Click MR16 LED Globe Failure Post-Mortem

Hi!

Interesting analysis. It would be good to examine the Vf of each LED too, because there is a school of thought that it is ok to parallel up LEDs without resistors under certain conditions.

Although the Vf variation can be large between LEDs, it is possible to get them binned from the manufacturer so that the variation is much smaller. In that case, the series resistance can be

much smaller. A point can arrive (in theory) where there is enough effective series resistance in the LED characteristic curve itself to be sufficient that the current delta (even under

junction temperature deltas due to the imbalanced current) is small enough to be within the rated spec for each LED.

It is hard finding this information though, it most likely needs to be tested. The best link I could find is here:

The Nuts & Bolts of LEDs—Part III | EDN

Notice however that still their conclusion is that so-called active control (i.e a constant current source) is preferred, but then the author does work for a semiconductor manufacturer - he's still

right of course, active is good.

However, that doesn't mean the series resistor method is bad, it just means a lot of care is needed, i.e. more expensive LEDs (since you're paying for the fact they are binned) and some calculations.

For example if you can bin to within tens of mV (this should be feasible, especially if they all came from the same wafer) and can adequately heat sink, and your LEDs happen to have

ESR in the right ballpark, then that means you could reduce or even eliminate the external series resistor. A lot of 'if's of course, but worthwhile the manufacturer doing if it saves them costs

and simplifies the circuit if they are manufacturing hundreds of thousands.

It could be that to save costs the low-cost China models are not binned, but if the LEDs are manufactured close-by then maybe they can purchase a whole batch, so that consecutive LEDs are

all from the same wafer and therefore naturally very well 'binned'. This is speculation though..

You are right - thanks for that link.

I did say that if the manufacturer did some work to "match" the LEDs, then it gives it a fighting chance in real life, especially if both strings have the same path over the MCPCB and thus can equalize their temperatures as well. The lack of any feedback, be it active (e.g. current mirror, etc) or passive (resistor) is a big drawback as even if you get the LEDs sufficiently close at beginning of life, you can't guarantee that the LEDs would always degrade at the same rate over time which adds uncertainty to the lifetime rating. Of course, with lots of LEDs, you get other failure modes, such as failure in the MCPCB itself and in the soldered connections between chips, so I do like single-series circuits with a moderate number of higher-powered LEDs but this then presents optical challenges in preventing hot-spots which cause glare.

It's a big balancing act when it comes to design, but when you come across many globes with up to five-string parallel configuration without a feedback/balancing resistor in sight, you really have to wonder if and how well those LEDs had been binned and how long it really lasts. After all, some of these globes do have more pessimistic projected lifetime (e.g. 15,000 hours vs 25,000 hours) and that might be because of this failure mode rather than any intrinsic lumen depreciation due to phosphor burn-out or accumulated junction damage.

At least it seems LEDs mostly fail more gracefully than CFLs which have occasionally smoked, sparked, popped and hissed. That being said, I'm not sure it's a great idea to have a lot of potential "strobe lights" hanging around - it could potentially trigger photosensitive epileptic seizures, or at the least, be a big annoyance to end users. Just another way things could be improved over time.

- Gough

Yeah I agree, things could change over time (I don't really know enough about Vf behaviour over time). It is interesting though that even well-known brands seem to do this, e.g. this Cree lamp teardown shows no resistors in sight, and apparently this is two parallel strings:

Given that every engineer knows (or should know!) about Vf and the variation of it, I suspect it is highly likely that the binning goes on (for reputable manufacturers at least), else it doesn't explain why so many different manufacturers do this practice.

Probably this is just for consumer lamps though, they can afford more exotic circuits for industrial lamps I guess.

You're right that we just don't know about the quality and binning as part of that, since failures take quite a while to occur.

Also, as a digression, Cree has some high-end lamps too, which have clever tricks to make sure the light looks "nice" for (say) cosmetics counters etc. Some of their lamps have white and red LEDs, and the brightness of the red LEDs are adjusted over time as the lamp warms up, to compensate for the changes in color that occur by using the phosphor method for white LEDs.

I bought a bunch of those to use in my lab, but not fitted them yet. Although with more circuitry, their failure rate may be just as bad/good as the simpler LED lamp designs (I've no idea : (

I suppose as Cree are LED manufacturers, they get the first pick about what goes into their lamps, so binning on tight tolerances is indeed a big possibility. As you mentioned earlier, if it's close enough, maybe you can get away with it.

I have a sneaking suspicion they may be relying on the residual resistance in the relatively thin PCB traces to help reduce the imbalance as a form of "inbuilt" feedback resistance.

