You’re welcome.
The first one is the Fireflies E12R, but all their new models have the same driver.
Edit : well not all of them, but all the ones with “High efficiency 6A constant current buck driver  (>90%   efficiency) with FET Turbo” : E12R, E07x pro, NOV-MU, T9R.

Unfortunately it doesn’t look good, both the buck converter and charger IC look to be out of stock with long lead time at major electronic vendors, especially the buck IC with more than one year lead time, they probably should have redesigned their driver with parts with better availability.

Edit : Ah it looks like they supplied banggood with NOV-MUs so they do manages to produces some drivers.

Pavlo wrote:

Merci beaucoup pour les test et résultats!

I would be super curious to see the same test results from an Emisar Linear driver.

De rien !

A linear driver is effectively a variable resistor, current in = current out, so the efficiency is just Vout/Vin, i.e. Vf/Vin , hence the efficiency is low when the cell is full and the Vf is low, e.g. low/mid current and/or multiple LEDs.

The efficiency is at its maximum (for a given current) when in dropout, meaning the input voltage is not high enough and the current starts to decrease, then the losses are just due to the resistance of the driver in its minimum resistance state. Which if I were to guess must be something like ~15mΩ for the Noctigon linear driver (just the board, not counting springs and wires), it’s the Rsense (10mΩ) + traces/vias + FET.

So if we take for example a D4v2 with 4 SST-20s, say with a sustained current of 2A, that means a Vf of ~2.8V and an average Vin of ~3.7V (30Q), that gives an average of ~2.8/3.7= 76% efficiency.
At 5A, roughly 3/3.6=83%.

The E12R driver would do something like 93~94% at 2A and ~91% at 5A in average. My buck driver 96~97% and 93~94%.
Ideally we would count the input side resistance from the springs and cell, which gives a slight edge for a buck driver since it draws less input current.

Another example that this time would give an advantage to the linear driver : an SFT-40 driven at 9A in a big host, used nearly exclusively in turbo, which would be regulated only for a small amount of the cell’s capacity, so it’s quickly in dropout and then the losses are just from the low resistance of the driver.

fogofwar wrote:

I’ll echo the posts above. I’m glad to see great contributions by knowledgeable individuals on here.

I hope you can test a cheap Convoy boost driver to see if it is also quite efficient.

I guess for a linear driver, the efficiency can be approximated without direct testing? Knowing the forward voltage and battery discharge curves should be sufficient to know how much power needs to be “burned off.”

I don’t have any DC-DC convoy driver unfortunately, but they seem to use high efficiency synchronous converters.
Agnelucio tested the XHP35 driver here , it’s based on the MP3431 just like my boost driver. Just a note, he converted it to 6V and bypassed the RPP PFET, so his numbers are slightly better than stock.

Haukkeli wrote:

These are always interesting to see. Could you shed some light(pun intented) from which background you jumped to designing drivers? Where have you accumulated your knowledge of electronics, if you don’t mind me asking? I’m just curious, you can PM if you don’t want to share publicly.

No background, just high school electrical knowledge, which I refreshed by doing a few Khan academy courses, plus some more beyond high school level.
I searched about making a constant current source and found circuits with a FET and op-amp, which I recognised in a couple of drivers, like Convoy and Noctigon, studied a bit about the op-amp, read a few app notes and applied that circuit to DC-DC converters.
Then learned how to use Kicad to design PCBs, spent a lot of time reading datasheets and searching for good components, and a lot more for prototyping and testing.

Thanks for the compliment

Zebralight H53 :

Uses the LTC3539-2, a synchronous boost converter with integrated switches and low start-up voltage (0.7V), 2MHz switching frequency.
~5×5×4.5mm ferrite core inductor.

I checked several times the measurements for a mistake but in the end the efficiency appears to be just… not great.
Though given the high resistance of the integrated FETs it’s not that surprising, ~90mΩ for the low side NFET and ~180mΩ for the high side PFET.

Edit : H53, not SC53, although I’m pretty sure they use the same driver, same lumen rating and there is a hole for a wire coming from below the PCB for the flashlight layout.

Zebralight H600c/d mkIV

Uses the TPS61088, synchronous boost converter with integrated switches, set at ~700kHz switching frequency, ~6V output.
Coilcraft XEL or XGL 4020 1uH inductor, 14.6 or 9mΩ
I couldn’t find the ref of the RPP PFET, but like in the SC64 it’s a 2×2mm package, 13~17mΩ.

Note that this is the driver only, the H600 series have a very high resistance negative spring around 50mΩ which substantially lowers the efficiency on high modes.

