December 20, 2003

Well, I've been fighting with efficiency on various different boost regulators over the past six months.

I tried lots of circuits, and got between 70 and 87% efficiency, which left me with the biggest loss factor, the shottky diode.(higher current ones ~1000mA)

Finally ended up on a dual MOSFET setup, where one MOSFET replaces the schottky diode, to drop the large losses of the schottky diode.

Today, I got the efficiency of it up to 90.4%. It should have been a lot better. Tried all sorts of things to get it up, changing inductor, inductor values and more. I tried higher current inductors (lower copper losses from DC resistance and B-H curve magnetic losses)(many Pspice simulators do not include magnetic losses), all to no avail.

Then it occured to me that unlike the previous circuits, this one ran at a rather high frequency (yes, high frequency = higher losses).

Looked up the Sumida CDRH127, couldn't find it's frequency rating, but they test it at 100kHz....humm

Dug up another inductor I had that was rated for 1MHz.

Bingo! Gained 6.79 % efficiency right off the bat.
Woohoo!

97.1946% efficiency boosting 3V to 3.6V at 350mA. Runs down to 1.7V, which with two Alkaline batteries, runs them down to 0.85V each, which at that point, ain't much juice left in them anyhow. BTW, it will supply 1A with no problem... ; )

Happy Happy, Joy Joy!

I think I've found my circuit. Now I gotta pick a flashlight to put it in.

BTW, in case some of you tech geeks out there bring it up, I have less than 20mV ripple on the input and less than 30mV ripple on the output.

[CPF-xbrite]
Dear Jarhead,

May I have the schematic diagram??

Thanks, YS Lee

[CM}

Sounds like a synchronous boost regulator which uses a FET instead of a Schottky which is one of the biggest loss contributors. Several manufacturers make controllers for synchronous boost circuits. Are you using one of these or is this a home brewed controller?

CM

[CPF-Steelwolf]

More info! Schematics! 97% efficiency! Holy Cow! I may never direct drive again!

Seriously, any chance of a look at the schematics?

Yes it is a synchronous regulator, rather common these days.

At these efficiency levels, losses in even small wires/traces can start to show up. Even the ESR of capacitors can start causing losses. Again, the choice of inductor is critical to getting the high efficiency.

The chip part number is TPS61030PWPR. My home brew hasn't reached this level, but I'll give it more effort as I have time. It is made by TI. It is a bit larger than some chips. There is a choice of TSSOP-16 or the RSA package (one of those small chip scale or DFN packages

http://focus.ti.com/docs/prod/folde...t/tps61030.html

For lower current levels, such as you'd use with 10 Nichia LEDs, also check out the TPS61025DRCR.

700KHz is alot faster than a lot of switchers, 6.8uH seems to work well, but in my application, I use a slightly lower value, which has a lower DC resistance of about 0.0023 ohms.

Since I use voltage feedback, I just tweak it for the specific luxeon. Yes, the luxeon's forward voltage will change, and thus the current, as it's temperature changes, but the better you heatsink it, the less of this effect you will see. This could be easily adapted to a sense resistor to provide a constant current regulator, but one needs to pay attention to start-up criteria, maximum load the chip can start up into, any pre-regulation cycles, and also gain-phase bode plots. Otherwise you will have issues, and a lot of dead chips or a quirky design. Small inductors that easily saturate can also cause issues.

With voltage feedback, I've ran it down to 2.0V on batteries, and haven't seen any change in the power delivered to the LED, I'll work on it more, with more details, once I fix another bench supply that gave up the ghost.

I'll probably rig up a current feedback setup when I get a chance, and see how it does.

No Star882.

A cpu switcher is a buck, not a boost regulator, and the circuit works a bit differently, and the inductor is moved from the power input to the power output.

For a buck-
During the first cycle the power flows through the top MOSFET, through the inductor, to the load on the first cycle. Once their is enough energy stored in the inductor, the top MOSFET is turned off, then the bottom fet shorts the input side of the inductor to ground, so the current flows out the inductor, through the load, though the MOSFET, and back into the inductor. Once the energy drops, the bottom fet is turned off, and the power is again applied to the inductor, "filling it up". And repeat.

Also, the shottky performs a little different function. Losses incurred during a brief period just before or just after the switching transition, during which the freewheeling MOSFET conducts the current with zero gate voltage. This forces the current to flow in the internal body diode and has a significant impact (1-3%) on the efficiency because of a much higher voltage drop across the device during this period. Putting a shottky diode across the lower MOSFET lowers these losses during this period.

However, a similar situation applies to a boost, but the schottky diode would go across the top MOSFET. In both cases, the body diode also stores charge, which can add additional losses during the switching time period. Schottky diodes store very little energy, unlike MOSFET body diodes.

At the moment, I am using some HC7-1R5 made by cooper electronics, in a series/parallel combination.

I was using a 10uH inductor, think it was a CDRH127 style.

Also tried some UP4B cooper inductors.

I'm waiting for a HC2-2R2 and a HC2-6R0 from them.

Panasonic also makes a decent part also, a ETQPAF4R8HF, which starts out at 6.8uH and drops to about 4.8 at like 10A or so, and it has a DC resistance of 0.0029.

