IGBT thermal questions

[mk_]

Hello Forum



we do have some troubles with understanding whats happens thermal in an DCDC-Konverter running with some power (25kW).
So here are some details and maybe someone can help us understanding a little bit:

The Primaryside of the converter (H-Bridge) switches something around +/-60A into the transformer with a short deadtime when under full power.

The IGBTs should handle this without any problem if cooling is well done. Our watercooler (surface 1mm nearby the IGBTs) goes up to 40°C so this should be fine and well within the limits required for good cooling. CE-Sat ist something around 2,2V (measured) so powerlosses (per IGBT) are below 70W, with free-wheeling-diodes and snubber-stuff we calculated 400W Ptot...

The system is running fine some time but:
what we see is at full power that part of the IGBTs get really hot, please see FLIR-Image,  SP5 (126°C).  Other parts of the housing are fine (SP1, 3, 4) and the cooler beside is (a little bit to low) shown with 26°C

Flir is calibrated to the IGBT-housing (not to the cooler), so temperatures are mor or less within +/-1°C.

Birdge_lowside_xxx shows the thermal IGBT with Rogowski mounted. The Rogowski was moved a little bit out of the line of sight in the FLIR-image. red and black wire are used for measuring CE-Voltage.

We assume that these hotspots are the area of the bondingwires and not the silicon-temperature. But we dont know exactly and what makes us wonder is - it`s only 60A, the device (IXYK140N90C3) is specified with 140A @ 110°C and even terminal current limit is specified with 160A....

Electrical isolation to the cooler is done with ceramic-pads with 25W/mK (meter-Kelvin, not milli-Kelvin) or 0,08°C/W.

So our question is - does someone have an other explanation how this high temperature is generated and - any ideas how this high temerature do age the IGBTs housing?

Thanks for reading and maybe some usefull ideas.

regards

Michael

[MagicSmoker]

Seems like you are ignoring switching losses (Eon & Eoff specified in mJ per switching event in the datasheet). Both losses scale with voltage, current and temperature (see Fig. 13 in the datasheet), so adjust as necessary before multiplying by frequency to calculate switching loss.

For example, at 60A, 450V and 25C Eon and Eoff are both ~2.3mJ, so ~4.6mJ of energy is lost per full switching period. At 20kHz that is 92W, which exceeds the ~70W contribution from conduction losses.

You are going to need to double or triple up on IGBTs per bridge leg, and I would also suggest using ones with co-packaged anti-parallel diodes, especially if this is phase-shifted full bridge, to minimize the transfer time of freewheeling or reactive currents from IGBT to diode due to stray inductance.

[dom0]

Besides other factors adding to the thermal losses, make sure that the packages have good contact with the heat sink everywhere, not just where the screw is (screw there <=> no die there). While TO-247 and TO-264 are not as susceptible as TO-220 to this it still happens, just to a lesser degree.

[station240]

For that sort of power (25kW) you should be using half bridge IGBT modules instead.
Much bigger surface area in contact with the heatsink, easier to wire than multiple TO264 packages

As you're switching an inductive load, there are also switch on/off surges.

[mk_]

Well, thanks a lot for suggestions...

Powerlosses while switching - yes,  forgotten because focused on hotspots...

Anyway - as far as I understand the flir-images the hotspots are not from switchicnglosses.. so the question still remains - why does the housing around the bondingwires gets so hot compared with the other part of the housing? And is ther knowledge here in this forum to answer the question independent from IXYS: how does this affect lifetime of the housing?

External freewheeling seems no problem at all, we measured this and everything is well within specs, even at 600V DC Inputvoltage..., distance Diodes<>IGBT is about 2,5cm, no problem at all.
btw: Snubber works also fine.

Halfbridgemodules - as suggested - are at least at the moment not possible, maybe we have to switch over but at the moment it is not possible.

(Any suggestions to a halfbridgemodule capable 120A/900V/20-50kHz switchingfrequency?)

Contact to the Coolingdevice seems ok, maybe we try ALN instead on AL2O3 but 0,08C°/W should be fine if we cann keep the Cooler cool...

regards

Michael

[Phoenix]

You need better mounting to the heatsink, or to use a module.

Putting a screw in the top will not make good contact from IGBT to heatsink (also more difficult to insulate properly). You need to make sure the IGBT is very flat and firm against the heatsink.

E.g. like this http://uk.rs-online.com/web/p/heatsink-mounting-accessories/5040592/
But also with an additional thin layer of insulating material (say, mylar) between the clip and IGBT.

May like this too http://au.rs-online.com/web/p/heatsink-mounting-accessories/1311238/

[MagicSmoker]

Anyway - as far as I understand the flir-images the hotspots are not from switchicnglosses.. so the question still remains - why does the housing around the bondingwires gets so hot compared with the other part of the housing? And is ther knowledge here in this forum to answer the question independent from IXYS: how does this affect lifetime of the housing?

How can thermal imaging differentiate between conduction and switching losses???

As for thermal imaging indicating that the emitter bondwires are the hottest part of the semiconductor device, they will have the highest current density and the least-effective heatsinking (from being encased in epoxy), so this is to be expected.


edit: imagine -> imaging

[dom0]

Contact to the Coolingdevice seems ok, maybe we try ALN instead on AL2O3 but 0,08C°/W should be fine if we cann keep the Cooler cool...

Phoenix basically said this above, the issue is not the insulator, but package bending. TO-220 is extremely susceptible to this, TO-247 and TO-264 not as much, but they still have it. Beside the spring holders Phoenix linked to there are other options, like using a U shaped metal bar. Bolt it to the heatsink between every device to have good contact and avoid issues due to the different height of the individual packages (+-0.1 mm package thickness tolerance or so). That will have better thermal contact, because the force acts on the package right where the die is and not beside the die.

[max_torque]

Paint everything matt black.  Retest, then get back to us!!

[mk_]

just a short update:

these rs-parts (http://uk.rs-online.com/web/p/heatsink-mounting-accessories/5040592/) didn`t change the situation very much, temperatures are more or less the same but we will keep them for other reasons.

We finally tried a better heat-distribution between IGBT and Insulator and some other tricks.

This helped a lot - even at 105% Overload (26,5kW Output), temperature went down about 10°C which is much better then expected and well within the Specs of the IGBT

thanks for the help

Michael

[krish2487]

Pretty much as others have said.
Plus
Surface area of the IGBT. TO220 and TO247 are not ideal for that kind of power delivery. It can be done but it just feels icky..


Half bridge modules are built to sustain pulsed overloads/shortcircuits for short duration.


Any specific reason why you are restraining from using one of them?


Fuji IGBTS are good. Look for the 150A versions. They are better than mitsubishi in ease of driving but slower than semikron in switching.
Semikron are good if you want fast switching and/or have low dead time between the upper and lower halves of the half bridge.


But mind you, the switching speed comes with a rider in the form of a beefier drive. Not easy to manage at high frequencies and power levels.
(Leads to higher switching losses).


Incidentally, a lot depends on the transformer design and the DC bus  too. We routinely deliver 20 KW of power in a 100A 600V IGBT half brick without the IGBT not even crossing 50C under full load.