cooling SMC triacs in D2PAK?

[max_torque]

I'm looking at a device that drives some single phase AC loads at various currents between 0.5A and 16A, and i plan to use Triacs as the load control devices.

Most triacs come in through hole packages, like the classic TO-220, which then need specific heat sinking. I had an idea to use SMC packages, such as d2pak, and arrange the triacs as a low side switch and to have a common neutral for all the devices, connected to the A2 terminal which is the tab of that package for most triacs.  With suitable via fields, i plan to sink the heat out the tab of the d2pak into a large top copper layer, and then through the pcb on the vias, and into a large copper pour on the back side, and then finally into a large(ish) heat sink screwed to the back of the pcb, probably using a thin thermal gasket for compliance. In this manner i heat sink all my devices together, and the heat sink is at Neutral potential, which would normally be close to Earth potential.   The entire device is inside a "finger proof" enclosure, so the heat sink cannot be touched, but it's better to be at Neutral potential, as that avoids issues with electrically isolation etc.

Being a low side switch, the loads are of course powered "all the time" on the high side, ie connected to Live, however the output sockets i am using are finger safe, and shuttered (and inside the overally unit, and only accessable via tools) and i have a non dynamically switching relay as a high side "safety" switch, which can turn off all the Hide side feeds to the devices (This is to prevent the process the loads are driving from going out of control should one of the low side triacs fail shorted.


Can anyone see a flaw in the plan?   

[MagicSmoker]

Without all the thermal vias and external heatsink the thermal resistance from junction to ambient for a single D2Pak device is about 40-45C/W so this may or may not fly depending on the total losses and how well your additional heatsinking measures work. Likely you'll have to actually build and test one to find out.

Otherwise, "low-side" switching with a triac is fine.

EDIT - I just noticed the 16A max - that's almost certainly not going to be doable unless you can get R_theta[j-a] down to less than 5C/W.

[T3sl4co1l]

Figure about 5W tops for a D2PAK with reasonably thorough heatsinking.  Maybe 2W for not much effort (a few in^2 copper pours, 2oz).  Maybe 10-20W with heroic effort (heavy copper, filled vias, low ambient temp e.g. water cooling).  So, whatever that comes out to, from the datasheet Pd spec, or whatever you can fill in for it with (Irms * Vf max? Don't forget switching losses too, if it's phase controlling).

Thermal vias with an insulator pad on the backside is a good way to do it, I've done it many times myself.  You can even fill the vias with solder, but take care that they don't end up proud of the bottom surface, or use a thick enough pad to clear them.

If you can ground the heatsink, that's even better and alleviates some of the finger safety precautions of the design.  If the mechanicals are fine as-is, obviously you don't have anything to gain this way, but just FYI.

Tim

[max_torque]

Triac is T3035H

https://www.st.com/resource/en/datasheet/t3035h.pdf

nominal max rating is 30A


the datasheet number (page 3) (caution, number might be cobblers as it's a datasheet number.....) is 0.8 degC per watt junction to case.

16A is worst case dynamic, steady state load is more like 12A (load is 3kW resistive heater, so some "cold inrush" does occur for a short period), again according to the datasheet that's a worst case heating of 12 watts (page 4, figure 1)

Max ambient is 60 degC, so to stay below 150 degC junction (Tmax) i need an overall thermal impedance of 7.5 degC/watt.

The pcb this is all mounted too is pretty large (330mm x 120mm) and the back side is pretty much all available for heat sinking.

Sounds do able to me?  (The enclosure does include a fan and is well vented as well)

[MagicSmoker]

Okay, so Vtm for this triac is 1.55V max (I assumed 1.4, which is a "typical" value, and an ambient of 50C) so for a 12A load you would have to get rid of 16.8 to 18.6W of heat, and if you went ahead with a 90C allowed rise (really not advisable) you would still need to get R_theta[j-a] to 4.8C/W or less... This just doesn't seem realistic to me.

