View Single Post
Old 02-28-2010, 09:51 PM   #3030 (permalink)
CamLight
EcoModding Lurker
 
Join Date: Aug 2009
Location: New York City, USA
Posts: 23
Thanks: 0
Thanked 6 Times in 4 Posts
Nice sensor setup! May have to use that one for a 1100A load I'm working on for a client this Spring.

I did a lot of research on the available thermal transfer compounds as I was really pushing the limits on how much heat can be moved out from a small volume. I picked Aavid's UltraStick phase-change compound as my #1 choice and have been very, very happy with it. It's a soft wax at room temperature but when warmed it turns to water consistency to flow and fill in every single micro-pocket in the metal. Going that runny also allows the FET to press against the metal as tight as possible. Ideally, you want absolutely no compound between the two component's high points of contact. You only want he compound filling in the low points to increase the surface area that's touching. When you remove the FET, the compound should be practically invisibly thin, ideally down to 0.1mils thick.

Digikey carries UltraStick (DigiKey Corp. | Electronic Components Distributor | United States Home Page) for about $21 a stick. But that stick will do hundreds, if not thousands, of FETs.

[Edit] To compare to other compounds and thermal pads, the thermal resistance of UltraStick with a TO-264 FET was about 0.017C/W....very low. The very good silicone compound I tested (Wakefiled Type 126) was about 0.03C/W. But, as mentioned below, the darn oil evaporated at the high temps the FETs were operating at, 140C junction, and condensed onto everything nearby in the circuit. A real mess.

I don't recommend the silicone thermal compounds as the silicone oil separates out when heated and the compound goes dry over time...very bad. Also, at the very high temps this controller will see the oil can actually evaporate and condense on nearby (cooler) components...also very bad. Lastly, the oil can flow onto components when hot too if too much is used. The stuff is inexpensive, but not a great choice for some applications.

65C internal ambient temperature? Let's do some math.
That IXFK230N20T FET has a junction-to-sink thermal resistance of 0.24 degrees-C/W. That means the temperature of the MOSFET's junction (the important part) will be ((Wattage) x (0.24)) degrees-C higher than the temp of the heat sink directly behind the FET's "rear" metal plate. If the FET is dissipating 75W, then the junction is 18C hotter than the sink temp directly behind it.

Assuming that a very good sink has a sink-to-ambient thermal resistance of 0.5C/W, then the total thermal resistance from the junction of the FET to ambient air is 0.24 + 0.50 = -.74C/W. If that one FET was on its own sink and dissipating 75W, then it would be operating at (75W) x (0.74C/W) = 56.25C hotter than ambient. Assuming that 65C ambient you mentioned, that's 65C + 56C = 121C junction temperature. That's a hot FET! But, well within its rated max of 175C and even below the "high reliability" max temp rating, which is 80% of the rated max.

But, it shows how easy it is for a FET to be practically glowing from heat even though the ambient air is cool and even if the heat sink is cool. This also all assumes that the heat sink is both very flat and very smooth and that the MOSFET is mounted well to the sink and a good thermal compound is used.

It can be a bit dizzying at first but a few rough calculations can really help to make a unit a controller a lot more reliable or properly sized without wasting weight or space.

As you get closer to the final design (heat sink, # of FETs, average current, max current you want it to survive, etc.), I'd be happy to crunch a few numbers. Then some quick tests of the temperature of the FET's case in the prototype will tell you the FET's junction temperature and could give you a good idea of how you might want to limit the throttle duty cycle or timing in software (or modify the cooling system design)?
__________________
John M.
Owner
CamLight Systems

Last edited by CamLight; 02-28-2010 at 10:00 PM..
  Reply With Quote
The Following 3 Users Say Thank You to CamLight For This Useful Post:
deodeo (03-23-2012), MPaulHolmes (02-28-2010), Sunny2 (07-14-2014)