current sink - sudden MOSFET death

[jmibk]

I'm working on a controllable current sink - see schematic at the following link:

https://audiowerk.at/downloads/temp/eev/current_sink_error/Schematic_2022-02-17.pdf

The system should have a range between 0 and 100A, max voltage 5V. It is based on a simple current sink with an opamp, a mosfet and a current shunt.
The voltage of the current shunt is amplified by 36 by the differntial amplifier and fed into an 16 bit adc.
An arduino based mcu reads the voltage and calculates the current through the shunt. In combination with a given current value (external pot) it
regulates a 16bit pwm to control the opamp that is enganging the load mosfet.

I tried the complete system with the TSD4M250F mosfet I had lying around - and it worked nearly perfectly up to 100Amps for a longer timespan (10 minutes).
Heatsinking was made with a big cpu cooler, temperature of the heatsink below the mosfet was never larger than 55°C. Source was a pack of 9pcs 21700 cells in parallel to get the current of 100 amps without any significant voltage drop.

Heres the datasheet of that mos device:

https://audiowerk.at/downloads/temp/eev/current_sink_error/TSD4M250F.pdf

After that I bought some different mos devices, because the TSD4M250F isn't available and it is on the edge of its rated data.
I found the IXTN660N04T4 that I ordered.
It also worked perfectly with lower currents up to 20 amps (coming from an power supply). After that I connected the current sink to the battery pack again and ramped up the current, approx 5amps per second. At around 80A the mos device goes boom and the magic smoke escaped from it. Measurements say, that is has 12Ohms between drain and source and 1 ohms between gate and (I think it was) source.

Heres the datasheet of the new mos device:
https://audiowerk.at/downloads/temp/eev/current_sink_error/IXTN660N04T4.PDF

I measured the circuit, there is no ringing or oscillating that can cause the problem. The sink current follows the setpoint directly and precise without any overshoot.

In the schematic there is also a second mosfet below the shunt - this one is for emergency shutdown if the main mosfet fails. This device isn't in circuit at the moment. Also the shutdown logic (schematic last page) is on the board but not connected to the circuit or managed by the software.

Maximum values in the circuit:
UDS = 5V, part has 40V
IDS = 100A, part has 200A (lead current limit)
PD = 500W, part has over 1000W
SOA gives 100A at 5V in the DC range, thats ok according to the chart Fig.12.
Switching Times doesn't matter, because the sink is controlled by an analog voltage (out of the opamp) and not via PWM.

Do you have any hint, what can cause that fault? The IXTN660N04T4 should be a much overrated part for the task than the TSD4M250F.
Thanks for any help and hints.

[PartialDischarge]

Do you have any hint, what can cause that fault? The IXTN660N04T4 should be a much overrated part for the task than the TSD4M250F.

Die temperature. That kind of power needs to be distributed in various mosfets, the specs are under very optimal conditions that you are never going to achieve.

Look at teardowns of Siglent or Rigol loads, you'll see massive heat sinks and lots of paralleled mosfets. The SOA is for pulses, I wouldn't trust it for DC.

[Ian.M]

Your first link (schematic) is borked - I doubt you meant to show us a 5V output buck converter!

SOA curves are almost invariably given for a mounting surface (or case) temperature of 25 deg C.   Did you derate for operation at 55 deg C?

Selecting a temperature for derating can be a bit of a chicken and egg situation as thermal modelling of real-world heatsink assemblies is often inexact due to lack of data so you cant be certain of the maximum temperature rise till you soak-test the device at full power, which is potentially destructive as you have found. To escape this trap, you may be forced to do a temperature rise vs dissipation test on the actual heatsink assembly e.g. with a metal case power resistor on a copper heat spreader with the same mounting surface area (and general footprint) as the semiconductor you want to use.  The sides and top of the resistor + heat spreader must be thermally insulated so virtually all the heat flows through the footprint.   Then there's the ambient temperature issue: Do you design to worst case (100 year heatwave, no aircon) climate data, or is your product strictly for use in a controlled temperature 'shirtsleeve' environment?  Also how much should you allow for local ambient increase e.g. inside a case or in a crowded rack?

