Yet another high density PSU design

[blueskull]

A year ago I posted a high density PSU design for my laptop (180W, 19.5V), but the project ended up being scrapped due to I sold that laptop and I no longer need it.

I decide to revive the project with a twist -- to shoot for much higher power, in a similar size, with active cooling.

This time, I'm building the PSU for my ITX build which is based on HDPlex 400W DC ATX. I need a 450W AC-DC module to extract the full potential of the DC ATX, but the highest power rated power brick is Dell 330W brick for Alienware laptops.

Also, I would like to get rid of that huge power brick, and to build a smaller one that fits inside the case of my ITX build (NFC S4M).

I talked with Larry from HDPlex, and he is actually designing a 700W DC ATX, and that opens a whole new possibility -- dual core GPUs in a tiny ATX.

To make that work, a 760W+ AC-DC PSU is needed to power the entire thing, and that's what I'm designing.

The target is simple -- to build a 750W+ PSU that fits in NFC S4M, I prefer the space between top of GPU and chassis wall. This allows me to install longer GPUs (compared if I put it on the rear end of the card), besides it allows for better cooling of the GPU.

From my measurement of my own S4M case as well as its SketchUp model, I determined the maximum size of my design: 150mm long*55mm wide*28mm thick. That's 3.3W/cm3 or 54W/in3 power density for a power supply with PFC, bulk capacitors and casing.

It's an ongoing project, so it may see major modifications.

So far, I have the schematics, some primitive calculations on power path and simulations on key circuits done.

I will need to assign MCU pins as well as to do the PCB artwork, but I would like to halt to listen from your suggestions.

FYI, the BOM is $112 at 500 units quantity. I don't plan to sell it for now, but since it's CC-BY-SA, I hope someone else can make it a product for my fellow mini ITX lovers.

[prasimix]

Any chance to attach "human readable" PDF schematics instead of Altium's .SchDoc?

[Rerouter]

Another hoping to see it in pdf, or an image,

Not dipped my toes anywhere near this stuff, but interested in learning about it.

[xani]

I wonder if someone tried to mod server power supplies for that reason they are pretty small and powerful

[jbb]

I saw 'high density' and Blueskull and I thought "This is gonna be good."

Those catalog planar transformers look really nice.  Any information on clearance, creepage and insulation strength??  Or will you have to do a teardown and assess for yourself?

[T3sl4co1l]

Any chance to attach "human readable" schematics

I got to this point reading this sentence, and thought to myself: you mean not packed like a game of Tetris?

Oh...

Sorry blueskull

Tim

[T3sl4co1l]

I saw 'high density' and Blueskull and I thought "This is gonna be good."

Those catalog planar transformers look really nice.  Any information on clearance, creepage and insulation strength??  Or will you have to do a teardown and assess for yourself?

Hmm, depends on construction, but if it's standard multilayer PCB, it should be a "cemented joint", i.e., clearance is through FR-4, only a dozen thous needed.  Creepage around the outside connections, of course, needs to be the normal ratings.

Creepage includes distance to the core, by the way, because ferrite is a little conductive.  Worth grounding the core for EMI reasons, too (if possible).

Copper tape over the transformers would take care of that nicely, as well as stray leakage, mmm.

Tim

[BravoV]

Subscribed

[jbb]

Thanks for the PDF schematic.

...

I have to agree with Teslacoil: that schematic is indeed packed in time Tetris bricks. What happens when you try to show component values or part numbers on it?

[NiHaoMike]

I wonder if someone tried to mod server power supplies for that reason they are pretty small and powerful

I have used quite a few surplus HP server PSUs in projects. And I can highly recommend them for DIY projects that need 12V at a high current.

I'm hoping that there will be affordable high power digital input audio amplifier modules in the near future so I would have a good reason to hack a server PSU to output 48V or so in addition to 12V and build a gaming PC into the case of an old A/V receiver and still have it work as a very good receiver.

[ali_asadzadeh]

Just visiting

[ali_asadzadeh]

Thanks for sharing, in which software are you simulating things? I wonder if it's OK for you to share simulation too, and please keep adding the Altium files, they help me way better than PDF, Also why choosing Cypress and PSoC, why not choose a more jellybean cortex M from ST or NXP?

[T3sl4co1l]

You get ZVS in a >100% ripple sync buck, but that may be undesirable for the very high Q inductor you need, and output ripple current.

I suppose you could add a pulsed commutator (a lower (but not zero) inductance path, that essentially shorts the supply into the switching node), which could be a smaller transistor than the main switch.  But IIRC, this ends up costing more losses (hard switching turn on + off, plus conduction loss during the 'gulp') than the switching energy saved in the main transistor, so it's not worth it.

Multilevel inverters can save a lot of switching energy, but cost a ton of complexity, which you probably don't have space for.

There are some resonant buck configurations, but I haven't played around with them.  I think they tend to be harder to control, with narrower variable range?

GaN is the best bet all around, assuming it's suitable for other reasons.  Overlap switching slightly to avoid "body diode" forward bias; or, considering what it is, maybe just go with it, since it doesn't cost recovery anyway.

SiC is good too, of course you need to avoid body diode operation in that case, for which schottky are just fine (or, again, shorting mode commutation).  Fsw probably won't be too impressive because of all the capacitance, but still better than Si of the same Rds(on).

Tim

[Wolfram]

Are you still using the PL300-104L primary as the PFC choke? Did you calculate the peak flux density seen by the core? A quick back-of-the-envelope calculation using a rough estimate for the core cross sectional area (2 cm^2) gives around 3 T at a peak current of 10A .

[tszaboo]

I have to agree with Teslacoil: that schematic is indeed packed in time Tetris bricks. What happens when you try to show component values or part numbers on it?

Trademark

Or perhaps OCD.

[DutchGert]

Really nice project 

But indeed, make your schematic a bit more readable and put some values in there

[ali_asadzadeh]

Thanks for sharing, in which software are you simulating things? I wonder if it's OK for you to share simulation too, and please keep adding the Altium files, they help me way better than PDF, Also why choosing Cypress and PSoC, why not choose a more jellybean cortex M from ST or NXP?

[TimNJ]

Blueskull, just curious, why are you rolling your own PFC/HBC controller (with the PSoC) when there are so many on the market?

[ali_asadzadeh]

Thanks blueskull, fair enough