[SOLVED] Failure analysis of broken MCU

[Lucast85]

Hi everyone,

I have a mixed analog/digital embedded board with a mirocontroller (STM32f407... 144 pin) in production and I notice 3 failure on a total of 1000+ boards. The component broken is the MCU (in a board is still working but it is very hot and sink a lot of current, more than the specification) but I cannot undertand what is the cause of the fail.
What is a good approach to reveal what pin/port sinks a lot of current? If I will find the hungry pin(s), there are good chance that the cause of the failure is the circuit close to that pins?

Do you have any suggestions?

Thanks

 

[wwfeldman]

The component broken is the MCU (in a board is still working but it is very hot and sink a lot of current, more than the specification) but I cannot undertand what is the cause of the fail.

The working 1000+ indicates this is an issue with the 3 bad boards
how detailed is the testing to determine good or bad?

is the MCU broken, or is the board working correctly?
is every aspect of the board working correctly?

check every aspect of the MCU and board. the section that doesn't work correctly may point to the problem

get a microscope and an x-acto knife with a sharp blade. put the MCU under the microscope. scrape the x-acto knife blade between every pair of pins.
there might be an invisible "solder splash" tying two pins together.

if you have a "bed of nails" impedance checking machine, apply it to a few good boards and to the three bad ones.
examine the locations where the impedance checks don't agree.

 

[andre_luis]

Portable thermal camera readers are often used to determine if there is, or what device, are hot in the board. If you get an affordable one, perhaps you could be able to - with certain degree of accuracy - determine at which region of the chip the heating starts and then have a clue of the candidates outputs that are sinking overcurrent. If you have luck, you could even determine what other devices/tracks on the board others than the microcontroller itself are hot aswell.

 

[Lucast85]

The board passes a production test made with help of AOI and ICT. The coverage is about 85%.
We already inpect the MCU with a magnifier but the pins seems ok and no traces of smoke are visibles.
All the "main" aspect of the board seems working.

We are sure that the guilty was the MCU because with the help of the thermal camera we see the hot surface of the broken MC. The but we are unable to distinguish the pin.

Fortunally we received the response from the failure analysis from STM. They opened the MCU and they find the burnt pin. It was a spare gpio with a 5 cm trace left unconnected. It acts as an antenna because of a noise DC/DC mounted in a mezzanine board.

Thanks for your suggestion.

 

[wwfeldman]

Lesson 1: never leave a pin unconnected or otherwise fixed

Question: why those three boards/MCU, and not any of the other 1000+
all the other boards have the same unconnected 5 cm trace

 

[Lucast85]

Lesson learned

Probabily it is due to the noise of the DC/DC converter used on a certain situation (we have only 10 boards used in the exactly same situation). The value of the input voltage and the output current of the DC/DC are the "worst" for that converter. I try to measure the voltage on the broken on a new board pin with an oscilloscope. I can see a DC value of about 2 V with some spikes on it.

**broken link removed**
**broken link removed**

After the measure, I see the pin is shorted to Vdd and stop to work as digital output (neither PP or OD). The MCU still woks and still not consume a lot of current but I think it is matter of time.
-The new question is: why I see a DC value? Maybe the non linear devices into the input interface of the MCU performs a sort of demodulation?

On another board I try to set the pin as OD and to force it to GND. The noise and the DC disappear and the voltage on the pin is now less than 30 mV.
-Waiting for the new PCB revision... this SW solution could be a good solution? Note that the DC/DC starts to work after the MCU is on.