Microcontroller fries with no apparent reason

[TNb]

I recently ordered some PCB. When I started soldering stuff to it, everything is working fine except the microcontroller.
I tried to solder just the uC alone with decoupling caps besides and power it with 3.3V from power supply - works OK.
But then, when I solder the usb port and 3.3 regulator something weird happens - instead of consuming 0.1A inrush current and then go to sleep with 0A consumption it just goes crazy and consume 1.3A and inevitably fry itself (it is not soldering short on the board because uC gets hot).
LM1117 regulator is working correctly, gives stable 3.3V, the voltage on power pins is 3.3V.
After soldering uC I thoroughly inspected it under a microscope so I am sure there are no sorts (unless they somehow made it under the package).
The layout should be correct, I triple checked the power pins numbers and the layout in itself is quite simple, rat nets don't allow me to connect anything by mistake basically.
I have no idea what is going on and I already fried couple of expensive chips.
Does anyone can give an advice what should I look for?
The weird thing is that uC works well without USB and LM1117 soldered, so it seems that there is a problem there, but I can't find anything wrong about it.

Here are top and bottom views of the PCB. The power goes from USB port on the left, converted to 3.3 with LM1117 and then goes to uC in LQFP144 package (STM32F429ZI). W6 and W1 are just test pads.

[TNb]

Another interesting thing is that it can keep going on 1.3 amps for a long time, i.e. it does not go crazy for 1-2 seconds and burns traces inside to open circuit. I can poke with probe for some time (I turn power off after 20 seconds not to melt it down on the table). Also there is no magic smoke coming out of it, so it is not completely burns. If that information may help...

[Mechatrommer]

schematics? small pictures of an unpopulated pcb is very little help. could be counterfeit 1117 that fail short giving 5V to uC, could be some short or floated on the output pin of the uC, could be wrong components to 1117 etc etc...

[TNb]

1117 is real I suppose, because all other chips from that batch work ok, and this works well too, giving 3.3 V. The voltage on pins is also correct - 3.3V on all of them, never 5V, as I wrote.
Here are schematics from KiCad (don't pay attention to ULPI labels, I don't use ULPI here, just forgot to delete labels):

[cowana]

Your LM1117 pinout looks incorrect. According to the TI datasheet (http://www.ti.com/lit/ds/symlink/lm1117.pdf), your LM1117 pinout is wrong - for the DCY (SOT223) package, the middle pin (and tab) are connected to Vout, not GND.

Edit: That was at a first glance assuming the top pour was GND. Looking closer, it appears the pour might actually be Vout - in which case you might be right...

[TNb]

Your LM1117 pinout looks incorrect. According to the TI datasheet (http://www.ti.com/lit/ds/symlink/lm1117.pdf), your LM1117 pinout is wrong - for the DCY (SOT223) package, the middle pin (and tab) are connected to Vout, not GND.

I have it the way you say. In schematic symbol "middle" pin has pin number 1, so it is not real "middle". I have both tab and middle pin connected to 3.3 net, i.e. it is Vout.

ah, ok, I seen your edit. Yes, top pour is 3.3V pour, ground pour is on the bottom side.

[tatus1969]

The LM1117 family is not designed to be stable with ceramic output caps. Check how the output voltage develops when powering on the circuit (preferably with a sharp rise time of input voltage), it is possible that it is only weakly damped and oscillates during turn-on. This way the micro may see overvoltage for a short time. Other than that, I cannot see anything obviously wrong.

[CJay]

I think you've got VCAP2 connected to VDD, which is a no no.

From the data sheet '•VDD should always be higher than VCAP to avoid current injection between power domains'

[Kjelt]

some other thoughts:

NRST is not connected how can you program it without connecting it, needs a small capacitor to ground
220 ohm Resistor to the crystal is also not as I expect it.

I think you've got VCAP2 connected to VDD, which is a no no.

+1 and Vcap1 and Vcap2 should also not be tied together, just every pin on its own a cap to ground.

[Kjelt]

Recommend to download the schematics from the nucleo_f429 dev board, just copy the Essentials as a solid starting point 
https://my.st.com/content/my_st_com/en/products/evaluation-tools/product-evaluation-tools/mcu-eval-tools/stm32-mcu-eval-tools/stm32-mcu-nucleo/nucleo-f429zi.html

[mariush]

Yeah, 1117 needs capacitors with some ESR on the output, most require between 0.1 ohm and 1 ohm esr on the output to be stable.  add a resistor in series with the ceramic capacitor or just use a better ldo that has no such requirements.
I'd suggest going with AP2210 , AP2212, AP7333 from Diodes Inc. , MCP1824 from Microchip  and lots of others ...

[TNb]

I think you've got VCAP2 connected to VDD, which is a no no.

From the data sheet '•VDD should always be higher than VCAP to avoid current injection between power domains'

oh shit... you are right... its a mistake obviously. But for some reason uC worked well without the whole regulator thing... though maybe it can draw 0.1 amps like that for 20-30 seconds and then something breaks and 1.3 amps go around...

Recommend to download the schematics from the nucleo_f429 dev board, just copy the Essentials as a solid starting point 
https://my.st.com/content/my_st_com/en/products/evaluation-tools/product-evaluation-tools/mcu-eval-tools/stm32-mcu-eval-tools/stm32-mcu-nucleo/nucleo-f429zi.html

I actually did like you said, I just f#$ked up the Vcap stuff, didn't pay attention and whack it in there...

Well, I am going to try to troubleshoot it somehow, though I need new PCBs anyway...

Thank you for advice

[Kjelt]

though I need new PCBs anyway...

Can't you scar some traces with a sharp knife to couple them loose from vdd just to make sure it is solved?
Darn expensive chips to loose though 

[CJay]

You should be able to isolate the VCAP pins by cutting the tracks with a scalpel, wire in the capacitors and try again, it'll prove that's the problem before you respin the board and find another problem.

May just be me but I find it odd that VDD is copper poured like that (not saying it's wrong, just seems odd to me), maybe make the copper under the chip ground and stitch it through with a number of vias so you can route ground pins to it instead?

[bson]

Some suggestions to help track this down...

Set the PSU to 10mA CC; this will starve it and only shorted paths will conduct.  Go over it with a sensitive volt meter (5.5d or more) and see which pins have Vdd with a drop of 1mV or more - there's the culprit.

Remove the uC altogether and see if the short is there or elsewhere.

With the uC removed, measure the voltage on each pad and make sure it's reasonable.

Keep in mind that many uC's reset to GPIO pins being output low and they can fight with Vdd if you tie them to Vdd.  They will do this until your firmware gains control and makes them inputs.  (I recommend only ever tying pins to their natural default state; if they come on low, tie them to ground, if they come on high tie them to Vdd.  If they come up as high-z inputs it doesn't matter what they're tied to.)