bonus it laser cuts very nicely, so custom shapes are easy.
-
Laird Tgard K52. It's a kapton film coated on both sides with a ceramic-filled phase change material. Available with 1-3mil thick kapton depending on how much cut-through resistance you need. Drawback is that the PCM is only 1mil thick, (0.5mil on each side) so it's far from a gap filler, but we use it between machined surfaces and find it works pretty well. As a bonus it laser cuts very nicely, so custom shapes are easy.
This is the correct answer :). Unless you're surfaces are rougher than anything you have any business trying to thermally interface, you an expect K52 to give results very close to a well-clamped thermal greased contact. It's priced very competitively with thermal interface compounds/grease too. Sure, it has way more electrical isolation than you need, but electrical isolation is like the cocaine of electronics: you'll always want more. 200V is enough, but if you can have more....it's not a bad thing. ^-^
Another option no one has mentioned is using heat spreaders, but I think that will be less coss-effective than K52. In the simplest form, you solder each TO-247 to its own heat spreader, which is just a wee little copper square. If done right, it will effectively increase the thermal contact area of the part. Not as if it were the size of the heat spreader, but it's still going to be pretty good, as if a bit smaller square of contact is there instead. This is due to the heat having to flow laterally through the copper to spread out. For situations like yours, they would work quite well though. Using the sil pad method that got you 150° junction temps would have more surface to work with and lower resistance, and get you to where you need to be. But not nearly as good as K52, and I *believe* not at lower cost, but then again, soldering bits of copper to parts is not exactly something high cost in the electronics industry ;). So I am not really sure.
I'd go with K52 because any possible added expense it might cause is well worth me not having to figure out if heat spreaders would be cheaper or not, so no matter what, I'd be getting what I am paying for ;). Anyway, I have used K52 several times, and it's always performed pretty close to using direct contact + thermal grease in my experience.
-
Also, consider spring clips or spreader bars instead of screws; TO-247 isn't too bad, but there's still concentration of clamping force towards the end of the package that doesn't need it. You can also use MAX247s in this case.
Things to consider about your circuit:
Are any/all of the tabs connected in common? If so, they can share a heat spreader plate.
A heat spreader can be floating (possibly but not guaranteed connected to the tabs), or electrically connected; but whatever the electrical case, the point is, you can get a nice low thermal resistance from case to heat spreader, then a much wider footprint between spreader and heatsink, where you can use a conventional (Sil-Pad, Gap Pad, etc.) insulator much more effectively. Mind the increased capacitance, if this is a switching application.
Can you rearrange it so the tabs are at a common potential? Example 1: MOSFET audio amplifier, source follower design: drains are at +/-V. Alternative: open-drain (rail to rail) type output, which has both tabs at the output voltage. Example 2: electronic load, where all the drains are connected together, to the live load voltage. Alternative: use PMOS so the drains are at GND.
Tim
-
If you're not scared of the toxicity and want the ultimate thermally conductive insulator, try beryllium oxide...
-
Downside, actually, is because it's inflexible, thermal resistance is still dominated by the greased interfaces, so it's not much better or different from metal (conductive, but you can use a big one as a heat spreader) or other thermally conductive ceramics (alumina, AlN -- which are also cheaper and, of course, not toxic when broken).
I suppose I'm not sure offhand what an optimal use case for BeO would be. It was lovely stuff back in the tube days, where the highest performance ceramic tubes were made entirely out of the stuff (no glass), using vacuum tight metal-ceramic seals.
There are RF transistors packaged on BeO substrates, with a smooth or polished surface for best thermal performance. That makes sense, because they're already $$$, and the reduced capacitance of a more-conductive footprint helps performance. I'm thinking, for most purposes, it would be hard to justify the slight performance edge of BeO over other ceramics, unless you've got something direct-bonded to it, like that. And even then, there are cheaper (DBC = direct bonded copper) mounting methods, such as for power switching devices (your average SOT-227s, brick-style transistor modules, etc.), which are perfectly happy simply using more footprint to compensate.
*Shrug* but I digress.
Tim
-
Get the best from Shin-Etsu, complete products list -> ShinEtsu MicroSi, Inc. (http://www.microsi.com/Products.aspx)
I believe pad is better, especially the two mating surfaces are not very smooth.
Few years ago bought from ebay, a really dirt cheap top tier thermal pads, looks like they're industrial excess stock from an Israeli's seller (its gone now).
Although not intensively tested, this one beats all other various branded thermal pads I got. Currently its the top performer thermal pad series from Shin Etsu.
Mine is TC-80BG , with thermal conductivity -> 7.3 W /m-K (even at 0.8 mm thickness :o) , and with dielectric breakdown voltage at 21 Kilo Volt, this should be more than enough.
Detail spec here -> High Hardness Thermal Interface Silicone Rubber Pad Materials (http://www.microsi.com/HighHardnessThermalInterfaceSiliconeRubberPadMaterials.aspx)
(https://www.eevblog.com/forum/projects/thermally-conductive-electrically-isolated-heatsink-compound/?action=dlattach;attach=136848;image)
-
Yes, "back in the day", BeO TO3 insulators were the bees knees for thermal conduction and dielectric strength. Most high power RF Valves were constructed of the stuff. I even had a tube of Beryllium heatsink compound with a box of disposable gloves :scared:.
As far as it compares with modern ceramics.... I don't know.
-
directly bolt the to220's to a wide strip of aluminium bar, then use the insulator to keep that isolated, the heatflux will be lower and that will reduce the temperature gradient.
-
I'd suggest looking on isolated high-power packages, like ISOTOP (SOT-227) if you need pull lots of watts into heatsink, while keeping everything isolated.
