+ Post New Thread
Results 1 to 18 of 18
  1. #1
    Member level 3
    Points: 540, Level: 4

    Join Date
    Jul 2011
    Posts
    67
    Helped
    1 / 1
    Points
    540
    Level
    4

    Calculating PCB copper area required to cool a MOSFET

    Hi

    I am trying to calculate how much copper area I would need to cool a surface mounted power MOSFET as heat sinks can't be used.

    I got this formula from wikipedia but I get awkward numbers,

    I know what my Rsa (sink to ambient) thermal resistance needs to be.

    Thermal resistance = kA/T ---------k is thermal constant for copper i.e. 401 W/mC (Watts per meter degree Centigrade)
    I am assuming a standard copper track is about 3mils thick i.e. (76.2micro meter)

    I want Rsa of 1.5 C/W

    Using the above equation to find Area I get a huge area...

    What am I doing wrong!!

    Most likely to be using 4-6 layered board with plated through hole vias for cooling....


    Thanks

    •   Alt18th December 2011, 05:41

      advertising

        
       

  2. #2
    Advanced Member level 5
    Points: 19,240, Level: 33
    Achievements:
    7 years registered

    Join Date
    Jan 2011
    Posts
    3,200
    Helped
    1134 / 1134
    Points
    19,240
    Level
    33

    Re: Calculating PCB copper area required to cool a MOSFET

    1.5C/W is an extremely low number (assuming you're talking about case to ambient), which is practically impossible with copper planes and vias alone. Here is a good document on the subject from TI: http://www.ti.com/lit/an/snva419b/snva419b.pdf

    Consider described options, such as case heatsinks and bottom layer heatsinking. Even with those, I doubt 1.5C/W is possible though, at least not without some forced airflow.



  3. #3
    Member level 3
    Points: 540, Level: 4

    Join Date
    Jul 2011
    Posts
    67
    Helped
    1 / 1
    Points
    540
    Level
    4

    Re: Calculating PCB copper area required to cool a MOSFET

    Thanks mtwieg,

    Great article, However I am a bit confused, how do I account for the fact that the PCB tracks will be getting hot due to current flowing through them and other components relying on tracks for cooling?? how do I put the whole image together?? any ideas??

    Another variable is the current that affects the track width, I have to make my track widths wide enough for that.

    I guess, if I can meet my cooling right, width should of track should more or less be adequate to carry the current??

    Thinking of using 4 layered PCB, 2 internal layers and mounting a heat sink on the back of PCB, would that do much help or vias limit the use of heat sink?? from that article you mentioned vias seem to have quite high thermal resistance...more vias to solve the problem?? (parallel)

    Thanks again



  4. #4
    Advanced Member level 5
    Points: 19,240, Level: 33
    Achievements:
    7 years registered

    Join Date
    Jan 2011
    Posts
    3,200
    Helped
    1134 / 1134
    Points
    19,240
    Level
    33

    Re: Calculating PCB copper area required to cool a MOSFET

    Quote Originally Posted by zuq View Post
    Thanks mtwieg,

    Great article, However I am a bit confused, how do I account for the fact that the PCB tracks will be getting hot due to current flowing through them and other components relying on tracks for cooling?? how do I put the whole image together?? any ideas??
    The dissipation due to trace losses should be very small compared to the loss in the FET itself. What package are you using for the FET, and how is it connected in the circuit? Hopefully its one where the tab is connected to circuit ground...
    I guess, if I can meet my cooling right, width should of track should more or less be adequate to carry the current??
    Definitely; at least on the track/planes connected to the thermal tab on the package.
    Thinking of using 4 layered PCB, 2 internal layers and mounting a heat sink on the back of PCB, would that do much help or vias limit the use of heat sink?? from that article you mentioned vias seem to have quite high thermal resistance...more vias to solve the problem?? (parallel)
    Yeah, just go crazy with the vias. If you're really serious about this (it's for a production design) you should speak with your PCB manufacturer. They may offer things like filled vias which greatly improve their thermal conductivity. Also their are aluminum backed PCBs which are made specifically for this sort of thing (though I don't have much experience with those). Also having heavier copper thickness helps a lot too.