The actual numbers might be slightly different, doing all this from memory.

There is a CEP125-6R0 that doesn't seem too bad either, from Sumida, with a DC resistance of 0.008(max) to 0.0066(typ.). (there are much better inductors on the market now, for one to choose from, fyi)

The Zetex ZXSC300, ZXSC310, ZXSC400 and such "regulators" used in many of the boost circuits talked about on CPF only run at 200KHz, which requires an even bigger inductor than the TI chip does, and for the same ripple as the Zetex part, this TI part requires much lower capacitor values. So, both the inductors and capacitor physical sizes can be lowered. For a given design, there is a sweet spot where you trade off size, frequency, and efficiency, tailoring to your specific needs.

Higher switching frequencies also equates to higher loses on the MOSFET gates, because you need to drive that gate capacitance (and other associated capacitances) around, and do it much more often, at higher frequencies. The faster you drive it, the more energy you waste there.

Power loss due to gate charge = Switching frequency * Total Gate Charge (Qgd_total) * Maximum Gate Voltage

Another advantage to the TI part for those that are spaced constrained, is that both the fets are internal to the TI part...

Comparing to the venerable LT1618 for a high current switcher, in a discussion with georges80 on CPF.

-doesn't have a 4.5A switch inside it
-plus it uses the external shottky diode (eats up your efficiency)
-and with the high switching frequency (eats up efficiency)

The LT1618 datasheet shows around 76% efficiency at load levels of 300mA....would imagine it gets alot worse fast, with heavier loads...

The internal "switch" Vce(sat) of 400mV * 2.0A = .8 W * DutyFactor = X Watts of heat alone at full load, just within the chip itself, just from it's switching element.

Toss in a schottky diode (lets use MBRM120), lets say an average of 350mA flows through it to the load. 0.35 A * 0.34Vf = 0.119 W up in heat in the schottky diode. Or, if you had a 1 amp load, (0.38Vf * 1A) you'd loose 0.38 W in just the schottky diode!!!

True, it does have the high frequency advantage so you can use smaller inductors/capacitors. Although, I have gotten
the TI part to work with sub 1 uH parts without issues, and since output ripple isn't a big issue...maybe I will fiddle with making a mini version.

BTW, I do have some self induced losses in my setup with the parts I have sitting around the house, but at 0.8A output, I'm seeing 94.9% efficiency.

[dat2zip]
Wow,

I like what I see. I agree with the issues with the LT part and for that matter almost any regulator has issues of one form or another. There is no magic 100% efficient converter and there is always going to be losses.

I'd be interested in working on this design to make a miniature version. I think the only stickler would be getting a small inductor for the 4.5A requirements. Other than that I think this part has a lot of potential.

Tame this into a CC design and you would have a nice converter.

Jarhead,

Can you do a couple of data points for us. Maybe, from 2V to 3V every 1/10V or so. I think the efficiency will stay high if the datasheets are correct.

Wayne

(Note: Since this time, Dec of 2003, dat2zip has produced a constant current regulated board based off this chip, known as the NextGen converter, and has gone through several revisions to fix some quirks. In his design, he optimized for size, which cost was a loss of efficiency, but it is still quite respectable, at ~93% at fairly high currents, and even higher at lower currents.)

LTC3402 looks interesting, 3MHz, 2A switch 0.18 ohms, efficiency dies very quicky at greater than 0.4 A loads.

I hooked up a standard off the shelf Cooper Electronics UP4B-4R7 inductor with a DC resistance of 0.0093 ohms to the TPS61030PWPR boost chip.

Input...Input..Watts.......Output.Output.Output...Efficiency
Voltage.Current............Voltage.Current..Watts
2.955...0.92...2.7186......3.466...0.768...2.661888..97.91%
2.541...1.16...2.94756.....3.466...0.808...2.800528..95.01%
2.262...1.34...3.03108.....3.466...0.808...2.800528..92.39%
2.031...1.544..3.135864....3.467...0.812...2.815204..89.77%
1.861...1.76...3.27536.....3.471...0.832...2.887872..88.17%
1.728...2.004..3.462912....3.479...0.852...2.964108..85.60%

All measurements with the same meter, and using current sense resistors.

BTW, the load was an LED, thats why slight changes in output voltage caused larger variations in current.

Current sense resistors used were Dale-Vishay WSL2512 Power Metal Strip® Resistors, low-inductance, SMT, 1% devices.

[CPF-georges80]
Jarhead, good info - can you run the tests with 1A approx into the load - presumably you can crank the output voltage up a 'tad' to push Luxeon up the Vf curve a little more.

That should push it into the current realm that will 'prove' the TI's metal vs the opposition

thanks,
george.

