TO263 measuring junction temperature

[multipla]

I am a bit confused about thermal datasheet values for TO263 MOSFETs, for example of Infineons CoolSiC Gen1: Datasheet

What exactly is the "Case" temperature of a TO263 MOSFET, is it the drain tab or is it the black plastic? In this App Note it says:

Heat Sink packages are the first group. The heat sink (chip carrier - lead frame) is soldered directly to the PCB. The thermal resistance of this packages between chip and heat sink is called Rthj-c (junction-case) and has low values.

And from it I understand that Rthj-c is the thermal resistance from the junction to the drain pad (heat sink directly soldered to the pcb).
That would also make sense to me, since the plastic mold is very thick and a bad thermal conductor. The thermal resistance from the junction to the drain pad is much lower than from junction to the plastic mold (is this the Rthj-a from Infineons datasheet then?)

I am designing a PSFB with these MOSFETs mounted on an aluminium IMS PCB mounted directly on a big heat sink, so the drain pad is coupled very well to the main heat sink.
From these points I would conclude that, since almost no energy (Rthj-c/Rthj-a is about 1/100) dissipates through the molding, the temperature of the plastic mold is very close to the junction temperature?
Is this correct?

[multipla]

Hmm, on the other hand, I just drew attached pic. The molding temperature only approaches junction closely when Rth(m-a) ist very high compared to Rth(j-m). Is this the case?
Rth(m-a) is only convection from molding to ambient, Rth(j-m) is conduction from junction to molding. Is it ok to assume that the convection part has a much higher thermal resistivity?

[AnalogTodd]

Your thinking on the thermals is correct. The plastic is a horrible conductor of heat, which is why there is that big metal tab on the backside. If you measure your case temperature at that point, you are going to get the most accurate measurement in terms of what your junction temperature is when calculating with the Rthj-c number.

Your best heat dissipation is through that path.

[mtwieg]

That value in the datasheet is definitely referencing the drain tab, not the plastic. The thermal impedance to the top of the plastic package is probably so high that it's negligible.

As for estimating the junction temperature based on the plastic temperature... hard to say. I would try an experiment where you dissipate a known amount of power in the FET on a test PCB. Measure the temperatures of the drain tab and plastic. Estimate the actual junction temperature using the thermal impedance in the datasheet and the drain temperature. Compare with the plastic temperature.

[multipla]

Thanks to both of you.
Actually that measurement would be easy to do yes, I will do it and report back, maybe it's interesting for someone else too.

[Wolfram]

A lot of Rth_jc is through the thickness of the transistor die itself. Therefore the top of the die will be significantly hotter than the copper leadframe when the part is dissipating power. The top of the package will be somewhere between these two temperatures

[multipla]

[...] the top of the die will be significantly hotter than the copper leadframe [...]

Yes, around Ploss * (Rth(j-c) || (Rth(j-m) + Rth(m-a)) hotter...
At least its average/hot spot/critical temperature? It also won't be uniform across the die surface.

[AnalogTodd]

When you're talking about a big MOSFET, you're probably going to have the hottest spot in the center of the die, but it should be relatively close to uniform until you get out towards the edges. Think of it from a cross section as a trapezoid with 45 degree sides with the widest part at the bottom of the die and the narrowest at the top (that's how most places calculate thermal resistance for a die).

A big MOSFET is going to be relatively uniform across its surface, if it isn't you can run into hot-spotting and thermal runaway.

[multipla]

The die is roughly 2 mm by 2 mm, I accidentaly split one case while trying to desolder one some time ago.

