MOSFET cooling in commercial controllers; how do they do it?

strantor

strantor

100 W
Joined
Sep 12, 2011
Messages
207
Location
Houston, TX USA
I am planning out my controller design for brushed motors. I am starting out conservative; 48V, 300A. Right now I am planning to parallel 6 IRFP4110PbF MOSFETs and I'm at the stage where I am doing my thermal design. I've calculated that I'll be dissipating 56W in each FET and that I need a heat sink with a thermal resistance of .77C/W. now, when I go and look up heat sinks with those numbers, the only things that come up are heat sinks the size of the Bible, or the size of a loaf of bread (and I need one of those for each FET). My other alternative is a smaller, fan cooled heat sink, with a 600CFM requirement. Now, I know Kelly, Curtis, et. al. sell controllers with these specs, and none of them have giant aluminum shark fins coming out in all directions or electric superchargers on them, so how do they keep their FETs at a safe (reliable) operating temperature?
 
I think there is an error in your math. Show the calculation and your assumptions about Rdson, derating, etc. and we will work it through with you.
 
Yes, I had an error.
I originally figured that for 3 MOSFETs in parallel, I would dissipate 113W each, and that was too much, so I did proportions from there to find the wattage dissipation for 4 FETs (84W), 5 FETs(67W), 6 FETS (56W). But it doesn't work proportionally like that. You need to do all the calculations for each combination of FETs, which comes out to 4FETs (63W), 5 FETs (40W), and 6 FETs (28W).
My math to find the wattage dissipation is as follows (I have made an excel spreadsheet that does all the work for me now):

Rds(on) max is 4.5mΩ (with worst case scenario) @ 25c junction temp. The temp VS. Rds(on) graph shows that at 120C, the Rds(on) will be multiplied by 1.75, which comes out to 7.87mΩ
Current limiting will be set at 360A
For 3 paralleled MOSFETs, the voltage drop across the MOSFET is going to be 120A (Package limit, worst case scenario) * .00787 = .94V. If passing 120A through it and it is dropping .77V, then P=IE, .77V * 120A = 113W (plus switching losses, negligible due to extremely fast switching times, calculated slightly above 1W)

For 6 paralleled MOSFETs, the voltage drop across the MOSFET is going to be 60A * .00787 = .47V. If passing 60A through it and it is dropping .47V, then P=IE, .47V * 60A = 28W (plus switching losses, negligible due to extremely fast switching times, calculated <1W)

Calculation for Switching loss:
When the MOSFET is in the middle of switching it is passing half of the current limit and half of the applied voltage is dropped across it. So, if it has 24V across it and 60A through it, then it is dissipating 1440W.
t = Q / A.
Qg = 210nC, Amp output from FAN3121 MOSFET driver @ 18V & 1.5ohm gate resistor is 11.4A.
210nC * 11.4A = 18.4nS
40,000 (2khz switching time) X 18.4nS = 737µS (the total time the MOSFET will be dissipating 1440W in one second). Since 737µS corresponds to about .000737% of 1 second, multiply 1440W by .0737 and get 1.06W

Calculation for heat sink:

Image111.gif

Where
Tj = Maximum semiconductor junction temperature
θjc = Thermal resistance, junction to case.
Tc = Case temperature
θcs = Thermal resistance, case to heatsink (see Hot Tips)
Ts = Heatsink temperature
θsa = Thermal resistance, heatsink to ambient.
Ta = Ambient temperature
values:
Tj = 120C
Ta = 50C
Q = 56W
θjc = .402
θcs = .10 (estimate, based on seeing 1C/W when no thermal paste used, and seeing .09 when good thermal paste used)
When I was figuring 113W per MOSFET would get I get 0.137C/W
When I was figuring 56W per MOSFET would get I get 0.768C/W
Now that I am figuring 28W per MOSFET would get I get 1.99C/W, which is MUCH more manageable.
 
So, with this new understanding that the #of watts dissipated per MOSFET decays exponentially with linearly increasing number of MOSFETs, I am thinking I should consider using a ton of cheap little TO-220 FETs instead of these big TO-247 High amp FETs; I could get away with using dinky little heat sinks on them. For example, if I increased the number of FETs to 20, I would only dissipate 2.7W/FET and I could use a 25.4C/W heat sink for each one (or bolt them all to one big heat sink). But then I have to make sure they all turn on & off at the same time. Do you think if I went this route and used a dedicated driver for each FET, that I would have a problem with the trace inductance of the PWM signal from my micro? It would be hard to make all the traces the same length.
 
You're drinking the datasheet kool-aid.

