Assemble a mosfet module for hundreds of kiloamperes

MrAi, consider posting some of the issues you have experienced working at high power.

I would find them helpful filling the holes in my knowledge. I worked with high power
Plasma and welder designers as a field engineer, and they were tight lipped, but figure
I absorbed maybe 5% of the issues addressed in MOSFET and general switching. Much
of their proprietary seemed to center on magnetics, and that is my Achilles heal.

So in short would welcome anything you would consider sharing.

Regards, Dana.

I will add the first one: ultra-high voltage spikes generated by the leakage inductance and the dI/dt.

Let’s run some numbers: For high efficiency, the rise time should be no more than 2% of the switching period. This means 200 nanoseconds.
If the current being switched is 100 kA, and a leakage inductance of 10 uH, L*dI/dt gives a voltage of 5 Mega volts.
And let me say that for such a gigantic array of Mosfets, achieving such a low leakage inductance of 10 uH may be impossible.
You say, well, let’s clamp it. But you are talking 50 kilojoules of energy contained on each spike.

v48371 said:

I need to use MegaJoules of energy per pulse, and the most compact solution is batteries.

Click to expand...

Compact?
Using lorry batteries, you may get 1000A each. That's at least a hundred very large lead-acid batteries. The connecting cables alone to link them all need to be massive to minimise voltage drop and the impedance of the whole setup crazy.
(Plus vehicle batteries can only stand about half a dozen full discharge cycles before being useless).


For megajoules per pulse, you first charge a suitable [non battery] storage device.
A large capacitor bank at high voltage is one common option.

Switching at high voltage keeps the current and conductor sizes manageable, for the same end power.

eg. 1000V at 1000A pulse is not difficult with off the shelf components and gives you 100 times more output per switch than at 12V, with a fraction of the losses.

schmitt trigger said:

I will add the first one: ultra-high voltage spikes generated by the leakage inductance and the dI/dt.

Let’s run some numbers: For high efficiency, the rise time should be no more than 2% of the switching period. This means 200 nanoseconds.
If the current being switched is 100 kA, and a leakage inductance of 10 uH, L*dI/dt gives a voltage of 5 Mega volts.
And let me say that for such a gigantic array of Mosfets, achieving such a low leakage inductance of 10 uH may be impossible.
You say, well, let’s clamp it. But you are talking 50 kilojoules of energy contained on each spike.

Click to expand...

Thanks for that, that's one of the first issues I was going to bring up even without doing any of the math. You outlined this very well.

I would think that this would have to be mitigated on a MOSFET by MOSFET basis, where each MOSFET had it's own little snubber or whatever it needed. It would be almost like putting two or more power supplies in parallel.
But putting a huge number of MOSFETs in parallel even without anything else like a snubber brings up some physical current path issues too. For example, the physical arrangement of the sources of all the MOSFETs, and the drivers would have to be in tight to the gates and sources, maybe one driver per MOSFET so it can do a differential type input to the respective MOSFET. That might mitigate the source current path issues.
Of course the drivers have to be able to do the required current level to switch the MOSFET on and off fast enough.

Cumulatively, the losses are going to be high, there's no way around that. Even with an efficiency of 90 percent the losses will be large. However, relative to the level of the output power it would be just like any other DC to DC converter. If we looked at a hundred 1000 watt 90 percent efficient converters running at the same time it would still amount to 90 percent overall and the same power wasted, so the power wasted is just the way life in the power industry is.

There's one thing going for him here, the voltage buss is only 12 volts as it sounds like.

This kind of thing may not be that rare as there are DC to AC converters being used in the grid in some places, but I don't know much about them without reading up on that. They must be handling a LOT of power though for that kind of operation.

I'll see if I can come up with another list.
Physical layout is high priority just like with some of the simpler 5 watt converters that run at 20kHz. The ground wiring has to be done right for example.

I am thinking that if we made a 100 amp converter and put 1000 of them in parallel, that would make up a 100kA converter. The wiring would be nuts though.

I was thinking something along these lines for building a simple supercomputer. Connecting 100 computers onto a network of some kind with one master computer. The wiring would be nuts for that too.

danadak said:

MrAi, consider posting some of the issues you have experienced working at high power.

I would find them helpful filling the holes in my knowledge. I worked with high power
Plasma and welder designers as a field engineer, and they were tight lipped, but figure
I absorbed maybe 5% of the issues addressed in MOSFET and general switching. Much
of their proprietary seemed to center on magnetics, and that is my Achilles heal.

So in short would welcome anything you would consider sharing.

Regards, Dana.

Click to expand...

Hi there,

Yes that's not a bad idea. Maybe I will make up a list of the most important points. Some have been talked about before this post which are some of the most important right off the bat, such as spikes and physical current paths.
It might boil down to getting one high power converter with a more reasonable output current spec to work then using many in parallel. That's the way a lot of higher power converters are done.
Driver circuits are top priority too, but the more I think about this the more it seems that the circuits would have to be done one at a time and placed in parallel. That would be more or less the way the wiring or PC board traces would have to be done anyway. If one circuit with one MOSFET can be done, then a million could be done, and then there should be some way to connect them in parallel.
The source wiring is also very important, but on a mosfet by mosfet basis it would be similar to a regular converter circuit.
Another important point is input and output capacitor ripple current. In high power converters there is almost never just one capacitor on the input and one on the output. There are several in parallel for both input and output so that the ripple current is not too large for any one capacitor. The capacitors would overheat if the ripple current is not right, and even explode in some cases. Next to worst case is the caps would lose capacitance and possibly develop high ESR, which would mean the converter would keep shutting down due to over/under voltage detection (which would also be required).
Maybe some monitoring circuits too to monitor the state of all the transistors as it is running continuously.

I also think it would be wise to look into designs used for the grid that do DC to AC conversion. They are already being used in some parts of the USA. I know almost nothing about these however.

Here is one link I found but have not read anything about it yet:

Not sure if that will help or not though.

MrAl

Back in the day someone built an array processor out of MOT 68000's. I thought
it was Thinking Machines (Cambridge MA) but not totally sure.

Regards, Dana.

This is an interesting video about a high power (144 MW) pulse discharge system:

rjenkinsgb said:

pulse discharge system

Click to expand...

What we built used banks of polypropylene capacitors not electrolytic. Ours had a shorter pulse.

danadak said:

MrAl

Back in the day someone built an array processor out of MOT 68000's. I thought
it was Thinking Machines (Cambridge MA) but not totally sure.

Regards, Dana.

Click to expand...

Yeah some cool stuff back then

I had some brief experiences with Motorola progressors but was mostly into custom made ones and Intel and Zilog.
I first started with custom processors before the chip type got real popular, then Intel, then Zilog. I really liked the larger instruction set of Zilog and had a TRS-80 computer which was based on the Z80, so I learned a lot about the instruction set and lots of applications. Floppy disk driver was one of the most interesting apps, and the first operating system of the TRS-80; loved it.