  5. #5
    Member level 3
    Points: 540, Level: 4

    Join Date
    Jul 2011
    Posts
    67
    Helped
    1 / 1
    Points
    540
    Level
    4

    Re: Calculating PCB copper area required to cool a MOSFET

    I am using D2PAK-7 it is a IRF2804S mosfet. My peak current demand is 200A and these babies can handle about 160A each, I will probably use 4 mosfets in parallel to reduce heat.
    I am using these in a synchronous chopper configuration so the drain (tab) will be on +ve for the high side.
    This is just a one off project and I am limited to what I can actually use to build the PCB.

    What do you suggest for hooking up the battery cables to the PCB??? It will be 2x 2AWG cables. I obviously cant just clamp the battery lug on top of the PCB, that would burn the top layer off.

    This is what I am thinking to solve the issue; have a big via that will connect all 4 layers of PCB together and connect my lug to that via, but I am not sure what the limitations are for vias and how much current they can carry etc etc.. suggest something please.

    Thanks for your help



  6. #6
    Advanced Member level 5
    Points: 19,240, Level: 33
    Achievements:
    7 years registered

    Join Date
    Jan 2011
    Posts
    3,200
    Helped
    1134 / 1134
    Points
    19,240
    Level
    33

    Re: Calculating PCB copper area required to cool a MOSFET

    Okay, if it's that much current then you will need to add some sort of busbar to the PCB. I thought at first you might be doing a high frequency SMPS, in which case trace losses aren't too much of an issue, but if you're doing a DC motor driver at 200A, then there's no way around needing busbar. What I would do is make a large copper pour for the drain tabs of the FETs, and make it a soldermask stop area (so that the plane is exposed). Have a few large holes in that plane, which you're use to connect your cables with ring lugs (quarter inch holes and hardware should do it). You will need to cut some busbar in roughly the same shape as your exposed plane. Thick copper sheet can work out well (and is cheap), 1/8" aluminum is better if you have a way to cut it. You'll lay that over your drain plane, and will be bolted in place. Make sure it makes good contact in order for it to work correctly, especially near the drain tabs of the FETs.

    The busbar should act to help cooling as well. You can also put an exposed plane on the bottom layer, connected with lots of vias (and the bolts as well), with a heatsink for added dissipation. Forced air may be mandatory.

    You will also need bus bar for the source connections leading back to the battery.



  7. #7
    Member level 3
    Points: 540, Level: 4

    Join Date
    Jul 2011
    Posts
    67
    Helped
    1 / 1
    Points
    540
    Level
    4

    Re: Calculating PCB copper area required to cool a MOSFET

    Genius, Thanks

    Slight confusion;

    So I solder the Mosfet drain tab to the copper pour which is on the PCB and I place the Bus bar where??

    Could you maybe draw up a quick picture in paint if you don't mind, please??

    and I guess, I can't perform much calculations in this case for a design like this?



  8. #8
    Member level 3
    Points: 540, Level: 4

    Join Date
    Jul 2011
    Posts
    67
    Helped
    1 / 1
    Points
    540
    Level
    4

    Re: Calculating PCB copper area required to cool a MOSFET

    Do you mean something like this?? see attached picture, they seem to have copper pour on the bus bar no PCB what so ever.. I cant seem to understand how the gate resistors (gate) is connected on the same bus bar!Click image for larger version. 

Name:	mosfets.jpg 
Views:	35 
Size:	161.6 KB 
ID:	65906

    Thanks a lot



    •   Alt19th December 2011, 14:34

      advertising

        
       

  9. #9
    Member level 3
    Points: 540, Level: 4

    Join Date
    Jul 2011
    Posts
    67
    Helped
    1 / 1
    Points
    540
    Level
    4

    Re: Calculating PCB copper area required to cool a MOSFET

    Do you mean something like this?? see attached pictureClick image for larger version. 

Name:	mosfets.jpg 
Views:	35 
Size:	161.6 KB 
ID:	65906, they seem to have copper pour on the bus bar no PCB what so ever.. I cant seem to understand how the gate resistors (gate) is connected on the same bus bar!

    I guess I can then attach a heat sink on the other side of these bus bars?

