MOSFET power dissipation. Can a heatsink take +300W?

Thread Starter

ballsystemlord

Joined Nov 19, 2018
102
Hello,
Some of you may recall that I was discussion making a load bank. It was suggested that I use 100W resistors, which is fine, but I thought I might be able to get a better and/or cheaper design by using MOSFETs (or maybe IGBTs).
I have since been searching what MOSFETs are available at a reasonable price point (even a small professional load bank (150W) will cost 100s of dollars), and I have discovered several possible contenders.

IRF3207 can dissipate 300W and it costs $1.59 per piece on arrow (R0JC 0.50C/W, R0CS 0.50C/W).
IXFK180N15P can dissipate 830W and it costs $9.88 per piece on arrow (RTHJC 0.18C/W, RTHCS 0.15C/W).

Both of these options, isolated from their cooling needs, are much cheaper than using 100W resistors. But even if this point were academic, the question still comes up...

How on earth do you cool a +300W MOSFET?!

Even powerful CPU coolers only do about 250W, and they have a 40mm x 40mm surface area for contact with the IHS (Integrated Heat Spreader), whereas these MOSFETs do not.

Thanks!

Links:
https://www.arrow.com/en/products/irfb3207pbf/infineon-technologies-ag
https://www.arrow.com/en/products/ixfk180n15p/littelfuse
 
Last edited:

DickCappels

Joined Aug 21, 2008
9,505
First of all, take the semiconductor device under consideration and figure the thermal resistance from the junction to the case (usually given in data sheets).

Next, figure in a small amount of thermal resistance from the case to the heatsink. (one to a couple of degrees per watt)

Knowing the power dissipation and the maximum allowable junction temperature you can calculate the thermal resistance from the heatsink to the case.

Simple in theory but maybe not really simple, for example, you might want to add a fan so your heatsink need not be as huge, but estimates of air velocity can involve large errors, so it is a good idea to budget for two or three go-arounds on the design.
 

Thread Starter

ballsystemlord

Joined Nov 19, 2018
102
Simple in theory but maybe not really simple, for example, you might want to add a fan so your heatsink need not be as huge, but estimates of air velocity can involve large errors, so it is a good idea to budget for two or three go-arounds on the design.
I kinda understood that much. I know I'd need a fan. What I was asking, in particular, is what type of heatsink would dissipate that much heat? What type of heatsink would be able to distribute that much heat fast enough?
 

crutschow

Joined Mar 14, 2008
31,097
What type of heatsink would be able to distribute that much heat fast enough?
You would need to look for a heatsink with around a 1/4°C/W thermal resistance to air.
You might look at some of the fan-cooled heat-sinks designed to cool high-power computer microprocessors.
 

Irving

Joined Jan 30, 2016
3,181
MOSFETs designed for high speed switching will fail dramatically when used continuously as a load due to spot heating. You need ones intended for the purpose like IXYS' Linear2 product line (see attached PDF). I use 8 IXTX110N20L2 in a 1500W active electronic load for 50Ax30v battery testing using a home made water cooled heatsink and the radiator off a motorcycle with 6 x 120mm fans.

300W air cooled with 2 such devices is relatively easy on a force-air- cooled heatsink. You can buy 250W e-loads on eBay or Amazon eg this one.
 

Attachments

LowQCab

Joined Nov 6, 2012
2,620

Thread Starter

ballsystemlord

Joined Nov 19, 2018
102
MOSFETs designed for high speed switching will fail dramatically when used continuously as a load due to spot heating.
I thought MOSFETs were single element items. That is to say, because there is only one switch, there is only one spot to get hot. So, how does spot heating come into play? Do you have a nice article or something I can read?

PS: I'll look into the above posts more later. I'm off to bed.
 

DickCappels

Joined Aug 21, 2008
9,505
If you split the load among more than one transistor you can spread the heat around the heatsink more easily. With a single transistor, the heat injected into the heatsink is injected in one place and spreads out from there, making the place the single transistor is mounted the hottest spot on the heatsink.

This can be important when the thermal drop between the case of the transistor and the heatsink becomes significant.

One potential work-around is to make the heatsink out of copper. Expensive but can be enough to allow you to go into production with an otherwise marginal design.
 

Ian0

Joined Aug 7, 2020
6,660
I thought MOSFETs were single element items. That is to say, because there is only one switch, there is only one spot to get hot. So, how does spot heating come into play? Do you have a nice article or something I can read?

PS: I'll look into the above posts more later. I'm off to bed.
It depends on the Vgs zero-temperature coefficient point.
If you use the older MOSFETs such as the lateral MOSFETs designed by Hitachi for audio, and made by Exicon, they have a positive temperature coefficient at normal operating voltages for linear mode. So, as the temperature increases, the transistor becomes less enhanced, and current flow concentrates in the cooler points, so current spreads out throughout the die.
More modern MOSFETs that are designed for switching have a negative temperature coefficient of Vgs, so as the temperature increases the MOSFET becomes more enhanced and takes more and more current; and this can happen on a small scale within the MOSFET because they are made like an array of tiny MOSFETs all connected in parallel. That produces local overheating and failure.
If you look at the safe operating area graphs, the earlier types have a graph which is square with the top right hand corner cut off. The later types have a two-slope line, similar to that of a bipolar transistor and its secondary breakdown.
There was a period of time before manufacturers admitted to the "secondary breakdown " effect, where datasheets still had the earlier graph.
You could always use the earlier MOSFETs but they cost a packet these days. To make the more modern ones work, add resistors in the source such that at maximum setting as much power as possible is dissipated in the resistors.
 