[langwadt]

Triac is T3035H

https://www.st.com/resource/en/datasheet/t3035h.pdf

nominal max rating is 30A


the datasheet number (page 3) (caution, number might be cobblers as it's a datasheet number.....) is 0.8 degC per watt junction to case.

16A is worst case dynamic, steady state load is more like 12A (load is 3kW resistive heater, so some "cold inrush" does occur for a short period), again according to the datasheet that's a worst case heating of 12 watts (page 4, figure 1)

Max ambient is 60 degC, so to stay below 150 degC junction (Tmax) i need an overall thermal impedance of 7.5 degC/watt.

The pcb this is all mounted too is pretty large (330mm x 120mm) and the back side is pretty much all available for heat sinking.

Sounds do able to me?  (The enclosure does include a fan and is well vented as well)

why not use the to-220 ?

[max_torque]

TO-220 doesn't help does it? All it does is focuses the heat on a small tab, to which i will have to interface a large heat sink  (thermal impedances are claimed to be the same in the data sheet from junction to case, both listed at 0.8 degC per watt)

Claimed Vtm is 1.25 volts at 12A (figure 9, page 5)  so 15w loss

The D2pak will be soldered directly to a copper plane of some size (i'd guess around 5 x 3 cm (15 cm^2) on the top side per device) and then i can put as many vias as necessary through the pcb to a backside copper plane that can be easily 10 x 20 cm (200 cm^2), and onto that plane i can bolt something like this:

https://docs-emea.rs-online.com/webdocs/1473/0900766b81473575.pdf

claims about 0.75 degC/watt for non forced convection.  I have 3 devices on at once at 15W (not continuously), so a total of 45 watts to get rid off, and i can arrange for the cooling fan for the unit to blow directly onto that heat sink.


T3035H says "specifically deisgned to operate at 150degC" in the data sheet, so i should be able to operate at, er 150 degC surely?

[MagicSmoker]

TO-220 doesn't help does it? All it does is focuses the heat on a small tab, to which i will have to interface a large heat sink  (thermal impedances are claimed to be the same in the data sheet from junction to case, both listed at 0.8 degC per watt)

D2Pak is merely a TO-220 that has been butchered modified for surface mounting, so it is unsurprising that R_theta[j-c] is the same for both of them. That said, it is R_theta[j-a] that is important, and it is getting heat from the case to ambient where the D2Pak performs significantly worse than the TO-220. Maybe a bunch of vias and a copper heat-spreader soldered to the bottom of the board will get you there, but you won't know until you build one. Note that it doesn't matter if you have a heatsink with a 0.75C/W thermal resistance (requiring how much mounting area on the heat generating device?) if the thermal resistance through the board is 5C/W (and that would be fantastically good unless you go with a metal core PCB).

T3035H says "specifically deisgned to operate at 150degC" in the data sheet, so i should be able to operate at, er 150 degC surely?

Sure, but what is the operational life at Tj = 150C? It might only be a few tens of hours for all you know, and I doubt ST will give you a number in writing as this is really considered silicon abuse.

[langwadt]

TO-220 doesn't help does it? All it does is focuses the heat on a small tab, to which i will have to interface a large heat sink  (thermal impedances are claimed to be the same in the data sheet from junction to case, both listed at 0.8 degC per watt)

Claimed Vtm is 1.25 volts at 12A (figure 9, page 5)  so 15w loss

The D2pak will be soldered directly to a copper plane of some size (i'd guess around 5 x 3 cm (15 cm^2) on the top side per device) and then i can put as many vias as necessary through the pcb to a backside copper plane that can be easily 10 x 20 cm (200 cm^2), and onto that plane i can bolt something like this:

https://docs-emea.rs-online.com/webdocs/1473/0900766b81473575.pdf

claims about 0.75 degC/watt for non forced convection.  I have 3 devices on at once at 15W (not continuously), so a total of 45 watts to get rid off, and i can arrange for the cooling fan for the unit to blow directly onto that heat sink.


