[ GXB172 - 50W Single Cell 17mm Boost Driver! ]

Would one 1210 cap be better than two 0805 caps?

one 10uF X7R 25V 1210 is definately better than 2 0805 X7R 10V 4.7uF

Only thing I do not see mentioned is being able to accommodate 1S or 2S input in the same configuration. Lights like the Convoy L2 that have an optional extension tube would really benefit from this.

I do not understand most of the thread otherwise but I cant wait to try out some of these new boost drivers

This I asked in Firmware topics and wrote PMs to Tom_E and Toykeeper
I got no real solution

It is no big deal if it gets integrated in Bistro and Narsil, but I am not capable of doing it

Welcome back Lone. I wrote you emails months ago and thought you’re already left us. But here you are with your crazy driver project. Great job Lone!

- Clemence

This looks awesome thanks for sharing the project with us.

When you solder the smaller components do you use a soldering iron or heat gun?

Answering comments:

Thanks FPV. Yes once I finish the firmware and done some proper field operation and stress testing, I'll put everything online so you can order your own PCBs and put your own together! However do be warned that you do need some sort of proper setup to either reflow / hot air this or put this together under magnification. Meanwhile I'll start putting up my GXB172 page on my website as I go along.

As for sales, we'll see how it goes! I certainly wouldn't want to put something up for sale if the firmware is not up to scratch or if the driver has issues. Suppose all is sorted out, I might do a small run, or send for PCBA which would be much nicer, but it depends. Morever Schoki / Lexel are already working on driver and it seems they have more history of releasing drivers, so you can go with that route as well! I just had no idea it was already in development when I made the GXB172.

KawaiBoy1428, whoops I meant to say 16V, but yes it's good for driving LEDs like the XHP35 (12V+), and by extension, also good for 9V operation. I do hope to make a 20mm version with even better performance (better passives) in the future to replace my GXB20, but I'm not planning to make this design any larger. The reason being that any larger would allow me to use an external FET solution and it's really much easier then to make a 50W or even higher power driver - no need to constrain myself with this particular topology. That's the good thing about the 'dev-board', easy to do tests like this.

Kiriba-ru, you are absolutely right! The driver and LED setup for any flashlight that can accomodate a 17mm driver would definitely overheat at 20W, much less at 50W! Drawing this amount of power from a small cell is also ridiculous. However, that's the premise of the project - to make a ridiculous driver. As I've mentioned, it's fairly useless from a practical point of view, but practicality was not the main point of the GXB172, at least not in 50W output. I have no illusions that 50W is anywhere near a practical drive power for a 17mm-driver flashlight.

Regardless, the driver should still be in the upper 90s% for most of the range from ~100s of mA to 2-3A output (will measure soon) so it's still a fairly efficient driver at realistic drive power.

For example, the Emisar D4 has a 3000+ lumen direct drive mode, but this lasts for only 10s of seconds, before the light drops down steeply to some ~300 lumens or so by thermal regulation (based on here https://www.youtube.com/watch?v=GVj7NnAj_Ps). Again depending on what you set the thermal limit to be and what the external temperaturate is, my guess is that the Convoy S2 can probably only handle on the order of 3 to 5W of continuous run time (e.g. headlamp), and likely not much more than 10W (handheld light) for any reasonable temperature limit (taking into account the whole system including LED, driver and battery).

And as Schoki said, because the driver is in fact much more efficient at lower drive power, this makes it a pretty good driver for 20W operation I think. In fact, in real life, I'm going to set my everyday mode limit to be around 25W operation. This is very easy to set and the 1024 current limit steps make brightness control still fairly high resolution even at lower powers (or you can swap out some resistors to make this even better!). I'll be able to do some more detailed thermal analysis later, especially when I get it into form factor.

Also Schoki, thanks for your comments, and I see the renders of the driver Lexel posted! Is that the driver you two are working on together?

Yes I agree the top (inductor side) layout looks similar, but I guess come to think of it it really falls out into place just based on how the (few) power components and traces needed to be laid out, and what we can fit into the same round PCB. I did start layout based on the reference layout in the datasheet, and used a single 1210 capacitor for input and output rails, but I couldn't get it to fit in a reasonable way, so I split them into combinations of smaller caps instead and that's basically the only way I could lay it out. The rest of the compensation network and feedback then just slots in the leftover space. I also initially use through-holes for the LED outputs but I switched them to large pads instead just to make routing on the other side easier. Also good catch on the max voltage rating, I meant to say 16V!

Its all black magic to me but know its all real good stuff. :slight_smile: :beer:


Lexel you're absolutely right! The XAL7070 and 7030 have essentially identical pads so I made the pads fit both inductors, so you can choose whatever you want. Some pills like the Convoy S2+ have a lot of space so it definitely makes sense to go with the big inductor. Others have less space and I've verified it works OK with the smaller inductor too. But like you mentioned, the 7070 helps in reducing DC losses (and thus also less likely to cause the inductor to drop inductance due to heat), and that we can use more inductance for less ripple and less RMS current losses and switching losses!

