Optimal location for devices on heatsink

[philibuster]

Hi,
I was thinking earlier, and somewhere I read that the devices should go near the geometrical center of the sink. As I thought about this, I realized that if you put the devices near the top, under natural convection it would be a sort of counterflow heat exchanger, with the bottom being the coolest and the top being the hottest. If the devices were mounted near the center, then the air flowing past the top would not take more heat from the sink because it has already passed the hottest center section. Is this not correct? Thanks for the help.

[pinkmouse]

Heat transfer is much greater between the device and the heatsink and the heatsink and the air, so devices should be mounted centrally.

The difference in air temperature between the top and bottom of the sink will be a few degrees at most, wheras the temperature between the heatsink and the air will be in the tens of degrees. The laws of thermodynamics show that conduction is relative to difference in temperature, so you want to get as much heat into the sink as possible, and to do that, you mount the devices in the middle. The effect of convection is purely to keep the air moving over the heatsink, the few degree temperature difference between air top and bottom will not noticably affect the amount of heat that the air takes from the sink.

[pinkmouse]

However, if you have a non laminar flow, such as from a fan blowing air onto the heatsink, it gets a little more complicated...

[AndrewT]

Hi,
if you have a single device put it in the middle across the width and about 40% to 45% up from the bottom.
For two or three devices form a row across the width spacing them out evenly with the edge gap about 40% to 60% of the centre to centre gaps.
For four devices the options increase. Two columns of two need each about 25% in from each side. The upper row should be about 60% to 70% up from the bottom and the lower row should be about 25% to 30% up from the bottom.

The reason for the slight downwards location is to make use of the cold air coming in from the bottom having most effect on the hottest part of the sink. The upper row have more sink area dedicated to them due to having pre warmed air coming from the bottom passing over their part of the sink.

A lot more data is available in a paper by Wakefield on their site.

[Geoff]

You can model the effect of different placings for the transistor on the heatsink at this site:

http://www.frigprim.com/online/natconv_heatsink.html

Input the dimensions of your heatsink and the expected transistor dissipation then move the transistor around and see what happens to the temperature gradients within the heatsink. Multiple heat sources can also be modelled.

[philibuster]

Thanks for the replies. I kind of forgot that aluminum conducts heat like a madman. Maybe if the sink were made out of steel it would work like I thought. Thanks again.
-phil

[darkfenriz]

Best to place power devices close to the middle but relatively remote from each other and temperature sensing (a thermostor or Vbe multiplier bjt) beetwen them.

[amplifierguru]

Hi Darkfenriz,

something like this - http://members.dodo.com.au/~gregball/guru_004.htm

The little ones regulate input stage current sources to track the Vgs of the MOSFETs. I would mount a compact module such as this (4" x 2") slightly below centre on a large heatsink and ventilated below on the inside of the chassis.

Cheers,
greg

[darkfenriz]

Hey Guru, your camera isn't that bad
nice photo of nice work indeed.
How do you mount to-92 devices on heatsink? Do you glue them? And for your specific killer design: do you use small power hexhets (irfd) in CCS to thermally stabilize it?

[amplifierguru]

Hi Darkfenriz,

Thanks. I'm REALLY pleased with how it's turned out.

The TO92's are actually diode connected BC546B/556B and they are simply bent forward to spring load their flat face against the heatsink when the MOSFETs are mounted. These 'diode' sensors are connected in the emitter sense leg on the 2BJT CCs on the input stages such that they have a comparable tracking rate for the MOSFET bias.

It works a treat.

Cheerrs,
Greg