Here is the part with ~0.5A load:

Vin.....Iin.....Watts.....Vout..Iout...Watts.....E fficiency
2.980 0.588 1.752240 3.328 0.508 1.690624 96.48%
2.801 0.624 1.747824 3.326 0.504 1.676304 95.91%
2.704 0.648 1.752192 3.326 0.504 1.676304 95.67%
2.598 0.668 1.735464 3.324 0.500 1.662000 95.77%
2.501 0.696 1.740696 3.323 0.500 1.661500 95.45%
2.400 0.724 1.737600 3.321 0.496 1.647216 94.80%
2.300 0.752 1.729600 3.320 0.492 1.633440 94.44%
2.208 0.789 1.722240 3.318 0.488 1.619184 94.02%
2.104 0.816 1.716864 3.316 0.484 1.604944 93.48%
2.003 0.869 1.722580 3.315 0.480 1.591200 92.37%
1.899 0.904 1.716696 3.313 0.480 1.590240 92.63%
1.799 0.969 1.727040 3.313 0.480 1.590240 92.08%
1.702 1.024 1.742848 3.312 0.480 1.589760 91.22%
1.616 1.096 1.771136 3.312 0.480 1.589760 89.76%

Here is the part with ~0.23A load, note that the data bounces around and is less linear, I assume this is the low power, energy saving burst mode kicking in:

Vin.....Iin.....Watts.....Vout..Iout...Watts.....E fficiency
3.001 0.264 0.792264 3.149 0.244 0.768356 96.98%
2.898 0.260 0.753480 3.140 0.236 0.741040 98.35%
2.810 0.276 0.775560 3.144 0.240 0.754560 97.29%
2.699 0.280 0.755720 3.140 0.240 0.753600 99.72%
2.600 0.292 0.759200 3.139 0.236 0.740804 97.58%
2.506 0.304 0.761824 3.138 0.236 0.740568 97.21%
2.402 0.316 0.759032 3.137 0.236 0.740332 97.54%
2.309 0.332 0.766588 3.136 0.232 0.727552 94.91%
2.210 0.340 0.751400 3.134 0.232 0.727088 96.76%
2.109 0.356 0.750804 3.133 0.228 0.714324 95.14%
2.009 0.372 0.747348 3.132 0.228 0.714096 95.55%
1.905 0.392 0.746760 3.130 0.228 0.713640 95.56%
1.806 0.412 0.744072 3.129 0.224 0.700896 94.20%
1.711 0.436 0.745996 3.128 0.224 0.700672 93.92%
1.618 0.460 0.744280 3.126 0.220 0.687720 92.40%

Here is the part with ~0.36A load (e.g. slightly over driven 1W Luxeon)

Vin....Iin.....Watts......Vout..Iout...Watts...... Efficiency
3.000 0.416 1.248000 3.247 0.376 1.220872 97.83%
2.919 0.428 1.249332 3.247 0.376 1.220872 97.72%
2.808 0.444 1.246752 3.246 0.372 1.207512 96.85%
2.717 0.456 1.238952 3.244 0.372 1.206768 97.40%
2.613 0.472 1.233336 3.242 0.368 1.193056 96.73%
2.502 0.488 1.220976 3.241 0.364 1.179724 96.62%
2.416 0.500 1.208000 3.241 0.360 1.166760 96.59%
2.308 0.528 1.218624 3.240 0.364 1.179360 96.78%
2.219 0.552 1.224888 3.239 0.364 1.178996 96.25%
2.100 0.588 1.234800 3.238 0.360 1.165680 94.40%
1.899 0.656 1.245744 3.236 0.360 1.164960 93.52%
1.813 0.688 1.247344 3.234 0.356 1.151304 92.30%
1.697 0.732 1.242204 3.233 0.356 1.150948 92.65%
1.594 0.780 1.243320 3.230 0.352 1.136960 91.45%

Ran the 1A test.
This time I used a little teenie tiny inductor, as georges80 asked in a PM, a Cooper Electric 2.2uH UP1B-2R2 DCR 0.0363, I loose a good chunk of efficiency (appx. 2% from the inductor alone).

Vin.....Iin....Watts......Vout..Iout..Watts......Efficiency
3.012 1.300 3.915600 3.497 1.052 3.678844 93.95%
2.898 1.364 3.952872 3.495 1.052 3.676740 93.01%
2.698 1.484 4.003832 3.493 1.052 3.674636 91.78%
2.615 1.540 4.027100 3.492 1.052 3.673584 91.22%
2.511 1.620 4.067820 3.490 1.052 3.671480 90.26%
2.409 1.704 4.104936 3.489 1.052 3.670428 89.41%
2.306 1.836 4.233816 3.488 1.048 3.655424 86.34%
2.197 1.916 4.209452 3.486 1.048 3.653328 86.79%

Here is the 4.7uH UP4B-4R7 at ~1 Amp out

Vin.....Iin.....Watts.....Vout..Iout...Watts.....Efficiency
3.002 1.268 3.806536 3.495 1.040 3.634800 95.49%
2.903 1.324 3.843572 3.493 1.044 3.646692 94.88%
2.805 1.372 3.848460 3.492 1.040 3.631680 94.37%
2.709 1.428 3.868452 3.490 1.040 3.629600 93.83%
2.604 1.496 3.895584 3.490 1.040 3.629600 93.17%
2.520 1.556 3.921120 3.490 1.040 3.629600 92.57%
2.417 1.632 3.944544 3.488 1.040 3.627520 91.96%
2.310 1.728 3.991680 3.488 1.044 3.641472 91.23%
2.210 1.828 4.039880 3.489 1.044 3.642516 90.16%
2.097 1.960 4.110120 3.490 1.044 3.643560 88.65%
2.018 2.076 4.189368 3.492 1.048 3.659616 87.35%
1.905 2.272 4.328160 3.496 1.060 3.705760 85.62%
1.818 2.492 4.530456 3.502 1.076 3.768152 83.17%
1.718 2.824 4.851632 3.508 1.100 3.858800 79.54%

Well, it looks like my new inductors/capacitors will not arrive until January 5th. I found some 4.8uH parts that have a DC Resistance of 0.0029 ohms, vs. the UP4B-4R7 DCR of 0.0093, which may possibly buy another percent per 1A of output, or and additional 2.5% efficiency at 2A of output.
But this is not the only factor in making an inductor (there are also B-H magnetic losses), so actual tests should prove to be very interesting.)

