Testing LED efficiency decrease at very low current

Earlier I came across these comments from flashburn , in quotes are the relevant part :

The lower efficiency of LEDs at very low current is something I’ve already read about somewhere, but seeing this comment made me curious about how much lower we’re talking and if that is something to be concerned about.

So I took an LH351D (rated 3A max) I had laying around and simply measured the illuminance by placing my spectrophotometer right in front of it, then I calculated the illuminance per watt relative to the maximum, luminous flux should be proportional to the illuminance so that gives us the relqtive efficiency :

Here is a log scale to see better at lower currents :

Maximum efficiency is around 175mA.
90% ~ 40mA
75% ~ 15mA
50% ~ 4.5mA
25% ~ 1mA
5% ~ 50uA

So it does decrease, a lot even, but only at very low current, where efficiency doesn’t matter that much, one could say this is even beneficial in helping to reach lower minimum lumens.

When using multi chip emitters, or multiple LEDs, like 4 LEDs as quads are pretty common, then

100% ~ 700mA
90% ~ 160mA
75% ~ 60mA
50% ~ 18mA
25% ~ 4mA
5% ~ 200uA

Still not a concern in my opinion, the gain in efficiency at higher current is much more valuable.

1 Thank

I haven’t seen this yet, from what I’ve seen before efficiency only went up when reducing current.
Note that you’re measuring current rather than power consumption. But this only reduces the effect that you’ve measured and shifts the peak a little lower .

I did measure power (If and Vf), the curve is relative lux/W vs current.

Table :

Edit : it’s lux/mW but no matter.

Oh, thanks for clarification.

I just used Samsung’s own calculator:

For LH351D CRI90 3000K D2 R2 (I don’t think that bin matters so I just picked some without thinking) and Tsp of 40C I see:
175 mA gives 174 lm/W
105 mA gives 176 lm/W (and the calculator doesn’t go any lower than that)….

I wonder how much the polar radiation curve changes at these currents considering the phosphor is operating in a different mode. Maybe ths could explain the mismatch in lm/w vs lux/w at a particular angle?

I measured at 60° and it’s basically identical :

So we can say that relative lux/W = relative lm/W

Wow. Thanks for the fast feedback freemam. Very interesting and much appreciated. :beer:

Indeed. :beer:

Nice measurements. Thanks.

I’m not sure the phosphor is the main explanation for the reduced efficiency at low currents, though. The efficiency of conversion of electron current to photon within the LED is called internal quantum efficiency (IQE). There are radiative recombinations and non radiative recombinations. Getting a good IQE involves balancing the different recombination mechanisms; maximizing the light producing (radiative) mechanisms and minimizing the waste mechanisms (non radiative). The radiative recombination rate is proportional to the square of the current (because you need an election and a hole to get a photon). There is a non radiative mechanism that is proportional to just the current. This causes the efficiency to drop at low enough currents. There is another non radiative mechanism which is proportional to the cube of the current, which contributes to efficiency “droop” at higher currents.

Here is a paper about this stuff.


Really interesting measurements! Strangely enough, I was looking for at sub-1ma data the other day while thinking about how feasible a keychain LED locator would be. Rather much like the ? Lamplighter keychain from CPF a number of years ago.

So using PWM to dim a 7135 should actually be more efficient than constant current at very low currents.


Not only that, you also avoid the tint shift by using PWM. What are the disadvantages of high frequency PWM? Am I forgetting something?

Does this make us rethink our desire for HDR switching drivers? 100:1 with a 100mA floor seems fine to me. Too bad there weren’t 100mA ‘7135s’.

Basically buck/boost down to 100mA, then pwm a fixed current linear regulator

Ideally you would PWM the ~100mA switched converter output to get very low modes. Because the low Vf (~2.8V) at these low currents means the linear regulator loses efficiency.

Hmm. I didn’t know you could pwm a switching converter that had a sense feedback. Wouldn’t this require a fixed (no feedback) buck circuit?

At low current there is barely any tint shift with the LH351D, and that’s probably the case for most LEDs.

350mA CCT = 3892K (Duv –0.0040)

175mA CCT = 3859K (Duv –0.0043)

100mA CCT = 3842K (Duv –0.0044)

50mA CCT = 3827K (Duv –0.0045)

10mA CCT = 3815K (Duv –0.0047)

1mA CCT = 3809K (Duv –0.0047)

Most of the tint shift happens at higher current, there PWM has a high efficiency cost.

Hypersonic PWM is difficult to implement with switching drivers, or at least I don’t know how to.

I think it would need a linear current source instead of sink, most linear led drivers are current sink (like 7135), I tried searching for some before thinking of the dual sense resistor solution, but I found nothing that was really suitable.

Plus it won’t work with boost drivers.

Anyway seeing those measurements it confirmed to me that the reduction in efficiency is not that important, so for me no, it doesn’t make me rethink my constant current drivers.

Good point, I don’t really know.

All good points here. I’m just thinking out loud. Enjoying the discussion. Thanks guys