estimating junction temperature of a power mosfet

er

r s

giciel antivirus Avast.

You know how much power it's dissipating, does the spec sheet give the ther mal resistance between the junction and the case? It's hard to believe tha t the top would be much hotter than the tab on the back side. You could tr y sneaking a little thermal couple in there... maybe on the drain lead (as you suggest.)

George H.

...

You could use the parasitic diode in the MOSFET to measure the junction temperature. You would first need to measure the forward voltage drop of the diode at some convenient and fairly low current with the device in an oven at various temperatures.

This would give you a calibration curve for the diode. Then use the device as a load with your heat sink until the temperature is stable. Finally, switch the device to your low-current source and very quickly (this probably needs to be within a few milliseconds) measure the parasitic diode forward voltage again. You will need a relay in order to switch fast enough.

This measurement can then be used to calibrate one of your other methods.

John

For a beast like that, the peak Tt is probably pretty close to Tj.

The classic technique to measure Tj is to temporarily (and very quickly) disconnect the fet from the circuit and use the substrate diode as a thermometer. That is sort of a chore.

I like to test the fets to destruction, to see how much margin we actually have.

You can't in general trust the heatsink's rated Tsa. The fet has a relatively small footprint, so there will be a hot spot there. Heatsinks are usually spec'd with a uniform heat load spread over the surface. Your thermal imager may have enough resolution to see the temperature gradients across the surface of the heatsink. People sometimes drill a tiny hole under the fet, to get a small thermocouple right up under the mosfet.

--

John Larkin         Highland Technology, Inc 
picosecond timing   precision measurement  

jlarkin att highlandtechnology dott com 
http://www.highlandtechnology.com

I like it! You could make a few measurements (in time) and extrapolate back to when the power was turned off... assuming a linear temperature decrease or something.

George H.

This is a steady-state circuit? You can measure heat by comparing the air inlet and outlet temperature (minor corrections with a barometer and calibration for airflow might be helpful). Then, make a few interruptions and check the substrate diode drop to get the true junction temperature after long operation (few minutes). That will give you a true value of 'thermal resistance' through case, grease, heatsink, to the cooling airflow.

If you need to improve the airflow or thermal resistance, or can benefit from redesigning the cooling hardware, this is the info that's required.

also IR readings can be in error due to the emmisivity of the material.

try using a small thermocouple

they make very small fine wire thermocouples that you can attach with a drop of glue, works well..

Mark

The epoxy used on such a package very likely has a very high emissivity, and good, well documented thermal conductivity data.

The rest can be extrapolated by how thick it is... & how far "up" from the baseplate the actual device inside is mounted, that internal mount's thermal coupling efficiency, etc.

Experiments could be done with a non-sink mounted device, in a very controlled, slow work application, and observe rates with IR, and collect data. Even though non-real, they can provide insight into thermal efficiency of the entire device's construction technique, which may differ from maker to maker, and from design to design.

But properly examined IR data can certainly be made useful, even on a real world sink mounted scenario.

Black whiteboard marker, or a bit of kapton tape, will get the emissivity close to 1.

--

John Larkin         Highland Technology, Inc 
picosecond timing   precision measurement  

jlarkin att highlandtechnology dott com 
http://www.highlandtechnology.com

You could try sneaking a little thermal couple in there...

--
Thermocouple. :) 

John Fields

Is it covered by the heatsink ?

I have seen some TO-3 transistors, with a small hole through the heatsink exposing the base of the transistor. A thermocouple or some IR measurement equipment was then used.

Of course, the hole reduces heat conductivity from case to heatsink, but at least it gives the worst case estimate of Tj.

Hello Thank you for your answers. I can give some more details on my application. The heatsink is a sort of tunnel filled with hollow fins, and there are two fans to force air flow through the fins. (the model is Fisher LAV7 for those who know this manufacturer). So it is not easy at all to put a thermocouple against the base of the MOSFET, through the heatsink. The idea of using the parasitic diode of the junction to measure temperature is very interesting, but the main problem is calibrating. I suppose that calibration must be done with THE mosfet of my device, and not another one. So I should unsolder the mosfet to put it in an oven as other parts of my device cannot support high temperatures. While not unfeasible, it's a little tricky... For upsidedown, the plastic of the TO264 package covers totally the metal tab, so once mounted on the heatsink, no metal is visible, and here is my problem ! I can measure (with the IR camera) the temperature of the screw, but this is closer to Ts than to Tc.

Here are results of my experiments: The mosfet dissipates 150W

some black plastic tape on the heat sink gives an emissivity between

0.9 and 1 the camera was calibrated for 0.95 as John said, Ts decreases as the distance from the mosfet increases

The datasheet gives theta-cs = 0.15 typical and theta-jc = 0.12 max. I take theta-cs = 0.2 as 0.15 is typical value.

mosfet) The emissivity of the black plastic of the mosfet case is supposed to be close to 0.95

There are two possible interpretations: - either Tt is close to Tj, and everything is ok - or Tt is close to Tc (and then theta-cs is around 0.3) : the future is bleak for my mosfet

I experimented with a reduced power, but I want the mosfet dissipate at least 200W. Measures with dissipated power = 100W give coherent results (measured Tt close to calculated Tj).

