32kHz oscillator on ESP-WROVER-KIT-VE not working

[istokm]

Sometimes I don't see the "BTDM_INIT: 32.768kHz XTAL not detected" message, but the 32.768k clock is unstable and causes the board to usually crash when entering light sleep.

I can confirm the exact same findings. I was not the one who tested the 10M resistor, a colleague did, but reported no change at all in the boards behaviour - though I haven't reported this because we didn't yet test this on the WROVER board, but a custom one.
As to the unstable clock, this only happened to me if I pushed the calibration cycles to a absurdly high value (8K+) & the board probably thought that the clock was running. This would either lead to RTCWDT panics (or something similar, I don't remember the name from heart), boards not waking up from a timed deep sleep, or something similar.

We'll continue to debug this, but as a simple sure fix (for custom boards), I would suggest just eating the cost of a MEMS oscillator instead of dealing with undocumented hardware requirements - this is our fallback plan as well, as we've been able to demonstrate that an external clock (ex. function generator) works reliably.

[therealergo]

I've re-tested everything on the WROVER-KIT with the 10M resistor between XN and XP, and it's been reliably running for an entire day now. (!!!) Maybe the flux/alcohol on the resistor hadn't fully dried when I first tested it. The XN/XP differential on the WROVER-KIT looks something like this, constantly:

Reference_32k_Clock.png (29.95 KiB) Viewed 4103 times

Unfortunately, the behaviour is still the same on the custom boards with either no resistor, a 10M resistor, or a 5M resistor. Looking at the scope, the oscillation slowly dies out over a couple seconds:

Bad_32k_Clock.png (33.2 KiB) Viewed 4103 times

As a simple sure fix (for custom boards), I would suggest just eating the cost of a MEMS oscillator instead of dealing with undocumented hardware requirements

A good idea, but I want to avoid that cost if at all possible
Considering that it now seems to be working for me on the WROVER-KIT, I'd think that this is solvable.

[m.madela]

the 10Mohm register in parallel to quartz works for me, thank you

[johnboiles]

So is it a bug in the hardware design that this 10M resistor isn't included on the PCB? The onboard crystal is useless without it right?

The hardware design guidelines (https://www.espressif.com/sites/default ... nes_en.pdf) seem to confirm it's necessary:

When ESP32-D0WD-V3 connects to an external 32.768 kHz crystal, the parallel resistor must be populated.