A little quick searching does look like there are better fet alternatives to solve this, need to compromise a little on rdson probably. Might require a minor footprint/pad change. (maybe soic-8, so pretty close I think) No time to pin it down right now.
And yeah, like TA says, drawing a basic board that will get electricity in one side and out the other, well probably isn't easy at all actually, but the details of how well all the parts really work together are MUCH more complicated in a buck than a DD or linear driver. We probably wont predict all the issues, which is a reason at some point to get on with it to see, but when it doesn't work you want to have some handle on as much as possible too so it's not just time to shrug. Some of these things are easily predictable (probably even some of the ones we'll miss), and staring at a board that doesn't make light will only give so much insight into it. Although it's been a little quiet here, I have been making progress on a few things slowly. We just never really looked into the FET in any significant detail at all until now. Not a big deal, but I have about 70 rows of calculations or inputs, none of which are a big deal. Just takes time to get around to thinking about them all. It will get there, probably.
No problem, I have been swamped myself so I understand.
A couple of Mosfets:
http://www.digikey.com/short/3bjdqf
http://www.digikey.com/short/3bjd8h
Both are very similar, same footprint to the previous one (better double check the pins)
Both about 20x less gate capacitance and about 10x more Rdson, which works out as a big win I think. (Notice the 74W one quotes gate charge at 4.5V and the 100W one at 10V, so they’re pretty similar once that's corrected). I kind of like the 100W one but the difference is miniscule, so whichever costs you less. Don't know what I was reading. They are pretty similar but in fact the 75W one has lower gate charge and I'd pick it probably. Plus it's cheaper.
There are some other options with better numbers for rdson and gate charge, but either bigger, less standard, or worse thermal performance as far as I found. I think these are a good compromise. I don’t swear the the best possible, but they bring down a 2.2W low power drain to now about 0.2W.
There is an Rdson hit at very high power, 5.3% loss at 20A, regardless of duty cycle. That's 14W in 1:1 (but that's in a 280W output mode, lol). It maybe a little less crazy to see this happening in 4:1 where that's only 3.7W because 20A 4:1 is only 70W total. Either way, it's still a 5.3% loss at 20A though.
At a more reasonable 10A 1:1 output it's 4x less loss (I^2 R) so 3.5W, not so bad, especially since we saved 2W in gate drive (and maybe more in other fet capacitance and speed issues I suspect; these are just lighter on their toes). I definitely think this is a good tradeoff. Unless you only ever run your light on meltdown mode, these fets are better, but the buyer can choose.
I've got new numbers about ready. I want to revisit input cap selection still. I was wrong aout Coff charging. On balance of all the numbers, we should leave it alone. Further improvement can only be done by selecting the RC for the desired output voltage.
We're close.
Those both look good, I agree either should work and just pick one based on cost and availability.
Well I got confused (too many digikey windows, dizziness sets in). They're both quoted at 4.5V, so I'd pick the cheaper 75W one then with lower gate charge: STL30P3LLH6
If there's any competition, it's maybe this:
http://www.digikey.com/short/3bj1h8
Half the Rdson, but twice as much space.
Those are only sold in 2500 increments, so I am gonna say those are not real practical due to that lol.
Although while it may be able to fit I think that keeping to the powerpak footprint is wiser.
oops: http://www.digikey.com/short/3bpccw
There are a ton of options (probably more actually, only stmicro seems to do the powerpaks) in these TO-252's as well (but this is the only one that's better on digikey at least), so still leaves choices, but I'm still leaning to the powerpak too for some reason.
I updated post 96, marked things as selected, and added model numbers in case links break.
I guess one mark against the powerpak is the same thing as the LDO drama, that people have powerpaks that look like these already here for other drivers (including possibly confusingly, some of yours), but trying to just use what you've got laying around is going to not work at best. Packages are just packages though. Mostly the powerpak is smaller, so can be used better when downsizing.
Yeah, anything larger then the powerpak would have to be changed for a smaller driver anyways. Might as well start out with something that at least has a shot of being used in a smaller version.
Ok, the price of that is about 2.5W of heat at 15A 1:1 from the extra Rdson. I'm ok with that. And it's already drawn.
2.5W more then the larger FET?
If heat is an issue we can always upgrade later as well.
Ok, here are the right numbers. have to slow down a minute to remember what my own numbers do.
It's actually worse than that once an estimate of heat dependence is included. By hand it's about 2.8W difference ( 15^2*0.015 *14/16.8)but the spreadsheet estimates roughly the effect of increased resistance at high temperature, 150C junction temp (not case).
at 150C junction, 15A 1:1 the 30mohm powerpak becomes about 49 mohm(based on some generic curve, but it seems to be reasonable) giving 9.6W
the 15mohm TO-252 is well half of that so 4.8W for a 4.8W difference when hot.
So to bracket it, between 2.8W and 4.8W difference at 210W total output, depending on heat sinking and how long you leave it in turbo mode.
Update: so for the powerpak 31 C/W junction to PCB 2C/W junction to case. At Tj 150C the Rdson is actually over 2x higher than at 25 so I'm still underestimating significantly for that particular fet.
The To-252 quote 50 C/W and 2.5 C/W for PCB and case but only a 50% rise in Rdson at 150C, and with fewer watts dissipated from the start so generating less heat anyway.
I don't really understand why there's so much thermal resistance to the PCB. Anyway, it's called turbo mode for a reason, won't last long at all.