Also ordered some 0.015 ohm ESR Aluminum Organic 220uf capacitors (the 0.012 ohm ESR panasonic ones cost a 30% more). Energy is dissipated in a capacitor which also adds to losses and loss of efficiencies...

And tossed in some 10uf X7R Ceramic caps as well as some 0.01uf COG/NPO parts for filtering on the input side.

Patience....

[CPF-Doug S]

Very nice work Jarhead. I share your interest in pushing converter design for high efficiency. The TPS61030 that you are working with has been on my list of stepup ICs worthy of using in battery powered Luxeon lights. I haven't actually done anything with it though. If McGizmo has been following this thread he probably has drool dripping from his chin at the prospect of circuits with the efficiencies you are obtaining being reduced to fit on a 14mm diameter board. I haven't the heart to tell him the size of the components involved
Well OK, maybe 14mm in diameter but 2cm tall, stacked boards.

Heheh, with the 4mm x 4mm chip (RSA package), since you need no extra boost fets, and no schottky diode, toss in two 0603 package resistors, and since you are not worried much about ripple, you could back off on the output capacitor say a little 1210 ceramic 100uF 4V, and one 0805 10uf 4V ceramic...

1 4x4mm RSA package chip
2 0605
1 1210
1 0805
1 inductor on the backside

so you'd have a *very* small footprint, smaller than
a SO-8 pin footprint...looking at in in my PowerPCB
program, alot smaller than your 14mm diameter target.

You could make it work in a 8mm (0.31") diameter board, with a tiny inductor I have sitting here, it would be
7mm (0.28") tall, and still push 1A well over 94.0% efficiency on a new battery. (found a new inductor
today, a Sumida CDRH6D38-3R3).

In otherwords, you could squeeze it in the base of a PR12 bulb and still have enough room to fit a Luxeon III inside the metal base of the lamp (think inverter and the entire luxeon within the metal base of a common PR12 or PR6 flashlight bulb base!)

Hows that for small?

: )

[dat2zip]
You using the 4mm package? Was looking at that. Nice. That's very small.

Well, keep us informed. I like hearing more about it.

How's the pricing on the IC?

Wayne

Right now, I'm just adjusting the voltage output for each Luxeon, to give the right current, after the Luxeon heats up a bit.

Maybe, if I get some time, I'll rig up a current sense scheme for it, since several folks seem to really want
a current regulator. Of course, it will hurt size a bit, making it just a smigin too big to fit in a normal flashlight bulb, but I do have some op-amps here that are rated down to 0.9V, that I managed to make work on a homebrew switcher, that would work perfect for a current sense setup. Oh, humm, it just occurred to me how I might make it work without an additional op-amp, but it would hurt the efficency significantly, so that idea is no go. Okay, got me thinking about it, so maybe I will whip it together.

(since this time, I discovered the pre-bias technique, which allows one to use a smaller sense resistor, and get rid of an op-amp. TI, Linear Tech, Maxim, and others now have application notes and variations on the technique. It also helps get rid of some of the gain/phase issues when using an op-amp, and some of the quirks some folks have had.

[CPF-Doug S]
If McGizmo has been following this thread he probably has drool dripping from his chin at the prospect of circuits with the efficiencies you are obtaining being reduced to fit on a 14mm diameter board. I haven't the heart to tell him the size of the components involved

Well OK, maybe 14mm in diameter but 2cm tall, stacked boards.

Jarhead. Great stuff!
I take it all back! I have an older version of the TPS61030 datasheet on my hard drive which only shows the TSSOP-16 package. Hell, the IC alone would not fit in an 8mm diameter circle. Well I just checked the TI site and they *are* bringing it out in the QFN-16 (RSA) package. Too bad I have neither the equipment or the nano-balls to attempt working with that leadless .5mm pitch stuff [img]/ubbthreads/images/graemlins/frown.gif[/img]
Now about fitting it in a PR lamp base, I am a bit skeptical that you can find an inductor that won't saturate on that 4A IC current limit.
You are quite right that for driving a Luxeon load you really do want the extra circuitry for the constant current regulation. Too bad they didn't bring out a comp pin on this IC. My limited experience with forcing PWM voltage control SWVR IC's into a constant current mode is that it is helpful to have access to the comp node which you can load way down with capacitance thus reducing the need for fast response of your current sense circuit. You can in fact input your CC circuit directly into the comp pin thus over riding the voltage control loop which typically can source/sink much less than 1mA.
Don: [who is on a quest to make the eyes bug out of Luxeons with only a single CR123] I think Jarhead is your man.

