LED series-parallel arrangement and reliability

[poorchava]

Hi,

I have a project where I have to drive 9 white LEDs (Seoul Semiconductor STW8T16C-Q0S0-CA) from a single Li-ion cell. LEDs are physically mounted on separate PCBs in groups of 3 and they can be connected wither in series (9x1) or parallel (3x3) or driven independently. LEDs are to be driven at up to 60mA. The size of the driver PCB is limited to 30mm (hard limit) x 30mm (give or take). Cost of the device is not an issue within reasonable limits.

Now to the problem: I've been searching for few days, but it seems there is no single chip solution that can drive 9 leds in series @60mA from single li-ion cell. Simple boost converter formulas show that maximum duty cycle would be around 92%, so it's not extremly high value. Switch current comes out to be 750mA + ripple = let's say 1A to be safe.

If it wasn't for the space limitation I would use 3 separate led drivers, but they won't fit.

I have found an IC, LT3591 that could THEORETICALLY do the job, in simulations it reaches about 53mA from 3V supply, but it hits the internal switch current limit all the time. I'm also not the fan on DFN packages because they are bitch when prototyping, but I can live with that.

Other solution would be to use LM3410. This one has too limited duty cycle to drive 9 in series, but assording to WEBENCH tool it can easily drive 3 strings of 3 in parallel. I like this solution much more than LT, but I'm kind of concerned about reliability of such LED arrangement. As it is now if one led in a chain fails short, then whole sting will most likely get toasted. Other thing is that I cannot really verify this, because TI doesn't provide spice model for this IC, and the simulator in WEBENCH is a sad joke. Do you know of any additional circuitry I could add to improve current sharing? I know about adding current mirror like in this paper: http://ledlight.osram-os.com/wp-content/uploads/2010/05/AppGuideCurrentDistributioninParallelLEDStrings.Web_.pdf, but this solution will fail if the shorted diode is in the reference string.

I'm open to any suggestions. If you can propose another IC, that's welcome too.

[kizzap]

Would I be correct in assuming that you are trying to use an LED driver chip to maximise the time that the LEDs get their full voltage?

If so, could you not simply make a switching controller to output the voltage of the LEDs and use some form of resistance in series with the LEDs to ensure LED current limitation? Surely something like the MC34063 would do the job?

-kizzap

[HackedFridgeMagnet]

Maybe MX841BETR  from IXYS might do the job.

One application circuit does 15 white leds but they are not current sensing all the leds. that looks a bit dodgy.

May you could fit 2 drivers on the board, one driving 5 and one driving 4 leds. That way you could get current sensing on all leds and if one goes short then you still only get the same current through the string.

[dannyf]

If you are concerned about reliability, you can do parallel-serial: parallel 3 leds (preferrably with serial resistors) and then serialize such modules. This combination will be more reliable than the typical serial-parallel connections.

[Bertho]

What about a standard LED driver:
http://www.onsemi.com/pub/Collateral/CAT4240-D.PDF

It can generate up to 38V and is a current-feedback step-up. Is is also cheap (less than $2).

[poorchava]

Would I be correct in assuming that you are trying to use an LED driver chip to maximise the time that the LEDs get their full voltage?

If so, could you not simply make a switching controller to output the voltage of the LEDs and use some form of resistance in series with the LEDs to ensure LED current limitation? Surely something like the MC34063 would do the job?

-kizzap

I don't know if I got you right, but I think you are suggesting that I could use a voltage mode boost controller and then introduce some sort of current sensing circuitry to drive the boost converter's feedback pin. That could be an option, but that would significantly increase circuit size. Also I'd like to minimize the series resistance to achieve as high efficiency as possible.

Maybe MX841BETR  from IXYS might do the job.

One application circuit does 15 white leds but they are not current sensing all the leds. that looks a bit dodgy.

May you could fit 2 drivers on the board, one driving 5 and one driving 4 leds. That way you could get current sensing on all leds and if one goes short then you still only get the same current through the string.

The IXYS device seems to rely on perfectly identical current sharing between strings. In case a led in the monitored string gets shorted, the voltage will decrease on other strings effectively turning them off. Besides, I don;t think I can get that in small quantities, even in DigiKey.


As for 2 assymetric drivers, that could be a solution, but it's not possible because of mechanical arrangement of the whole thing: led boards contain 3 leds each and I cannot make a 4-5 connection. 6-3 connection could be possible though. I will check if I can fit 2 LT3591 on that PCB.

