IGBT Temperature

[AstRii]

Hey guys, a quick question..
I've just ran my DRSSTC and after each few minute run i measured the temperature of the heatsink with IGBTs.
But now i'm thinking.. is the heatsink temperature even telling me something?
From what i understand, in DRSSTC operation, the temperature of the IGBTs only rise during the short enable pulse and then lowers again.
And it is this fast temperature rise during the enable pulse which can destroy them.
Therefore isn't it useless to measure it like this?

Thank you,
Mark.

[Max]

It's true, during an ontime the temperature will rise quickly, and decrease slowly after the ontime as the heat generated during the ontime dissipates. And you're right, this temperature peak on the IGBT die itself can't be measured by checking the heatsink.

Think of it in an electric way (if it helps). Temperature is voltage, thermal mass is capacitance, thermal resistance is electrical resistance (surprise). When electical current flows through the IGBT die, it generates heat, which you can see as current source. This source charges the parasitic (thermal) capacitance of the die. If it was completely isolated, its voltage (temperature) would continue to rise until the part is destroyed.
This does not happen because the die is connected to larger masses, each with a resistance and a capacitance. Effectively the whole system is a low pass filter. And in the end your short pulses from the IGBT die "charge" the large capacitance of the heatsink. The heatsink "voltage" would rise, too, if there was not the environment, which acts like a resistance to ground. It discharges the capacitor. The bigger the heatsinks surface, the lower is that resistance, the more power it can remove from the capacitor.

Eventually the system will reach a stable state, where the average power coming from the IGBT die equals the average power dissipated to the environment. And if you've done things right, no voltage (temperature) in the entire system is over the limit at this point.

So... is it useful to measure the temperature of the heatsink? Yes. It tells you if your cooling is sufficient. The temperature spikes on the die itself are merely affected by the size of the heatsink; it's "too far away". Edit: not quite true. See daveknis reply below. But for the average power dissipation it does matter. Even if the losses are low, they do exist. If your heatsink is too small your IGBTs will eventually overheat. Most smaller and mid-size DRSSTCs don't have this problem because their heatsinks are small enough to easily handle the average power. But if you're playing with things like pulse skipping or building a higher power system, you should definetly keep an eye on the temperatures of the heatsink - and the MMC and the wiring, and the primary, etc.

Btw. if you knew the thermal resistance between your system (heatsink, IGBTs, etc) and the environment, if you knew the thermal properties of the IGBT die itself, and because you know the duty cycle, you should be able to calculate the peak temperature of the die based on the heatsink temperature.


Kind regards,
Max

[davekni]

Heatsink temperature does matter, as that is the baseline.  The junction temperature pulses above that baseline.  The higher the baseline, the higher the pulse top.

The junction temperature rise can be estimated based on the IGBT's specification graph for transient thermal resistance and calculated power dissipation (Vce times Ic plus a small amount for switching loss at close-to-zero switching current).  This rise is on top of the heatsink baseline.  (That presumes low IGBT case to heatsink thermal resistance.)

For most DRSSTC designs, the transient (pulse) temperature rise is larger than heatsink temperature rise.  If pulse rise is say 100C, and heatsink is at 50C, then peak junction temperature is 150C.

[Mads Barnkob]

I have a calculation example in my DRSSTC guide that should give you an idea. http://kaizerpowerelectronics.dk/tesla-coils/drsstc-design-guide/igbts/

You say that the die heats up fast in a pulse and then that heat disappears... Where do you think it disappears to?

[AstRii]

You say that the die heats up fast in a pulse and then that heat disappears... Where do you think it disappears to?

I was thinking that the IGBT heats up to let's say 100C and then transfers the heat into the heatsink, which with it's larger volume only increases a few C in temperature (therefore not saying anything about the peak temperature).

My heatsink is getting just a few degrees above the ambient temperature, which i guess then means that i'm either safe with low switching losses or my thermal contact to the heatsink is poor.
I'm sure there are many ways to calculate the IGBT peak temperature, but can you actually measure somehow if your transistors are going to die? Or you just have to push them and see?

I'm slightly worried about my IGBTs, since my switching is only near zero current crossing (imperfect phase lead), for more details check my thread https://highvoltageforum.net/index.php?topic=1196.0

[davekni]

Measuring IGBT temperature can be split into two questions.  For average power dissipation (ie. whether your IGBT-to-heatsink thermal resistance is low enough), measuring temperature on the top of the IGBT case centered on the die gives a fairly close approximation.  That's for TO247 packages.  I've measured ~140C on the top of the case when die temperature was 150C.  (For IGBT bricks, I have no information on measuring temperature, other than that some bricks include an NTC thermistor in the package for that purpose.)  Measure case temperature with a thermal camera or a fine-wire thermocouple.  (Thick wires read lower temperature because heat conducts down the wires.)  Coil operation will disrupt most thermocouple meters, but look at the first good reading immediately after stopping.

For transient thermals during an enable pulse, most people rely on calculations.  However, there are a couple measurement options.  One is to measure some IGBT parameter, such as Vce at a known current, that depends on temperature.  First characterize your specific part for that parameter vs. temperature separately without a heat-sink.  Use very-low duty-cycle waveforms so that measured case temperature roughly matches die temperature.  Then measure the same parameter during operation, using the previous static measurements as a transfer function from the measured parameter to temperature.  (I've used this technique only once at home, but several times at work.)  The biggest down-side of this technique (besides complexity) is that it gives average die temperature.  It typically shows lower internal thermal impedance than listed in specifications for the IGBTs.  That's because specifications are for the die's hottest point, say on top of a small void in die-attach solder.  Keeping the average die temperature below 150C or 175C may not be sufficient to avoid damage if the hot spot is much warmer and fails.

The other option is half measurement and half calculation, perhaps the best choice in your case.  Leave the IGBTs in their normal mounting with fan running.  Disconnect from GDT and run a known current and voltage to the IBGTs.  (Something like a zener from gate-to-collector and a current-controlled power supply.)  Measure heatsink and IGBT case temperature rise.  To first-order, temperature rise will be linear with power.  Calculate the power required to match the measured temperatures during operation.  Then scale that average power by 1/duty-cycle to get power during pulses.  Use that pulse power with the data-sheet's transient thermal resistance graph to calculate IGBT internal temperature rise.

[Weston]

Another important thing to consider is that the magnitude of the delta T matters, not just the temperature. Temperature swings stress the attachment between the die and the heat spreader and can cause failure. At great enough temperature swings it can only take a few thousand or tens of thousands of temperature cycles.

 I suspect that this mechanism contributes to a lot of the latent failures seen DRSSTCs

[Hydron]

Delta-T cycling also affects the wire bonding I believe.

From my rough calculations delta-T becomes problematically large far before you hit Tjmax, which has driven quite a bit of my QCW design.

I believe that IGBTs used in traction applications (e.g. metro trains) also die (or more likely are preemptively replaced) due to this too - they go like the clappers every time the vehicle accelerates then cycle back to a much lower power state when running at a constant speed or stopped. Especially on a metro system running all day with a short time between stops this adds thermal cycles pretty quick!