Another nice option:
http://www.digikey.com/short/3bprcz TPN4R712MD
4.7mohn rdson
the 20V max is a bit scary though ( bucks ring)
http://www.digikey.com/short/3bpr78 (I'm liking this)
with 30V max and 8.4 mOhm Rdson
Twice the gate charge of both will mean about 0.4W 0.22 W (updated because it's quoted at 10V) minimum draw at 1Mhz. (Higher switch over to PWM helps optimize this out)
But at 3.5 x lower rdson compared to the previous powerpak, (2.8W 1:1 15A, when hot) a good option for a turbo oriented build. Total power dissipation handling is less, but it won't need nearly as much either.
(These thermal resistances are hard to reconcile. This one states 50C/w junction to ambient. Same package as previous one, so if I assume still 30 C/w j to case, that's more delta T between metal/silicon to metal than insulated metal to air. That's hard to believe. It's hard to really nail things down at this level, just have to choose one.)
I'm liking this one.
Good numbers, that last one is the best so far IMO. It not only has the best numbers but is also the cheapest interestingly. This driver is going to cost enough as is, anywhere we can save money and get better performance is a very good thing.
So when y’all get to where you need it idiot proof tested, I’ll try to build one. 
Richard says if there’s a tiny miniscule chance of something going wrong, I’m the guy that stumbles into it. lol So I hereby volunteer to be a guinea pig for this project.
LOL, I will keep that in mind. I think these have a good chance of working well but not of being cheap.
Once we nail down a parts list and design I will release everything open source.
Now all the numbers: (the third image is my favorite if you only look at one, but the component assumptions are in the top of the first image)
This time I’m showing everything across the board at 70W and 3.5W for apples to apples. That’s a bit low for 1:1 (5A) and very much pushing the limits of 4:1 (20A), but is a pretty good high power number for 4 xp-l’s in any configuration.
I’ve set the first image up with a fixed Coff RC for all voltages, tuned for 300khz and 1Mhz at 7V output. So this is what happens if you use the same setup at different voltages. However, I have changed the sense resistor so that max is 70W for all voltages (so not the same resistor). A 200W max setup will have less sense loss than this at 70W, but may have worse control at 3.5W if it can go that low at all.
You can see that this implies much lower frequencies (see “coff derived frequency”) for both 1:1 and 4:1. That results in large ripple current. Anything over 200% is discontinuous mode (which actually might not be such a bad thing, but I haven’t tackled it yet and would prefer to avoid it for now). Anyway, the numbers become meaningless for ripple much over 200, so basically the low power 14V results there are useless. It turns out this effect actually gets mostly worse with coff charged from a fixed voltage. The moral is, a driver should maybe be optimized for one configuration.
The next image is back to fixed frequencies, so a different Coff RC in each case. That’s more of a fair comparison of what each voltage output is capable of:
Finally here are configurations setup for constant high power current, 12A across the board. That seems a little too civilized for 4p xp-l at 3.5V (but may be all we can get anyway), is living on the edge for 14V xhp35, and is probably great for 2s xp-l at 7V if you’re a little brave, but anyway much closer to reality for all three. For low mode I still stuck with 3.5W because low mode doesn’t have power constraints and if you only want 3.5W of light, then that’s what you want.
This is my subjectively most real-world relevant set of values.
Some observations…
Frequency:
Mostly low frequency increases inductor ripple, important at low current/power, entering DCM sooner, and increases input cap voltage ripple at high current/power. However lower frequencies greatly reduce gate drive losses, improving low power efficiency, assuming things stay under control otherwise.
Cap ripple:
Output cap ripple and losses still look great after much checking of the math. (Note RMM said he had stability problems with such high output capacitance. I hope this is related to the hysteretic control on the Max, we’ll see.)
Input cap ripple. Fixed some bugs here. It’s not that bad, but could be a little better. Power loss is now fine, but there is some significant voltage ripple at 300khz. This assumes that the battery provides a constant average current and the cap provides the full pulsed ripple to the fet, possibly true (assumption used many places including by Ti). I don’t know what impact this has on output ripple voltage. I would think not too much actually. I guess it will drop max output voltage by 0.1V.
Since ESR is no longer a big concern, I could look for bigger caps, but they’ll have larger size. Maybe one large one is better than two small ones. None of these larger caps are tested/rated at high frequency though, and I just don’t have experience to guide this. I’ll see what I can figure out though.
Mosfet:
The higher rdson of the new mosfet shows its teeth in the 20A 4:1 (because it’s 20A not because it’s 4:1). Gate drive losses are down to 0.22W minimum at 1Mhz though, 10x lower than our original choice. Notes on mosfet losses are long and complex. This is just an estimate to help choose the best fet we can find.
Biggest effects:
The biggest differences from 14V to 7V to 3.5V are still from the inductor resistance and diode Vf*I A major change in performance requires cramming in through hole inductors, finding a suitable synchrounous IC, and using some crazy high side fet driver and rsense options that are expensive, more complex, and possibly open up other cans of worms, things for another version, if at all. I’m also finding some interesting low-side switched IC’s, but at this point will save that discussion until this version is finished off.
So the main thing if any to polish off is the input cap (it’s not so bad though), and then maybe just getting a systematic listing of caps and resistors for some desired setup.
(edited for spacing only)
Numbers look good. We could always meet in the middle for frequency with something like 500khz to try to get the best of both worlds as well but that should only change values and not component footprints.
So if we can nail down a final parts list and layout changes I can release this thing and see if it works!
Thanks for all the hard work Flint, I would not have had remotely enough time to sort all of the parts out, I franky don’t enjoy it and as such don’t find much time for it lol. PCB layout is not so bad, it is “direct interaction”.