Harvesthor wrote:
> My thoughts are that the gates of the MOSFETs are not charged/discharged
> fast enough.

That's right.

> The gate charge in my case is 50nC and the gate current is
> limited by a 100ohm resistor down to a maximum of 120mA (driver can
> handle 130mA).

The driver can handle 130mA DC - the pulsed current may be much 
higher, and without any problem.

> This should not be a problem for 31kHz, does it?

It does. Try gate resistors of 22 ohms or less, then it should work 
right.

> My thoughts are that the gates of the MOSFETs are
> not charged/discharged fast enough.

Yes, I would also check that at first.

The Gate has a capacity of about 1.5nF, the Gate resistor in 100 Ohm. So 
the charging time would be a somewhat more than 750ns (R*C*5).

31kHz frequency has a pulse width of 16µs.

So gate charging time should be a very small fraction of that 16µs which 
is the case. Also 4A is not that much.

It seems that you have an issue with the charge pump or wrong resistors. 
I assume that only the PWM side becomes hot, right?

The blocking caps (C1 to C4) should be paralleled with small value low 
impedance caps - maybe foil or ceramic. Most importantly the caps for 
bootstrap, C1 and C3. Place some 100nF to 1µF caps beneath them.

If you don't want to use such low frequencies any more, but stay at 
30kHz, then C1 and C3 can besmall value low impedance themselves - the 
electrolytics with 100µF are not necessary then.

And which type of diodes (D1, D2) did you use for bootstrap? They should 
be rather fast for 30kHz.

https://www.infineon.com/dgdl/an-978.pdf

When I look at example schematics in the internet I see that most of 
them use 10 Ohm resistors in front of the gates.

But what's wrong with my calculation? I wrote above that the chargint 
delay due to the R and Gate capacitance would be somewhat more than 
750ns. Lets assume 1ms. This is a small fraction (1/16) of the pulse 
width.

The maximum thermal load would occur when the transistor hals half of 
the load voltage, which would be 10V at (estimated) 2A, so 20W.

20W * 1/16 = a little more than 1W. With a small heatsink, the 
transistors should be able to handle that.

Or am i wrong?

Try again with 
https://www.infineon.com/dgdl/an-978.pdf?fileId=5546d462533600a40153559f7cf21200

You should use a driver with separate inputs for low and high driver and 
a controller with dead time generation for both complementary PWM 
outputs, e.g. Attiny85 or Attiny861.
Then adjust the dead time to the shortest value, where without load the 
lowest current was consumed.

Thank you for your answers.

D1 and D2 are SB 160 Schottky diodes. I just looked for Schottky and 
bought the first type which appeared.

First I will try out lower gate resistors, but there is a question. In 
the datasheet of the driver is written "Output high short circuit pulsed 
current: Min 130mA Typ 210mA, Test Conditions: VO=0V PW<10us". For me, 
this is the maximum current, not the average current and a 22ohm gate 
resistor will reach with the 12V a peak of 545mA. Average with the 50nC 
and a switching frequency of 31kHz I will be a DC current of irrelevant 
low 1.5mA, so not sure if they really mean DC current in the specs.
However, I already ordered new driver (IR2183) with a peak current of at 
least 1.4A. Before they arrive, I will try out a lower resistance than 
recommended with the old driver. If they get damaged its not a huge loss 
(the circuit is connected to a current limited supply, so there should 
be no damage to the expensive parts)

If a lower gate resistance is not the answer, I will add 
non-electrolytic capacitors to the bootstrap. In future, I will swap 
them for sure, but now I would wait for more Items needed, instead of 
paying the shipping just for a couple of capacitors.

With a heatsink, I guess I can deal with the temperature, but that would 
be the last option because the circuit should get as compact as 
possible. The maximum output current I want to reach is about 10A and 
that only for a few seconds until the TEC has created a relevant 
temperature difference.

Use separate high and low inputs with an extern dead timer is not 
necessary I guess, so I would prefer to use a comfortable version with 
everything integrated.

Also think, that the problem is a shot thru, there is a short periode, 
where both FETs are "open". (Peter D.: dead time) You can try to solve 
it by an "analog" way, a R-D-C-combination. Use the diode to discharge 
the gate, R to charge and small C to GND to fit the time constant.
greetings
Achim