Don: Translating part of this foreign language flik: You can kick your drooling into high gear

Fitting this in a PR type bulb.

Well, if you look up that Sumida part (CDRH6D38-3R3),
you will see that it has a DCResistance of 0.0015 ohms and it is actually rated for 3.5 A, and its maximum dimension diagonally is 9.5 mm (0.374") and today I busted up a PR3 bulb I had, it's internal dimension is 8.6868mm (0.342"), so it won't quite fit. Dang.

Ah, but there is a CDRH5D23R that is 3uH and a DCR of 0.0015, a maximum diagonal dimension of 8.2mm (0.323") which would fit, and it is rated for 2.3A.

Humm, so if we assume a Luxeon III has a Vf of 3.9, and draws 1 Amp (since I don't have one), and at one amp output, we should have 94% efficiency with this little inductor at 3.0V, it starts out at 3.0V at 1.378A, and as the battery voltage drops, the efficiency drops to around 90%, we are looking at 2.2V at 1.95A out of the battery, which is well within the rating for the inductor. (should be in the ballpark without having the inductor in hand).

My question, is on a little CR123, will it put out the power, and for how long?

Second question is, if we backed off on the output, how long will a CR123 put out 1A (from the samples of the chips I have here, they will run down to 1.610V or so, but you start loosing more efficiency)? And, since the Vf of the LED drops, you don't have to boost the output voltage as high...

Now, if you don't constrain yourself to a tiny inductor, I did run some tests today that give me 95.5% at 1A on a new inductor I just received, when the input voltage is 3.0V. But this inductor would only be suitable for a C or D cell flashlight, since it is 17.2mm on a side, and 9mm tall.
(datasheet for the inductor is here, http://industrial.panasonic.com/www.../AGL0000CE2.pdf )

Can anyone spin me up on small sized batteries and their output abilities? I'm used to having aircraft power...

[CPF-Doug S]

Duracell has the most useful of the CR123 datasheets I've seen:
CR123
Also, you might find this thread of interest:
Heavily loaded 123
As you can see from the Duracell datasheet, available energy really takes a dive at high power drains.

Okay, please tell me about NiMH batteries then.

[CPF-georges80]

They'll supply a couple of amps without too big a hit on their rated amp/hours. They have very low internal resistance. The problem is most of the torchies (flashlight nerds as I refer to them) like 123's because they are a) light, b) don't self discharge, c) are good at low temps, d) are small, e) are cheap these days, etc etc...

I myself like nimhs because I can recharge them while camping and they give good runtimes in high current draw situations.

Again, you can visit eveready's website and you can download the specs of their nimhs and get an idea of how well they handle higher current draws.

george.

Added a 10uF (yes uF) ceramic capacitor to the input (already had a 200uF low-ESR tant), and I can't even see spikes any on my 350Mhz bandwidth o-scope on the 50mV setting, plus gained more efficiency.

Vin.....Iin.....Watts.....Vout..Iout..Watts......Efficiency
3.022 1.312 3.964864 3.502 1.084 3.796168 95.75%
2.509 1.620 4.064580 3.501 1.084 3.795084 93.37%
2.199 1.904 4.186896 3.501 1.088 3.809088 90.98%
2.011 2.140 4.303540 3.505 1.088 3.813440 88.61%

Aadded another change, which gave me yet better efficiency.

I just got in some Aluminum Organic 0.0015 Equivalent Series Resistance 180uF capacitors, plus added a compensation capacitor (which is now over compensated because all I had was a 27 pf), bought me another 0.5 to 2.0 % more efficiency, depending upon where on the graph you look.

Vin.....Iin.....Watts.....Vout..Iout..Watts......Efficiency
2.999 1.332 3.994668 3.510 1.092 3.832920 95.95%
2.507 1.616 4.051312 3.500 1.088 3.808000 93.99%
2.205 1.836 4.048380 3.478 1.080 3.756240 92.78%
2.002 2.012 4.028024 3.444 1.060 3.650640 90.63%
1.807 2.184 3.946488 3.399 1.028 3.494172 88.54%
1.616 2.368 3.826688 3.336 0.992 3.309312 86.48%

The cap datasheet is here: http://rocky.digikey.com/WebLib/Kem...0Data/AO%20.pdf

These AO are nice compared to Tantalum and Nibium Oxide, in that you can over voltage them, even 3X and they don't blow up, nor go up in smoke/flames, and often they even recover.
I figured at higher currents, the ripple current in and out of the capacitor would go up, and a lower-ESR part would dissipate less energy, and my graphs here show it to be so.

Check these numbers at 0.57 Amps output...

