First time using these, because they are supposed to transfer vastly more heat to the heatsink than mica or silicon insulators. The question I have is... Do I still use a quality thermal grease like DuPont with these? Or do you mount these insulators dry? They are a porous crystalline solid dielectric but with very high heat transfer, so wondering if the grease helps or hurts. Debating what to do, not much on Google about which way to go.
https://www.aliexpress.com/item/1005001350971494.html?spm=a2g0o.9042311.0.0.dcd44c4dEiogOD
https://www.aliexpress.com/item/1005001350971494.html?spm=a2g0o.9042311.0.0.dcd44c4dEiogOD
You DO need grease with ALL solid/hard insulators, that includes Mylar, because you need to fill surface imperfections with other than air, and I mean transistor case and heat sink surface imperfections.
Only Silpads and similar can avoid grease but that because they are very flexible (rubber) and fill the voids.
Eye/mind blowing microscope images: anodized aluminum, the most common heatsink finish.
To our eyes: flat/polished:
Besides that, heatsinks are extruded , which creates its own set of imperfections, you can easily see "scratches" with your naked eye.
Only Silpads and similar can avoid grease but that because they are very flexible (rubber) and fill the voids.
Eye/mind blowing microscope images: anodized aluminum, the most common heatsink finish.
To our eyes: flat/polished:
Besides that, heatsinks are extruded , which creates its own set of imperfections, you can easily see "scratches" with your naked eye.
Last edited:
Thanks, makes sense. The porosity of these was throwing me off, in my mind thinking what if they are just big sponges absorbing the grease and maybe losing their characteristics if they get "clogged up". Overthinking it probably.
In addition, a thermal compound will help spread the strain, and reduce the risk of breakage with such stiff/brittle materials
Look into "Berquest Sil-pads" Not as god as mica and paste when new, but does not change. Very reliable and you know if it is there, it is correct. You design for the thermal resistance
If you can tolerate a "hot" heat sink, then graphite pads are the very best. They actually get better with time.
Good old paste ( used to be Be oxide,) is that you never know if the correct amount was used, the oil picks up dust and disrupts airflow and over time, thermal cycling pumps the conductive particles out from under it greatly degrading performance. It took a while as we had some poorly designed exceeding best practice boards that had to use paste, but we boned it from everything else. Along with star washers.
There are also the Kapton pads I have no experience with.
Elvee points to an important issue, but the wrong solution. Packages should be mounted with a compliant hardware stack. These days it is common to see a leaf-spring arrangement. In the bad old days, we went to stacks of bevel washers. I have seen small coil springs used. The absolute worst, common on cheap junk, was a pop rivet.
Lastly, hardware is for mounting, not current.
At least that was what I learned about 30 years ago when I was in Failure Analysis. I have seen nothing new to change any of it.
If you can tolerate a "hot" heat sink, then graphite pads are the very best. They actually get better with time.
Good old paste ( used to be Be oxide,) is that you never know if the correct amount was used, the oil picks up dust and disrupts airflow and over time, thermal cycling pumps the conductive particles out from under it greatly degrading performance. It took a while as we had some poorly designed exceeding best practice boards that had to use paste, but we boned it from everything else. Along with star washers.
There are also the Kapton pads I have no experience with.
Elvee points to an important issue, but the wrong solution. Packages should be mounted with a compliant hardware stack. These days it is common to see a leaf-spring arrangement. In the bad old days, we went to stacks of bevel washers. I have seen small coil springs used. The absolute worst, common on cheap junk, was a pop rivet.
Lastly, hardware is for mounting, not current.
At least that was what I learned about 30 years ago when I was in Failure Analysis. I have seen nothing new to change any of it.
I disagree. A competent engineer or DIYer does the math. Not a problem or very hard. This has been well understood for 70 years or more. Sorry, no hints as the last time I did it was 30 years ago, but a failure on the bond can produce HUGE differences at the die. Rule of thumb, Every 10 degrees C at the die is half life. Never run above 135 C on the die. 90 is a better target. That is die, not case. Cases need to be more like 35 C.
So when my heak sink hits 110°C, that's a bad thing? LOL
I mean the IC has thermal shutdown, right? 😀
I mean the IC has thermal shutdown, right? 😀
Aluminum nitride has thermal conductivity close to aluminum metal, and half of copper.
The advantage to using it directly under the semi package is that it spreads the heat before it reaches the heatsink.
(pardon my use of inches as units)
For example, if you use a transistor like the IR450, with a die size of 250 mils and a package with a bottom copper of 60 mils, the effective transfer area to the alumina would be a square of 250 + (2 times 60), or 370 mils by 370 mils. This would be the number used in the thermal calculation package to insulator through the grease. An insulator of aluminum nitride will expand that effective area also based on it's thickness. With a .040 inch (1mm) insulator, the effective area of thermal transfer at the heatsink will be a square of 370 + (2*40), or 450 mils square.
