Achieving 98% efficiency in a boost converter

What's the solar cell source unloaded output voltage?

With 600mV loaded, I'd expect >900mV?

If so,

RL

Right. Quite a few boost converters claim to achieve 95% under certain circumstances. I purchased eval boards for about 10 of them, but wasn't able to get better than 90% for my 2.0 to 2.4-volt boost. One did 92-94% at 25mW with 0.8 volts in, but strangely deteriorated at 2.0 volts in. In many cases it appeared the eval board inductors were too small. First, too low a value, and second, too much core loss. It seems the manufacturers just want to prove how small their overall used space will be. So I have footprints for about a dozen inductor types. I also wound an RM8-core inductor, that measured to fit into a 98% budget (next post), so I think that part of the scene is possible.

Another big part is the MOSFET's tradoff, for Ron and Coss. Low Ron goes along with high Coss, and thus high C V^2 f losses.

Having a variable frequency with load, as the Maxim parts do, helps out with that problem.

--
 Thanks, 
    - Win

This may be one of the rare occasions where GaN FETs might make sense.

Unless there is zero stock a few weeks down the line :-)

--
Regards, Joerg 

http://www.analogconsultants.com/

From the 2nd-page of my document on DropBox:

600mV input and 2.4V output at 10mA load = 24mW. Boost converter freq 250kHz. An RM8 inductor, with a low-loss 100uH winding, can work well. Measured DCR 0.018 Ω for 0.03mW calculated loss, and AC res 1.3 Ω, for 0.06mW loss. TS3A24159 CMOS switch, Vcc = 3.3V, for Ron 0.10 Ω, loss 0.16mW. Switching 2.4V with its Coss=150pF, loss 0.21mW. This gives a calculated continuous loss of 0.46mW, which is 2% of 24mW."

To compare the MAX1721: Ron 0.06 and 0.15 ohms. The switching capacitance isn't given. With an efficient inductor could be a winner. I haven't the new rev of the board yet, maybe I can add it.

--
 Thanks, 
    - Win

The 17224/5 switches are 31 and 75mR typ.

When 10mA load current was plugged into their EEsim tool, it spat out a 10mR 10uH inductor, but the sim doesn't function for inputs below

950mV, if it functions at all (I couldn't get anything sensible).

There's a dev board with a 'buy' button in the GUI that is non-functional. It's photo shows the kind of crap inductor expected.

It wouldn't be possible to separate the control cct losses, but internal start-up load is supposed to look like 3KR. This load condition is off their efficiency charts, but most non-isolated commercial applications consider anything over 90% to be creditable.

I don't know if you've noticed, but the 8pin bobbin for RM8 is no longer offered by the usual vendors. There's a Polish source on EBay.

The kit here is gathering dust......

RL

something like EPC2014 offers 16mR with total gate charge of 2nC. only 1/3 of that is Qgs.

The trouble would be getting it to run at 600mV.

At least in this app, there are no heatsink considerations.

RL

You need a helper converter to make a few volts at very low current. If it is a requirement to start at At least in this app, there are no heatsink considerations.

:-)

--
Regards, Joerg 

http://www.analogconsultants.com/

Yes, but with higher Coss losses. But I like the highly effective variable switching frequency scheme. But they are offering a family, which makes them more attractive.

Yep. OK, I thought to replace U13, the TPS61070, with the MAX1722x family, but the uDFN package is so small, I was able to shove some things around and add it as an additional selection. I managed to find room for one more 0805 1% resistor, to set the output voltage.

It's time for me to go ahead and order the PCB now.

--
 Thanks, 
    - Win

It looks like MAX17222 is the only actual part you can buy, at the moment (digikey zero stock but short leadtime).

A sales rep followed up on the funny EE-Sim performance, with no knowledge of the original query. I mentioned your project, but didn't try to button-hole him on availability of samples of the other flavours.

RL

I think the problem is ill-defined. When the only requirement is efficiency, a boost converter can be made as simple, large and expensive as anybody sees fit.

Without the size restriction one could go to a very low switching frequency, ruling out all switching losses. Conduction loss at low frequencies is a problem of cost (amorphous iron, film caps), and size (put components in parallel).

The power needed for the controller goes away when you don't add it (no specs. given on that).

There are peculiar effects that still might make optimization interesting. E.g. losses of an inductor are lowest around 85 deg C, and diode losses go down with temperature.

-marcel

Core loss data, at the low flux levels required here, would be exceptionally difficult to collect.

Extrapolating from the published charts by two orders of magnitude in one direction is a somewhat wishful expedient, as the charts themselves only use straight line approximations for the actual curves generated by empirical data and modeling.

Core loss equation exponents actually shift with each order of magnitude of both frequency and flux density. At low power, worst case losses might be expected to occur at lower ambient temperature due to published high flux ntc below 60-85C, but this information is not actually available at lower flux levels.

Low-power-circuit efficiency differs from low power efficiency in medium-power circuits, as you point out, because it should profoundly affect the choice of topology. It's not just a matter of smaller parts. Characterizing the source impedance can be important in making such decisions, as decoupling can only go so far in masking it.

RL

Why do you think the flux level is low (I agree it would be the case for high efficiency)? If we make the core smaller, flux levels will increase. Although the flux level would decrease for bigger cores, the volume of magnetic material increases and may drive up the loss again. Without a size (and EMI) restriction one would use an air-core :-)

-marcel

I think it's a physical scaling thing caused by the winding, that limits the core size reduction. Winding length decreases linearly, but xsectional area decreases by the square....

RL

The boards have arrived. I sent Rob Legg his set, next I'll send Don Kuenz his pair, and check my email for sea moss' mailing address. I'm not sure, but I think James Arthur is out, too busy. Did anyone I missed opt for one? Any other new customers?

The RIS-767 files have been updated: Now there's more discussion, top and bottom pcb-traces, 2-side assembly drawings, plus the orginal four example versions. The DropBox folder now includes Altium CAD files + Gerber.

--
 Thanks, 
    - Win

OK, Don, your set of boards is on the way. I chopped up one set of 20 flip-chip daughterboards, you can use your fancy saw for the other set.

I've added searchable parts-location assembly drawings.

--
 Thanks, 
    - Win

My metal circular saw blade works great with sheet metal. FR4 may require an abrasive blade to keep the FR4 from splintering. :0)

Thank you,

--
Don Kuenz KB7RPU

Yep, too busy, but thanks Win.

Cheers, James Arthur

Yep.

Dan, your RIS-767 with 20 chopped-up flip-chip daughterboards is on its way.

--
 Thanks, 
    - Win

Great. Thanks for taking the time to chop up the boards. I forgot to ask what the required tolerance is on the 2.4V output... how about 10%, too loose?