    Thanks a lot



  10. #10
    Advanced Member level 5
    Points: 19,240, Level: 33
    Achievements:
    7 years registered

    Join Date
    Jan 2011
    Posts
    3,200
    Helped
    1134 / 1134
    Points
    19,240
    Level
    33

    Re: Calculating PCB copper area required to cool a MOSFET

    well that is a more advanced method where there is no PCB and they just solder on top of the busbar. I was thinking of something where the components are still soldered to the PCB, but the busbar is shaped to sit on the PCB very close to the components.

    It looks like all the busbar is on top of a larger heatsink, separated with a bunch of Sil-pad. If you have the tools and materials required, and don't mind having the actual PCB with control stuff being separate, then that approach should work fine as well. The gate resistor(s) probably are out of the picture, meaning that each FET doesn't have it's own gate resistor. For a low frequency DC chopper that is probably fine.



  11. #11
    Super Moderator
    Points: 238,808, Level: 100
    Awards:
    1st Helpful Member

    Join Date
    Jan 2008
    Location
    Bochum, Germany
    Posts
    41,362
    Helped
    12620 / 12620
    Points
    238,808
    Level
    100

    Re: Calculating PCB copper area required to cool a MOSFET

    I can tell, that I have implemented high current (> 100 A) PCBs with thick copper outer layers (about 6 ounces/210 µm). But I would probably prefer today a multilayer sandwich with e.g. 70 µm copper. Vias need to be implemented as arrays anyway. You'll also possibly need a large amount of thermal vias for the FETs.

    In so far, I agree, that the busbar hybrid is an interesting idea. If your high current path is exclusively through the MOSFET, conventional packages with screw connection like SOT-227B should be considered as well.



  12. #12
    Member level 3
    Points: 1,428, Level: 8
    thunderdantheman's Avatar
    Join Date
    Jan 2010
    Location
    Australia
    Posts
    57
    Helped
    10 / 10
    Points
    1,428
    Level
    8

    Re: Calculating PCB copper area required to cool a MOSFET

    first things first, you need to try and estimate the power loss you need to dissipate... At what frequency are you switching the FETs? as with any high Id FET the trade off is often high Qg meaning slower switching speeds and hence longer time in the linear region = heat! (not to mention much higher gate drive current = more heat!). Also, depending on what it is you're driving, the intrinsic diode can also be a heat producer. (this is something to note when driving PWM inductive loads). Now... sinking heat into the copper alone is probably not going to work for you.. well, maybe for a while but not continuously, although your copper track will sink heat well, it aint going to get rid of it... it'l just keep getting hotter until you let the smoke out. To get 1.5degC/W you will need significant surface area in free air. (hence the massive copper area you come up with in your equation). Defiantly require a bus-bar to feed each FET at that current, even with a 6oz copper and allowing for a 50degC temp rise you would need a 70mm track, or 35mm top and bottom with solder filled vias. (Maybe a copper strip soldered along the tab?). Also, be very careful about driving high current FETs in parallel unless you have some method of ensuring current sharing... (i know people say they automatically current share... 5 FETs, 4 start warming up.. their RDSon goes up so they draw less current... fantastic right... wrong, the poor old 5th FET starts taking up the slack.. BANG... and this will all happen before you even register a temp rise on the case!) Yes... big round vias (80mil) and fill with solder. don't know if I've helped or hindered.. good luck, let us know how it goes!



  13. #13
    Member level 3
    Points: 540, Level: 4

    Join Date
    Jul 2011
    Posts
    67
    Helped
    1 / 1
    Points
    540
    Level
    4

    Re: Calculating PCB copper area required to cool a MOSFET

    would it be easy to make a layout like the one i showed you?? making the copper pour on bus bar?
    I think this way I can get rid of the complication of working out how much copper I would need in PCB for cooling ...
    If I go this way..I can work out Rsa of the metal bar, by adding heatsink at the back of bar, is that considered to be going in parallel or in series (reduces the Rsa or increases it?)

    Thanks



  14. #14
    Advanced Member level 3
    Points: 4,522, Level: 15

    Join Date
    Jun 2010
    Posts
    724
    Helped
    223 / 223
    Points
    4,522
    Level
    15

    Re: Calculating PCB copper area required to cool a MOSFET

    Is this for a commercial product or hobby? I have faced a similar need for hobby when I needed to run some fairly large DC motors on 12V. One was for a heavy duty, 6V Ford "long shaft" starter motor and the other was for a treadmill motor rated at about 1.5 HP. The pictures below show how I did it. The methods might not be appropriate for a commercial product.

