STM32H753 initialization failure when in contact with the chip physically

[melody_kingdom]

Our application is to use STM32H7 to finish the communication to FPGA/LAN9252/PHY/SDRAM. The power supply of VDD3V3 is generated by an LDO with Tr = 2.4ms.

During the test, we found that if we get in contact with STM32H7 (not pins but the package) by ANYTHING - hands (with ESD protection), plastic, paper, whatever without much pressure, we will meet the initialization failure, and if not, the initialization as well as the program work quite well.

Is there any explanations or suggestions to this phenomenon? The final product will contain a cover to avoid any directly contact to the chip. But we are still worried about some unknown risks or problems.

[Tesla DeLorean]

Do you see similar behaviour on NUCLEO and EVAL series boards, or just your own?

Sure you don't have floating grounds or disconnected islands/nets on the board?

Issues with capacitors and supply regulation? Issues with soldering or traces?​

[melody_kingdom]

Thank you for your reply!

Yet we don't get any EVAL series boards from the distributor, and it is the first time that we make application using STM32H7 series. We fabricated 10 boards this time and all of them have the same problem.

However, we made a lot of sucessful products based on STM32F1~F4 series applied to industial motion control. So the peripheral circuit design is reliable. The possible reason might be the match-up between peripheral circuit and MCU, or the misconnection of some configuration pins.

Some further details:

1. PDR_ON pin is connected to VDD3V3; We already tried to use 10k pull-up resistor, and connect this pin to GND (with or without 10k pull-down), but it doesn't make sense.
2. ​VCAP1 & VCAP2 pins are connected to GND via two 2.2uF low ESR capacitors. Vvcap = 1.2V, but the rising edge has a step (find details in attached figure 1).
3. BOOT0 is connected to GND via 100k pull-down, and BOOT1 is float.

Figure 1. Vvcap rising edge

Figure 2. VDD3V3 rising edge

[Tesla DeLorean]

I have NUCLEO H7 boards here, I can run them North of 500 MHz while touching them without them crashing or getting hot.

Are you taking static precautions or is the test environment particularly dry?

A​s it is effecting multiple boards I would double check the netlist make sure you don't have any odd or misnamed nets causing islands, especially in grounds, ie hidden pins in symbols, separate RF, Analogue and Digital grounds which are all bonded at some point.

Also inspect unpopulated boards for issues with metalization/etch, and do continuity testing while flexing.​

[Tesla DeLorean]

>>Yet we don't get any EVAL series boards from the distributor

These things are available at multiple places, the NUCLEO's significantly more prolifically. You might have to pay for them.

Have the FAE supporting your account, either at the distributor, or ST, work to understand the problem you are describing.

I can touch the parts on the H7 EVAL and NUCLEO without them having kittens, so a more rigorous review of what you are doing is in order.

Make sure parts place on the board have the expected values (resistors, inductors, capacitors) whether that's due to the wrong reel in the pick-n-place, or defective/counterfeit parts.

Check for soldering issues, issues with solder mask, thickness of traces, and other quality issues. Try pushing the part with a wooden stick, does flexing the board yield similar failure?