The point is the difference between Tt and Tj, and it's difficult to get an idea on the subject (John vs George...).

packages (power-pack,...) using the PCB as heatsink. But may I extrapolate for TO264 on a large heatsink? Till now Ixys didn't answer my questions. I understand that a parameter as Psi-jt is more difficult to garantee for a TO267 package as it depends on the characteristics of the heatsink.

--
Alain Coste

That's the way to do it. A DVM, on its "diode" range, can do the measurement, but you'd have to calibrate the fet, or at least one from the same batch. Close enough.

It often makes sense to use more fets, spread out over the heatsink surface, especially if the baseplate part of the heat sink is thin, namely has high thermal spreading resistance.

This uses copper heat spreaders to transfer the heat into the aluminum sink.

If the hot spot Tt is only 100C at 150W, you should be fine.

It would be fun to somehow remove the epoxy and image the actual chip, to see its temperature profile.

I removed the epoxy from a bunch of mosfets, but the process was sort of violent.

I think there are organic epoxy removers that might not trash the silicon.

It could be measured, for a given fet, but it would be an all-day project.

We have found the Ixys mosfets to be good for surviving linear-mode high-dissipation pulses, out in the northeast corner of the SOAR curve. We blew up a lot of fets to learn that.

--

John Larkin         Highland Technology, Inc 
picosecond timing   precision measurement  

jlarkin att highlandtechnology dott com 
http://www.highlandtechnology.com

wo

ure

e.

tle

,

en

es

f the

re

t

an

s

giciel antivirus Avast.

Thanks for the update. At ~$25 a piece for the FETs, I would try pushing o ne and see where it fails. That could give you some measure of comfort.

I measured the diode drop on ~10 IRF820's in my part box and they were all within 1 mV of each other. (Not bad) (Don't touch with hands while measuri ng.)

George H.

That's impressive, less than 0.5 C.

--

John Larkin         Highland Technology, Inc 
picosecond timing   precision measurement  

jlarkin att highlandtechnology dott com 
http://www.highlandtechnology.com

Solder a test transistor to a piece of copper block, big enough to attach a thermocouple to. Clamp the transistor-block asssmbly as normal, it can be just a couple mm thicker. The block should be at almost the exact temperature of the transistor metal heat conductor, and the junction will be just a little hotter than that. This should get you the info you need, unless I misunderstand the problem.

The soldering between transistor and copper block needs to be a full-area joint, but tinning both parts and then collapsing the solder while molten until close contact is made should do it.

Jon

Alternately, run the fet in free air, or even sitting on a bit of styrafoam or bubble wrap. Dissipate a watt or two and compare the top and bottom temps. The bottom will be Tj, and the top is, well, Ttop.

--

John Larkin         Highland Technology, Inc 
picosecond timing   precision measurement  

jlarkin att highlandtechnology dott com 
http://www.highlandtechnology.com

Thank you John for the interesting information. The use of mosfets in linea r mode is not very common, and it's more difficult to find data than for switch mode.

Now I see what to use _more_ fets means... For my electronic load I could have used more transistors, but this increases the number of current sense resistors and operational amplifiers

to control them. For the power I wanted (400 .. 420W), I thought that two mosfets was a good compromise.

If I limit the power to 300W (two mosfets in //) I have enough confidence that Tj max will not be exceeded. But when I designed the electronic load I had counted on around 400W. Well, 300W are not so bad for my needs, but I would like to know more precisely "how far I can go too far".

Now I see what _violent_ means... Considering the price of my fets, I didn't really want to try it...

Yes, my selection of Ixyx mosfets owes nothing to chance. If you want at th e same time very low theta-jc and guaranteed SOA for DC (and not only for switch mode) the choice is rather limited. International Rectifier and Infineon had some fets with guaranteed SOA for DC, but for theta-jc Ixys wa s the best. The counterpart is the price, which doesn't encourage to destructive tests...

--


http://www.avast.com

e two

rature

at

one.

my

little

tal

my

rew,

tween

eases

x.

e of the

be

uture

t

d Tt

et an

r as

on

logiciel antivirus Avast.

g one and see where it fails. That could give you some measure of comfort.

ll within 1 mV of each other. (Not bad) (Don't touch with hands while meas uring.)

Yeah the diodes (transistors) I use for temp sensors have ~5-10 mV of variation at room temp. Hey, I've got a dewar of LN2 next to me.. I'll dunk some in and see how well they track each other.

later... George H.

Given that heat sinks and huge mosfets are expensive, and opamps and resistors are cheap, it often makes sense to use a lot of small fets, to spread out the heat. And dump as much of the power as possible into resistors.

A switchmode load box would be interesting. Fets switch and dump the power into a big resistor.

--

John Larkin         Highland Technology, Inc 
picosecond timing   laser drivers and controllers 

jlarkin att highlandtechnology dott com 
http://www.highlandtechnology.com