3kW non-isolated dc/dc converter

[skn96]

Hi All
I am planning to design a DC/DC converter 3kW, 300V input /150V output with no need for isolation. Frequency around 100khz is desirable. On the paper synchronous buck with 400V or 600V Mosfet can do the job. However, at this power range and this frequency I only see people use LLC or phase shifted full bridge. I understand that hard switching at 100khz and 3kW and 300V is very difficult due to reverse recovery current and switching loss. But if I want to use phase shifted full bridge or LLC , I am paying for isolation that I do not need in this application. I do not need transformer and also transformers are lossy too. So what should I do? Is there any other topology that can be used something like a non-isolated soft switch topology?

Thanks.

 

[CataM]

300V input /150V output

PWM resonant Cuk converter (patented obviously). I guess you can use it for academic purposes.

 

[asdf44]

Right so the story is that for a few reasons (safety among them) many converters of that power level need isolation and that necessitates a transformer and that necessitates AC current/voltage and AC current/voltage is a key ingredient for ZVS switching which means that it's (comparatively) easy to go LLC or phase shift to realize the benefits of ZVS.

But don't let the disproportionate amount of literature on those subjects throw you off. Hard switching buck converters, boost converters, class D amplifiers and inverters are still very common at high power levels - they're just not as 'cool' as the resonant topologies so you'll see less literature on them. But in-fact the most common hard switched topology in the 3kw range is the PFC boost converter that's inevitable feeding all those resonant isolated converters.

Non-isolated topologies can ZVS as well but typically you pay a comparatively higher cost. For example a regular buck converter can ZVS if the inductor ripple current is so large it goes negative every cycle (without this buck converters ZVS for one switch transition but not the other) and if dead-time is manipulated to benefit from this. But that's a high cost.

So anyway, you don't need to add isolation to achieve ZVS but more importantly you also don't need ZVS. Here is a great example of a hard switched 3kw converter using GAN fets that claims 98% efficiency. It's a full bridge classD inverter but a full bridge classD is just two buck converters. I strongly suggest you take a look at these GAN fets first from Transform and next from GAN systems which can drastically reduce hard-switched losses. These two companies target 650V devices which are a good fit for your application (EPC just came out with 350V parts but that's cutting it close for you).

Transphorm 3kw inverter:
https://www.transphormusa.com/evaluation-kit/tdinv3000w050-kit/

Gan systems half bridge demo:
https://gansystems.com/gan-transistors/GS66516T/#board

 

[Easy peasy]

If your input 300V is relatively stable then you can have an H bridge driving an auto-transformer ( 100kHz) with 49% duty cycle to give soft switching on the H bridge devices ( with a soft start )

You then need very little or no output L, can go straight onto the output C (again with soft start to limit current at start )

The output reg (with load) then depends only on the R ( and L leak ) in the auto-Tx.

The output diodes, you only need two if the centre of the auto-Tx is the 0v output point, are also soft switched.

You can also do this with a push pull ( 800V or higher mosfets needed ) and if you want the 150V referenced to the same 0v at the input put the fets at the top and tie the centre tap of the auto-Tx to 0v ( gate drive Tx's needed ) - but again can get soft switching ( caps across FETs ) at fixed freq and soft switched diodes too...

We have built many converters on this principle.

A straight Buck Converter is entirely feasible too, if you use two of them 180 deg out of phase then you will have very little current ripple at the input or output - if the buck switch fails tho -> 300V at the output ... SiC output diodes make reverse recovery much less of a problem - easily cured with good layout and snubbers...

good luck ...

- - - Updated - - -

I strongly suggest you take a look at these GAN fets first from Transform and next from GAN systems which can drastically reduce hard-switched losses.

What happens when you turn off 10A very fast without de-coupling caps of very high quality within 1-2mm of the fet switches? the volts fly up to > device ratings and bang ...! suggesting GaN to a newbie is a bit like putting a person who just got their car license into an F1 to do the shopping - bound to end in upset ...

 

[asdf44]

Designing a 3kw 300Vin converter is difficult. You make suggestions and people are going to sink or swim. My thought is that if GAN let's you do a simpler topology then that's the easiest path.

Specifically here Transphorm devices are a hybrid silicon/gan which have identical gate drive requirements as silicon and also come in standard packages. With a demo board that can test this exact application I think that's a pretty reasonable path to success.

 

[mtwieg]

A hard switched buck converter should be ok for 3kW/300V/100kHz, especially if you use two interleaved phases and use SiC FETs (GaN is overkill at 100kHz). I wouldn't go for anything fancier unless you need efficiency much greater than 90%.

 

[asdf44]

Sorry to nitpick but I don't see any way these GAN devices are 'overkill'. The development board linked above is targeted at 50-150khz and the cascode configured GAN fets use standard gate drive techniques compared to SiC which wants positive/negative supplies.

Also my understanding is that SiC advantages don't really come into play until above 650V.

 

[Easy peasy]

I have made very many 3kW converters, would I use GaN - no, why? because std Si mosfets are cheaper, easier to use, easier to mount, far more rugged, and I don't need 5nS turn off speeds which would require very tight layouts with expensive SMD caps to soak up the turn off current...also as I have to design product to meet EMC standards, the noise from a 5nS turn on requires a lot more esoteric filtering arrangements than for slower transition Si mosfets ...

- - - Updated - - -

respectfully, your understanding of SiC is fairly obviously not based on practical use. Because they allow far less reverse recovery current peaks they are used from 100V upwards in 100kHz + power circuits to reduce noise and dissipation...

 

[asdf44]

Yes the switching speeds are faster which is good for efficiency but bad for EMI. I haven't brought my GAN designs through EMI testing so wish me luck on that.

 

[Easy peasy]

We routinely have clients that spend 30k - 100k+ on getting an EMC pass for their power products, none of them are keen to jump into GaN when they already have Si working at 500kHz with controlled EMC. If you are below 50V you can get away with a lot more than if you are operating off the 230/400Vac line and have to keep conducted below 46dBuV and radiated down all the way to 30GHz ....