A unit that resets once a day, or fails to start one time in twenty, is hard to debug once it is screwed into a panel. On ESP32 designs, two hardware causes account for a large share of these reports, and both can be caught on the schematic.
Strapping pins
A few GPIOs are sampled at reset to decide how the chip boots. On the ESP32-S3 they are GPIO0, GPIO3, GPIO45 and GPIO46. On the original ESP32 they are GPIO0, GPIO2, GPIO5, GPIO12 and GPIO15. Check the datasheet for your exact part, because the list and the effects differ.
The trouble starts when something external pulls one of these pins the wrong way at power-up:
- A button, a sensor output or a transistor base on GPIO0 that happens to be low at reset puts the chip into download mode. It looks like a dead unit until someone power-cycles it.
- On the original ESP32, GPIO12 high at reset selects 1.8 V for the flash. With a 3.3 V flash chip, the board will not boot. A pull-up on that pin, or a peripheral that drives it high, is enough. GPIO45 has a similar role on the ESP32-S3.
- Peripherals that are still powered while the ESP32 resets can back-drive a strapping pin through their own outputs.
What we check: every strapping pin, what is connected to it, and what state that connection is in during the first milliseconds after reset. If a strapping pin must be used, the external circuit should be high-impedance at boot, or the flash voltage should be fixed in eFuse so the pin no longer affects the boot.
Brown-out
The radio draws short current peaks of several hundred milliamps when it transmits. A regulator that is fine on average, but cannot supply those peaks, lets the rail sag, and the brown-out detector resets the chip. It shows up most on weak USB supplies, long thin cables and battery packs near empty.
What we check: a regulator rated with margin for the transmit peaks, bulk capacitance close to the module, and short, wide power traces. We also check the RC delay on the EN pin, so the chip does not start before the supply has settled.
Disabling the brown-out detector in firmware is not a fix. It turns a clean reset into flash corruption or a hung radio.
Make the firmware tell you
Read the reset reason at every boot (esp_reset_reason() in ESP-IDF) and send it with the first message to the cloud. A fleet that reports “brown-out” from twelve units on one site tells you about the site’s power, not the code.
If your ESP32 units reset in the field and not on the bench, book a discovery call. A schematic review usually narrows it down.