Regarding the measurements:
I sent 8,000 A through the body diode, which resulted in a voltage drop of 3,6064 V after having run for over an hour, which results in a power dissipation of 28,85 W.
The MOSFET is soldered to an aluminium IMS PCB, which is mounted with good thermal grease to a heatsink with a fan.
It was quite difficult to measure the drain tab temperature and I'm still uncertain if the measurement was accurate. Anyhow, I scratched on it a bit and painted it black so it doesn't reflect as much, then I used something reflective which doesn't heat up too fast (a brushed aluminium plate) to mask the body, so the IR camera "doesn't" see the body. It's not perfectly accurate, but I got following temps:
Ambient: 24,5 °C
heat sink and alu PCB: 31,0 °C
drain pad: 71,2 °C
plastic molding: 88,2 °C
(see also attached pic)

Also, I don't know how the heat dissipation splits between the paths (drain-heat sink-ambient) and (molding-ambient).
Assuming 90% leaves through the drain pad and I measured the drain pad temperature accurately, that would put the junction temperature between 92,5 °C and 99,8 °C (Rth(j-c) is 0,82 K/W (typ) or 1,1 K/W (max)), 4,3 K to 11,6 K higher than the plastic molding temperature.

With the 10/90 split, it would put Rth(m-a) to 22,08 K/W and Rth(j-m) between 1,49 K/W and 4,02 K/W...

I wouldn't have expected the drain to be so much hotter than the heatsink, I thought the soldered connection would cost only a few degrees, not 40 K. But all in all it adds up not too badly.

[mtwieg]

I sent 8,000 A through the body diode, which resulted in a voltage drop of 3,6064 V after having run for over an hour, which results in a power dissipation of 28,85 W.
The MOSFET is soldered to an aluminium IMS PCB, which is mounted with good thermal grease to a heatsink with a fan.
It was quite difficult to measure the drain tab temperature and I'm still uncertain if the measurement was accurate. Anyhow, I scratched on it a bit and painted it black so it doesn't reflect as much, then I used something reflective which doesn't heat up too fast (a brushed aluminium plate) to mask the body, so the IR camera "doesn't" see the body. It's not perfectly accurate, but I got following temps:
Ambient: 24,5 °C
heat sink and alu PCB: 31,0 °C
drain pad: 71,2 °C
plastic molding: 88,2 °C
(see also attached pic)

All looks reasonable to me. Coincidentally, I ran across this video from Nexperia which seems to agree with your results (Tj about 5-10% higher than the top case temperature):

If you're so inclined, you could try measuring the junction temperature more directly by exploiting the known temperature dependence of the diode forward voltage or Rdson. This requires careful characterization of the parasitic resistances to give valid results though.

I wouldn't have expected the drain to be so much hotter than the heatsink, I thought the soldered connection would cost only a few degrees, not 40 K. But all in all it adds up not too badly.

You didn't really explain how the PCB+heatsink are set up for these experiments. I'm assuming the drain tab is soldered to the copper foil, not the aluminum substrate. I'm betting a large portion of that thermal impedance is due to the dielectric between the two. Also your results will depend on where you measure the heatsink temperature.

[multipla]

Thanks for the video, yeah it seems to say the same thing. However, what always makes me think, what is "10%" more of a temperature? 100 °C + 10°C (110 °C)? 212 °F + 21,2 °F (111,8 °C)? 373 K + 37,3 K (137,15 °C)?
For me it only makes sense using absolute temperatures, but I'm sure in the video he means degree Celsius. Whatvever.

I described the setup briefly above, but more detailed:
The aluminium IMS PCB consists of an aluminium substrate (1,5 mm thick), an isolating foil above (100 μm thick, ca. 2 W/(m*K)) and then a 35 μm copper foil, which the drain pad is soldered to.
The PCB is screwed to a heat sink (90 mm x 90 mm x 150 mm) with a layer of thermal grease inbetween, a fan is blowing air through it (air flow is around 78 m^3/h).
The heat sink temperatures were all within a Kelvin, both the big heat sink where the PCB is mounted and also the top side of the PCB itself.

I'm okay with the measurements though, I'll assume in worst case the junction temperature will be around 40 K higher than the plastic molding (if it doesn't get any direct air flow on it), should be good enough (TM).