Aside from that, my experience with tearing down Philips (professional series) luminaires is that they tend to design differently - even if they have two strings parallel, they're often made of Luxeon chips which may be rated and specified at both 350mA and 700mA each with an absolute maximum of 1.5A per string, but in practice, they're actually running the two strings with a combined forward current of either 700mA or 1A. This means that the margin for imbalance is quite big, as say with 1A combined current, if one string was at 700mA, the other would only have to be soaking 300mA and everything's still "in spec" although there will be obvious differences in brightness. Better still, if one string fails, the cascading failure takes more time, because the resulting current still remains below the absolute maximum.

Having two strings seems fairly common even amongst quality lights - that is consistent with my experience - but the other thing I suppose is that two strings is easier to balance than three, four or five. The reason I say this is because the "margin" between the full driver output (i.e. worst case, every other string is dead) and the absolute maximum current of a single string of LEDs changes and in some cases, the designs are set such that a single string failure is ultimately catastrophic (i.e. no margin at all). The integrity of the luminaire depends on the combined reliability of every LED element - so I suppose the other benefit is that fewer strings often also coincides with fewer LEDs and fewer solder joints etc.

The paralleling business is likely to ensure the output voltage remains low enough - for example, mains powered "double insulated" or SELV globes say for countries running on 110-120v AC, they will need the globe to be operational down to about 84v AC in the worst case and giving the driver loss leaves us with maybe 80v to play with. At 3.5v/white LED, that leaves a maximum series string of 22 LEDs. If built with 5630 0.5W modules, that gives us about 11W of output only. Of course, larger LEDs will mitigate this, but I suspect another possible reason boils down to the insulation in the LED and MCPCB assembly itself given the substrate is metal and electrically conductive. If they want the output to maintain SELV status and stay under 48v (even with step-up) then that cuts the number in half.

Overall, I have no doubt that they probably know what they are doing. After all, as a very young child, I managed to run LEDs in a "voltage mode" rather than current mode than I should have by ensuring the series combination had a voltage drop somewhat above the maximum anticipated voltage and living with reduced power output. It's not a good idea, but that doesn't mean it can't work for almost "forever" (e.g. 7 x series green 2.1v LEDs = 14.7v nominal operating Vf, run from a 12v (maximum) supply with a 2.7v margin and accordingly reduced output).

- Gough

I suppose as Cree are LED manufacturers, they get the first pick about what goes into their lamps, so binning on tight tolerances is indeed a big possibility. As you mentioned earlier, if it's close enough, maybe you can get away with it.

I have a sneaking suspicion they may be relying on the residual resistance in the relatively thin PCB traces to help reduce the imbalance as a form of "inbuilt" feedback resistance.

Aside from that, my experience with tearing down Philips (professional series) luminaires is that they tend to design differently - even if they have two strings parallel, they're often made of Luxeon chips which may be rated and specified at both 350mA and 700mA each with an absolute maximum of 1.5A per string, but in practice, they're actually running the two strings with a combined forward current of either 700mA or 1A. This means that the margin for imbalance is quite big, as say with 1A combined current, if one string was at 700mA, the other would only have to be soaking 300mA and everything's still "in spec" although there will be obvious differences in brightness. Better still, if one string fails, the cascading failure takes more time, because the resulting current still remains below the absolute maximum.

Having two strings seems fairly common even amongst quality lights - that is consistent with my experience - but the other thing I suppose is that two strings is easier to balance than three, four or five. The reason I say this is because the "margin" between the full driver output (i.e. worst case, every other string is dead) and the absolute maximum current of a single string of LEDs changes and in some cases, the designs are set such that a single string failure is ultimately catastrophic (i.e. no margin at all). The integrity of the luminaire depends on the combined reliability of every LED element - so I suppose the other benefit is that fewer strings often also coincides with fewer LEDs and fewer solder joints etc.

The paralleling business is likely to ensure the output voltage remains low enough - for example, mains powered "double insulated" or SELV globes say for countries running on 110-120v AC, they will need the globe to be operational down to about 84v AC in the worst case and giving the driver loss leaves us with maybe 80v to play with. At 3.5v/white LED, that leaves a maximum series string of 22 LEDs. If built with 5630 0.5W modules, that gives us about 11W of output only. Of course, larger LEDs will mitigate this, but I suspect another possible reason boils down to the insulation in the LED and MCPCB assembly itself given the substrate is metal and electrically conductive. If they want the output to maintain SELV status and stay under 48v (even with step-up) then that cuts the number in half.

Overall, I have no doubt that they probably know what they are doing. After all, as a very young child, I managed to run LEDs in a "voltage mode" rather than current mode than I should have by ensuring the series combination had a voltage drop somewhat above the maximum anticipated voltage and living with reduced power output. It's not a good idea, but that doesn't mean it can't work for almost "forever" (e.g. 7 x series green 2.1v LEDs = 14.7v nominal operating Vf, run from a 12v (maximum) supply with a 2.7v margin and accordingly reduced output).

- Gough