Consider implementing an over-temperature shutdown so you actually have a max. heatsink temperature you can safely design to.

Edit: as Wolfram points out below, the mounting surface temperature can significantly exceed the heatsink temperature.  If trying to determine it experimentally with the suggested power resistor + copper block you need to measure the temperature of the copper block at the center of the footprint, which may be difficult to achieve for smaller footprints. 

[Wolfram]

500 W dissipation in a SOT-227 is optimistic, and requires a lot of care to get right.

Rth_jc for the IXYS device is 0.144 k/w. They specify Rth_cs as 0.05 k/w, but this can be hard to reach in practice, maybe if you lap the heatsink and use a very good thermal compound. Realistic figures are in the 0.1 to 0.15 range, potentially more if your heatsink flatness and surface finish are not ideal. Let's say this gives you a total Rth_js of 0.25 k/w. At 500 W, this gives you a delta-T of 125 degrees between the die and the heatsink surface right under the part. The heatsink can also have a significant internal thermal resistance, so the temperature right under the device can be significantly higher than the rest of the sink. This puts you above the maximum junction temperature of the device for any practical heatsink temperature. Also consider the impact of ambient temperature. , which can further reduce your thermal budget.

To reach 500 W in a SOT-227, I'd go for a lapped copper heat spreader at least 10 mm thick, and a very solid heatsink. To reach the advertised 1000 W, the whole package likely needs to be immersed in nucleated boiling freon.  Splitting the dissipation across multiple devices is likely cheaper and easier, and allows you to add margins to account for transients, high ambient temperature and so on. 250 W in a SOT-227, 75 W in a TO-247 or 30 W in a TO-220 are the upper limits I usually budget for, beyond this the cooling effort tends to be more expensive than just adding more devices, though this does vary a bit depending on the application.

[fcb]

The last big CC load I built was 1kW.  We ended up with 8 parallel MOSFETS, each with their own heatsunk source resistor (we sampled the control current current from one resistor only).

We used one of those expensive fan-tunnel heatsinks - that kept all the MOSFET's at close to the same temperature and they shared the load fairly evenly (Vgs was fairly well matched), my biggest concern was linearity as we were only controlling the system from one source resistor - it was more than linear for our needs, but I doubt would meet your requirements without a different arrangement to close the loop.

When it comes to Pdisp, treat MOSFET datasheets as lies. Only under the most perfect conditions is your -redacted- MOSFET going to survive 350W...

[Vovk_Z]

Heatsinking was made with a big cpu cooler, temperature of the heatsink below the mosfet was never larger than 55°C.

Use two MOSFETS instead of one, and I guess it'll be fine inside SOA (it was said by many about 25C rated temperature of a case in the SOA). Of cause, with two current shunts and two opamps.
I attached an example how to parallel it right.

[MarkF]

Lots of MOSFETs for that kind of power. 
The HP6060 Electronic Load uses eight MOSFETs.

[jmelson]

I'm guessing that the MOSFET that failed was Q1 on the 3rd schematic page?  What limits current?  You are using PWM, but running the power transistor in linear mode.  100 A?  Looking at the safe operating area curve, you are olny allowed 4V D-S at 100 A.  No idea what your battery voltage is, but I can see why things went boom.  Modern MOSFETs (except those specifically designed for audio amp use) are designed for hard switching, and not operating in the linear mode.
Your design is REALLY pushing things in the wrong direction.
Jon

[jmibk]

Thanks for all replies till now!

The thing with the soa should fit the purpose (not meaning other specs are the false decision!). The operating area of the circuit is marked red.

TSD4M250F


IXTN660N04T4



A collegue wants to caracterize the temperature thing here.

