Reliability of LED arrays with series or parallel configurations

[tom66]

Here's a question for any statisticians out there.

If an average LED has a useful mean time before failure of 50,000 hours, and 50 of them are tied in series, what is the expected lifespan of the array (assuming that one failure of an LED would bring the whole array down.)

My guess would be that you'd expect on average one failure every 1,000 hours... so the array would not be expected to last longer than 1,000 hours. But can this really be true?

Is this different from parallel LED configurations (with the current in each LED regulated appropriately)

I've seen quite a few LED backlight TVs failing recently, all with open circuit LEDs, and wonder if the 50,000 hour expected life cannot be considered accurate for series strings.

[Marco]

Impossible to say without knowing the pdf.

[n45048]

I've seen quite a few LED backlight TVs failing recently, all with open circuit LEDs, and wonder if the 50,000 hour expected life cannot be considered accurate for series strings.

Can you provide us with data (as this is very interesting). I'm contemplating a direct backlit LCD (vs an edge lit LCD). How do they compare?

[tggzzz]

I've seen quite a few LED backlight TVs failing recently, all with open circuit LEDs, and wonder if the 50,000 hour expected life cannot be considered accurate for series strings.

Is it the LED or the connection to the LED or the driving circuits?

What's the failure mechanism leading to the figure of 50000 hours? That question is more relevant to LEDs used for general lighting, where although the LED itself might be OK, the surrounding circuits might fail, or the phosphors in the LED could dim or colourshift.

[ConKbot]

50,000 hour lifetime isnt the MTBF, so the statistics work out different.  50k hours is (generally) the time for the output of the LED to halve. 

And it depends on what you mean by in parallel.  50 in parallel, when one pops, the current on the rest increases, and expedites their failure too.  So it depends on how "the current in each LED regulated appropriately" is achieved. Balancing resistors? Individual linear constant current regulators? etc...

And TVs are enclosed environments where the manufacturers want as much brightness for as cheap as they can get, with minimal heat sinking or airflow. Under such conditions, no, 50k hours isnt accurate, heat will kill it way before then.

[wagon]

I repaired a centre brake light for a motor scooter recently.  It had 4 groups of 4 in series.  They were running them at around (measured) 35mA per string!  One in each string was open circuit.  I expect they were using cheap LED's and pushing more current through them to get higher brightness.  I replaced all 16 and modified the value of the series resistors to get below the 20mA limit per string.

Now, I wonder how many manufacturers push more current than they should, to save money?  Does this change the life much?

[TonyStewart]

LEDs weakest link is the ultrasonic bonded gold wire to the surface pad,

It is made in small um diameter to block as little light as possible.  This bond is prone to thermal shear stress from customer IR reflow profile, EOS in reverse direction >> 50uW , and depanelization warp stresses .

Parallel arrays work well from same batch otherwise, adding series R to normalize the values prevents thermal runaway from current-hogging and NTC of -200mV/60'C.  This NTC parameter can be used to measure Tj of junction in real-time with a shunt pulse to bypass LED current down and not affect CC regulators.  I choose ~10% of rated current to make a linear estimate of threshold voltage to compute Vth, whet Vth+If*ESR=Vf  (@ rated If current and 50% variations of ESR)

I have discovered all LEDs follow an Internal ESR of 1 Ohm per Watt of rated power per LED, thus an array of xSyP is  x/y times an single diode for ESR at rated power.

[dannyf]

If an individual led has a failure rate of p, a series of n such less has a failure rate of 1 - (1 - p)^n. For sufficiently small p, this approaches n x p from the above.

the same less, in parralel, have a failure rate of p^n.

[AG6QR]

If an average LED has a useful mean time before failure of 50,000 hours, and 50 of them are tied in series, what is the expected lifespan of the array (assuming that one failure of an LED would bring the whole array down.)

My guess would be that you'd expect on average one failure every 1,000 hours... so the array would not be expected to last longer than 1,000 hours. But can this really be true?

A mean time before failure doesn't tell you enough.  You need to know more about the distribution.  In particular, it doesn't tell you to expect on average one failure every 1000 hours.

For example, if LEDs all failed between 49,000 and 51,000 hours, and the failures were uniformly distributed over that interval, that would be a mean time before failure of 50,000 hours.  But a 50 light string would last at least 49,000 hours.

If a different batch of LEDs were such that half of them failed at 100 hours, and the other half failed at 99,900 hours, that would also be a mean time before failure of 50,000 hours.  But you'd expect that a 50 light string would be very unlikely to last more than a hundred hours.

Neither of these distributions are realistic in practice, but the point is that there is more to it than just the mean time to failure.  The nature of the distribution plays a big role.

The real life distribution is not likely to be uniform.  It's more likely to be some sort of "bathtub curve", with a few early failures in the infant mortality range, a period of time where failures are exceedingly rare, and then a later period where failures become more common due to aging.  But it's not safe to make many assumptions about that without more data.

[TonyStewart]

If an LED ran for 24x7  with a reported 50khr MTBF  it should last a mean life  5.7yr  But in theory they should last for decades, just get dimmer or experience Lumen Maintanance often called the LM70 or LM50 life test when they say MTBF.

But theoretical MTBF is based on ideal design and ideal process controls and avoidance of ESD handling or excess thermal shock from a poorly thermal profiled assembly or from poor solder joints.   

LED products aren't even 5 yrs old before they are obsolete and confidence factors for MTBF values are not even given.

Most stats are meaningless unless supported by accelerated life tests.

to have a 80% confidence of achieving 50kHr, as I recall, we had to run a lot of units under different conditions failure-free.

Now how many car LEDs have you seen malfunctioned? I see about 1% field failures on the road every day.

In the end it depends on brand name reputation  and quality test verification with highly accelerated life testing with great process controls and great designs in high volume to sort out yield issues.

But getting back to the root of the question. Most LED failures are open circuit, so parallel operation gives more redundancy but at higher currents.