High brightness white LEDs damaged by custom switcher

How about asking Microchip Inc. if PIC's are just as reliable as dedicated smps PWM controller IC's. D from BC

I'm sure they would "stand by their chips", but I think the concern was that the software could go awry and damage could be done before it could react. That's why I am considering a separate fail-safe mechanism, such as a fast acting fuse, or an NPN transistor that would turn on, when the voltage on the 1 ohm sense resistor reached Vbe, and turn off the MOSFET. Maybe an SCR would be better, requiring a complete power down to reset.

Paul

(snip history)

Did the manufacturer examine the entire string, and if so what were their conclusions (if any) about the diodes that *hadn't* failed? (It is easy to say that a "current surge" blew the shit out of the failed ones., just like most house fires always seem to be caused by an electrical fault.)

Arrghh..I'm forgetting lots of my PIC stuff.. Then there's the PIC watchdog timer... IIRC ..it's to help with "getting stuck in the hole" type glitches"... The uP will recover however I can imagine that the recovery period might still cause some smps external damage.

The method of OC protection will depend on worst case over current rate of rise. MegaAmps/picosecond??

Alternatively, Polyswitch devices (PTC) are used for overcurrent protection but I don't know if I'll protect LED"s in your app. It's a datasheet exploration... D from BC

Perhaps for development you could use a power NPN that directly did a limiting clamp of the current through the LED strings. Also a sensor that flags when the NPN is conducting.

This would allow any fault to remain long enough for investigation, without damaging the LEDs.

Why not use something like an LM317 in the (series) feed to the LEDs as a programmable current source. That would facilitate (a) avoiding possibly damaging current surges (sorry, tongue in cheek), and (b) allowing the stepping of the LED current through 0/100/700mA or whatever steps you want.

Pierre (who loves 34063's and 317's)

That's a little naive. There are plenty of switching power supply out there being run by microcontrollers these days, and on ones where it's preventing catastrophic failure is important, you certailny to see various current limiting/fail-safe devices... ***just as you do on analog designs, since there's plenty to fail on them as well***.

You don't necessarily lose the dual-brightness feature -- somewhere an analog design still has a reference voltage (or similar) that it's trying to match, and you can usually find some means of changing that reference, PWMing the output, etc. to get variable brightness.

I usually use two-transistor current limiters--you know, the classical two-terminal ones--rather than ICs for protecting sensitive devices like diode lasers. I'm much more confident that I understand their transient response than the IC's, especially in areas like coming out of thermal limiting.

Cheers,

Phil Hobbs

Thanks for the suggestion, but the power required for the clusters is at least 21 watts for 7 and 39 watts for 13. Even if the device could handle the voltage and power, it would be inefficient and unusable in the application, which is a diving flashlight. The efficiency appears to be in the order of 90%, and 4 watts is not excessive, especially when the water acts as a huge heat sink. However, efficiency and battery life are important.

I think the additional transistor or SCR will work. I can't add too many parts, as the entire board is only 0.95" x 2". The transistor would effectively form a linear current regulator, but would subject the MOSFET to excessive power. If I can get a small SCR, or make one from two transistors, it would latch the MOSFET off, but then the 12 volt gate drive would be applied across the 20 ohm gate resistor.

Maybe I will use the TI UCC27321 MOSFET driver which has an enable line with a Schmitt trigger. Luckily it is an active High with internal pull-up, so the saturated NPN transistor would disable the drive. Adding a capacitor should hold the output off for long enough to produce a "flashing" effect, which would be very noticeable.

It needs a redesign anyway, and I'm a little leery of the Microchip drivers. The TI products have a combination MOSFET and bipolar output designed to switch more efficiently through the on threshold.

Additional suggestions are welcome!

Paul

Hmm, are you trying to maximize thermal-gradient stresses?

I was trying to minimize the possibility of overshoot. In the simulation, with 12 VDC input, an 80% duty cycle produces a peak of 2 amps at 55 volts in 600 uSec. The desired 700 mA is reached in about 300 uSec, and the first

A 45% duty cycle, on the 7 LED cluster, reaches and holds 700 mA at 27 volts in about 210 uSec, and a 65% duty cycle PWM reaches and holds 700 mA at 46 volts in about 1.2 mSec.

The circuit is certainly capable of generating enough current to damage the LEDs, particularly the cluster of 7. However, I am reading the output current at a 1 kHz rate, so I think I may have discovered the problem. The PWM starts at 0, and can only increment by about 1/63 per mSec, so it should take at least 30 mSec for the PWM to reach the 45 or 65% level needed for the target output current. However, if somehow the PWM is set too high, it will take less than 600 uSec for the current to shoot up to as much as 4 amps, which would probably cause the damage seen.

I could easily increase the sampling rate, at least for the output current, to 100 uSec, as the conversion time is 51 uSec. It seemed like the slow rise of PWM would have been enough, but this is very likely the culprit.

Another simple thing I could do is generate an interrupt when my proposed NPN overcurrent sensor indicates more than 700 mA, and then shut down the PWM (as well as lowering the duty cycle) until it clears. I will probably want to change the sense resistor to about 0.68 ohms so that it will only trip over about 1 ampere, which is still safe. The same logic signal could be tied to the enable line of the driver as well.

OK, back to the drawing (and coding) board...

