Ultra-efficient bucks

On a sunny day (Mon, 16 Dec 2019 09:24:12 +0100) it happened Piotr Wyderski wrote in :

I think the question in the experiments you did is "where does the heat go?" did you measure temperature increase (thermocouple) of the various components? Why the paranoia about a few percent?

I don't have a FLIR. Touching the converters with a Sevres-certified Prototype Finger indicates the CFFB is uniformly cool, particularly the MOSFETs. The trafo is a bulky 27mm N87 toroid core. It gets warm when the converter is heavily overloaded (7A or so), most likely because of ohmic losses in the windings. Efficiency drops to 80%, but it bravely keeps its 3.37Vout. :-)

OTOH, the TPS is hot. I adore the sense of humour of the designer who put such a multi-amp circuit in a TSOT23 without a thermal pad. COme on, there are thermally-enhanced MSPO10 and the likes.

Will try to do that later, but if the finger doesn't show anything obviously hot, there is no point in measuring that exactly. The components are big, so natural radiation/convection cooling is enough to make tracing the power loss issues hard.

For no important practical reasons, it just seems that there is this inpenetrable 92% barrier and I don't know why it hangs that low. Consider it a scientific problem at the verge of engineering. I would like to understand the whys, not to make a sellable gizmo, although a reference implementation proving the validity of that knowledge would be valuable.

High-power LLCs can go to 97%, totem-pole boosts to 99%, so why most likely the simplest converter of them all cannot break that mediocre 90%?

Best regards, Piotr

I've found that jacking up the bootstrap pin voltage (using a resistor to a higher voltage supply, if available) can help quite a bit.

Cheers

Phil Hobbs

On a sunny day (Mon, 16 Dec 2019 10:02:54 +0100) it happened Piotr Wyderski wrote in :

Yes, transformer, and do not forget core losses. If you switching frequency is high, the switching edges will contain a lot of high frequency components, you need a good core material.

I think Rob Legg could do it. We worked together for several months on creating efficient low-voltage boost converters, using my PCB breadboard, RIS-767, making extensive mods, trying various parts, switch ICs, and discrete MOSFETs. He managed to exceed 98 to 99% in several instances, and over 100% in one case. We left the work with the ball in my court, needing a new PCB version, with lower trace loss, and better measurements.

I recently thermal imaged a TPS54302 switcher. The tiny chip was cool, and the much larger inductor was warm.

Well then, buy one.

The TPS seems pretty good up to about 2 amps. 3 amps is, well, optimistic for this part.

I use fat traces on all the leads. I'm not sure on which pins the most heat comes out of, so I fat-up all that I can when the current might get above, say, 1 amp.

Too small an inductor can make the TPS get hot, too.

Since it is only 10 watts output, I am guessing that the losses are mostly idle tare loss. Are you including the auxiliary power supply as well if there is one ?

The supply may be drawing 1 watt even without a load ?

The TPS54302 being hot may be part of it. How hot does it get with really small load ?

Yes, no cheating.

The CFFB idle power is 45mA@11.69V, that is 520mW. This is mostly due to the transformer-based gate drivers. Fully +/-10V swing with ~50.5% duty cycle. Four secondaries per transformer: two for the bridge itself, one per synchronous rectifier after the main transformer (for the 3.3V and

The TPS wins hands down here, negligible idle current. A milliamp or so. But it has much worse load regulation than the CFFB. I mean, Vout is still acceptable, no complaints, but the current-fed converter is incredibly stiff.

I am going to check the buck stage only efficiency of the CFFB and replace the LM5101/SQJB80EP with LMG5200 just to see what happens. At mere 250kHz this GaN half-bridge should have negligible switching losses, in spite of the extreme hard switching conditions.

Best regards, Piotr

TPS54302 goes into burp/burst mode below about half an amp load. Switching losses (in the chip and in passives) go way down, and ripple goes up.

The data sheet shows silly high values for the feedback network, probably to demonstrate very low supply current at light loads.

Cute. That could be useful at higher currents.

...

This is the Art of Electronics, you should have known that.

All kidding aside, efficiencies close to 100% are ridiculously hard to measure, so the result of 101% is perfectly fine if your error bar is 2%.

Best regards, Piotr

I have just discovered the LM5041 doesn't have any deadtime generator in the buck stage and LM5101 relays the signals to +/-3ns accuracy. That shot-through would explain a large fraction of the idling losses.

Best regards, Piotr

As I said, the ball is now in my court, to update RIS-767 PCB, with lower trace loss, and better measurements. At present, the measurement errors and corrections are evidentially at least 0.5 to 1%, and haha, perhaps much more. I made a "copy" of Legg's circuit, but struggled and only observed 97% efficiency, using two SMUs, one each at the input and the output.

Unlike Legg's circuit, I didn't have enough measurement points to apply the corrections. He also used a better, homemade, inductor.

RIS-767 was discussed on s.e.d., Sept 2017.

You can always trade size for efficiency.. Try switching at 20-50kHz, use FETs with ridiculously low Rdson, use a large and expensive inductor (lots of turns to reduce core loss, large gauge Litz wire to reduce DC and AC winding loss)

Right, but Win was stating 98 and 99%. Subtract a 2% error and it looks a lot less magical.

You might go for a gapped inductor to minimise core losses, and a higher frequency ferrite material for the core.

Traditionally Manganese Nickel ferrites were good for about 100kHz, and the much higher resistance Nickel Zinc for up to about 10MHz, but when I last poked around, somebody was offering a new ferrite material aimed at 300kHz operation.

You beat me to it ! Dead time was going to be my next question.

High frequency is great but doesn't give a lot of time for DT.

Good deal if that IC has separate SR output drive and can be taken care of with just separate secondaries to the up and down FETs.