I have done this too, but it comes with a high price. The control loop will be really slow. Might not be a problem for a LED lighting application tho.
The problem is that you lose the in-cycle current limiting an analog controller usually offers. This might end really really bad for the switching transistor unless you take one with very high current ratings, combined with an inductor so large the current cannot rise to a critical level at maximum duty cycle. (or you add external circuitry for peak current limiting).
This might be a little to much for a 17mm board.
Flashlight applications have a very well defined and very slowly varying load. That suits MCU based controllers very well. You can characterize the required PWM setting for a given current (and battery voltage) and use that as the initial condition for the control loop.
Texaspyro, dave_, all valid points. The coil would need to be 5A for a 3A driver, at least. Driving at 1MHz is not feasible for an MCU, so the coil size would have to increase. Direct driving a FET above 50kHz is not efficient at high currents.
I’m not too concerned about the speed of the control loop. For this application, running as fast as the ADC can go should be fine. The supply is nice and stable, and with hard limits on the PWM duty, it can avoid catastrophic failures in loop response.
Now, if we can find a way to get both the MCU and the boost controller on there, I’m all for it. I was just suggesting a possible alternative that might allow us to cram 9lbs into the 5lb bag instead of cramming all 10lbs. 
Thanks to everyone for your efforts. PLEASE dont give up in these discussions. You can make it happen! Also, feel free to invite others to this thread if you feel they might be able to contribute. You’re all off to a great start…
I somewhat doubt a converter of that power level will fit on a single sided board in that tiny area togehter with the µC. I probalby would build it on something like 0.5mm RF4 and give the controller it’s own board. But from what I read many people here dislike multi board drivers. 
I’m not so much worried about control loop instability but more about the current peaking within a cycle. A inductor that would physically fit into the little space at a current rating of ~5A would be in the range of maybe 5…10µH. Just quickly estimate the time constant of that inductor in the head. On time would have to be really short to avoid to high current. With adequate switching frequency, high resolution pwm and extensive testing/data collection it might work out tho. I really like digital control, but ultra fast analog control has it’s charme too. 
I whish I had time to play around with that, but that sadly has to wait a few months before I get some more freetime. 
Current spikes are an issue, but remember, this is a boost converter. As long as we limit the maximum on time of the FET to no more than the theoretical saturation time of the coil at VinMax (say 4.3V), Anything else is just Direct driver through the coil, blocking diode and into the LED. Until Vin-losses drops enough to prevent direct drive (no FET switching) from delivering the current we want, the FET doesn’t turn on at all. Once we see current drop below the desired level, we start switching to boost things up where we want them again. This goes on until the battery reaches the low threshold….
OK, I just realized that I missed something when removing the boost controller. There’s no way to create modes without adding another MOSFET into the path. because a boost driver has the battery connected to the LED via the coil and diode, there needs to be a way to switch this with PWM in order to create modes. Even with the boost controller, the extra FET may still be necessary. Hmm…
Any chance a dual board setup can be used?
One would be a single sided board with the contacts and some components, and the second could be a double sided one.
I think there has been some significant development over at LPF, as most diodes require higher forward voltages that what a single li-ion can supply.
Here’s one for example, but it’s limited to about 2A:
Edit: The main reason why LPF doesn’t need 3A+ drivers (yet) is because laser diodes with the current technology can take only 1.8-2.5A at a Vf of 4.5V before dying. However, I’m sure we can get this working, and that would be awesome.
Sounds good. Don’t want it to let it burst into flames when someone tries out his new ultra low resistance cells.
Why not use buck-boost/sepic/cuk topology and then implement modes through set point of the control loop?
A buck-boost would do it, but that also requires two FETs. No free lunch today.
I’m not exactly a hardware designer expert, more of a firmware guy with peripheral knowledge of how hardware works (get the hardware working, then fix it in software ).
There is a single switch variant of the buck-boost topology. It inverts the output voltage, but that should not be a problem.
Hmm, firmware guy . . . you are not by any chance into DSpics and desperately want to implement synchronous rectification in SW? 
DSPics? I just threw up a little when I read that :Sp
Never was a Microchip fan, I have used them enough (16 and 18 series) but never really appealed to me.
These days I’m a little biased though, current employment and all
Sorry that I disturbed your stomach.
I’m not a Microchip fan either. But some of the DSPics have dedicated hardware for smps. The only others I know with those features are within the Ti c2000 family. That’s even more exotic, at least to me. Synchronous rectification is so nice I would even touch a Microchip part for that.
Are there constant current boost drivers that can deliver 3A from 2xAA?
Last time I looked (several years ago) you had to use a high current fixed voltage boost regulator, sense resistor, and an op amp to make it work as a constant current source. You multiplied the voltage from the sense resistor with the op amp and then used that as the feedback voltage for the regulator.
Maybe I’m missing the target here? 3-6V 3A to power multiple LEDs in series or what? 3-6V to power a single LED with 3A needs a buck/boost.
If possible, include compatibility for 2 & 3 NiMH/alkaline and you'll really have something special here. 3 NiMH shouldn't be a problem, and even if it has reduced output with only 2 NiMH, that'll still be far ahead of what's available now.
I’m not aware of any. Most capable of that current have an undervoltage lockout of at least 2.8…3V.
But one could convert a voltage regulator in the way you described. Would probably fit on an 16mm veroboard cutout if you have a calm soldering hand.
Yes, perfectly right, the driver needs buck/boost capabilities. (but hasn’t to be of the buck-boost topology)
This one from Fasttech has the specs to do 3A, but it’s been out of stock since January. I’d like to get one, just to see if it can do it. I suspect that it can only do 3A when running on a full charged Li-Ion. Since it is output voltage adjustable, there has the possibility to run two emitters in series.
Too bad it looks like they will probably not get any more.
Relic, I just submitted a request to FT for the driver you mentioned.
Its been awfully quiet over here. Does anyone have any more ideas or strategies? Please dont give up trying.
What about that 1A boost driver that was looking to scale to 3A in a future iteration?
Hey, who’s talking about giving up?
I jotted down a few ideas I want to try out on my ‘when I finally got some freetime’ list, but unfortunately I will not be able to build anything within the next 6…8 weeks.
If this is not done by others in the meantime, I will happily contribute what I can after that time.