Vin.....Iin.....Watts.....Vout..Iout..Watts......Efficiency
3.000 0.672 2.016000 3.500 0.568 1.988000 98.61%
2.818 0.724 2.040232 3.501 0.572 2.002572 98.15%
2.504 0.816 2.043264 3.498 0.568 1.986864 97.24%
2.213 0.936 2.071368 3.495 0.572 1.999140 96.51%
2.003 1.032 2.067096 3.495 0.568 1.985160 96.04%
1.806 1.176 2.123856 3.496 0.572 1.999712 94.15%
1.599 1.364 2.181036 3.500 0.572 2.002000 91.79%

And now we have 0.38A numbers

Vin.....Iin.....Watts.....Vout..Iout..Watts......Efficiency
3.007 0.456 1.371192 3.508 0.388 1.361104 99.26%
2.813 0.488 1.372744 3.506 0.384 1.346304 98.07%
2.510 0.544 1.365440 3.503 0.384 1.345152 98.51%
2.211 0.624 1.379664 3.501 0.384 1.344384 97.44%
2.003 0.700 1.402100 3.501 0.384 1.344384 95.88%
1.810 0.780 1.411800 3.500 0.384 1.344000 95.20%
1.610 0.884 1.423240 3.498 0.38 1.329240 93.40%

Check out that +99% efficiency...too bad we can't use these
big honking parts in a miniature version...But fine with C and D form factor...

[CPF-Doug S]
Very impressive

Any estimate on the error bars for those efficiency values?

I'm going to swap the 1% current sense resistors from the input to the output and vice versa.

The measurements are much better with the last round, as with my 350MHz o-scope and on the 50mV scale I see no ripple, and I see some 2-4 nanosecond 10mV spikes on the output, but the DC level isn't shifting. I may have to dig out my 1X probes... The same voltmeter was used for both the input and output, for both voltage and current. When I swap the sense resistors, that will be the "proof" for me. Using the same measuring instrument for all measurements gets rid of the errors from using multiple meters. In fact the meter could be purposely set wrong, and you'd get the same %

I'm using much better parts than TI uses in their example, where they show 96% efficiency. Their CDRH124 6.8 uH has 0.023 ohms of DCR and a saturation of 4.9A, I'm using the Panasonic ETQ 4.8 uH with 0.0029 ohms of DCR and a saturation of 10.6A. At 1A, that alone should buy me over 2%.

I'm using some super low ESR caps, which are a factor of probably four times lower than they used (especially looking at the date on the datasheet...), so I don't loose as much energy as it goes in and out of the capacitor with each pulse. I'd figure the part they were using had a ESR of 0.060, mine are 0.015.

Additionally, they probably used 1 oz. copper or 2 oz. copper, I'm using 4 oz. copper planes for traces.

Of course, my last home project was a 1200W switching power supply (right after I did a 200W one at work), so my typical construction techniques and materials are a little different. You should have seen my load bank of resistors bolted to the 1/4" thick copper plate with thermal compound underneath. Trust when I say the load bank got real hot.

I'll swap the resistors and let you know.

[CPF-Doug S]
Jarhead, I like your style!
It is interesting that you are getting so much benefit from the added capacitance on the input side. Assuming that your power supply is fairly stiff and the leads reasonably short, I would not have thought that you would need so much capacitance on the input side. I would have been tempted to do this build with only the 10uF ceramic on the input side.
BTW, you are doing your measurements across your sense resistors with an *average reading* meter and not a true RMS meter? This is a case where using a true RMS meter is not appropriate.

Got to thinking.

Which gave me one of those stupid simple ideas. Wire up the two resistors, dump alot of current through them, then measure the voltage drop on each resistor. The difference
between the two WS series Dale resistors was 0.19%. So, the resistors are nearly matched. We did one of those measure 1000 of this part, at work, and found all of them to be within 0.5% rated value, and most to be within 0.2%.

Probably couldn't have found a better set of matching resistors if I had tried... Hows that for luck!!!

Dale-Vishay WSL2512 Power Metal Strip® Resistors, low-inductance, SMT.

So, since the same meter was used for all measurements, and if it was off, the same amount off would apply to all measurements, nullifying most of the the calibrated accuracy concern of the instrument, since efficiency is a input/output ratio number.

With no DC shift in input or output, that gets rid of another concern. The 2-4 nanoSecond spikes of 10mV from the MOSFET switching/inductor in a 1,600 nanoSecond period would add nearly a non-measureable difference.

Looking at the internal MOSFET resistance, of 0.055 ohms,
and the DC Resistance of the coil at 0.0029 ohms. And for the 0.38A numbers, for this TI part, I take 0.38 Iout *(3.5 Vout/(3.0 Vbatt * 0.8))= 0.554167A average inductor current.
I get a loss of 0.01778 Watts. Because of the "large" value of inductance chosen, the inductor ripple current is under 5%, and I am using a way over spec'd inductor, I expect very low losses from this. The chip does not mention it's current draw during operation.

So, we have inductor and MOSFET losses of 0.01778 Watts and input power of 1.371192 Watts. 98.7% efficiency + plus some additional loss for the power to operate the chip. You could probably add in a little tiny bit for the losses in the super low ESR capacitors, and since I use 4 oz. copper planes, the trace resistance would be quite low.

But, there is a chance that the internal MOSFETs have a lower on resistance, since the part probably has gone through some improvements. In talking to one of the TI
Engineers on another TPS series part, they had lowered the on resistance a good amount.
If so, this would drop the losses and raise the efficiency.

I also allowed 5 minutes warm up time, before I started measuring anything.

So, in the end game, these numbers are fairly decent, I couldn't see them being off much more than 1% on the efficiency numbers.