Thanks,

Paul

is device to device parameter variations being accounted for?

most of these LED's are not anywhere near being matched pairs/triples or other, unless you get the manufacturer to custom mount them from the same die and then do a parametric test on each and then sort them and mark them, (Very costly)

even clusters are assembled from random devices!

MIL specs may offer you some leeway, but even controlled groups will be bulk tested only for basic params.

if spiked current is failing some devices, they may have beeen borderline to begin with, as most will just heat up and dim before catastrophic failure occurs. usually over time.

your blown bonds and similar symptoms tells me that you had EXTREME current.

rise time between units may force some to go into failure befor the adjacent units absorb the available current also.

as with all design, the simplest is the best. trying to parlay your education of PICs and neatsy circuits willl yield expensive and hard to manufacture systems.

take some lessons from the chinese,tiawanese and others, cheap simple and basic.

stay within some easy set standard limitations and you may have good results.

the

test

without

to

the

Paul E. Schoen wrote:

I don't recall seeing the previous posts on this project so really don't know what your circuit looks like. The most reasonable current limit topology for almost any kind of switching power control is one which truncates the power flow upon over current. In your case this would mean a small sense resistor + comparator monitoring the source current of the MOSFET and truncating the gate turn-on pulse as qualified by a fixed threshold corresponding to a level somewhat below the peak current capability of the LEDs. This will provide protection against an instantaneous over-current while the PIC will handle the averages. It is most likely that your design should include protection against the PIC in addition to the usual analog protection measures. On the analog side of things there could be several causes of the over-current if this is in fact a failure mode. Some possible causes are: 1) the inductor transiently saturates for some unanticipated reason having to do with the PIC or possibly a bad rating selection, 2) the filter capacitor displays too much ESL and ESR relative to the LEDs+ 1 ohm sense resistor resulting in higher peak current transients being driven into the LEDs than planned, 3) too high a gain of dIleds per unit percentile of duty cycle corrupting the PIC control algorithm into a hunting condition or resulting in a faulted steady state which is an alias of a fault condition. This last may require that you employ some sort of dual modulus PWM scheme for the duty cycle generation to obtain better resolution on your control. After reading your other replies, it looks like your sample rate is too low by a factor of ten. There is such a thing as an effective Nyquist criteria that must be satisfied for applications such as this.

Hmm . . . saw something similar in my own experience. Bought Cree one watt leds and connected them in series and blew one up - linear supply with a dropping resistor.

Problem turned out to be the little aluminum heat spreader pad they were mounted to and multi layer board. I used something like a #6 machine screws to fasten it to the heat sink and the screw threads contacted an exposed bit of conductor on the white insulating board. Apparently the copper multi layer board uses large traces to aid heat removal and there's no provision in manufacture to prevent the edge of the copper from contacting mounting screws. Once the screws were tightened down the copper deformed and made a permanent short to the heat sink - the problem was solvable in the others (that weren't destroyed) by taking a drill bit and reaming the hole a little to break the bond, then using smaller insulated screws.

Check for shorts to the heat sink?

Amen!

Many years ago I did some analysis one a power switching circuit which used two transistors in the primary of a center tapped transformer. The transistors would switch alternately to produce a square wave on the secondary which was then rectified to produce a higher voltage.

Modules which failed had one blown transistor on them. As it was blown, it was not possible to read the parameters. Analysis of stock transistors showed that their characteristics, although within spec, varied. When the characteristics of the transistors were matched and they were used in matched sets, the problem disappeared.

The problem disappeared until a couple of years later, when a new manufacturing guru wondered why they were going through the expense of checking all the transistors. He abolised the practice, and voila, problem reappeared.

Al

Of course there was no "engineer" there to ponder WHY? And fix the frigging problem the right way. Selecting parts is the weenie's way of "engineering :-(

...Jim Thompson

Al a écrit :

Bad design to start with. Bad answer to go on: for the cost of matching, you could easily pay for the components cost to fix the problem.

What's your circuit? In details...

With 7/13 leds you're having about 28V/52V from your strings of white leds. Your circuit is a boost, so you're reasonably safe from some POV, except this: at 100kHz, L=10uH, E=12V, and Io = 700mA you're in CC mode with: @ Vout=28V : duty ratio= 0.57, Imin=1.3A, Imax=2A => Irms=1.1A @ Vout=52V : duty ratio= 0.77, Imin=2.6A, Imax=3.5A => Irms=1.5A

depending on how you've filtered (which kind of bypass cap) your output the led might see a big chunk of the pulsating current: the 47uF may not be as good as you think and the dynamic impedance of your leds will be very low at 700mA. Thus the Leds bonding might really see an effective 1.1/1.5A rms, not even speaking of any regulation problems involved by the low loop crossover frequency (for ex. what will happens if you have a small +0.5V step on your supply once the circuit is running full power?).

And why messing with a pic (everything implying a pic is a mess, anyway) while you could easily do this with a cheap dedicated SMPS controller?

Just so I'm reading this correctly, you are not paralleling strings of leds. That is, you only drive one string of leds. If that is not the case, I would expect problems.

When I think of the effort it takes to make a bullet-proof DC/DC chip, I just shake my head at the idea of doing it in software. In a chip, events take place simultaneously, while a uP is a step at a time using polling.