SKV89 wrote:

Wow the Fireflies driver is impressive! I’ve been using the Nov-MU on a daily basis and absolutely love it. Too bad they seem to be having difficulty sourcing components for the driver for the past year.

It’s pretty good yeah, what is disappointing is that there not enough input capacitance and when a higher DCIR cell is used like a 50E or an older cell then the output can be unstable, which when it’s pronounced will lower efficiency, output and generate noise which several people have reported, it does it with my P42A that is not even that old.
It uses the amount of capacitance suggested in the datasheet (2× 10uF 0603) but in my experience of building drivers the suggested input capacitance for buck converters is usually not enough.
I added more caps, two 10uF 0603 stacked and one 0805 22uF farther away by removing some solder mask, but I haven’t reassembled it yet to fully test it.
Another (small) issue is the max current of 5.5A, the traces resistance on top of the sense resistance were likely not taken into account when deciding on the Vsense.

xevious wrote:

Seriously great topic, thefreeman! So fascinating to see how efficiency changes across the output selections. I’m impressed at how well the Skilhunt H04 does. In my anecdotal experience with my H03, I felt it was an efficient flashlight based on how long the battery charge would last with usage at a variety of brightness levels. The Zebralight is SC64c LE is rather disappointing… I’d been under the impression that it’s a nicely efficient design. I’m tempted to do a crude head-to-head test of mine with a few other 18650 flashlights all running the same 18650 cell spec, and see how it does.

Thanks
I haven’t tested the SC64c LE (3V), only the SC64 (12V), the former will likely be more efficient, I’m not sure what is exactly the issue with the TPS61088 used in the SC64 but the efficiency drops at 12V output, boosting to higher voltage (12V vs 6V) is a bit less efficient but not that much.

The H04 is in my opinion not very good, it only goes above 90% in dropout (i.e not stepping down the voltage), they’re cheaping out on components : high resistance path (about 220mΩ) and asynchronous topology.

Pavlo wrote:

Any chance you have and could test the timeless Noctigon Meteor M43 flashlight driver?

I could if I had it

upsidedown wrote:

This data is really interesting, thanks for providing it. Seems like the E12R and SC700 do quite well. Do any of the other ZLs have the same high resistance negative ground spring issue, or is that just the H600? Does SC600?
SC64w HI also seems pretty decent. Do you know how SC64c LE would do? I wonder, since I would guess that would be a 3V buck or buck/boost, in contrast to the boost used by the other 3 models here.

Why do you think some of these drivers have lower efficiency than some of the driver designs I’ve seen around here on BLF (such as yours)? Are there particular design limitations involved for these companies?

The SC600 III/IV and SC700 have pogo pins, I don’t know which part exactly they use but similar pogo pins from Mill-Max have a max resistance of 20mΩ, and with the amount of them the total resistance should be quite low (6 in the SC600IV for example = 3.3mΩ max)
The SC64 does have a thin steel spring like the H600 but there is one less spire and maybe they touch themselves when compressed, in the the H600 spring doesn’t compress a lot with unprotected cells, in practice the tailcap can get quite hot in turbo with the H600, showing that a significant amount of power is wasted, but it does not seem to be the case with the SC64 (but it also draw a bit less current so…).
Personally I bypassed the spring on my H600s, virtually eliminating the resistance.

I haven’t disassembled my SC64 LE yet but I will measure it in the future, it’s a 3A synchronous buck driver based on the TLV62085, it should be more efficient.

Regarding drivers having better efficiency it’s mostly about using better components : lower resistance PFET (for the reverse polarity protection), inductor, sense resistor, switching/rectifier FETs, for converters IC they’re integrated in the IC, so it’s about choosing a good converter, and one with synchronous rectification i.e the rectifier is a FET and not a diode (asynchronous).

Convoy FC40 22mm driver :

I upgraded my measurements setup, before I used 2 Uni-t UT210E current clamps (2%+3) , an Uni-t UT139C (0.5%+2) and a RDtech ”precision” voltmeter.

Now I measure the current across 0.1% current shunts with a BM867s (0.03%+2), Aneng AN870, Owon B41t+ and a RDtech precision voltmeter, the 3 later voltemeters are calibrated against the BM867s and are within 0.02%.

Well power losses are not linear (which would mean constant efficiency), a large proportion are conduction losses (I^2 x R), but you can see from my graphs (I should redo them with my new setup) that efficiency points aren’t linear either, but especially I think it’s the temperature increase that greatly increases power losses at high currents and tanks the efficiency, because running simulations without temperature modelisation gives a fairly straight efficiency curve.

Anyhow the Emisar boost driver efficiency is lower than that according to Hank’s measurements :

Which isn’t surprising since the inductor is smaller.
Edit : Vout is also higher.