What the best way to mount HIGH voltage semiconductors to grounded coolers?

[001]

What the best way to mount HIGH voltage semiconductors to grounded coolers (radiators)?
For example TO-247 transistor at 600VDC?

[NiHaoMike]

Use insulating thermal pads.

[Ian.M]

There are several things you have to get just right for long term reliability.

Burr free and flat heatsink surface.

Spring clip across the package so you don't have to worry about creepage distance from the metal pad in the package base to the mounting bolt or sleeve the bolt with an insulator.  The clip provides spring pressure on the center of the package that does not vary significantly as its thermally cycled.

Mica insulating sheet, which if a good grade of natural Mica is typically more reliable long term than a silpad.

Light smear of heatsink compound on both sides of the mica sheet.

If the Mica thickness is appropriate, the above is good for over a kilovolt for many years.  An appropriately rated glass-fibre reinforced silpad (no heatsink compound needed) is better for automated assembly and will be good  for a few years, but I wouldn't bet on its reliability a decade later as  silpads can fail by deforming under pressure as they age then pin-holing at or near a package corner.

[Gyro]

For higher voltages (probably 1kV+) you can get solid Aluminium Oxide washers. They have surprisingly good thermal performance but you need to have some tension control on the screw (crinkle washers etc) to avoid the risk of cracking them.

eg. https://uk.rs-online.com/web/c/?sra=oss&r=t&searchTerm=aluminium+oxide+washer

[kony]

Alumina is shit compared to aluminium nitride ceramics - that has comparable heat conductivity to the alloy heatsink is extruded from. So if you are running out of the thermal resistance budget, this is a very good fix.

[001]

aluminium nitride ceramics

how it look? is it expensive? I never seen one

[kony]

It depends, I had seen some TO220/247 spacers from chineese vendors, other option is to get them custom made for the purpose from local ceramic hybrids manufacturer. Regular ceramic alike material with ivory to grey color. It is not really a comodity solution, but when you are after utilising most of the power handling capacity of the semiconductor device yet want to keep it with safe galvanic insulation, this considerably outperforms all other options.

[NiHaoMike]

Something else that can be done, if the thermal conductivity of a thermal pad is not good enough, is to mount the transistors or whatever on a block of aluminum or copper, then use a larger thermal pad to couple that to the heatsink.

[Wolfram]

Standard 96 % alumina has a thermal conductivity of around 25 W/mk, which results in an added 0.25 k/w thermal resistance per millimeter of thickness, given a TO-247 metal pad area of 160 mm^2, while the number for aluminium nitride is less than 0.04 k/w in the same scenario. The problem is that it's very easy to add ten times this thermal resistance just from non-optimal thermal compound application. So to fully benefit from AlN insulating washers, it takes a lot of care in surface preparation, thermal compound selection, thermal compound application and optimization of the level and distribution of the clamping force. For cases where you need thicker insulating pads, the advantage of AlN is more dramatic, of course.

[floobydust]

Surprised to see Wolfspeed using: Kapton Thermal Interface 22mm x 29mm , Fisher Elektronik, KAP 1 P
"polyimide-carrier foil with silicone-free phase changing thermal conductive layer completely coated on both sides"
0.15 K/W [at 1 inch2; = 6.45 cm2; = TO 3 (KAP 3)]
0.45 W/m·K (substrate)
material thickness 0.077mm (substrate 0.05mm)
dielectric strength 7.8 kV

[Wolfram]

Those kapton foils are nothing but trouble in high voltage applications, especially in cases with high dv/dt. I optimistically used some to isolate some SiC MOSFETs and diodes in a 750 V buck converter, and they failed after a short time in operation. Further testing gave consistent failures of this material. The manufacturer of this material (Aavid in this case) blamed the surface finish on our heatsinks, but they don't actually publish any information on the recommended surface finish for kapton foil insulators. Swapping them with alumina pads solved the problem completely.

[floobydust]

I would look at what the automotive industry is using.
Anyone know what Tesla Motors used here in older Model S?  I think it's a modified TO-247 IGBT package.  I couldn't find pics of the other side.

[jbb]

+1 for spring clips. Having a hole in you insulator through which you put a metal screw is really bad for creepage & clearance. Also some TO-220 devices very thin metal tabs which deform (causing increased thermal resistance) when the screws are tightened.