    The starter controller (immediately below) was my first project in many years. It was kind of crude, but worked even when the starter was stalled. I used 0.025" copper plate (25 mil) for the conductors. Mosfets were soldered directly to the buss bars. Gate resistors were done as shown. I used a relatively hard setting thermal adhesive (Loctite 383) to attach the bus bars directly to the heat sink. The heat sink was anodized and I used no other electrical insulation than that and the adhesive. Power cables were connected to the bus bars using bolts. As to your question about ease of working with the materials, cutting the copper plate was the hardest part. It was done with a small scroll saw and cleaned up with hand files.



    The other images are for the lower powered controller. For this, I just used 4 ounce PCB and etched the tracks. It was certainly easier to make than the one using the copper plate.



    John


    1 members found this post helpful.

    •   Alt19th December 2011, 19:31

      advertising

        
       

  15. #15
    Member level 3
    Points: 540, Level: 4

    Join Date
    Jul 2011
    Posts
    67
    Helped
    1 / 1
    Points
    540
    Level
    4

    Re: Calculating PCB copper area required to cool a MOSFET

    Hi jpanhalt

    Very interesting design, I guess this is the sort of thing I am after. I am not building a commercial product.
    How did you attach heat sinks to Power Mosfet bus bars?? did you have three heat sinks, one on each bar?? cant use one big one otherwise you will short the DC bus??

    How did you work out how big your copper bars had to be to carry current and for cooling etc?

    By the way the first picture in your post did not show up, could you please attach that again, I only see three pictures. Thanks



  16. #16
    Advanced Member level 3
    Points: 4,522, Level: 15

    Join Date
    Jun 2010
    Posts
    724
    Helped
    223 / 223
    Points
    4,522
    Level
    15

    Re: Calculating PCB copper area required to cool a MOSFET

    I will try to answer each part:

    Quote Originally Posted by zuq View Post
    I am not building a commercial product.
    How did you attach heat sinks to Power Mosfet bus bars?? did you have three heat sinks, one on each bar?? cant use one big one otherwise you will short the DC bus??
    A single heat sink was used. Anodized aluminum is an insulator; however, the coating is very thin and one must be careful not to scratch it. The adhesive I used is also an electrical insulator. Be sure not to squeeze it completely out. A few small mica spacers may help prevent that. The mosfets were initially soldered to the bus bars before assembly to the heat sink. (Soldering after assembly can be done for repairs, but is more difficult.) I used typical "hat-shaped" washers to insulate any attaching screws from the bus bars. Page 3 of this application note shows how: http://www.irf.com/technical-info/appnotes/an-1012.pdf. They are called "insulating bushing" in the AN. If I were doing it again, I might just use insulated bushings and screws for each mosfet. If your current can be handled by 4 oz. or greater PCB (if you can get it), I would prefer that method of assembly.

    How did you work out how big your copper bars had to be to carry current and for cooling etc?
    It is embarrassing, but I did no calculations, except to determine that standard PCB traces would have to be quite large at an anticipated current in excess of 200A. "Looks about right" was all I used at the time and nothing got hot. It was a single project and has worked since it was made in about 1997.

    By the way the first picture in your post did not show up, could you please attach that again, I only see three pictures. Thanks
    For some reason, the image attachment for the first design didn't work like it did for the second design that was done on PCB material. Try clicking on the small thumbnail to the farthest right and see if that make it viewable. If that doesn't work, reply back here and I will re-load that image.

    John


    1 members found this post helpful.

  17. #17
    Member level 3
    Points: 540, Level: 4

    Join Date
    Jul 2011
    Posts
    67
    Helped
    1 / 1
    Points
    540
    Level
    4

    Re: Calculating PCB copper area required to cool a MOSFET

    Thanks everyone for your contribution. I am doing some calcs at the moment to work out the amount of copper/metal I need.
    According to my calculations 10cm by 5cm (0.1cm thick) copper bar can get me a Rsa of about 10C/W. That is if sink is at 110deg and ambient is 35. Most of cooling is done by radiation.

    From what I have read is that, (from equations) it is better to make the surface area bigger and keep the thickness of metal smaller hence smaller Rsa (radiation).

