DC transformer

They call these SMT transformers, but they should call them low-voltage transformers, because both input and output voltages are so low, probably limited by low turns count.

--
 Thanks, 
    - Win

Here's one of my inductors, hand-wound from #14 magnet wire on a Sharpie, heat sunk to the PCB hence the cold plate below.

Given the RMS current, that wire size seems entirely unjustified. It needs the surface area to reduce skin loss and get rid of heat. The coil spacing reduces proximity effect and further helps cooling.

Now I'm thinking that this could be a "planar", namely PCB-based, inductor. Better cooling, less skin loss per pound of copper, saves wear and tear on Sharpies.

--

John Larkin         Highland Technology, Inc 

lunatic fringe electronics

As switcher frequency goes up, skin depth starts to make extra-thick copper not worthwhile. Very roughly, 2 oz copper is good enough around

1 or 2 MHz sinewave. For switcher waveforms, the tradeoff further favors not wasting copper.

Exotic PCB material probably doesn't help much. The main cooling is air at the surface. [1]

I'm guessing that adding more loops on more internal layers doesn't help. There's basically no cooling, and proximity effect will be serious. It might help if you need a lot of inductance.

[1] how about a rising-sun copper pattern on some inner layers, to conduct heat out?

--

John Larkin         Highland Technology, Inc 

lunatic fringe electronics

Maybe not a litz, but I am sure it would be possible to get a really decent interleaving. Pity the PWB technology is so dead... :-/

Best regards, Piotr

What exactly is that thermal goo? I think I may need something like that in my recent project.

Best regards, Piotr

It's a soft sheet material,

3GSHIELDING TW-T600-2MM

It feels like used chewing gum. Compress it to about half its nominal thickness for good heat transfer. Use thinner stuff if you can. I have high voltages so went with the 2 mm thickness.

Ask for a sample. They are good about that.

I use biggish pads of that between my PCB and the cold plate, and a scrap between the inductor and the PCB. Thermal vias conduct the heat from the inductor to the bottom of the board, then another pad below to the cold plate.

--

John Larkin         Highland Technology, Inc 

lunatic fringe electronics

PCBs are like piston engines, very old but still the best way so far.

--

John Larkin         Highland Technology, Inc 
picosecond timing   precision measurement  

jlarkin att highlandtechnology dott com 
http://www.highlandtechnology.com

Playing with this lately,

note the planar coil L1 with the red coil above it. Currently L1 is cut out of circuit to try a high Q coil.

The small-signal measurement of L1 shows a rather poor Q of 14 (at 3.5MHz), which would account for about a watt at the ~15VA I'm throwing at it (depending on operating point). Efficiency seems similar with the good coil though.

The biggest single loss seems to be the >1W burned purely in gate drive, a combination of the LDO supplying it and the MOSFET's ~15nC gate charge being driven at 3MHz. I should've given that a double-check and went straight to GaN instead. (Which can pretty much drop in, too, since the gate drive is only 6V. Swap the 78L06 --> 78L05 and there we go. Shame about the GaN footprints being different.)

Looks like this in the CAD,

4 layers, 2oz, and, think it was something like 15-20-15 mil dielectric stackup (62 mil overall). So it's very much like one of those edge-wound flat wire inductors you see so often, sans core.

The transformer does very nicely, seems less core loss than I designed it based on. 3F46 material.

It's too bad there's no high frequency (>1MHz) resonant controllers out there yet, all that area (the discretes) would be easy to save.

Tim

--
Seven Transistor Labs, LLC 
Electrical Engineering Consultation and Design 
Website: https://www.seventransistorlabs.com/

They are most cost-effective and very repeatable, but "best" in purely technical terms -- no, I don't think so. The limitations imposed by this technology are severe.

Kind of the best description of socialism I know of: a system bravely fighting problems unknown in other systems. Exactly the same with the PCBs.

Best regards, Piotr

This layout is a beauty. Is a single VIA sufficient?

Why "very much like"? Those are 1mm thick or more, so the 2oz does not even come close. Or do you mean the entire stackup?

Anyways, I like it.

Best regards, Piotr

Can you show us the schematic? How did you pick 3MHz?

--
 Thanks, 
    - Win

Thanks! Yeah, only a few amperes. Design spec 10W total output. The chip inductor is 6x6mm if scale isn't apparent. The vias there are pretty big (1mm?).

The traces and vias under the MOSFET are tiny, 0.15 and 0.3mm respectively. But it's short so it's not going to burn out.

Not like the ones you use, at least? :-) I've seen them made with nearly paper thin sheet, not sure how they manage to bend the stuff without tearing or scrunching it, but they do.