A Sil-Pad style tends to add about 1 deg C / W of thermal resistance which must be taken into account.

Regarding the Tesla inverter; those look like a modified TO-247. By not having the hole you can get more silicon into the same package. IXYS make some standard parts like that, but at Tesla scale they could be custom devices.

[wraper]

Mica insulating sheet, which if a good grade of natural Mica is typically more reliable long term than a silpad.

Light smear of heatsink compound on both sides of the mica sheet.

Silpad should be more reliable because it's thicker and won't crack. However mica + thermal grease have better thermal conductivity.

[Ian.M]

Mica  only cracks if grossly abused, and is very stable under pressure.  Its also virtually immune to ageing as its millions of years old to start with!   If you start with a good clear piece without any signs of delamination or cracking, and don't mangle it during installation, or have any burrs on the heatsink or device surfaces, its very reliable and fully reusable.

Silpads however tend to degrade with ageing and high temperatures, and if they've been in use for any significant length of time should *always* be replaced if disturbed.   Their only advantage is convenience on the assembly line for low cost products, as their resilience makes the package base and heatsink flatness, and initial clamping pressure less critical, and if you specify self-adhesive ones, they can make automated assembly much easier.

[wraper]

Mica  only cracks if grossly abused, and is very stable under pressure.  Its also virtually immune to ageing as its millions of years old to start with!   If you start with a good clear piece without any signs of delamination or cracking, and don't mangle it during installation, or have any burrs on the heatsink or device surfaces, its very reliable and fully reusable.

If all of those conditions are met, which often are not. FWIW I've seen plenty of MICA insulators shoot through.

[Ian.M]

Certainly, it only takes a small surface defect or bit of debris for it to suffer crush damage during assembly.  Also there is the problem of obtaining high enough grade mica sheet for HV work - you absolutely cant tolerate significant non-mica inclusions, and I'd be very wary of built-up (composite) mica sheets, so in large enough pieces to provide sufficient creepage distance at the edges,  its likely to be expensive compared to other solutions and any attempt  made to reduce costs either or materials or by using less skilled labour has a high risk of introducing defects that can  cause it to breakdown under electrical or mechanical stress, either rapidly or much later when the heatsink compound has deteriorated enough for it to cease to entirely fill the defect.

[SeanB]

Having taken apart a few power blocks, they all use a large aluminia insulator, with the power devices soldered to the ceramic on silver plated pads, and the bottom side soldered to a large copper heat spreader. Good for 2kV of isolation, and very reliable.

You probably will want to get large alumina plates ( at least a lot larger than the devices for creepage clearance), and use a spring clip or a top clamp with Belleville washers to provide the correct spring clamping to the devices.  Then appropriate thermal compound and the manufacturer of the devices recommended clamping force per device. Note with the aluminia isolator you do need to have a machined burr free heatsink mounting surface, as extruded is not going to work, it has to be machined to be as close tolerance as the ground aluminia surface to prevent cracking. Remember that the clamp as well needs to be as rigid as the heatsink, so as to apply even pressure, so you will likely either have a bar across the top ( probably with a kapton tape to provide higher isolation) and a clamp assembly both ends, or a longer bar with multiple devices being clamped by it. When doing multiple devices with a single clamp make sure they are all from the same lot number, as they might have slight differences in the thickness of the top epoxy, and tolerances add up.

https://en.wikipedia.org/wiki/Belleville_washer
Basic info, use with a nylock nut to provide the anti loosen requirement, the washer stack provides the even clamping force with changes in operating temperature.

http://www.bellevillesprings.com/belleville-washers.html
UK company, I have bought from them before, good service even international, and very friendly as well. Good quality items.

Worst thing are power SCR pucks, they have a special heatsink assembly that you must use to get the lifetime, and generally no warranty if you do not use it either or use incorrectly. Those things do not let magic smoke out, they instead turn into shrapnel powder with a loud bang, then blow the expensive FF semiconductor fuse.

[coppercone2]

Something else that can be done, if the thermal conductivity of a thermal pad is not good enough, is to mount the transistors or whatever on a block of aluminum or copper, then use a larger thermal pad to couple that to the heatsink.