Irving

Joined Jan 30, 2016
3,181
To make the more modern ones work, add resistors in the source such that at maximum setting as much power as possible is dissipated in the resistors.
That works, but severely limits the voltage range the device is suitable for; eg if it's LiPo battery testing the losing 2v in the resistors and 0.8 - 2.2v in the MOSFET is doable. If for 4-quadrant power supply testing that can be severely limiting if you can't get near 0v at high current.

STM, Infineon and IXYS all make specialist linear devices for this purpose that have extended and guaranteed SOAs. All good commercial e-loads I've looked at use linear MOSFETs.

Even so, heatsinking is an issue with high current, low voltage, e-loads. You need to get as close to a cold wall as possible. Forced-air-cooling runs into problems above 500W or so without some exotic high-volume tunnel designs that are hard to manufacture and thus expensive, upwards of £150 per 100mm of heatsink and typically you need 200 - 300mm per pair of devices at 200W each.

I've been playing with CHT (conjugate heat transfer) simulations to look at the best layout for 4 devices on a cheap eBay heatsink that's 350mm W x 500mm L X 50mm D, so two set fin to fin inside a rectangular tube takes 3 X 120mm high-flow fans and 8 devices. Even that is limited to around 190w per device at 50A overall, 0 - 30v.
 

Ian0

Joined Aug 7, 2020
6,660
That works, but severely limits the voltage range the device is suitable for; eg if it's LiPo battery testing the losing 2v in the resistors and 0.8 - 2.2v in the MOSFET is doable. If for 4-quadrant power supply testing that can be severely limiting if you can't get near 0v at high current.
Agreed. If the TS stated the actual purpose, the I missed it somewhere along the line. The one I made was for load testing 12V lead acid batteries and dissipated 9V in resistors, with the MOSFETs attached to a CPU cooler. I remember being rather impressed by the performance of the CPU cooler in the application. I have a 3-channel 70kW per channel dummy load at work, which looks like a forest of Immersion heater elements with a 600mm fan at the end. One tries to avoid using it in the summer!
 

Janis59

Joined Aug 21, 2017
1,523
Teoretically can, practicaly is hard to. Theory says that R(th)=400/A(cm2). Thus, if may permit to alter T for say 30C then Rth must be about 0.1 thus area around 4000 cm2 or 65x65 cm. Isnt this bit large?? Other problem is that if heat impact is coming is one single point, the thermal flow may not travel so much to aside. Thus one will have there over-hot radoator center and fully cold perimeter. Thus, effective cooling is expected to be very weak - the cure is to take radiator abnormally thick, but over some 8...10...12 mm it seems nonsense. Thus, one cure is to use the fans. Plethora of computer fans have 10...20... 30 cfm thus to have the 0.1 C/W one need to use 200 in^2 at 200 cfm... sounds like a farming tractor. At more normal values 10 in^2 and 40 cfm gives 0.6 C/W Generally the fan may be calced by cfm=1.76*N(W)/ dT (C). Other example 20 cfm at 4.5x1.5x3 in (80 in^3) gives 0.58 C/W but at 10 cfm 0.77 C/W. In air without fan, 170 cm2/W will give the dT=10C, 65 give20 C, 26 give 40C, 10 give dT=70C. So, Pemntium radoators may laid slightly less than 100 W, but mre modern round type with copper core may laid 200W or with extra strong fans from both sides near 300 W. Generally, from 100W and above to kilawatt range the custo is to use water cooling. Just drill the holes in 1 cm thick aluminium and then screw the channels closed where needed. If not good, may use the heatpipe - them right cost near 100 USD but are highly effective until half kilowatt, and demand no water. More over, them are ultimate solution in cases when high power to area load is necessary (W/cm2) for semiconductor, as an example Peltier cooler platelets.
 

ronsimpson

Joined Oct 7, 2019
2,515
In data sheets there is a line about how many watts and a small footnote about "infinite heatsink". There is no such thing. This year I made one with a large block of aluminum and a cooling element and controller that kept the heat sink at 25C.

In reality we "make" one with math. By knowing that the thermal resistance is we can easily derive what will happen. Start out with a 25C heatsink, thermal resistance, and 150C silicon temperature, = how many watts.
 

Thread Starter

ballsystemlord

Joined Nov 19, 2018
102
300W air cooled with 2 such devices is relatively easy on a force-air- cooled heatsink. You can buy 250W e-loads on eBay or Amazon eg this one.
And not make my own? I wanted to make my own, in part, to learn about these things.