Yes, they are little pricey, but thermal performance is up to price as well.
-
it's the oxide which is the killer; we used to use machined beryllium as x-ray windows (admittedly it was machined in Russia....we didn't ask where!)
-
I note when I was repairing an old HP 6644A power supply , HPs solution was to electrically isolate the heatsink, made for interesting times when working on it as it had about 80V DC.
-
The other thing to check is that your heatsink and the transistor tabs are absolutely flat. Cheap heatsinks tend to be anything but flat, and the tabs on some TO packages I've seen were comical (possibly counterfeit devices).
-
Perhaps it would be effective to solder/attach the device to a small copper or aluminum plate then isolate the plate from the heat sink, that will give you more surface area for the less thermally conductive (electrically isolated) interface.
-
Hi,
I dont know if this has been mentioned yet, but the thermal resistance is just like an electrical resistance in that the 'longer' the medium the higher the resistance. In the case of the thermal resistance, the 'length' is actually the thickness of the material between the two pieces, the heat producing device and the heat sink.
So the key is the thinner the medium the better the heat transfer, regardless of material. Some materials are better of course, but it is surprising how well a simple material works as compared to an exotic material when the layer it kept very thin.
I did an experiment a few years back with some relatively normal epoxy and some Arctic Alumina epoxy. Two high power LEDs on an aluminum heat sink, one glued down with regular high temperature epoxy and one glued down with Arctic Alumina. The temperature rise with equal power in each LED showed that both substances were nearly equal. I can not explain why, except that the most important part is the thickness of the substance, it must be kept thin, and that's what i did for this experiment. Even a thin layer of Dinosaur dump would probably work :-)
LEDs a are pretty nice for testing these materials because they can be mounted to a heat sink easily, and their voltage change follows the temperature change, so with similar LEDs we see a similar forward voltage change with constant current. To test the two to make sure they are similar, use paste first and see that both voltages change in the same way when mounted to similar heat sinks with the same type of paste.
Of course you can use heat sink mounted diodes too, or anything like that where you'll see a change in operating point with temperature. That means you dont have to measure the case temperature to perform the comparison.
One last quick note:
Although the two substances above had shown the same characteristics, they could be temporary results that do not hold over long periods of time because the aging of the two materials could mean one gets much worse over time. For this reason i would use the material made for the job rather than some made up compound, unless i was willing to test and retest it several times over many years.
-
With a thin enough layer, thermal resistance depends less and less on the TIM (thermal interface material) and more on the device construction itself (internal RthJC and so on).
It's true even for no liquid or solid material alone; a very flat surface with moderate clamping pressure, filled only with air, has still usable thermal resistance (ballpark figures seem to be around 50-100% of RthJC for typical cases, versus < 10% with grease). It's not disingenuous to call such a joint "air filled", as the thermal resistance is significantly worse (200-1000%??) under vacuum: as bad a conductor as air is, it's still a considerable contributor to the conductivity of ungreased surfaces.
For a large part, "high performance" greases are absolutely useless. Properly mated surfaces squeeze an ordinary silicone grease down to microns or less; indeed, the only case where the grease does matter is when the surfaces themselves are exceedingly poor to begin with: warped, scratched, pitted, uneven surfaces that don't mate well.
The direct analogy is trying to improve the performance of a linear power supply by replacing the wiring with silver instead of copper; yes, you are unequivocally reducing the resistance, but you're absolutely barking up the wrong tree, considering there's four ohms in the transformer (or whatever) and about 100 microohms difference between those choices of wire.
Or a programming analogy: optimizing the main menu with assembler code, rather than the deep inner loops that are actually using 99% of CPU time.
Tim
-
The tgard looks about right. I have a sheet on order and will be due in by the end of the week...
The sil pads we have here are pretty crap. I believe they are made from kool pad and have a thermal resistance of about 4 c per watt. Perfect for their intended use in switching power supplies but useless for dissipating more than 10w!
Unfortunately the devices I am using the tab is connected to individual voltage rails so I cannot weld them to a heat spreader although this would be the best solution.
The devices are all mounted on a 10 mm thick ali plate which is then secured onto a hefty black anodized heatsink. I cannot change the mechanical or electrical properties as these have been designed now.
100w thru two devices and standard thermal compound overnight had a junction temperature of 60 degrees which is my baseline figure. I want to obviously keep the junction temp as low as possible (max of 100) and the tgard material should be fairly close to this figure.
-
This topic is old, but wanted to mention another possibility depending on available materials and immediacy of need.
Standard electrical tape has thermal conductivity around 3 W /m-K at 0.7 mm thickness with dielectric breakdown rated at 600 Volt and max temp rated 80C.
If you can get tape rated for 150C..
There was a GPU application here with high limit of 80C needing electric isolation, where I successfully used the following approach for limited card (not heatsink compound) testing:
The tape was stretched in both dimensions circa 1.5:1 to 2:1, considerably thinning the material and theoretically raising thermal conductivity to about 10W/m-k at 0.3mm and perhaps isolation to circa 60V.
To maximize thermal transfer and minimize possibility of creating a conductive path, graphite with high thermal conductivity was applied to connecting surfaces:
1. The chip surface and heat sink connecting surface was coated with a layer of graphite (using No.1 lead pencil).
2. After tape was stretched, it was applied to chip's thermal transfer surface.
3. Graphite was applied to tape's upper surface. Using No.1 lead pencil and liberally writing in a circular motion from the center of the tape outward to avoid creating air bubbles under the tape.
cheers,
[edited to clarify that graphite was also applied to relevant surface of heat sink]