I have a design using two TPS92512DGQs, each one driving five (5) high-power IR LEDs. A total of ten (10) IR LEDs are located each the PCB.
Each regulator block is as identical as possible (a reused schematic block, with small but not significant differences in the layout). 35 µm copper in ground and feed planes, so the IR voltage drop should not be a factor.
We have manufactured about 30 of these boards, and only about 10 of them work as intended. A yield this low is of course not acceptable. And it seems very strange that some units work, and some don't with identical implementations.
Sometimes, one of the LED drivers work, but the other does not, on the same board.
A typical problem is that the maximum current will follow a sawtooth shape: From 100% of the design current, which will fall down to about zero after about 600 ms, after which the current jumps back up to 100%. I.e., we have LEDs that blink with about 1.6 Hz.
Other units won't ever get up to the full design current.
Some design parameters:
- VIN = 24 V.
- VOUT = 16-20 V according to LED datasheets (i.e., 5 * Vf) per LED driver.
- Output current (when PDIM is high): Set by constant voltage to IADJ, 220 mohm sense resistor => up to about 1.3 A through the LEDs.
- Inductor: 10 uH
- Output capacitance (across LEDs): 4.7 uF.
- Switching frequency: About 1.8 MHz (56 kohm from RT/CLK to GND).
I have measured various voltages and don't find any obvious differences. I have followed the formulas in the datasheet, and can't find any item that is out of specification.
However, when reducing the regulator switching frequency, all problems disappear. 82 kohm instead of 56 kohm on RT/CLK works, this is about 1.3 MHz.
To me, it seems as if the TPS92512 has some kind of internal resonance at this operating (switching) frequency. And minor manufacturing differences from unit to unit makes some devices function as intended, as others not.
If lowering the switching frequency, which will increase the inductor ripple current (and heat), fixes the problems on these particular boards, what assurances do I have that the next batch of TPS92512 regulators won't have a resonance at a slightly different frequency?
In other words, how to I make sure the design is stable, even when accounting for component tolerances?
And why did this problem appear in the first place?