Here is a big chart, works best on 1600x1200 resolution:

Some of the gyrations are due to lack of proper compensation, I don't have any 10pf caps to compensate
for the ultra low ESR cap I put in. I added a 27pf cap after the purple measurement line, but it still overcompensated.

Oh, forgot to mention a really cool feature of this part, it has a input voltage measuring pin, and a output pin.

A fella could wire that up to make a little surface mount LED light up to indicate when the battery is getting low, to warn the user that the voltage has dropped to X, to indicate there are only X mount time time left...

Adds to the "cool" effect!

(another thing I have used this for is to delay the connection of the load, to get around start up issues, like where under certain battery and load combos, the part gets stuck in it's pre-regulator start-up phase.)

[CPF-Doug S]

BTW, you are doing your measurements across your sense resistors with an *average reading* meter and not a true RMS meter? This is a case where using a true RMS meter is not appropriate.

First, the benefit of the 180uf AO 0.015 ESR cap is due to it being put on the output, not the input. I already have a 220uf 0.045 ESR Tantalum on the input, plus the 10uf ceramic on the input.

Okay fella, time to call your bluff on your average reading meter comment. I pulled out the 1X probes to get the 5mV scale.

Note the 2mV of DC shift. Now, tell me, just how much error would this induce in the meter reading???

; P

[CPF-McGizmo]
Q from the peanut gallery:

Am I still drooling or are we up against a brick wall if the 123 cell is the source of power? If I am following this correctly, this circuit is dang capable of passing through most of the power to the LED. On a single 123, what would be a reasonable current level to target for max sustainable illumination; say in on cycles of a minute or less if that makes any difference.

[CPF-Doug S]
Don, drool away! With the efficiencies that Jarhead is obtaining with his circuit, even allowing for some reduction due to the restraints of reducing the size to a form factor you would like, I think that 1A into a luxeon with typical Vf binning would be achievable with your 1X123 in 1 minute bursts criteria.

Here is a look at the prototype I'm working with now.

[AilSnail]
looks like a regular nuclear war zone. whats that finned big thing?

In response to PMs... It's nice to hear from other EE's that they are seeing higher than 96% efficiencies when they build the TPS61030 based design.

[CPF-LED_ASAP]
I hate to pour a bucket of cold water on this hot IC, but I wonder if others have noticed this <font color="red">feature</font> of the TPS61030 IC:

Until the output voltage is reached, the boost switch current limit is set to 40% of its nominal value to avoid high peak currents at the battery during startup. When the output voltage is reached, the regulator takes control and the switch current limit is set back to 100%.

In other words, if you connect a LS to the output and then start the power, you can only run it at 40% of the maximum capacity because the IC will be clammed down at the "soft start" mode.

I noted this problem on my prototype: If I connect the ls and then connect the power, I measured 3.26V across the LS. If I connect the power, then the ls, I can get 3.49V across.

I haven't figured out a good way of hacking this feature, except maybe build something that have the boost converter permanantly connected, and hope the quiescent current will be as low as the datasheet claimed so it won't drain the battery down too much. Then I can put the switch on the LS side

LED mods As Small As Possible,

Actually, I've built two of these without this issue, one is running in my flashlight off a 3W luxeon, adjusted for 1A from the converter, and I am using 2 AA low-capacity NiMH and Alkaline with no problems at all. Another EE has built one, and has it running fine too. If you have used Y5V ceramic caps, say like the 6.3V ones, the capacitance will reduce to 20% of it's value (a 10uf will become a 2uf at 60% of the voltage rating), and the loop will go unstable and keep trying to restart. If you dropped below 20uF, you will have to compensate it to get it to start properly. If you used ceramics, you may need to adjust the recommended 10pf cap mentioned for ceramics (read as very low ESR parts). If anyone is thinking of using Y5V/U, you are going to have a small headache. I'd never ever use a Y5V/U when doing smoothing/filtering, stick with the X5R or X7R.

Now if you used the TPS61020, I could see you having an issue, as it has a spec on the sheet of max load of 125 ohms at Vin 1.2V during startup, and I have seen your issue.

There are a few tricks if you went outside the datasheet recommendations, drop by the chat room and talk to me.

BTW, I ran this part through all my testing for three weeks with a load on it, never failed to start under load, even a 1.3A load.

Again, it does work fine if done right.
(another trick is to use an external MOSFET to switch in the load, after it is started-fyi.)

[CPF-Burnt_Retinas]
Jarhead,

I agree, but I'm curious if the 61030 will run out of huff before I reach my target. Target = approx 3V in, tad under 4V out @ 1.5A (a 3W at 1.5A from Li-Ion all the way to battery being depleted). Though the 3402 gets relatively poor efficiency results at high currents, I figure it may go beyond the 61030's capability? I don't know. I don't have both. Perhaps soon though. BTW I was disappointed at the performance of the 3402's switching FET's performance - surely they can do better than a poor 0.16R. Needs a diode too to to gain any real efficiency. Didn't think I'd have any option though.

Thanks guy's

Chris

Hey Chris,

I've got it in a flashlight with 2AA NiMH, doing 3.742V at 1A, with no issues.

I've also ran it up at 1.3A, and it was plenty happy, even down to 1.8V and it slowly drops below 1.8V.