Regarding the attached paper, I'm quite curious now how these companies came to decide that their MOSFETs could handle X amount of watts, given that they're obviously not testing the MOSFETs using them in linear mode. In fact, the pdf noted that in order to get the specified power dissipation you actually have to use a pulse width of 1ms or less, unless I'm reading it wrongly.

Article attached.
Awesome!
I can't read both tonight. I'm leaving this one half read.

It depends on the Vgs zero-temperature coefficient point.
...
Oh, that is interesting! Thanks!

I've been playing with CHT (conjugate heat transfer) simulations to look at the best layout for 4 devices on a cheap eBay heatsink that's 350mm W x 500mm L X 50mm D, so two set fin to fin inside a rectangular tube takes 3 X 120mm high-flow fans and 8 devices. Even that is limited to around 190w per device at 50A overall, 0 - 30v.
Cool! Send me a PM (to the forum posts?) when you get it working. I'd love to see pictures.

I have a 3-channel 70kW per channel dummy load at work, which looks like a forest of Immersion heater elements with a 600mm fan at the end. One tries to avoid using it in the summer!
Can you post a picture or two?

Teoretically can, practicaly is hard to. Theory says that R(th)=400/A(cm2)....
I'll have to come back to this later when I can think and read it over several times.

In reality we "make" one with math. By knowing that the thermal resistance is we can easily derive what will happen. Start out with a 25C heatsink, thermal resistance, and 150C silicon temperature, = how many watts.
Yes, yes, but how do you "add" fins to those sorts of equations when obtaining your thermal resistance? So far, I've just been calculating total surface area, but it's not that simple.
 

ronsimpson

Joined Oct 7, 2019
2,515
Yes, yes, but
You cannot have an "infinite heatsink". I made themby adding liquid cooling and keeping the temperature at 25C when 1000 watts was added to it. An infinite heatsink does not need fins or airflow. It is stuck at room temperature no matter how much energy you add. You can't have one. It is math not reality.

After saying you can't have one, I build them. I build machines that test power transistors so we can make data sheets. We have a heatsink that can be set to any temperature from -60C to +200C and will hold that temperature even when the power transistor is trying to lift the temperature. Our heating/cooling machine is the size of a washing machine. It takes huge amounts of power. It could heat and cool your house. I am testing transistors that power the drive motors in cars and trucks. I know that before the test the heatsink is 25C and after the test it is at 25C. Most data sheets are made using math not real-world tests. But if you don't trust the math, I can prove the numbers.
IRF3207 can dissipate 300W
Do not run the part at 300W. The inside temperature is at 150C and about to explode. (the heat sink is at 25C!) This is an imposable thing. This is why they make power resistors.
 

Irving

Joined Jan 30, 2016
3,181
And not make my own? I wanted to make my own, in part, to learn about these things.
Whole-heartedly agree... was just pointing out that cost to make >> cost to buy!

Regarding the attached paper, I'm quite curious now how these companies came to decide that their MOSFETs could handle X amount of watts, given that they're obviously not testing the MOSFETs using them in linear mode. In fact, the pdf noted that in order to get the specified power dissipation you actually have to use a pulse width of 1ms or less, unless I'm reading it wrongly
Not sure which datasheet you ere looking at, but the IXYS datasheets (attached is that for the device I use) have extended SOA down to DC and that's critical for linear operation. Here are the relevant charts for the IXTX110N20L2 at case temperatures of 25 and 75degC

1668601310370.png
As already pointed out by @Ian0 & @ronsimpson the 25degC chart is of academic interest. The 75degC chart is the one to use. From that we can see the safe operating area for DC - the yellow bit (inside Rds(on) and DC lines). That suggests peak dissipation is theoretically around 580W (approx 78A x 7.5v) but that's not the full story.

From the data sheet, the thermal resistance, junction to case Rthjc for the plus247 package is 0.13degC/W, so for a case temperature of 75degC and a junction temperature of 150degC the maximum dissipation is (150 - 75)/0.13 = 576W, ie the same as the SOA chart and the SOA example in the datasheet (200v, 2.88A, the RH limiting line on the SOA chart). But is that case temperature realistic? Well, for an ambient of 25degC the thermal resistance of of the thermal paste and the heatsink together must be less than (75 - 25)/575 = 0.087degC/W. The thermal resistance of a good thermal paste, a few microns of Arctic Silver 5, is around 0.05degC/W so we are looking at a heatsink of around 0.037degC/W! Such heatsinks exist, for a price, but as always there's a catch; heatsinks are rated assuming the heat load is across the whole surface not at one point. The upshot of all this is you actually need 3 devices to hit around 550W.

Here is a CHT simulation of 8 devices on a pair of identical heatsinks, each device handling 150W for a total of 1200W with a junction temperature of around 80degC with an airflow of 7.5m/s.

1668614282087.png
 

Attachments

Ian0

Joined Aug 7, 2020
6,660
There's an interesting paper on Ixys's website saying how they achieved it, though I think I'll have to read it several times before I understand how it can be thermally stable whilst still having a negative temperature coefficient of Vgs(th).
It's a bit suspicious that they recommend them for Class AB audio amplifiers, but there are no P-channel versions.
 
Top