Improving thermal interface of transistor to heatsink.

[Circlotron]

One of the perennial problems of the amplifier builder is keeping those power semiconductors cool. All the heat goes out the flat face of the power device to the heatsink., and it is a bottleneck that if it isn't performing properly then all the heatsink in the world isn't going to help you. I myself am a believer of directly connecting the power semi to heatsink wherever possible, and running live heatsinks as a consequence. This gets rid of the insulating washer and improves things quite a bit. Another must is to have a flat heatsink surface so the heatsink paste can be squeezed into the thinnest layer possible. Then just today I was looking at the back of some TO-247 shape devices and some were shiny and some were matte finish. I thought the shiny ones would be almost perfectly flat, but when I looked in the reflection they were far from it. I couldn't really tell with the matte ones.

Then I got this idea: I got some 1200 grade wet and dry sandpaper - really fine stuff - and laid it on a good flat surface and then ran water from a tap continually across it. Then I got one of my trusty hexfets and rubbed the rear surface across the sandpaper. Straight away I could see just how un-flat it's surface was; the high and low spots were obvious. I continued for about a minute and then used somewhat less pressure and it really began to smooth out and felt like it was gliding across oil. Important to have plenty of water flowing. Maybe underwater would be a good method. Anyway, pulled the fet out and had a look and it had an almost perfect mirror finish! Copper colour too. Really cool it looked, it did.

I did a similar thing with the copper block that the fet is going to go onto and now the pair have a noticeable suction between them when you draw them apart. Any thermal paste I now put between them will be squished into the thinnest layer imagineable and therefore offer the least thermal resistance.

That's the good thing about diy - there are almost endless tweaks to do that commercial manufacturers for the most part don't bother with because they only want to satisfy the masses.

GP.

 

[AudioFreak]

Lapping

Yes, this works well. It's called lapping and is done frequently by serious / extreme, computer overclockers / modders.

 

[mrfeedback]

The 6 P's

Graham, I use an inch thick block of glass as the surface backing the wet and dry abrasive paper.
I also counter-sink the mounting screw holes before surface grinding the heatsink.
Even with mica insulating washers, carefull preparation (reparation) of all heatsink surfaces pays big dividends for long term reliability certainly.
I agree that most device mounting surfaces are poorly finished and this simple step improves thermal coupling markedly.
I always do these steps on customer repairs and this helps to eliminate workshop warranty issues.
I also find devices where the epoxy encapsulation is lower (slightlty) and actually lifts the device off the heatsink.
I have repaired a Kenwood reciever (80'smodel) that had two isolated heatsinks (one at B+, one at B-), and built several ETI-480 kit modules with isolated heatsinks where the collectors are connected at the output point and the output transistors directly mounted.
An improvement may have been to directly mount them to a copper busbar mounted to the heatsink.
I think heatsinks isolated from the chassis is a perfectly valid technique, although seldom used.

Eric.

 

[Circlotron]

Yesterday I did a little experiment where I was pushing 100 watts of heat into a 3" x 1/4 " x 2 foot copper bussbar with a T0-247max (no hole), and I was was using Thermstrate TC which is a crash hot thermal compound and no washer and I was getting about 22 deg C drop across the junction with unprepared surfaces. I'm really keen to see what I get now.

Also I just read about another cooling method - imagine a square pipe with a hole in the side and cold water flowing through it. Instead of putting your thumb over the hole like a certain famous Dutch boy, cover the hole with a power semi with a gasket around the edge. The cold water makes direct contact with the back of the semi so there is one less thermal junction. You could probably even hold the case at the mythical 25 deg C quoted in the data sheets and actually utilise the full rated dissipation. Wow! That'd be a first.

GP.

 

[Steve Eddy]

Any thermal paste I now put between them will be squished into the thinnest layer imagineable and therefore offer the least thermal resistance.

The purpose of the grease isn't to form a layer between the device and the heatsink. While it has fairly high thermal conductivity, it's not nearly as good as that of metal.

The purpose of the grease is only to fill any voids between the two surfaces where the thermal conductivity of the grease is far greater than the air which would otherwise occupy the voids.

se

 

[mrfeedback]

Beer Class Amplifier

What about at dry-ice temp ?.
If you were to cut the bottom panel out of an esky and replace it with the heatsink panel and fill it with dry-ice and beer, you could have a real winner.

 

[Circlotron]

Class Ale, Class Beer, even Swigging amp

I just repeated the thermal experiment with both fet and heatsink polished. At an ambient of 26.7 deg C the heatsink @ 100w dissipation was 61.4 deg C and the fet tab was 69.4, giving a junction drop of 8 deg for 100w, an improvement of about a factor of 3.

Heatsink = 61.4 deg C, fet = 69.4 deg C
Rth of junction = .08 deg C/W,
Heatsink temp rise @ 12v fan = 34.7 deg C
Rth of heatsink = 0.347 deg C/W @ 12v fan.

Heatsink = 72.5 deg C, fet = 79.0 deg C
Rth of junction = .065 deg C/W (that's weird, gets better as it gets hotter!)
Heatsink temp rise @ 5v fan = 45.8 deg C
Rth of heatsink = 0.458 deg C/W with fan running at 5v.

GP.

 

[mrfeedback]

Class Drunk Amplifier

Graham, those are significant changes.
I always knew there was a reason that I do this heatsink lapping to customers amplifiers.

Eric.

BTW - How are you making the measurements ? (Looking for reasons for the change in Rth).

 

[mrfeedback]

Class Frigid Amplifiers

Has anybody done experiments with amps regulated at room temp, or chilled and noted sonics ?. (Sort of on topic ?).

Eric.

 

[Circlotron]

I did the experiment by connecting a 27 volt zener diode between gate and drain (and 10k resistor between gate and source) of a chunky 450 watt mosfet and applied 33 volts dc at 3 amps current limited. The fet works as a big zener. I used a couple of thermocouples for the actual measurement along with a Fluke thermocouple meter that has been calibrated within the last 12 months. I measured the fet tab temperature on the drain lead where it exits the plastic package. I let the fet tab get up to 150 deg C so the silicon would be at about 178 deg C (wasn't that naughty!) and guess what? The heatsink was only 2.5 degrees C less than the tab! Not bad for 100 watts passing through it. This is the paste stuff I used http://www.powerdevices.com/thermtc.htm I read the other day about some thermal compound meant for CPU heatsink that was loaded with silver particles. I couldn't find any specs on it but it sounded good.

GP.

 

[AudioFreak]

The PC thermal paste you refer to is Arctic Silver but i'd stay clear of it because it exibts electrically capacitive qualities.

 

[peranders]

Circlotron, did you have isolation between the fet and the heatsink?

 

[Circlotron]

Semiconductor teabags and rowboats.

AudioFreak, yes, Arctic Silver, that's the one!
peranders, no, just metal to metal with a bit of slops in between.

Just now I moved one of the thermocouples to the centre of the tab and got a cup of water at 10 deg C and immersed the fet, and 100 watts dissipation as before. In still water the tab gets to 48 deg C above water temp, but with the fet on the end of the wire getting jiggled up and down like a teabag, and more particularly side to side like a rowboat oar, the tab only gets to 9 deg C above the water! Some of this may actually be the water cooling the thermocouple directly, I cant really be sure, but when the fet was stationary some steam bubbles would start to form on it, reducing the active area. Maybe the mythical 25 deg C device temperature for full rated dissipation is within reach?

GP.