Let me give you a reality check. An Altrax controller uses 28 x TO-220 mosfets in parallel for its 450amp 72v controller. They are each IRFB4110pfb mosfets.

It gets scalding hot and sometimes burns up if it runs for over 10mins at full throttle.


If you drink the datasheet kool-aide, you would think it could do thousands of amps and only make a few Watts of heat.
 
Which, unfortunatley, is a problem even with the commercial designers. ;)

Most manufacturers don't bother to warn you that their "rated" specs for the controller are not for continuous use, but only for a burst or momentary use, intermittently. Some actually do rate them for continuous use, but they don't usually advertise using that rating, becuase it would look like they suck and are extremely overpriced, compared to everyone else's super-high ratings (that are only burst).


A number of bigger controllers are made so that their heatsink surfaces can be (and should be) bolted to a much larger heatsink than the controller surface itself can provide, preferably with active cooling as a part of it.


Even our little ebike controllers work way better (and/or longer and more reliably) if they're actively cooled, when we are pushing them to their component limits. But even then, sometimes it is not enough and things still fail, because at the edge of the limits per the spec sheets, stuff doesn't work the way it does in the middle of the range.
 
I've tried to avoid the datasheet koolaide. I used the RDSon for a junction temp of 120C instead of the "rated" number, and using 20 FETs I wouldn't think I would be going anywhere near the max amps/fet. Are there any more thumb rules that I can apply to avoid disaster?
 
I'll take things a step at a time. Luke's data on commercial controllers is like gold. It needs to be processed. Not everyone can buy kilobuck controllers and dissect them!

You need to get your PhD in paralleling FET's before you even pencil your design. Scour the App Notes of at least two prominent FET manufacturers and post what you find. If you don't find the right one's I'll chip in. After you get the PhD, you will start your apprentice ship. The price of entry is the sound of popcorn FET's in the morning... :mrgreen: Hint: You will likely not be successful with one driver per FET. It is all about load sharing. When you go for the post doc, you may want to study texaspyro's posts on how he matched FET's for his welder. Good stuff from the man in Texas that Lights them Up.

PS: I hope my teaching style doesn't put anyone off... It's just that Luke, Methy and I apprenticed under a grand master of technology, known as the Ol' One... he never just gave us the answer, he made us dig and think, and rotate that rock in the light until we saw the gem inside. He not only taught us the technology, he taught us how to think in technology. I am forever grateful to him! He wanted us to be able to carry on without him... I share that goal in my efforts to pass it forward.

PSS: There is a plethora of people wanting to "do controllers" these days. My comment to them, it will be a long, hard journey. Working at these power levels is not easy. This is not Kansas, and it is not vanilla digital design. Much of what you need to know to be successful in high power electronics is NOT taught in college EE programs... that said, have at it. It's a great challenge to sharpen your skills and abilities on.
 
TylerDurden said:
You can check out oil-cooling...
http://endless-sphere.com/forums/viewtopic.php?f=6&t=25286&start=30
Click to expand...
That's a fantastic idea! and a sweet bike too. even with the oil just sitting in there, no circulation, it would help conduct heat to the controller case.
http://en.wikipedia.org/wiki/Thermal_conductivity
The thermal conductivity of mineral oil is .138W/(m·K) and the thermal conductivity of air is .025W/(m·K). that's a 550% improvement from the case being filled with still air to the case being filled with still mineral oil. the inverse of 550% is 18%, do you think I could get away with using a heat sink 18% the size of that required for still air convection if I filled the case with mineral oil?
 
No, because you still have to move the same total amount of heat away from the case itself. ;) It's just that now, that heat will make it to the outside of the case faster than before.

I think you might be thinking of a heatsink *inside* the case for soemthing, but I wouldn't do that at all--I'd physically connect all heatsink-requiring devices to the case itself, as directly as possible.

There are things that will still get hot inside there that can't be mounted that way (caps, resistors, PCB traces) that the oil might help if you're not ventilating the case with forced-air cooling.
 
bigmoose said:
I'll take things a step at a time. Luke's data on commercial controllers is like gold. It needs to be processed. Not everyone can buy kilobuck controllers and dissect them!

You need to get your PhD in paralleling FET's before you even pencil your design. Scour the App Notes of at least two prominent FET manufacturers and post what you find. If you don't find the right one's I'll chip in. After you get the PhD, you will start your apprentice ship. The price of entry is the sound of popcorn FET's in the morning... :mrgreen: Hint: You will likely not be successful with one driver per FET. It is all about load sharing. When you go for the post doc, you may want to study texaspyro's posts on how he matched FET's for his welder. Good stuff from the man in Texas that Lights them Up.

