Before I start with the actual problem let me note that I went through the available topics for LT3045 in search for a situation similar like mine. Since I was not able to find information that could point in any direction I decided to open this monstrous post. So, let the games begin.
A little background first. Earlier this year I designed two boards that are intended to stack together (on on top of the other). The boards were manufactured in China with a successfully passed electrical test (no rejects in both batches) and a controlled impedance requirement. The smaller board holds some sensor circuits, a laser diode driver, an EEPROM chip etc. Nothing fancy. Let's call it a sensor board for reference and simplicity. The slightly bigger board holds voltage regulators, a level shifter, auxiliary oscillators. Again, nothing special. Let's call this one the power board.
The start of my journey. After some basic troubleshooting (solder bridges on neighboring pads, wrongly calculated resistor values) I managed to successfully run all four power boards. All voltages were finally with the right values, snow was falling and it was an electronic romance. So, I moved on to the sensor board and I started doing some obvious rework (fixing solder bridges, replacing tombstone components - all the fun stuff) in order to bring it to a state for testing. At this point I should throw in that every circuit block on the sensor block has a ferrite bead in series with its power pin. This position was unpopulated during assembly. And with that being said the first test of stacking the boards and powering them up was a cosmic success. I made a voltage map of every circuit block that was connected to certain line and all values came out correct. At this point there were no overloads, no significant drops and the benchtop power supply has not yet reached its current limit.
The strange occurrence. After having all the correct voltages reaching the correct circuit blocks, I decided to move on with testing. I added the missing ferrite beads to every circuit making them all connected to power and active. I repeated the above test (stack the boards together and add some power supply) expecting that I'll have some circuit blocks to play around with. And this is where the +3.3V regulator (LT3045) took off to maximum current drawing and extremely hot QFN top surface. This happened the moment I enabled the supply's output. Not more than a second or two. Now, that's some fast transient process. Anyway. I disassembled the stack and check the sensor board for shorts between the voltage rails and ground. Nothing came out. I check the power board on it's own and nothing came out there as well. Except the fact that the +3.3V regulator was outputting 0.8V and it was still drawing maximum current.
The next clever thing I did was to take another power board and repeat the test. This time +9V (LT3045) went wild (nothing on the output and extremely hot). At this point I stopped making mad scientist experiments.
Some technical details. Here are some numerical details regarding the power board.
VIN: 15V - 18V done with LTC4365.
VOUT: +1.8V, +3.3V, +5V and +9V done with LT3045; +15V done with LT1764A; -5.7V and -7.5V done with LT3093.
PG pins are not used and they don't go anywhere.
Here are some technical details regarding the sensor board.
There are two separate ground planes on the sensor board - DGND and AGND. The common ground is achieved on the power board where there is only one plane - PGND.
So, what am I missing? Why are LT3045s randomly burning out? I've done something stupid and obvious, but I am missing it.
If you need additional information that might seem relevant for the problem but I've missed, just ask.
Any feedback would be highly appreciated.