If you are concerned about reliability, you can do parallel-serial: parallel 3 leds (preferrably with serial resistors) and then serialize such modules. This combination will be more reliable than the typical serial-parallel connections.

If one 3 led's in parallel fails, then the two ramaining ones will take current 50% greater. As a matter of fact, the absolute maximum rating on those LEDs is 100mA, so one failure wouldn't be catastrophic, at least for some time.

What about a standard LED driver:
http://www.onsemi.com/pub/Collateral/CAT4240-D.PDF

It can generate up to 38V and is a current-feedback step-up. Is is also cheap (less than $2).

That look promising. I'll check that out.

[Marco]

Why does it have to use an integrated switch?

[poorchava]

Becase it takes least space. Can you suggest a discrete circuit that could do this and fit in 30x30mm area? I'm not trying to be rude or anything, I just discarded this idea at the beginning based on the assumption that it will be larger than an IC based solution.

As for the CAT4240 it seems that it will be even more on the edge than the LT3591. LT3591 was hitting the current limit in the simulation almost all the time (I suppose LT gives pretty accurate models of their own stuff with LTSpice).

[dannyf]

If one 3 led's in parallel fails, then the two ramaining ones will take current 50% greater.

I am not sure what "one 3 led's" means but doesn't sound like you got the idea of parallel-serial connections.

[Marco]

Becase it takes least space. Can you suggest a discrete circuit that could do this and fit in 30x30mm area?

That's not what I mean, I just meant using an IC with an external mosfet ... which would take maybe 10 mm^2 of your 900 available ones.

PS. how about the ZXSC400?

[poorchava]

Becase it takes least space. Can you suggest a discrete circuit that could do this and fit in 30x30mm area?

That's not what I mean, I just meant using an IC with an external mosfet ... which would take maybe 10 mm^2 of your 900 available ones.

PS. how about the ZXSC400?

Seems like a viable option. The datasheet mentions only 8xWLED  / 30mA, but I think the only limit in this topology is getting an external transistor with high enough current gain.

I don't know if I understood the datasheet correctly, but the mode of operation seems to be "fixed off time". In this case to achieve required ~92% duty cycle the frequency would drop to ~40kHz. Smells like big inductor and capacitors. Too bad Diodes/Zetex doesn't provide a model. All in all - this should work though.

If one 3 led's in parallel fails, then the two ramaining ones will take current 50% greater.

I am not sure what "one 3 led's" means but doesn't sound like you got the idea of parallel-serial connections.

In the left one:
-one led in left chain fails short -> most of the 180mA of current goes to that one chain -> other two leds in that chain are fried -> if other two also fail short then whole led cluster is dead. If one of them fails open then other two chains get 90mA per chain instead of 60 (but still below absolute maximum of 100mA) if current sharing is perfect.

-one led fails open, other two chains get 90mA each (same story as above -> this will most likely work if current sharing is close enough to perfect

You meant something like the picture on the right, correct?

Edit: that would be perfect http://www.linear.com/product/LT3003. Too bad that MOQ in DigiKey is 550pcs (i need about 100 at most right now).

[dannyf]

In the left one

Go through the same analysis for the right one.

[poorchava]

In case a led fails short, the ones in parallel with it will not work, current regulator will drop the voltage and the appropriate current will be maintaned, so their lifetime will not decrease.

In case one of them fails short, the other two in parallel get +50% of the current. Rest of the leds work on the regular current.

Ok, seems like a better solution. At least the chance of whole cluster going dead is much lower. Thanks for the suggestion. I would guess that current sharing is better, because differences in forward voltages of the series strings do not sum up (in series-parallel if it happened that 3 leds with lower than average Vf ended up in one string, this string would take much more current than the others)..

Seems like right now I have three options:
LT3591 and 9 leds in series: operates near the maximum of specification in terms of switch current, but there seems to be active current limiter for the switch (for the period when current ramps up). Effectively immune to diode failing short, entire cluster dead on diode failing open.

LM3410 and 3x3 parallel-series matrix. Very far from chip limitations (~2.3A switch current). One led shorted = 3 leds not operational, others work without problem. One led failing open: two have increased current, rest work normally

ZXSC400 and 9 leds in series. reaction to led failing open same as LT. Non-fixed frequency (and a very low one). No model or simulation tool available.

So far it looks like LM3410 and 3x3 matrix is the best.

[dannyf]

At least the chance of whole cluster going dead is much lower.

More importantly, the "more than N leds are on" probability is much higher with the parallel-serial connections.