Would it damage the micro to cycle the power relatively frequently so you can sense the magnetic field change with the step in current? From my vague awareness of power glitching attacks in reverse engineering, I get the impression they are fairly resilient to this kind of repeated abuse.
A Hall sensor might be too insensitive. Maybe a magnetoresistive sensor? https://www.farnell.com/datasheets/3211980.pdf has a 2x2mm package that might approximate to a probe. I made a temperature sensor like this and a long thin PCB works fine as a handle/probe with a small package like this at one end.
It sounds like it might be what you need for this, to be honest I have only done some quick measurements with it. I see there is an EEVBlog review maybe has more details (another untested recommendation, just saw it came up in search results).
Looking at that totally reminded me. I have the Little Bee B1, but I have never really put it to use. The CrowdSupply page compares it to the i-Prober 520. Unfortunately, it’s showing no longer available on CrowdSupply and Mouser. Not sure if you can still find it anywhere…
This is my first suggestion as well, but unfortunately Weston could only get 3k of the core AMR part he used. It’s still available, but only at an MOQ of like 25k parts or something absurd, so he’s expecting never to produce more. I guess the i-prober is probably the right call.
That said, my usual go-to is thermal. Of course, you say you already tried it and couldn’t find any hotter-than-average traces, so maybe not.
It is a Cortex M3 micro, and it looks like the current enters the device from the 4 corner VCC/GND connections (faint thermal line seen in each corner)
So right now I am guessing it is a higher current due to a floating pin. So will try to connect each pin in turn to VCC with a 1k resistor and see if the power dissipation is reduced
This board is manufactored at a EMS, but the soldering is bad:
I would like to find the fault before re-soldering all pins, which could cause a short and a dead device
Interesting idea. Or if it’s steady state DC, just put the DMM positive on the rail and positive on the top of the VCC pin to see relatively how much drop across the PDN there is on each one.
I’d be willing to bet that’s just the shape of the leadframe conducting more heat out of the core:
I think it’s pretty common for leadframes to have stubs of metal out to the corners for mechanical handling during manufacture, and if so, those would be the most thermally coupled to the core.