And yeah the stackup, hard to follow the colors, but it's about 0.9 turn per layer, then a via to the next layer and so on. Trace width probably doesn't need to be so wide -- and Q may well be higher with less, despite the reduced cross section, because of eddy currents -- I just figured on that width as a starting point for the overall geometry.

I started by calculating a wire helix of 100nH, then drew a planar version of similar dimensions -- guessing that the mean current path is not quite the ID nor the mean diameter, but a bit inbetween, due to current crowding and skin effect. Ended up with about 90nH, not bad!

Tim

--
Seven Transistor Labs, LLC 
Electrical Engineering Consultation and Design 
Website: https://www.seventransistorlabs.com/

Uh, sure what the hell--

Some component values changed since proto testing, of course it's an exercise for the student to find out which ones. :^)

L2 is only there because the planar core isn't gapped, of course (Np = 2t, Lp = 2.7uH).

Really I just wanted to use the ELP14 because I had already made some footprints, and I wanted to push up to 15V out of them (3V/turn). For a total 30V output (full wave doubler). Which comes to something like 2MHz for comfortable core loss. Although because of economy of scale, I could push the core loss quite a bit higher: 1MHz or even maybe 500kHz would be reasonable, which is within reach of many controllers/regulators, no need for custom.

As mentioned, pushing silicon quite this fast is something of a mistake, but GaN would drop in with little change, giving a significant boost in efficiency.

Which by the way, is only about 70% here, rather embarassing for a resonant converter. :^)

Controller behavior is roughly based on L6599 which I did a project with earlier this year; I think I get the idea. Resonant control may seem intimidating, but I think because controlling it precisely would be both very intensive (e.g., state space control?), and not very useful at that (the Q is low so the feedback doesn't need to be very gradual; except when it is, but *hand waving*), they usually settle for a simple burst mode with CW operation under heavy loads.

Main difference here being, I didn't opt for the 74HC74 to latch the enable, so it just starts and stops whenever, gladly making runt pulses. Likely far from optimal, but it's under light load at the same time too, so... meh?

Oh hm, I should feed in a bit of CLK to the comparator so it gets injection locked, duh. That'll solve that...

Tim

--
Seven Transistor Labs, LLC 
Electrical Engineering Consultation and Design 
Website: https://www.seventransistorlabs.com/

PCBs have been replaced by welded cordwood modules, hybrids, flip chips, SLT modules, multiwire, all sorts of now-gone stuff.

Piston engines (pistons, rings, crank, cam, poppet valves, spark plugs, clutches, gears, 100 year old tech) have been replaced by turbines, many sorts of rotary engines, steam, all kinds of now-gone stuff.

--

John Larkin         Highland Technology, Inc 
picosecond timing   precision measurement  

jlarkin att highlandtechnology dott com 
http://www.highlandtechnology.com

Copper is a pretty good thermal conductor. As John says, the main thermal barrier is the air at the surface, and there's not going to be steep thermal gradient inside the windings.

Printed circuit board substrate material aren't as thermally conductive as copper, but they are still roughly a thousand times better than air.

John Larkin isn't great at quantitative thinking.

--
Bill Sloman, Sydney

Thank you, John! Very useful info.

Best regards, Piotr

I would happily use something like that, but now this is my integrated magnetics LLC prototype trafo (~100A):

Technically correct, but I don't feel comfortable with that. I think I am just prejudiced and should work on that. :-)

I use what is available and have never seen anything thinner. In the transition zone between the hair-thick PCBs (2oz) and that 1mm slabs is some room for laser/water-cut copper plates. Press-punched in quantity. Downside: you can make just two turns that way. More require welding, riveting or another origami, neither very attractive.

But yes, these flat wires are fun. Just like the microwave waveguides, they have E(asy) and H(ard) modes. At least if it comes to bending. ;-) Here is another E-mode prototype for 100A high efficiency bidirectional buck-boost:

This is a wide swing one (22uH@0A) to allow deep CCM ( I've seen them made with nearly paper thin sheet, not sure how they manage to bend the stuff

Are you sure it wasn't a plate varnished after cutting?

Best regerads, Piotr

Hm, at 10W why do you even bother? It's a flyback realm or push-pull if needs to be quiet. LT3999, LT3439 and their inferior brethren. Knowing your skills there needs to be a good reason for that.

I am happy with my 150kHz DC, have fun boldly going where no man has gone before! ;)

Best regards, Piotr

No totem-pole PFC controllers either, fortunately I have already had an FPGA there.

Best regards, Piotr