Using a very thin insulator will not increase the effective thermal transfer area much, so this semi would have a transfer path area to sink of .37 * .37, or .136 square inches.
Using a .040 inch alum nitride, the effective transfer area is .45*.45, or .202 square inches.
So the thermal drop in the grease directly under the package is the same in both structures, but the thermal drop at the heatsink interface will be .136/.202 for the alum nitride, 67% of the thermal drop that a thin insulator would provide.
In other words, the thicker the nitride used, the cooler your semiconductor will be because of lateral heat spread. I detailed this in Linear Audio volume 9, published April 1 2015.. (And no, it wasn't an April fools article..)😉
Use thermal grease to fill voids as mentioned previously by Jim Fahey. But do not go overboard, as the grease is not really good at thermal transfer but does assist where there is no solid contact due to surface finish.
jn
BTW...many vendors neglect the die size in their specification of thermal resistance but instead use a value based on the package size. This value is worse than what would be realized using the 45 degree spread model, but is safer for everybody in the long run even if not optimal. The reason is, TO packages are not well designed for retaining contact pressure directly under the die if you just use the screw to hold it in place. As TVRgeek mentioned, a leaf spring clamp can provide pressure directly over the die, and will retain pressure over a very long timeframe. I find I have to retorque simple screw hardware every 10 years or so due to thermal cycling loosening the hardware.
The advantage to using it directly under the semi package is that it spreads the heat before it reaches the heatsink.
(pardon my use of inches as units)
For example, if you use a transistor like the IR450, with a die size of 250 mils and a package with a bottom copper of 60 mils, the effective transfer area to the alumina would be a square of 250 + (2 times 60), or 370 mils by 370 mils. This would be the number used in the thermal calculation package to insulator through the grease. An insulator of aluminum nitride will expand that effective area also based on it's thickness. With a .040 inch (1mm) insulator, the effective area of thermal transfer at the heatsink will be a square of 370 + (2*40), or 450 mils square.
Using a very thin insulator will not increase the effective thermal transfer area much, so this semi would have a transfer path area to sink of .37 * .37, or .136 square inches.
Using a .040 inch alum nitride, the effective transfer area is .45*.45, or .202 square inches.
So the thermal drop in the grease directly under the package is the same in both structures, but the thermal drop at the heatsink interface will be .136/.202 for the alum nitride, 67% of the thermal drop that a thin insulator would provide.
In other words, the thicker the nitride used, the cooler your semiconductor will be because of lateral heat spread. I detailed this in Linear Audio volume 9, published April 1 2015.. (And no, it wasn't an April fools article..)😉
Use thermal grease to fill voids as mentioned previously by Jim Fahey. But do not go overboard, as the grease is not really good at thermal transfer but does assist where there is no solid contact due to surface finish.
jn
BTW...many vendors neglect the die size in their specification of thermal resistance but instead use a value based on the package size. This value is worse than what would be realized using the 45 degree spread model, but is safer for everybody in the long run even if not optimal. The reason is, TO packages are not well designed for retaining contact pressure directly under the die if you just use the screw to hold it in place. As TVRgeek mentioned, a leaf spring clamp can provide pressure directly over the die, and will retain pressure over a very long timeframe. I find I have to retorque simple screw hardware every 10 years or so due to thermal cycling loosening the hardware.
Last edited:
JN,
So what is the difference between a package sitting on an aluminum heat sink and one with a "spreader" under it? Seems to me you are just adding one more break in the transfer path.
We had an engineer who claimed polishing and anodizing the heat sink would give the needed electrical isolation and a better thermal path. Maybe if the transistor was flat! Anyway, it was a total failure. The root problem was they needed two devices, not one. It was a bad design.
We also used some be-ceramic insulators under big TO-3 packages. Even though they were very high conductivity, it meant additional rigid surfaces that had to have compound. A sil-pad had a better total performance. There may be newer films that perform better.
The biggest problem is insufficient de-rating. Modern CPUs really push these limits. Cray had a solution back in the 70's. Emphasis in "solution" as the Cray II was immersion cooled.
Darn you, now I have to send Jan more money. Just re-read 0 through 6. 🙂
So what is the difference between a package sitting on an aluminum heat sink and one with a "spreader" under it? Seems to me you are just adding one more break in the transfer path.
We had an engineer who claimed polishing and anodizing the heat sink would give the needed electrical isolation and a better thermal path. Maybe if the transistor was flat! Anyway, it was a total failure. The root problem was they needed two devices, not one. It was a bad design.
We also used some be-ceramic insulators under big TO-3 packages. Even though they were very high conductivity, it meant additional rigid surfaces that had to have compound. A sil-pad had a better total performance. There may be newer films that perform better.
The biggest problem is insufficient de-rating. Modern CPUs really push these limits. Cray had a solution back in the 70's. Emphasis in "solution" as the Cray II was immersion cooled.
