Ygtjki

I want to build a DC to DC converter with:

25 V input side

400 V output side

4000 watts of continuous rating

That means (obviously) 160 amps on the input side and 10 amps on the output. I can’t find anything remotely close, that has such a high voltage gap with such a high wattage rating.

1. Is this feasible? I can’t see why not.


2. What would be the best way to go about this, or a reference design somewhere that I can scale up? I don’t mind spending the money on quality products, but I prefer to DIY it so I can make changes etc.

I have an adequate voltage source for the input side that can handle the current so that’s not the issue, but I’m failing to find more info and even big companies like Vicor, Lambda, etc. all have 400-600 max rated units. Therefore I have to build it.

160A at 25V will not give you 4kW out. If it is very well designed, you'll get around 3.2kW. The rest is wasted as heat. As you're just setting out to do this, and you're trying to design it yourself, you need to model it well and simulate to work out where your losses are going to be, and how you're going to cool it.

This is a perfectly do-able boost converter project. I have done a 5kW output DC-DC (admittedly that was 48V), and that required a full automotive style liquid cooling system. That was using a standard synchronous DC-boost converter, 48V came in, and we got up to 200V out.

First things are cooling and component ratings, those are the hard bit to do. 25V is low for 4kW, so you'll quickly see (once running the numbers) why it is that higher voltages are chosen for these kinds of power.

At 25V in, 4kW out, 80% efficient means around 200 Amps in, plus a 50% overhead safety factor for your FETs, so you need to find FETs rated to 300A, 800V (high voltage due to high output voltage requirement). Don't forget to de-rate for temperature, and check your simulations for junction temperature rise. I say your FETs, but it doesn't have to be FETs, I've used FETs and I've used IGBTs for this kind of work, some people say GAN transistors would also work (but I've never used them). Depends on the voltages, currents and switching frequencies you're looking at. But that in turns depends on your application, budget, size constraints, development time allowed etc.

You'll also need an inductor, rated at similarly. But these are probably easier to find.

You could split the power down, using multiple channels in parallel, each channel doing a part of the current (I used 3 channels on my 5kW system). But still, cooling will be your biggest challenge.

You can then build it, find out what fails, what mistakes you made in your simulations, and then repeat the process until it works how you want it to work.

In summary:

• It will get hot


• You need to simulate it


• Don't underestimate how hot it will get


• You can just scale up a standard DC boost converter


• Watch your cooling


• Simulation is vital

The inductive coupling must be modeled. As well as eddy currents.

200 amps switched in 200 nanoseconds (for high efficiency, fast switching must occur) and wired to be 1cm away from a 1cm-by-4cm servo-regulator loop, will induce this error voltage:

Vinduce = 2e-7 * Area/Distance * dI/dT

Vinduce = 2e-7 * 1cm*4cm / 1cm * 1Billion amps/second

Vinduce = 2e-7 * 0.04 * 1.0e+9

Vinduce = 2e-7 * 4e-2* 1.0e+9 = 8 volts.

To be completely accurate, you need to write the integrals and extract the equation that uses NATURAL_LOG. And you need to model the eddy currents.

At these levels of dI/dT, ground planes will NOT be ground planes. There will be large differences in voltages across the plane, because of eddy currents.

The math suggests shields and planes (VDD or GROUND) will have EIGHT volts of gradients.

I was brought in to diagnose the failures on a 15,000 horsepower speed controller. A loop of wire, to sense the magnetic field , held near the ground-plane, indicated 2 volts per square-inch. Ground Was Not Ground.

A programmable voltage three phase inverter with a full bridge (or two that are wired with a 30degree phase off-set) (to gain minimum ripple) with some inductance and capacitance to follow that are designed for 24V input might just be able to be found semi custom.

Some crazy variable speed drive system with a 24V supply and programmable voltage might even work out of the box.

Starting to design one from scratch for a once-off will cost much more than you think, hoping to save money this way is futile. I would spend a week making calls to every VFD and inverter manufacturer with your loosest tolerable specifications and see if anyone can configure OTS parts.