PS: I hope my teaching style doesn't put anyone off... It's just that Luke, Methy and I apprenticed under a grand master of technology, known as the Ol' One... he never just gave us the answer, he made us dig and think, and rotate that rock in the light until we saw the gem inside. He not only taught us the technology, he taught us how to think in technology. I am forever grateful to him! He wanted us to be able to carry on without him... I share that goal in my efforts to pass it forward.

PSS: There is a plethora of people wanting to "do controllers" these days. My comment to them, it will be a long, hard journey. Working at these power levels is not easy. This is not Kansas, and it is not vanilla digital design. Much of what you need to know to be successful in high power electronics is NOT taught in college EE programs... that said, have at it. It's a great challenge to sharpen your skills and abilities on.
Click to expand...
I fully expect it to be a long hard journey, and Ill take whatever help I can get; I'm not picky and my feelings don't get hurt. I'm not an EE, never been to college, just a blue collar guy with a hobby that's recently turned into a mission. I'm in it for the experience.
I'll give you a breakdown on the documents that I currently have printed out (with things that I consider important underlined):
http://powerelectronics.com/power_management/motor_power_management/Current-limiting-eases-design-motor-drives-PET.pdf
http://www.ti.com/lit/ml/slup169/slup169.pdf
http://homepages.which.net/~paul.hills/SpeedControl/SpeedControllersBody.html
http://robots.freehostia.com/SpeedControl/SpeedControllers.html
http://robots.freehostia.com/SpeedControl/Mosfets.html

Microchip AN898
Microchip AN799

PS. I'm new here; who's Luke?
And thank you all sincerely for all the help thus far and furture.
 
amberwolf said:
No, because you still have to move the same total amount of heat away from the case itself. ;) It's just that now, that heat will make it to the outside of the case faster than before.

I think you might be thinking of a heatsink *inside* the case for soemthing, but I wouldn't do that at all--I'd physically connect all heatsink-requiring devices to the case itself, as directly as possible.

There are things that will still get hot inside there that can't be mounted that way (caps, resistors, PCB traces) that the oil might help if you're not ventilating the case with forced-air cooling.
Click to expand...
I guess I was thinking that since heat transfers to aluminum even better than it transfers to mineral oil, that you would want at least a small heat sink on the mosfet submerged in mineral oil. But, yes I agree, if it were in an aluminum enclosure, it would be better just to go ahead and bolt them to the enclosure. BTW, if the mineral oil conducts heat better than air, does that mean it would transfer more heat from the MOSFETs to the capacaitors & other components than in air?

bigmoose, thank you for that. I just printed those off. I will read them (or, start reading them) sometime in the AM.
 
I'm Luke, and BigMoose is Dave, who is one of the kindest and smartest humans to ever grace this planet, and I'm not kidding.


So, for starters, FETs don't like the share (even though they share much better than IGBT's), they can only share as well as you allow them to share.

They also don't like to switch On together at the same time, don't like to switch off together at the same time, and when the flyback comes they have to clamp, which ever FET has the lowest clamping voltage body diode eats most of that cake for himself.

Layout is everything. Passed 2 fets in parallel, getting a design that shares current gets very tricky. Having 100 fets in a row is only marginally better than having 20 in a row, and having 20 in a row can easily be worse than having 6 or 10 setup in a radial or axial array and bused to permit sharing current effectively.

Second thing to consider when working the Rth numbers is that your heatsink will only start out at ambient temp, you need to work your thermal calc's by first estimating the FET heating (which a good rule of thumb to compensate for switching losses and flyback heating is to triple RdsOn conduction losses), use this heating value with the sink you plan on using's degC/watt, and use 15degC over the highest ambient temp you plan on operating in as the air surface temp for the outside of your sink. Now that you have this temperature, which will be something awful hot, at least 70-90degC, now you can use that number to calculate the package Rth sink to package case (don't forget your isolating medium if you need it, brushed won't necessarily need it if you design around it), and add your junction to case Rth.

At this point, you've got a decent ball-park guess on the real-world FET thermal picture.


This is why the IR rep I worked with who had some very small very good silicon die FETs capable of 375amps on the datasheet actually were calculated out to be capable of... ready for it? Ready for it? 11amps continuous. Yes, the 375amp direct FET package in my application was determined by IR to be capable of contributing 11amps. Virtually ANY mosfet with an effective thermal path to a heatsink is capable of higher continuous current. You don't get to see the 11amp reality until you dig pretty deep and put a lot of the pieces together for how the system actually transfers heat to your cooling medium.
 
strantor said:
BTW, if the mineral oil conducts heat better than air, does that mean it would transfer more heat from the MOSFETs to the capacaitors & other components than in air?
Click to expand...
Unfortunately, yes, it does. However, that just means they get hot faster--they would probably eventually reach the same temperature anyway, if used long enough at power levels where you need to get the heat out of the case.