I like to use multiple devices in parallel. I got a lot of boards with IGBTs on it - its the IRG4PC50S device. They are not the optimal choice but they are here and I got them free of charge - 80pcs.
Here is the datasheet:
https://audiowerk.at/downloads/temp/eev/current_sink_error/IRG4PC50S-p.pdf

What do you think about using 6 to 8 of this IGBTs in parallel for that purpose?

[Gyro]

Your highlighting on the SOA graphs nicely demonstrates the pitfalls and traps in datasheets...

Looking at the TSDM4M250F SOA, your used area is clearly on the bleeding edge of the DC curve. The manufacturer hasn't included the Tj on the graph (as they have on the lower one). Looking earlier in the datasheet, the device has an Rthj-case (thermal resistance, junction to case) of 0.25'C/W. At 5V and 100A this would give you a junction temperature rise of 125'C before you even get to case temperature. It also states an absolute maximum junction temperature of 150'C. This implies an absolute maximum permissible case temperature (at it's mounting face) of 20'C. That's before you consider the thermal resistance of the mounting to the heatsink and the thermal resistance of the heatsink itself. 10 minutes sounds about right for lifetime.

Now let's look at the IRG4PC50S. Your highlighting shows that you are more comfortably within the DC SOA. This time the manufacturer has unhelpfully put on the temperature conditions... Tj=175'C and Tc of 20'C. This again demonstrates how huge the thermal resistance is between the junction and the case, and how low the case temperature needs to be to meet the SOA conditions. Looking back at the beginning of the datasheet, we see that this device has a thermal resistance, Junction - Case of 0.64'C/W, much higher than the previous device. You can also see that, with this being a smaller plastic device, when perfectly mounted to flat heatsink surface, it has a Case to Heatsink thermal resistance of 0.24'C/W (note Typ!), again, before you get to the heatsink thermal resistance. A reallty big, and not really forgivable trap is that they put Tj of 175'C on the SOA graph when, if you look at the beginning of the datasheet, Tj max is shown as 150'C Absolute maximum.

Without going into how many devices you need to parallel, these figures demonstrate how easy it is to be misled by the datasheet figures and graphs. It shows how, although you though to were within the SOA curve on the Mosfet, you were in fact way outside.

Hopefully this will help you to decide how many devices you actually need to parallel, and also the required heatsink thermal resistance and device mounting issues (for instance, insulating washers will significantly affect mounting thermal resistance. You also need to decide how much you want to de-rate the devices. All of the figures above relate to the Absolute Maximum ratings. For long term reliability, I would de-rate by at least 25%, maybe 50% in this case to account for some unknowns like exact thermal resistance in mounting (heatsink surface finish, knowing the exact thermal resistance of the heatsink, ambient temperature etc.).

[Vovk_Z]

I like to use multiple devices in parallel. I got a lot of boards with IGBTs on it - its the IRG4PC50S device. They are not the optimal choice but they are here and I got them free of charge - 80pcs.
...
What do you think about using 6 to 8 of this IGBTs in parallel for that purpose?

It should work with IGBT too (despite they have larger minimal collector-emitter voltage). Typical TO-247AC package can work with up to 80W watts dissipated continuously (at 'average' ambient temperature). Having this number you may calculate needed number of packages. Or you may calculate how much you are safe with 6-8 packages.
6x80 = 480 W, and 8x80 = 640 W continuous.
So, if you need to dissipate 500W then 6 TO-247 cases possibly is ok (that's 83 W/case). But 8 cases is better, because of possible uneven current distribution between transistors.

[CaptDon]

What was the voltage across your 'single' mosfet at 100 amps?? V x I = W  At your 100 amp current and voltage drop the sq/mm of a single die will never survive, for Christ sake it was probably the same heat per square or cubic millimeter as a welding electrode. As said by others, more mosfets and lots of fan cooled heatsinking. Don't really know how you expected it to not go boom.

[jmibk]

What was the voltage across your 'single' mosfet at 100 amps??

It was a 1S9P lithium battery nearly fulled charged at 4,1V at the beginning.