But, without protection devices built-in to the LEDs, won't you, the lowly circuit designer, just blow them up??  They put those zeners in there for your own good, you know.  The paternalistic LED manufacturers know better than you :-)

Actually, the Zener makes sense for strings of LEDs.

These 4014 are used for backlight of large LCD panels for laptops (that's where I encoutered them first in 2009). Efficiency dictates to make long strings that are driven by a constant current DC/DC step-up driver.

If one LED breaks, the whole string goes  black.

The Zener prevents this "worst case" scenario. The Zener will pass when ore than (approx) 4V is applied across the broken LED, bridging the current and keeping the other LEDs working.

This is also countered by interleaving 4 or 5 shorter strings.

The reverse diode is a side-effect of the Zener. Like with MOSFETs, this diode can be an inconvenience or a feature...

This is all true, but if they're driven correctly, the lifetime of the LEDs should far exceed the usable life of the laptop.  They mostly fail if overdriven or under-heatsinked to save cost.

Don't fool yourself.

LEDs fail.

Unpredictably.

Expect from 0.05% to 1% failure rate depending on whatever conditions you care to observe.

When you have one LED, there is one chance in 100 that it fails so it's deemed "reliable".

When you have 50 LEDs in a backlight, one in two screens will have a broken LED in the coming months or years. This is significant. So it makes sense to add a 0.001$ worth of Zener to each LED, and reduce the costs of shipping, repairing etc. "broken screens". Not all device makers are like the Chinese ("copy, manufacture, sell and forget").

But what's the failure mechanism here?  An open bond wire?  That could just as easily happen to the zener, too.  Maybe something else?

I thought the zeners were there primarily for ESD protection.

I'd like to find some good info on LED failures.  The stuff I've seen makes it look like everything else fails first, so you hardly see the LED failures (for example, failing capacitors in household LED bulbs).

Tomorrow, I'll go again to a workshop where I repair "deficient" LED screen modules. Think 100×50cm tiles with 64×128 pixels. In average, 5 or 6 faults, half are broken from misuse, the others miss a color (sometimes it reappears when you push on the package.

The most usual fault here is not the bond but the sticking of the die on the substrate, which uses conductive adhesive. A bad dose or a tiny bubble of air is enough to "break" the LED.

Zeners protect somewhat, but I fail to see how, except for reverse charges, since the LED conducts in forward.

Anyway when you have 48K LED dies per m² (16K×RGB), the odds of a broken one is "probable". It's a matter of statistics and big numbers, not of individual longevity.

Do these failures happen during a "burn-in" period, in the initial part of a "bathtub" curve, or are they the failure mechanism at end-of-life.  My gut feel is that these would be early failures, and once you shake them out, the lifetime is fairly long.

Not counting abused LEDs, of course.

I'd say, some die in the first hours or days of use, then sporadicly.

I don't have reliability numbers, it's probably proprietary (some Chinese do great engineering and support work, you know).

In another project, I have first-hand experienced failures caused by inductive parasites on the mains network, no idea how they could have propagated to the secondary of the AC/DC switchmode PSU but it happened before my eyes. Failures were reduced when the system was powered through a backup generator for computers.

You can guess why I'm so interested.  I'm thinking of all sorts of things that can go wrong keeping an LED lit for years (and eventually decades), and haven't really considered the lifetime of the LED itself.  Of course, I'm de-rating the average current by four or five orders of magnitude, but still...if the LED fails, the light goes out. Period.

I understand. But reliability is a question of scale. If you're doing a hundred #TritiLED then there is one chance that one goes bad in the coming year. Use two in parallel, "just in case" ?

I thought about having a 3-chip "RGB" LED case filled with 3 cyan chips for redundancy, but I don't really want to drive more than one of them.  Ideally, I could sense if one was working and fail-over to one of the others.  Redundancy is hard!

You know, usually, when one fails, it is just thrown away...

the triple-LED assembly would cost more than the rest, I suppose.

Yes, practically, you just throw them away.  But I don't really buy in to practicality...you know, making logic out of impractical things (diodes).  You know the type, Mr. germanium relay computer :-)

Germanium and relays are old trades, unlike your innovative designs :-D