Problem with 7135 in 2 cell light

Into my Convoy L6 I built a driver consisting of a Blf X6 driver (FET+1) and a 105C with 8x 7135 (380 mA) as slave. When setting the 7135 PWM to 100% it outputs the calculated 3400 mA (8x 7135 with 380 mA on the slave and 1x 7135 with 350 mA on the FET+1). Very nice.
But after about 10 seconds lumen output and current consumptions goes steadily down, at the end it pulls about 2400 mAh and stays constant then. My guess was that the 7135 starts throttling due to the heat on the board. So I set the PWM to about 65% and indeed lumens and output stayed almost constant at start level.
In order to achieve better efficiency I removed 3 of the 7135 and PWMed the remaining 6 again with 100%. I expected the current being constant at about 2250 mA. But surprise - after some seconds the current went down again to about 1700 mA.
So I thought every single 7135 is at 100% PWM overloaded with the higher voltage of the 2 cell setup.
But guess what - using only the 7135 on the FET+1 works as expected: no lumen or voltage drop.
The only difference to the 7135 on the 105c: its a 350 mA version and on the 105c are 380 mA versions.

Any idea?
Are the two versions of 7135 different in this respect, or do I miss something else?

Is it Zener modded?

The 7135 has a automatic step down to 250mA at a certain temp. I don’t remember the step down temp, but it sounds like what your slave board (guessing it has a limited thermal path) is doing.

What led are you using with two cells.

Yes, the board is zenermodded.

I also believe it’s the stepdown, but why does it stepdown with 5x 7135 at 100% PWM and not with 9x 7135 at about 65% PWM? Both variants pull a similar current, and the heat on the board should be not that different.
And if every single 7135 decides to stepdown due to the individual heat on the chip, why only the 380 mA versions on the slave and not the 350 mA version on the FET+1.

Ok, perhaps it’s just the 30 mA more on the 380 version which makes the difference.
Did anyone else see this behaviour?

This is interesting. The loss at each chip might sum up to the total loss I observed. Do you have any link regarding this? I looked into the specs but couldn’t find any exact values to the stepdown.

It’s the original XHP70.

or

AMC7135 Datasheet(PDF) - ADDtek Corp?

Input voltage up to 7V ??
Output voltage up to 7V ??

I found different datasheets of the 7135.
In a datasheet from 2007 thermal protection is not mentioned, in a datasheet from 2016 it is.
I wonder if there are really different versions of 7135 or if it’s only the documentation.

Edit:
Ok, and found another datasheet from 2006 which mentions thermal protection.
Confusing.

Not all chips are created equal.

I think this is the thread where I first learned about the 7135’s thermal fold back feature. Since then, I have witnessed it several times. I think I had a data sheet that had some info on it, but I lost a bunch of files a ways back.

Regarding your first question above. I’m guessing 5 chips at 100% duty cycle will get very hot if there is a big difference between Battery voltage and emitter Vf. Where as, 9 chips at 65% duty cycle will be much cooler. Just a guess.

Regarding the second question: The single 7135 can dissipate its heat directly into the driver and host. It is probably mounted on the ground ring of the driver which makes direct contact with the host.

Aktually the 105C has the better heat dissipation. It is glued directly to the empty original PCB with Arctic Alumina, lying on the backside of the 4x 7135 at that side. The FET+1 is mounted askew, with the 7135 up in the air. (The original PCB has a diameter of 30 mm.)

I wouldn’t be surprised if the 380 mA variants make the difference, more current to burn there, as I learned from the linked thread (thanks!). Perhaps I will try a 105C with 350 mA chips one day, but for now I stay with the single 7135 on the FET+1.

I wonder if the difference in efficiency is really a big one - using 9x 7135 for about 3400 mA at 100% PWM or using the FET with about 38% PWM (the Keeppower protected deliver about 9 Amperes when full). The output efficiency of the LED is lower at higher currents, but on the other hand I am burning a lot of power in the 7135, especially in a 2 cell configuration.

^

That might be enough thermal path, but it sounds weak to me. If I follow you correctly, your driving the heat going through the top of the chips into an emptied driver board that will have layers of solder resist, electrical insulation, and fiberglass (or whatever).

Personally, I like a more direct path when driving 7135's with 2S cells or more. Something like below. Like you, the chips were folding back fast before doing the below. It was overkill, but I needed structure too for other parts that went in the light.

The single 7135 still has a whole driver to disapate heat into and that driver probably has thick copper wires connected to it. Just guessing again as you have left out a lot of details.

The 105C is glued to the original board, where parts, solder and copper have been removed. I guess the heat transfer is near zero. Didn’t think about it much when I built this driver mess. Just wanted to point out the heat transfer of the 7135 in the FET+1 driver is even worse, and this one does not step down.

Anyway, I’ll try the FET+1 without slave and check if and how much the efficiency decreases compared to the 9x 7135. Will post the results here in short time. Have to finish my first momentary firmware before.