T3035H says "specifically deisgned to operate at 150degC" in the data sheet, so i should be able to operate at, er 150 degC surely?

the to-220 can bolted directly to the heat sink, you could probably use the isolated tap one and still be better that d2pak through vias and pad

and pcb/planes have substantial spreading resistance so making it big doesn't help so much

[T3sl4co1l]

Can easily get 20-30W through a TO-220 and thermal pad, up to maybe 50W with thin, conductive pad and spring clamp.  Even with greased contact (live heatsink?), I would be very cautious about more, despite what the datasheet might say (that's a rant for another thread however).

The problem is 1. yeah you've got good sized copper pours, but they're all thin, and 2. most of the meat of the board is FR-4, which is terrifically bad thru-plane, it's doing almost nothing.  Obviously you can address these to some extent by fab (more via cross section, heavier copper), but that's pushing into "heroic" territory and one wonders if it would've been cheaper to use the bolted package in the first place?

You can run some guesstimates based on via barrel thickness, copper conductivity, board thickness (you might opt for 1.0 or 0.8mm stock, incidentally), and filling (solder or whatever as applicable).  If the estimate is questionable give or take a factor of 2, you may want to order a quick proto and test it (e.g. with a transistor or resistor).

Solder -- typical result is, because it has such worse conductivity compared to copper, even a solid solder-plugged via is about half the Rth of an empty one.  Not nearly as amazing as it maybe looks like it should be.  But also nothing to sneeze at!

Incidentally, pure tin is better than pure lead, and their alloy is worse than either; lead-free is actually beneficial here.


Or you can have it both ways, SMT and THT -- a TO-220 can be mounted over a hole in the board, so it can mount to the heatsink nearly flush with the board while the leads are soldered in at a modest distance (or even SMT'd against the board, but I'd maybe recommend not doing that at mains voltage).  Obviously the heatsink will need some clearance to the thru-pin tails, if applicable.

Tim

[langwadt]

Can easily get 20-30W through a TO-220 and thermal pad, up to maybe 50W with thin, conductive pad and spring clamp.  Even with greased contact (live heatsink?), ..

The triac mentioned even has a version with isolated (2500V rated) tap, at the cost of being 1.6C/W vs. 0.8C/W

[max_torque]

All good points!   

Having a large hole through the pcb, and machining an alluminum block that sticks through that hole and allows the use of TO-220s direct to that block, and thence to the heat sink on the other side of the pcb is certainly doable.   The current package, due to connectors, has 30mm of "free" space behind the pcb, with is what i'd like to utilise for the heat sink.

Looking at the overall power rejection number (45w or so) it's certainly going to need some forced convection, so i need to go back to the CAD and work out the best location for the fan, which may then drive the heat sink location etc

I'll report back when i understand the package envelope a bit better for round two   

[T3sl4co1l]

Can easily get 20-30W through a TO-220 and thermal pad, up to maybe 50W with thin, conductive pad and spring clamp.  Even with greased contact (live heatsink?), ..

The triac mentioned even has a version with isolated (2500V rated) tap, at the cost of being 1.6C/W vs. 0.8C/W

Yup, which puts it in the thermal-pad class.  Which is perfectly acceptable given the ratings (imagine that ).

A shame there aren't any affordable insulators with good conductivity.  AlN is less toxic than BeO and almost as good, but it's still hard to find (by itself).  Al2O3 and AlN are commonly used for DBC (used for packaged insulators, like the iso-tab TRIAC).  Thin insulators (mica, polyimide) are atrocious and only work because they are so thin.  Thermal pads need a resin or rubber filler so can't do much better (though there are some quite attractive formulations available, but again, for a price..).

Would be cool if a metal-slug process were feasible for PCB fab + reflow assembly.  Could actually run D(n)PAKs at ratings.  It's not at all impossible, but I'm not aware of any real production application..?  (There's also heavy metal PCBs, probably made with a combination of milled bar and electroplated foil, but, that's into the deep end of "heroic effort"...)