Darn you, now I have to send Jan more money. Just re-read 0 through 6. 🙂
If you don't mind a hot sink, then not using a spreader is best, with a layer of graphene used as the thermal interface.
Silicone grease is a terrible conductor of heat, and if it is put on thick and squeezed out, it will produce horrible thermal numbers. if the total area conducting heat is small, the numbers go up, and a thin mica or silpad doesn't change the effective footprint much whereas a 40 mil thick nitride ceramic will increase the area by the addition of double the thickness to both dimensions of the effective area.
Surface roughness and flatness is also a big factor, if the case or sink have surface features in the 2 to 3 mil range (.002, .003 inch), the layer of grease will produce bad results.
Lapping the surfaces, then wetting them with grease using a single edge razor to skim it off will leave grease in the low points. Lapping BeO is obviously not allowed, and a TO-3 package also prevents lapping due to the pins. Lapping sinks will get expensive in a production enviro.
I also dealt with -220's and -247's which had non flat bottoms due to the stamping process used to make the leadframe/case bottom. For those, either lapping of the bottom is warranted, or the use of sil pads for the compliance.
In the article, it mentioned that my excel 45 degree spreadsheet was available for free download, but that was a long time ago, not sure if it is still available.
The article does give examples however.
John
Silicone grease is a terrible conductor of heat, and if it is put on thick and squeezed out, it will produce horrible thermal numbers. if the total area conducting heat is small, the numbers go up, and a thin mica or silpad doesn't change the effective footprint much whereas a 40 mil thick nitride ceramic will increase the area by the addition of double the thickness to both dimensions of the effective area.
Surface roughness and flatness is also a big factor, if the case or sink have surface features in the 2 to 3 mil range (.002, .003 inch), the layer of grease will produce bad results.
Lapping the surfaces, then wetting them with grease using a single edge razor to skim it off will leave grease in the low points. Lapping BeO is obviously not allowed, and a TO-3 package also prevents lapping due to the pins. Lapping sinks will get expensive in a production enviro.
I also dealt with -220's and -247's which had non flat bottoms due to the stamping process used to make the leadframe/case bottom. For those, either lapping of the bottom is warranted, or the use of sil pads for the compliance.
In the article, it mentioned that my excel 45 degree spreadsheet was available for free download, but that was a long time ago, not sure if it is still available.
The article does give examples however.
John
My main objection was the word "vastly".
As for doing the math: that rarely happens. Here, and even in commercial planning.
Are you this guy? You made me remember that once there was a Philips K9 TV set that used such a ceramic plate to spread the heat from its horizontal deflection drive transistor to an aluminium heatsink and now i get why.Never thought that deep about horizontal spread ...I saw a few computer cooling solution compared on youtube and the most interesting thing was that thermal grizzly kryonaut based on some sort of aluminium(they say it's just nano particles of aluminium) outperformed arctic silver by far.Then the motive seem to be the higher surface available for heat transfer in all directions and probably the horizontal spread is the key there.Why is that I can't find commercial brands selling aluminium nitride based paste, only solid fillers?
Attachments
Last edited:
I used these for two F5 Turbo. Much better with a thin layer of grease but in my case it was somewhat tied to the rough finish on the heatsinks.
Sitting there, so full of myself....then the photographer said to me...lean forward so you don't look so fat.
ffffsssss.... air rushing out of a balloon.. sigh...
jn
It is a small test pattern for a longer coil. It is a serpentine pattern that creates an octupole magnet.
I start with a small test pattern, as the process has almost 70 variables which require optimization for new patterns.
It is exactly like flying in a plane.... 95% boredom, with 5% gut wrenching terror. In my case, the straights are boring, the ends are the problem areas, so there is no need to waste wire in the straights for tests.
The poster behind me is a false color image of a quadrupole, 4 poles 2 north, 2 south. It also has shield coils to cancel external fields, I turned off the shields on the left of the pic, and on in the right side.
John
It's not much power going through old CRT TV horizontal deflection transistor (one often see these H transistors mounted on a flimsy piece of aluminium sheet), I think it's more a matter of high voltage, Mica isn't reliable material for high voltage due to micro cracks (I have seen mica insulators with carbon build-up in between its fine structural layers which shorted the transistors backside to the heatsink) so that may be a reason why some TV manufacturer used ceramic, but many horizontal transistors came in fully isolated plastic packages (metal tab inside the plastic mold) meaning the backside was insulated with a thin layer of plastic so these transistors didn't require any additional isolation/insulation, some H transistors have been made in both 'normal' packaging and fully isolated, and the power rating goes down quite a lot only due to the poor thermal conduction properties of fully isolate packages, in the end it simplified and reduced manufacturing cost assembling a fully isolated transistor over one that needs additional parts.
Last edited:
Some great information in this thread! If you use Sil-pads, buy the good ones. They make many types and the good ones, though expensive, are substantially better than the cheap ones.