thermal interface materials for transistors

[Toster]

I'am thinking about thermal interface materials for current stage in my diy class AB power amplifier.

Output transistors are TO-264.

With Bergquist Sil-Pad 2000 for example I can get 3.5 W/m-K of thermal conductivity while Aluminium Oxide provides arround 20 W/m-K and Aluminium Nitride 180 W/m-K.

I got quote for Aluminium Nitride insulators custom made as follows:
Quantity: 50pcs USD2.00/pc
Quantity: 200pcs USD1.70/pc
Quantity: 500pcs USD1.50/pc
Quantity: 1000pcs USD1.40/pc


How fragile are Aluminium Oxide and Aluminium Nitride insulators?
We are talking about 1mm thickness.

Anyone have experience with these materials?

What do you think about using anything better than Sil-Pads?

Mario

 

[BesPav]

Anyone have experience with these materials?

Hi, Mario!

I'm using custom made AlN substrates, they are cut for 4 TO247 cases.
The most critical question is a heatsink surface polishing and take on force.
With using some kind of AAVID MaxClip clips you can be sure in proper and constant force.

With bad heatsink surface improper force or very tight screw fastening can break substrate while device become hot. But even with cracks those substrates are perfect in terms of conducting heat from device to heatsink.

What do you think about using anything better than Sil-Pads?

You better to simulate your output stage devices heat dissipation and your case heatsink capability, then you can calculate needed heat conduction speed or equivalent heat resistance in terms of ouptup transistor crystals safe temperature rise or SOA.

Please, show your OPS scheme with supply voltage, load crossover scheme with rated speakers impedance and estimated average power.
Then we will try to estimate something and you could decide what to use.


Next thing to know - is a parasitic capacitance between transistors substrate and heatsink.
With Silpads you will use thinner layer, so heat transfer would be mostly the same, but with Al2O3 or AlN you can use thicker layer with same heat transfer, but much less parasitic capacitance.

 

[Toster]

Hi, BesPav!

Thanks for some insight into ceramic substrate insulators.

I am building Jens Rasmussen Leach Amp Clone.
https://web.archive.org/web/2007012..._Clone/10_TRANSISTOR/Leach_ver_5.6.10_sch.pdf

But version with 6 pairs of mjl1302a/3281a output transistors.

I will be using +-72V supply voltage. Each channel is mounted on 0.18C/W heatsink.
Low sensitivity 8ohm loudspeakers for home listening, but ocassionaly 4ohm load would be hooked up and high power delivered on party nights in the garden, probably side ventilators blowing.

For Al2O3 or AlN insulators what thickness are usually used?

 

[PH104]

Haven't used the Aluminum-based insulators but have used BeO insulators. Make sure you tighten evenly to torque specifications.

 

[BesPav]

I am building Jens Rasmussen Leach Amp Clone.
https://web.archive.org/web/2007012..._Clone/10_TRANSISTOR/Leach_ver_5.6.10_sch.pdf

But version with 6 pairs of mjl1302a/3281a output transistors.

Not bad, but it can be better.
Just replace yellow marked resistors with current mirrors, blue marked resistors with a good current source and tune frequency correction accordingly. This will drop THD by an order or slightly more.

I will be using +-72V supply voltage. Each channel is mounted on 0.18C/W heatsink.

Ok, let's estimate max output voltage as +-60-63 Volts peak-to-peak and ~40 V RMS.
This is due to triple EF OPS, it needs some headroom to be saturated.

Low sensitivity 8ohm loudspeakers for home listening, but ocassionaly 4ohm load would be hooked up and high power delivered on party nights in the garden, probably side ventilators blowing.

The main thing, mostly undiscussed, is a reactive impedance of the speakers.
First, they have instrinsic source of the electric-drivind force - moving mass + coil + magnet.
Second, voltage and current aren't syncronized, there are a phase lag, so with 0 V at the output there can be sensible current.

I will simulate your case with some simplification. But you are free to deep inside in that way.

Your load can be estimated with a simple circuit like this:

This is not a loudspeaker model, but it have a resonant peak and needed phase lag ahead it. You can use different frequencies between 120 and 200 Hz to simulating and compare results.

First of all let's put OPS devices, bias, stimulus and supplies:

Feel free to try, there is nothing complicated.

Let's check output:

Nice +-60 V sinewave, we are going right, aren't we?

Let's add transistors currents:

Oh, they aren't inphase with voltage! There are good phase lag between output voltage and current. That's the point usually undisclosed!

Now, we can put voltages and currents in background and estimate devices dissipation:

Zoom in:

So, dissipation maximum at near 15 V output with a current around 4 Amps.
Our OPS devices will be at 57 Volts and must conduct ~4 A.

Let's see SOA graph:

Looks like for 8 Ohm load we can use one transistor and add second for be safe and sure.


Now, 4 Ohm. Set R3 to 3,2 Ohm as Re of the speaker:

From the SOA perspective this is 8 A at 70 V, and you need at least three OPS devices pairs to be inside SOA.
Five pairs would provide bulletproof OPS in terms of SOA.


Next you can average heat to dissipate and estimate your heatsink temperature rise. Based on estimated rise you can choose to use active cooling.

For Al2O3 or AlN insulators what thickness are usually used?

Usually 1 mm. This is thick enough to be strong and thin enough to conduct heat well.
0,5 mm are very sensitive but wery good conducts heat.
2 mm are very strong, but have less thermal conductivity and used mostly with high demands to lower parasitic capacitance to heatsink.

 

[sajti]

Just replace yellow marked resistors with current mirrors,

Don't do that! It's a common mistake, which was problem many times. Nothing will define the VAS current, if You replace that 4 resistor with current mirror.

Sajti

 

[BesPav]

Nothing will define the VAS current

Hi, Sajti!
Thank you for your comment.
May be, i need to simulate overall circuitry.
But from first sight, i suppose, that R35 and its complement can do the trick:

Aren't they intrinsic VAS bias?

 

[sajti]

Hi, Sajti!
Thank you for your comment.
May be, i need to simulate overall circuitry.
But from first sight, i suppose, that R35 and its complement can do the trick:

Aren't they intrinsic VAS bias?

Hi,

R35 is there to protect the VAS transistor if the protection working. Your solution will work in the simulator, for sure. But there are no ideally same transistors in real life.

Sajti