new MosFETs on the block...

[darkfenriz]

Eva, what you describe happens only with perfectly trimmed dead time.

[ViennaTom]

Originally Posted by Eva

What you fail to understand is that with external diode there is always a diode recovery process taking place, and conduction losses are increased substantially. On the other hand, with Rds-on shunting the diode, reverse recovery only happens during high current peaks when driving low impedances, and only for the amount of current that the diode is shunting from Rds-on, not for the full current. The amplifier I mentioned can do full power into 4 ohm without diode recovery, which is only required for 2 ohm operation.

In terms of efficiency (power efficiency, cost efficiency, space efficiency), the Rds-on shunting method wins with current state of the art 200V and 250V MOSFET. For 500V/600V I would still go for external diodes or IGBT.

I see. But isn t there a risk of thermal runaway? Hotter FET means more RDSon means higher percentage of peak current in the body diode means more recovery losses means more temp means more RDSon... you get the idea..

I think the hypex UCD2k does have external diodes - don't kill me if I am wrong..

[Eva]

Diode forward and reverse recovery processes are symmetrical. The charge that takes 100ns to be removed from the diode took another 100ns to be stored... Oh, wait, was the diode allowed to conduct full output current for so long? Obviously no.

Damn, hasn't nobody learned this by investigating and optimizing bipolar transistor switching circuits before going into MOSFET?

UCD2k is not as advanced as the lower power versions, yes, it has external diodes. It seems that they are much more creative for small signal stuff than for KW power switching.

Optimizing switching timing only makes sense when it's the major source of non linearity, once the modulator becomes the dominant source of distortion, it's pointless (without improving the modulator too).

I don't want to advertise directly the product where my amplifier module is being used, but IR2110 timing is good enough to require no trimming at all (+/-10ns). Consistency between units is very good, there are a few hundred modules working out there now without problems. Idle losses are substantially lower than in UcD2k, yet almost no dead time is used. Remember that di/dt is finite, so even in the worst scenario a 20ns error at 750A/us results just in 15A excess current during 20ns. But in my circuit di/dt is shaped to be lower at the beginning of turn on process, and I use a simple but effective trick to speed up turn off of one MOSFET just when the other is starting to conduct. Source lead inductance helps a lot

Like all you, I had many fears at the beginning, but I found a way to synchronize an oscilloscope to the switching events with high precision (pre-trigger), and then I investigated all the processes taking place in a N-channel output stage down to the nanosecond level. I have already published waveforms a few times.

[ChocoHolic]

...yeah, seeing ahead the sweet spot discussion of the dead time ....
..leading to the question if it shall optimized for low idle losses or optimized for lowest losses at max current... leading to the discussion of most fortunate ripple current in the output filter and the related component loads and of course the transfer function...
I guess we all went through all this already multiple times in the related threads. Detailed examination in real life seems to be less popular.
(Eva, your way of displaying all load situations in one screen shot is great for efficient design work. But for teaching, it would be better to show just one load situation in one screen shot, and a second load situation in a second screen shot.)

Anyway, the IPPs are looking promising, but I also dislike data sheets without specification of Qrr as a function of temperature and di/dt.
IMHO for designing class D amps the missing specification of max. dv/dt is even worse.
As long as Infineon does not specify this - my favorites will remain:
IRFB4615, IRFB4227, IRFP4668.
Erhm, yes and if I remember right there were also pretty nice Fairchild types.

P.S:
Threshold voltage of the IPP appears to be pretty low, which is more demanding for the design of the gate driver and layout.

[theAnonymous1]

Originally Posted by ChocoHolic

Erhm, yes and if I remember right there were also pretty nice Fairchild types.

Yes, I like FDP42AN15A0 and FDP2552, but they are only 150V. I don't think Fairchild makes any good 200V FETs, but I could be wrong.

[ChocoHolic]

Yup, the FDP42AN15A0 was the one I was formerly thinking of.
But the data sheet does also not specify Qrr as a function of di/dt.
Also max dv/dt is not specified.
Except the short comings of the specification, it could be similar to the IRFB4615.

Sorry for messing up this thread, which is about the new promising Infineon types.
Anybody out there who works for Infineon and could help to get a more detailed specification?

[ViennaTom]

Originally Posted by ChocoHolic

Yup, the FDP42AN15A0 was the one I was formerly thinking of.
But the data sheet does also not specify Qrr as a function of di/dt.
Also max dv/dt is not specified.
Except the short comings of the specification, it could be similar to the IRFB4615.

Sorry for messing up this thread, which is about the new promising Infineon types.
Anybody out there who works for Infineon and could help to get a more detailed specification?

I don t work for Infineon, but I know a sales person very well, she is selling CoolMos by the millions to our company and I can approach her by mail or phone (if I don t forget because of 100 things to do once in the office)...
So to sum up: We need Qrr as a function of di/dt and max dv/dt, right?

[ChocoHolic]

Yes, max allowed dv/dt.
And Qrr depending from di/dt and from temperature.

[darkfenriz]

... and Qrr versus previous diode forward conduction time

[ViennaTom]

I have sent an email - hope to get more information soon...