That's why (in general) that capacitors end up dying so quickly inside many power supplies of all sorts of devices, from chargers to controllers to TV sets and computers--the heat ages them, electrolyte heats up and maybe even boils off, certainly leaks out.
 
I have a controller with similar target application to what you want to do in my roadmap. I had the pleasure of having rithee05 discuss it a little with me in the middle of this thread, and I'm designing something much smaller first in order to build experience. I'm aiming at a smaller number of FETs and big margins, trying to avoid as much trouble as I can from the start; TO247 FETs for sure, maybe IRFP4368, 4 for each switch of the half bridge (half bridge to support regen) totaling 8 for the controller. Using the smaller TO220 is really tempting, but the complications that it may bring...
I think 120A per TO220 FET is very optimistic. IRF has a paper on package current limiting and the difficulty they have in setting a value already shows how tricky this is. I would start setting for some 75A max per TO-220, and not continuous. Have you seen the thickness and number of the wires that connect a led to the silicon? And the MOSFET legs thickness? The tests they use to spec many of the limiting values are at the temperature limit of the parts (I'm not even going to mention the 25ºC temperature) and you sure don't want to work near the limits; even a bit below the limit is too much, because having things hot already brings energy loss. And Luke is right of course, in that the heating is always above ambient temperature (*heatsink* ambient temperature, not "outside air" temperature) and you need to take into account that it will be hot inside the controller box.
And make sure you listen bigM also :)

p.s. Heck, altrax is giving 32A per FET. Got to factor in the switching losses.
 
You know, I was looking at the current source circuit with an ampop and thinking if we could get an ampop plus a transistorized power output to "force" a bunch of FETs to equally share current. We could loosely measure the current across the FET and compensate with temperature to avoid using the shunt. We could just then toss it FETs without matching, and I guess that would also forgive some layout "sloppiness". The reference for the ampops would be square wave with controlled raise and fall times, which would control the FETs to turn ON/OFF in a "synchronized" way. Could we make switch under 1us this way without expensive hardware and would it be worth it?

update: humm, it's an inductive load... not sure if the switching phase can be made without an external shunt. back to the books...
 
This thread has some advice on how to fail at calculating your FET's thermal load and cooling and heat generation.

It does mention a few good things that are commonly over looked though, making it worth reading, despite it missing the entire big picture.

http://endless-sphere.com/forums/viewtopic.php?f=7&t=32316
 
Thanks Luke, Dave, amberwolf, njay.

njay; I am reading all these documents that were linked to, plus a bunch more supporting documents that I came across. It's slow work because every paragraph sends me back to the internet to learn more about something I don't understand. my book of references is outgrowing it's 2inch binder! I read your thread about your controller but really can't comment on what you've suggested, as I am learning just how much I don't know what I'm talking about. I have a 1 week out of town vacation starting tomorrow, during which I won't be hitting the books, so give me at least 2 weeks (possibly more) before I'm ready to say anything remotely intelligent.
 
liveforphysics said:
This thread has some advice on how to fail at calculating your FET's thermal load and cooling and heat generation.

It does mention a few good things that are commonly over looked though, making it worth reading, despite it missing the entire big picture.

http://endless-sphere.com/forums/viewtopic.php?f=7&t=32316
Click to expand...

I'm glad you linked to that. That document that he copy/pasted into the thread is one of the ones I printed out. Seeing your comments on it makes me think maybe it shouldn't be in my binder (or at least with your comments attached).

EDIT: it's official; your comments are in the binder behind that document.
 
Great thread guys!
 
Are you glad you're only starting with a brushed controller yet my friend? :)
 
You must log in or register to reply here.

Similar threads

S
Serious FOCer (84V VESC 6 based controller)
8 9 10 11 12
Replies
299
Views
69,297
shaman
S
S
Cheap FOCer (VESC 4.12 based design)
5 6 7 8 9
Replies
213
Views
45,813
slaphappygamer
S
L
Direct Drive Vs Geared Hub at 48V possible replacement
2
Replies
32
Views
5,197
wesnewell
wesnewell
amberwolf
The HI-Lebowski: a Lebowski SMD brain running a zombified Honda IMA Inverter: *a HOW-TO guide*
2 3
Replies
64
Views
20,533
amberwolf
amberwolf
S
Ebike Hub Motor overheating cooling solution - HUBSINK OFFICIAL THREAD
2 3 4 5 6
Replies
133
Views
46,018
ahmet
A
Back
Top