As for the capacitors, yes definitely larger and higer voltage rating with better dielectric is more desirable. In fact I did change some of the pads to 1206 before I sent it off for fab last week. Since the inductor is so fat, it's also possible to stack MLCCs on top of each other. Currently I have 94uF on both the input and the output at 16V, though we could do something like a 138uF 10V on the input and 66u 25V on the output, and more with stacking. Finally, I also found some nice bronze springs from Kaidomain which have a ~5mm top and ~8mm bottom. The high drain cells all have fairly large flat tops so I'm planning to try to use those springs in reverse (since my spring pad is very small to save space), but I suppose it doesn't really matter since we'll need to bypass them anyway.


Well the way I have it set up should work just fine I hope with perhaps a change in a resistor or two. The LDO max input is above 2S, and my battery voltage divider also takes this into account so it's a simple firmware change to have it running with 1S or 2S with auto-detection. The downside however is that because this driver is fairly specialized for the application, for best performance ideally one really needs to configure it for either 6-7V out or 12-14V output. For example, if I really wanted to drive a 6V LED hard, I'd choose input and output caps specifically for that purpose which would not work well for a 2S input. Likewise for 2S operation, I can choose a larger value inductor for less ripple and a more regulated output with lower peak switch current. So in short, it can work, but it will not be near optimal. That said, if you're just planning to run at fairly low power levels (<20W), I see no major problems.


Hello Clemence! Yes thanks also for your help in getting Nichia LEDs to the community! I ordered some from you recently so hopefully I'll get to pair one of the 144s with this project.

Thanks for your kind words! For soldering, it depends! Typically for stuff like this I use a stencil and reflow it in an oven, but for one-off small boards like these (at least for now), I typically use hot air for the stuff with pads underneath, or in this case, the big indutor too. The rest I either use hand-placed solderpaste with a needle and hot air it, or use a tiny tip soldering iron and needle tweezers. Typically I do this under magnification.

Thanks MRsDNF! I hope to get this inside a flashlight soon when the host and boards arrive, and when I finish whatever firmware I'm working on right now.

In which mode are you running the boost IC?

Wow, so much reading to catch up on this thread, hope I can find time to catch up at some point.

If you want to get my design files for private use only I have it in Diptrace the others seem to use Eagle for their drivers

Schoki, I tested both pulse frequency modulation / pulse skip mode as well as ultrasonic mode (USM) - one of the cool features of this IC. Once the output current (with a 6V LED) drops to about 75mA or less, the pulse group frequency drops to below 20kHz (with my set-up). As a result, running it in really low modes or moonlight will result in audible noise with PFM/PSM. For example, at 10mA output drive, the pulse frequency is well under 2kHz. So anything below ~60mA or so results in a (quiet) but definitely audible sound. This is absent in USM mode where the group frequency is kept above ~23kHz but at the expense of only slightly lower efficiency at those low values.

After some evaluation I decided to just go with USM mode since this allows us to float the mode pin (fewer components), and I'm pretty sure nobody wants a moonlight / nightlight flashlight beside your bed making this tiny high pitched sound, even if it has better battery life. For example, based on the datasheet with a 4.2V input 9V output and 1mA output, PSM vs USM is about 60% vs 40% efficiency. At 10mA, this becomes 93 vs 83% efficiency. It's a significant efficiency decrease, but I think it's worth the trade off, and still reasonable in practical run-time duration.

good news as my design has same floating ground for USM mode

could you test together with an MCU already if the chip needs the LVP pin populated with 2 resistors, this is really a pain to integrate with all 0603 components
I had later a design adding 2 resistors but then later have to make a wire bridge as absolutely no space to place a via for V+

While well out of the loop I agree with your decision, that is the trade off I would have made.

When you are talking about a mode that will run for months in either setup, it really doesn’t matter much lol.

BTW can anyone share the MP3431 datasheet with me? I got troubles getting it from Monolithic…

Impressive driver! What host could even handle that much heat? A C8 light is the biggest thing that comes to mind.

ICSK, from a practical point of view, I am the first to acknowledge that the driver doesn't make sense , since heat will be the biggest issue. However, this is intentional! Moreover, it will be able to drive 50W for a short period of time, before thermal regulation kicks in; and if you turn it down to say 10-20W, then you've got a much more practical but still powerful driver. I've got a prototype PID thermal control firmware tested and it seems to work OK, just need to put it in an actual light to fine-tune the variables. However, you're right that the C8 is likely one of the larger lights that still uses a 17mm driver. I'd imagine that having a C8 and swapping out the XPL HI with a XHP35 HI would produce some pretty nice results , even if the XHP35 HI die is a little bigger than the XPL HI.

Yes, the XHP35 is a great upgrade from the XP-L in every way. More lumens, more throw and generally better tints from what I have seen.

I don’t agree it makes no practical sense.

  1. If you lack distance to shine it where you want, you can use a turbo or walk. I consider the former to be a practical option.
  2. Aside from ultimate power it offers superb efficiency.
  3. Larger variants will find use in larger hosts.