I really like this idea because the peak pressure on the dielectirc material is much smaller since it has much more surface area to distribute it especially if its something like a TO-220 screw. If its a heat sink clip or a press bar it might not make as big a difference but it should still be better because you can use nice grinded or lapped parts that really make a good fit rather then trying to get equal stress from a very cheap part.  I think I will do this from now on.

[wraper]

Something else that can be done, if the thermal conductivity of a thermal pad is not good enough, is to mount the transistors or whatever on a block of aluminum or copper, then use a larger thermal pad to couple that to the heatsink.

Example of this in old Soviet amplifier.

[coppercone2]

I am pretty sure though that you want to use a tensioning bar or one of those special heat sink clips to tension the package rather then a screw for optimal heat conduction

[Wolfram]

I would look at what the automotive industry is using.
Anyone know what Tesla Motors used here in older Model S?  I think it's a modified TO-247 IGBT package.  I couldn't find pics of the other side.

That's a Roadster inverter phase leg. This inverter used some polymer based insulation between the IGBTs and the water cooling blocks, and these are failing in large numbers: https://teslamotorsclub.com/tmc/threads/roadster-pem-failure.88028/ . The Model S uses 0.5 mm alumina pads and I haven't heard of them failing. I'm not aware of any pictures of them as the whole phase leg assembly is potted in silicone gel and removing the transistors is a pretty destructive affair.

[Zero999]

Something else that can be done, if the thermal conductivity of a thermal pad is not good enough, is to mount the transistors or whatever on a block of aluminum or copper, then use a larger thermal pad to couple that to the heatsink.

Yes, that works very well. In the past I soldered an LM317 to a piece of copper bar which I then fixed to the case with an electrical insulator sandwiched in between.

I was still learning when I built this, so don't give me too much grief about the messy construction.


I repeated the project but made it dual rail, tracking and fixed the LM317 and LM327 with screws to copper blocks, again mounted to the case with electrical insulators. It's a little neater, than the previous project, but could've been better.


Obviously for 600V, the above examples are inadequate. The insulators need to be larger than the copper blocks to ensure adequate creepage distances.

[coppercone2]

I don't think solder is permitted for earth ground connections officially, because of mating cycles and flex of the ground pin if the cord is fucked up. The solder lug on the chassis is a bit more acceptable because that point is not being flexed, but you obviously still need excellent strain relief so the motion does not transfer through the ground wire from the plug (which is less if you use multi strand wire).

and the earth solder joint in the lower picture looks down right suspicious. I think you should at least make a half hook through the hole to solder it with good clearance to the pin if you insist on soldering it, it looks like its bearly  attached. Did you clean and flux the joint (perhaps with a brass brush?). The wetting on your solder joints looks horrid but I am not sure if its a camera thing.


Also, where are the fuses at?

cant hurt to put the levitating rectifier on a bracket either.

Also would recommend a softer dielectirc between the torroid and the tensioning washer.

also are those earth ground lugs star washered?

This one is unnecessary but there is a kind of cheap trick to wire-pcb junctions, where you drill a hole away from any conductors and zip-tie the wire to the PCB through the holes or between the hole and the edge of the PCB. It works very well I had a hobby project that got yanked very hard and it held up fine without damage to the PCB.

[Gyro]

Something else that can be done, if the thermal conductivity of a thermal pad is not good enough, is to mount the transistors or whatever on a block of aluminum or copper, then use a larger thermal pad to couple that to the heatsink.

Yes, that works very well. In the past I soldered an LM317 to a piece of copper bar which I then fixed to the case with an electrical insulator sandwiched in between.

I was still learning when I built this, so don't give me too much grief about the messy construction.


I repeated the project but made it dual rail, tracking and fixed the LM317 and LM327 with screws to copper blocks, again mounted to the case with electrical insulators. It's a little neater, than the previous project, but could've been better.


Obviously for 600V, the above examples are inadequate. The insulators need to be larger than the copper blocks to ensure adequate creepage distances.

"I was still learning when I built this, so don't give me too much grief about the messy construction."


Wouldn't dream of it, but I think you owe it to the forum to post the picture in the 'ugly-but-flawless-project' thread for posterity - if only for the 'securely bolt the lightest components and glue the heavy ones' methodology. Sorry, I really couldn't resist 

https://www.eevblog.com/forum/chat/show-us-your-ugly-but-flawless-project/