STM32F4 custom board design problem

Question

I designed a PCB with STM32F411RE and a couple of sensors.

Before and after mounting components on the PCB, I checked the Vcc and GND pins for short circuit and I did not see any problem.

I am using SWD programming and I connected my ST-Link V2 to my board and successfully loaded the program.

Everything was normal for a few minutes. Approximately 3-5 minutes after, the MCU disconnected from the PC and then I checked the Vcc and GND pins and observed a short circuit.

Interestingly, as time goes the resistance between Vcc and GND pin increases by itself.

After that, I just mounted STM32F411RE, reset pin configuration and Vcap. I connected my board to PC and loaded the program successfully.

After a while the power pins shorted again. I just uploaded a simple LED blinking test program.

My board schematic and PCB images are below. Please help me to find my mistake.

Answer 1

First off: that is a terrible schematic; no ground or power symbols, no logical "flow", arbitrary boxes for things like the FET... the layout fragment looks good but you should spend some time working on improving your schematics because it takes considerable time to try to decipher them -- both us and you.

Moving on: one thing I notice right away is that your MCU is running off of 3.3V while your sensors are running off of 5V; this is going to cause trouble and it could be taking time for the regulators to finally punk out on thermal cut-off as they "fight" with each other through the sensors and MCU.

You've been asked by several people if anything is getting hot; something should clearly be getting quite warm, and my suspicion is going to be the regulators. A good rule of thumb is that if you can hold your finger/hand on the part, it's under 40C; if you're not particularly heat sensitive the part could be as hot as 50C.

I feel that you are not giving us all of the information. Take your time and methodically work the problem. When the board is cold, the system works. After a while, it stops and you keep mentioning that the rails are shorted. I'm willing to bet that nothing is shorted and the output is simply 0V. Feel around. Something will be getting hot. Probe around; watch the 8V, 5V and 3.3V rails -- something is going to move -- either suddenly or gradually. Put LEDs with appropriate resistors on the 8V, 5V and 3.3V rails -- what do they do, do they suddenly go out, or do they slowly dim?

Like Sherlock Holmes says: "Data, data, data; I cannot make bricks without clay!"

Answer 2

I could not solve the problem yet. I have reduced the solution to two different main distinction. I assume that internal voltage regulator shorts the power pins.

1- A physical interaction breaks the regulator. 2- Too much current is taken by internal units.

First one might be related with soldering issues. I saw some voltage regulator IC's that is broken after soldering with higher temperature and out voltage gets short to gnd.

Second one is only possible through SWD pins because only swd pins are connected to controller in my last setup.

I'm planning to measure the current difference between SWD connected setup and VCC-GND only setup but I dont have unused controller anymore I need to wait a few days.

If I can verify that power consuption is increasing dramatically after connecting SWD, we can say that we have a SWD interface design error.

I have tried a lot of stuff. Nothing else I have.

If you try something and get any kind of result please let me know.

Good luck.

Answer 3

Probably I solved the problem finally. I removed the LEDs and did not use flux while soldering the STM32F411 and changed the ST-Link. As a result, the board is working fine, however I did not mount all of the components. For now, I mounted STM32F411, HC-06 Bluetooth and reset circuitry and it works without any problem.

I think that the problem may be caused by the LEDs or the ports that connected LEDs or maybe the ST-Link that I used is damaged or there is something. In addition to that I doubt that maybe particles (dirt) in the flux may changed the conductivity of the flux.

Tomorrow, I am going to complete the board.

I will give a feedback from here. Thank you everybody for your contribution.

Answer 4

I don't believe that you have solved anything. The board is 99% to blame. It's not fixable, any physical boards you got can be tossed. Given everything else, it's likely that the MCU or some other part connected across power supply rails (like voltage regulator perhaps) fails due to electrical overstress caused by high inductances on the power supply circuits and voltages and/or currents beyond the limiting values. It's also possible that a regulator starts oscillating once it gets warm enough, and then runs away when it's presented with higher voltages it generated by pumping the L-C tank formed by all the long traces and the solitary capacitors stitching them here and there.

1. The schematic is missing key decoupling components.

2. You have not actually observed the behavior of the power supply rails - you have to look at each with a 100MHz digital oscilloscope (at least), using proper probing techniques.

3. The 5V I2C bus is connected to a part with 3.3V I/O output swing (the MCU). This may not cause any damage, but is not how it should be done and you're cutting it close with margins for logic levels on the 5V side of it. In the ADS1115 datasheet, the \$V_{IH}\$ specification is given as 0.7VDD minimum. 0.75V = 3.5V. The MCU is not really driving 3.5V onto the I2C pins, thus you're using the A/D converter outside of its spec. Note that the 0.7*VDD does not refer to the voltage on the pull-up resistors, but to the supply voltage to the ADC itself - the ADC does not know nor care what voltage is on some resistors, all it cares about is what voltage results on the I2C pins. The pin protection diodes will shunt the pull-up voltage to MCU's VCC (3.3V), with a fairly small voltage drop since there'll be only about 100uA flowing into the MCU's pin when the levels are high. If you were to measure the high-level voltage on the SDA and SCK signals, it'd probably be right around 3.4-3.5V, either below spec or with no margin left. Even if it "works" now, it'll be reliability headache later. It's not causing damage, but if you get junk data from ADC (even occasionally), you'll know one reason why (your firmware may supply other reasons).

4. It'll be very hard to see what other blunders are there on the PCB without you sharing all layers in full. From what I can see, there's no consideration given at all to the requirement of low impedance supply connections to the MCU. The supplies are routed as if they were some non-critical signals. Lots of potentially destructive ground bounce and VCC bounce is likely present. If some of the vias shown are meant to be ground plane stitching, then at least 3/4 of them are missing.

5. There are zero decoupling capacitors visible in the vicinity of the CPU. Power supply and decoupling (e.g C5) is routed without regard for loop length. It's the loop that counts, after all - a circuit is literally a circuit. You have to be able to go round it with a finger tip or a tip of a pencil when someone asks - the layout must be, after all, driven by such considerations. For example, for C5, to investigate one should draw (red pen!) the path from pin 30, through the capacitor, to the ground plane (which seems to be on top), and then back to the related VSS pin (31) - it's unnecessarily long, and the capacitor looks like a fairly large package, too. But in that picture we should see either all of the decoupling capacitors, or direct vias to the capacitors mounted on the bottom layer directly underneath the VDD+GND pin pairs on the MCU. Those pins are paired for a reason: to minimize the loop length from chip's VDD ring through bond wires and lead frame and PCB to the capacitor, and back to the GND ring in the chip. Ideally, that loop should be <10mm long, on higher performance processors it'll be 5mm long or less.

Basically, the board is laid out mostly as if all copper was an ideal conductor and even more importantly, as if inductance was absent (even superconductors have inductance!).