One of the keys that helps keep it going as the battery gets weak is to have some bulk capacitance on the input, so as the inductor grabs a chunk of juice, it doesn't pull the input voltage down below the cut-off voltage.

Also keep your inductor resistance down at these higher power levels, don't go above 0.015 ohms DCR.

With a nice big inductor and 1.7 Amps out, and a inductor with a DCR of 0.009, I'm well over 90% efficient at 3V in.

[AilSnail]

Eureka! After having went through six 61030 chips, and a host of 0402 resistors and capacitors, I recycled one of the unfeeted chips, and finally made it work!!

setup is as per jarhead's cirquit diagram, deadbug on one of wayne's emitter boards, using x5r caps and 3.3uH sumida cdrh8d28r.
Since I don't have any low inductance sense resistors nor a high res dmm I won't be taking any efficiency measurements yet.

I added a Lusense pressure sensor with another resistor in parallell (between FB and R2), now it is adjustable by finger pressure!

[CPF-Burnt_Retinas] Any advancements? Any pics of completed projects?

I just got my samples of LT3402 (took forever!) chips and was wondering if I should bother building or just wait a little longer and get some 61030's.

Chris

The LT3402 is definitely less efficient. Especially at higher currents, it becomes quite significant. The LT3402 has three times the resistance in the internal switches. Been there, done that.

Example, say you have 2A average current. Lets average the two internal switch resistances of the LT3402, 0.17 ohms. I^2*R=W 2^2 * 0.17 = 0.68W of heat just from the resistance of the internal switch.

TPS61030, internal resistance 0.055 ohms. 2^2*0.055 = 0.22 Watts of heat.

The LT3402 creates three times the heat, which is power lost, heating up your flashlight. The real calculations are much more technical, but this simple example will get you in the ballpark. There is additional losses for higher frequency drive.

Just from resistances, the LT3402 will have 3x the losses. With over 1/2 a watt of losses inside the chip, you might think about thermal sinking it. When it gets hot, it's internal switches will go higher in resistance, and the losses will increase further.

The LT3402 is a nice choice for a single cell flashlight, since it will run down to 0.5V, stop it with the LBI/LBO if you are going to use rechargables, at around 0.9V, or you risk cell reversal issues.

Please keep in mind, proper layout is absolutely critical at these high currents. It can make the difference from something that goes poof, to something that works like a charm, even with the same current. Everyone I have talked to that has problems, a significant part of their problem has been in layout/ground loops, causing things to go unstable. Pay attention to the datasheet *very* carefully. (you can buy a finished converter with this chip from dat2zip, aka NextGen, I think he has most of his bugs worked out now?)

Update:

The 1uF I mentioned, I just realized it happened to be a Y5V from the old package I had. The interesting part of Y5V high density capacitors, is their capacitance drops to 10-20% when you get near their rated voltage. I just built another one, and I used a X7R, and it has been driving me nuts, it would pulse during the first second that you turned it on. Once I realized what was going on, I switched to a 50V 0.1uF X7R, and all is fine now.

[Chimo]

Just noticed this thread.

4sevens, thanks for the bump.

Newbie, this was a good read and it was interesting and informative to see your design iterations. Thanks,

Paul

Thanks Chimo.

Since that time, I've further changed how I did things, and added a little circuitry myself, and have also made it do constant current.

It was a fun first project that actually ended up in a finished light(actually several now), after constructing several dozen other regulators to evaluate the performance.

Has anyone else had a chance to work with this chip any further?

The power paths are very critical, and don't take their layout suggestions lightly.

[JasonC8301]
Haha, whose gonna give me a electrical engineering degree in advanced flashlight 101 after reading this thread?

[CPF-glire]

What about the TPS63000 ?

Not a bad choice for lower currents, like 350mA, but efficiency suffers a bit due to the buck-boost mode:
http://www.ti.com/corp/docs/landing...ble+PA+tps63000

Datasheet is here:
http://focus.ti.com/lit/ds/symlink/tps63000.pdf

Though it might not be that bad for Li-Ion use.

FYI, the TPS63000 chip forms the basis for dat2zip's GD series of converters that you can purchase from the sandwich shoppe.

Here is another technique for doing a current regulated buck supply (which can also be done with boost supplies), by using a pre-bias:
http://www.edn.com/contents/images/6409620.pdf

Here is another technique for doing a current regulated buck supply (which can also be done with boost supplies), by using a pre-bias:
http://www.edn.com/contents/images/6409620.pdf

TI has some app notes where they also discuss the technique.

[UncleFester]
Hmmm.... I wonder if a diode could be used to generate the offset. It's fixed voltage might improve regualtion over using a resistor?? Dunno, I have't fully thought it through.

There are plenty of techniques one can use, TI, Maxim, Linear, and others have some app. notes if you look around.

Check here for some of them
Switching Regulator chips, buck, boost, buck-boost, sepic, and tricks and techniques for making constant current, as well as CC without an op-amp... . (Mirror Site)

Just how much difference can an efficient converter make in a flashlight? Here is an ARC4 retrofit I did, with some plots as I worked on it's already great performance, against another ARC4:

Main site
Mirror:
Main site mirror