Video: TLC5947 leakage current problem

[allanw]

I have a board with TLC5947's (http://electronoblog.allanw.org/rgb-stick/) where some of the pins have high current leakage, causing LED's to dimly glow when they should be completely off. In around 5% of the outputs, I get around 1uA of leakage current. Some pins even have 5uA+ which is very visible. The current rapidly decays when sinked but doesn't go to zero, instead to something like 0.5uA to 1uA, so the glow never goes away. After disconnecting the pin and waiting a while, the leakage current seems to "reset" back to the original value and starts decaying again. The effect happens on the same pins everytime. The leakage current is roughly linear with the supply voltage. Replacing the TLC5947 will occasionally fix this issue.

The datasheet says the maximum current leakage is 0.1uA. What could be causing this? The resetting and RC-like behavior makes me think it's some unwanted capacitance somewhere on-chip which tries to dump some fixed charge into the load. Q=CV would indicate it's linear with supply voltage as I've observed. As Q decreases from the current, the current also decreases, producing RC behavior. Would heat damage from soldering cause this?

This is with Rref = 2.4k. I have a nearby decoupling cap and bulk capacitance on board. I have a video demonstrating the issue if anyone's interested:

[mikeselectricstuff]

Have you tried thoroughly cleaning the  PCB - flux residues can cause leakage.
 

[allanw]

Yeah I've tried cleaning the board thoroughly with alcohol and it still behaves the same. I put a 1 Meg ohm resistor in series with an LED and measured the voltage across it. So 1V = 1uA. Here's a scope capture with 1 sec/div:

[Rufus]

I would say dirty boards.

You have a small gap between the package pins and grounded thermal pad which would see plenty of flux and be difficult to clean. Alcohol may not dissolve salts which conduct especially when damp. I would also wash with water (preferably de-ionized) and bake afterwards.

I would expect the 0.1mA max leakage specification to be conservative. If it is silicon leakage it is almost certain to rise quickly with temperature so hold a soldering iron to the chip and see what happens.

Contamination leakage may go up or down with temperature. It may also depend on moisture so breath on the board and see what happens.

It is unusual but not impossible for boards to be contaminated during manufacture so if you have some blanks check them for leakage with a meter.

If you are really stuck making and keeping the boards clean put bleed resistors across the LEDs.

[allanw]

Hi Rufus, thanks for the detailed answer.

Heating up the chip causes the leakage current to increase by a lot. I guess this doesn't really prove that the boards themselves have leakage.

Could it be capacitance from the pads to VDD? I only have 10mil clearance from the pads to a VCC pour. If it was just board leakage, would the leakage current behave like what I observed, where it decreases like an RC circuit to some constant value? Here's my layout:



Putting a resistor in parallel would solve the problem... but it makes the PCB design much more complicated and would increase the cost a lot. There's 24 outputs per chip.

Do you think I should wash and soak the entire board after it's assembled? It's not that convenient for me to get purified water though... I'm sure tap water wouldn't make it any better.

[tecman]

Capacitance is not your problem.  Parasitic capacitance on a board is in the pF and not enough to give you your symptoms. 

Bets way to wash a board is put it in your dishwasher.  No soap, just hot water.  Hotter the better.  Remove and air dry.  Sounds crazy,m but it works very well.

paul
 

[mikeselectricstuff]

A quick way to verify would be to lift a pin and measure the leakage direct. More leakage at a higher temp would suggest silicon and not PCB.

[deephaven]

Are your tests with the blanking pin low? If so, you could try setting it which should turn everything off. Might give you a clue to which direction to look in.

[Leo Bodnar]

This is bizarre.  I have used a lot of SPI constant sink LED drivers and never seen this problem.  But I have used simpler ones without dot-correction and brightness control.

Are you sure you are actually shutting them off?  

Check your SPI code and as suggested earlier, turn on the blanking and see what happens.

P.S.  You have huge pulsing currents flowing from Vled in PWM mode and yet you don't use a single decoupling cap?

You are usually required to put at least one good ceramic cap per IC between Vled rail and GND.  If Vled is the same as Vdd then even more so!
Most other manufacturers stress this in their datasheets but TI seems to have skipped over this.

[mikeselectricstuff]

I was wondering along similar lines - If VCC is being dragged down, it could be that the logic drive signals are injecting current into the die via the ESD input protection diodes. There may also be ground bounce issues.

[allanw]

I do have decoupling caps and I pointed them out in the beginning of the video.

The SPI code definitely shuts them off. Trying to output 1/1024 outputs 20uA as expected, so measuring 1uA of current can't be from the device being inadvertently on.

I can do huge load switching by pulsing them on and off and there's no problems there. When testing leakage, I have all the outputs off anyway so everything is static. Note that in the video I have no LED's connected. I even turn off the micro and turn blank on.

I'll try lifting one of the pins to see if it's silicon leakage or my board, but it's difficult to do.

[Leo Bodnar]

I do have decoupling caps and I pointed them out in the beginning of the video.

Thank you for pointing out that you have pointed "them" out.
Good luck sorting your problems.
Leo