    Having said that, help me understand this: How would placing a heatsink on the back of my copper bar then help cool better? the thickness then simply becomes larger?

    Or

    Is it the fact that I start getting more cooling due to convection as there are now fins on the heat sink? which has a stronger effect than total increase in thickness that decreases cooling due to radiation.

    **Excuse my presentation skills, hopefully the question makes sense** Cheers everyone.



  18. #18
    Advanced Member level 5
    Points: 14,946, Level: 29
    mrflibble's Avatar
    Join Date
    Apr 2010
    Posts
    2,724
    Helped
    686 / 682
    Points
    14,946
    Level
    29

    Re: Calculating PCB copper area required to cool a MOSFET

    Hey, pretty useful info jpanhalt, thanks!

    And on the subject of vias ...

    Quote Originally Posted by zuq View Post
    I guess, if I can meet my cooling right, width should of track should more or less be adequate to carry the current??

    Thinking of using 4 layered PCB, 2 internal layers and mounting a heat sink on the back of PCB, would that do much help or vias limit the use of heat sink?? from that article you mentioned vias seem to have quite high thermal resistance...more vias to solve the problem?? (parallel)
    Quote Originally Posted by mtwieg View Post
    Yeah, just go crazy with the vias. If you're really serious about this (it's for a production design) you should speak with your PCB manufacturer. They may offer things like filled vias which greatly improve their thermal conductivity. Also their are aluminum backed PCBs which are made specifically for this sort of thing (though I don't have much experience with those). Also having heavier copper thickness helps a lot too.
    Luckily my current design has a more modest thermal requirement than all the high power stuff in this thread. But it being a bunch of led clusters it still pays to keep it as cool as possible. So if by paying extra attention to my layout I can extend the useful life of said leds, then it's worth a shot IMO. :)

    Power dissipation is ~ 1.5 Watt per pcb. This is just a bunch of low power (~ 75 mW) smd leds bunched together. So no rocket science whatsoever. I'm just new to this whole thermal design stuff. Up until now for motr driver designs I just did "increase track size to Big&Fat" and "when in doubt slap on a big heatsink". Worked so far... But I thought I'd try to do it "proper" this time, just for the fun of it.

    So, 1500 mW on a pcb as small as I can reasonably make it. The idea is to use a 0.8 mm double sided FR4 board with HASL finish. SMD components mounted only top side, and then thermal vias to the bottom side. Slap it on a heatsink and we're done. Only I'm not quite done yet, because I have a few doubts on what is best. :P

    Took a look at this application note from Avago, and decided on using the arrangement in figure 7a on page 6.



    So additional copper on the top covered by solder mask. And then on the bottom leave the additional copper pad without solder mask.

    Now regarding the via's ... I was thinking on the top side do tented via's, and on the bottom side do open via's. The big question being: are these open via's on the bottom going to give a problem? This being a board with HASL finish, is solder going to flow into those via's during the hot air leveling process? And if yes, is this going to cause a problem with outgassing and whatnot?

    To prevent that sort of problem I could of course tent the via's on both top + bottom side. And while I'm at it just put solder mask on the entire bottom copper pad. That would take care of any problems with via's, but that bottom side solder mask will not exactly help in getting the heat to the heatsink. Any ideas on what's best here?

    Another unsolved problem is electrical isolation between the board and the heatsink. This thing has multiple led strings on it, each being PWM-ed individually. And as such different voltages. So lets say I have several copper pads on the bottom side, all at different voltages. Oh yeah, it's relatively low voltage. Max voltage difference between pads is going to be about 17 Volt, probably more like 15 Volt.

    Now how the hell do I place this on a heatsink without shorting things out? I could do as jpanhalt, and use anodized + some adhesive. Now an anodized heatsink was the plan anyways, but I'm not sure on the interface material. Lets say there are 10 seperate pads on the bottom side all at voltages between 0 V and 17 V. Can I get away with the anodized aluminum heatsink + some thermal adhesive?

    Possibly this may be a bit overkill for a measly 1500 mW (certainly when compared to the above MOSFETs), but lower temperatures for leds is nice.

    Besides, I also have some HTSSOP packages in there somewhere with thermal pads that each have to get rid of 3000 mW max. And it's much the same problems + questions.

    Any comments, ideas, app notes, etc would be much appreciated! :)



--[[ ]]--