Looking at the overall power rejection number (45w or so) it's certainly going to need some forced convection, so i need to go back to the CAD and work out the best location for the fan, which may then drive the heat sink location etc

If saving heatsink cost or package size is valuable, you might consider MOS SSRs.  Possibly discrete (hand made) at this power level.  (Not sure offhand if any commercial units of comparable rating beat a TRIAC.)  Or, to be perfectly honest: good old fashioned relays, though I'm guessing you already have reasons to avoid those, hence the TRIACs in the first place.

Tim

[max_torque]

I had thought about a parallel Triac and relay, i have microprocessor control on this board, so i could use the triac to modulate, then switch the relay on when it "full on" which would remove most of the heat flux on the longer duration burns, but that's  a lot of extra effort!

http://www.farnell.com/datasheets/1723975.pdf

There is that ^^ BTA26 which is available in a TOP-3 package, with an insulated tab and is 0.9 degC/W, and in that package has nice widely based (good clearance/creepage!) strong pins, which could make the "hang the device in space over a hole in the pcb" type afair a flyer?

It has a slightly lower peak current rating, but looking at the data sheet numbers that seems mostly an arbitrary limit, and it has slightly lower dv/dt capability, but i'll snubber it and it's driving a broadly resistive load, so i don't think that's an issue. Cost is about par with the previously mentioned triac.  Vtm is, as you would expect, pretty much the same, so i'll still have 15w to deal with.

[max_torque]

Ok, i think i am converging on a solution that stands some chance of doing the numbers !

1) Swapped to TOP-3 packaged Triacs, as they are chunky and easy to work with, and have a big tab and surface area.  BTA26 from ST has insulated tab (2.5kV) so that helps make the heat sink interface nice and easy

2) Triacs now mount from backside of pcb, and are held (with spring clips) to a pair of heatsinks, also mounted to the pcb, but with little spacer to allow air to flow under the sinks, and for the sinks to clear the back of the pcb for isolation (might end up with plastic insulator in this gap, water jetted out to clear as necessary)

3) Using decent sized heatsinks, each with a pair of Triacs on, means max thermal load into each heatsink is 30 watts worst case

4) Powerful (ish, 16w rating) 240VAC fan sits with 3d printed duct on top next to the pcb stack, and therefore shoots it's air down behind the pcb, and over the heat sinks, ensuring a decent airflow and hopefully not too many stagnant areas


Fitting everything on two sides of the pcb, with the I/O connectors, the piggyback logic pcb and mounting into the assembly was quite a challenge!  How people did this before decent 3d CAD i'll never know     

[Jeroen3]

I would go for a lower height, higher fin density heatsink that mounts the TO package between the board and the heatsink.
It does require bending of the TO pins, but it immediately screws to the heatsink in an area where you can actually tap thread.
Plus, strain is put on the TO leads bend instead of the FR4. Cracked joints are far less likely.
But I do not know the application.

However, if you require cheap manufacturing steps (which my way is not) you should look into heatsink that solder into the board and use clips to press against the TO package. As often in ATX power supplies.

Eg: Fischer SK 525 of SK 573.
You skip the step to mill and tap the heatsink. And you don't need to screw things during assembly.

[max_torque]

Heat sink is already drilled for mounting by the manufacturer. Triacs will be held to sides of heat sink with "C" spring clips as mentioned (constant force, but allow for some movement)  and in the vertical orientation the legs of the Triacs are left long so they can flex if necessary rather than stress the pcb.

Heat sink is rated at 11W for a 60DegC deltaT with natural convection or 2.0 degC/watt at 300 ft/min blown air.

[T3sl4co1l]

That duct is tiny, is there no way to get a more direct flow path?

Could for instance use two half size, higher performance fans, and shrink the board away from them so there's more room to turn the flow around.

Or use a turbine style (a low performance one would still be quiet -- think laptop versus leafblower) that naturally blows to the side.

Tim

[max_torque]

The duct is actually just over 65% of the swept area of the fan, so it's fine, it looks smaller than it is!

(and the exit duct is not yet final, it's just a placeholder in the CAD to see how things sit together)