My GPIOs are testing/characterizing now. My last problem: The JP6 jumper on the nucleo board was mashing into the spring pins of the nucleo hat board from the bottom. This was short circuiting a few signals (like gnd and CARAVEL_CSB) and causing problems. Solution: Put the hat board on but only halfway. If you don't put it on enough, you will get a Winbond SRAM not found error, if you put it too far it shorts out the pins. First try:

===================================================================
== LOW chain FAILED.   Valid IO = 0 thru 00.                      ==
== HIGH chain PASSED.  Valid IO = 19 thru 37.                     ==
===================================================================

FYI, once you've got it running the configuration, if it's not passing, the two best things to do are (1) try other chips, and (2) try a lower voltage (down to about 1.43V; lower voltages tend to stabilize hold violations that are on the boundary) (

make run PART=xyz VOLTAGE=1.43

).

Thanks. I was just going to try another chip, but I'll try the voltages first.

Obviously if you're trying to characterize SRAM, you don't want to be limited to running at 1.43V; but you would only need the lower voltage right after reset to make sure the board conditions match the calibration. After that, you can set the voltage to whatever you want to using the Micropython routines on the Nucleo.

Another great suggestion. Thank you.

The openlane team did, which is how we landed on the current default value of 1.6V (which is also what the on-board LDO is set to on the caravel development board, if you switch the supply headers from "HAT" to "MAIN"). More chips could be tested at 1.6V than any other. But that was early work and I don't know if it applies any more. Anecdotally, I've found that generally, lower voltages work better, which makes sense, as lower voltages make the circuits run slower, which trends in the direction that resolves hold violations. Somewhere below about 1.43V, everything stops working.

So maybe a bisect search between 1.4 and 1.6 🙂

Most of the scripts were written by @Marwan Abbas, but I've been regularly running

mpremote

and figuring out how to run the routines manually, so I can help.

@Matthew Guthaus I would suggest to add retrying the GPIOs several times when you implement sweeping, since I am afraid that you will get lots of noise when you try the GPIOs just once per voltage step.

@Philipp Gühring I bet we could heat it up too actually... Kinda joking. 🙂

At least please please log the temperature of your tests, so that you can compare later.

Is there a temp sensor on the nucleo board? I actually don't know what it is in this room

I was also thinking about doing temperature sweeping, but my hot-air station unfortunately only allows to define the temperature between 100°C and 500°C, not below 100°C, and it also does not allow to configure it by software yet.

On the subject of voltages, if we made a custom board for the rest of the chips, is there any problem forseen with running the caravel domain at 1.5V/1.6V and the user domain at 1.8V, assuming we have no unmitigatable hold violations in the user project?

@gatecat: No, there should be no issues running Caravel at a lower voltage than the user project.