Hello. After switching to NCS v2.0.0, we started observing a strange I2C issue - some sensors did not communicate whatsoever, and some sensors worked just fine. Further analyzing this, the problem was that there was a high voltage for the log. 0.
See the waveform and the "third state" on the green line (SDA):
This is an I2C read sequence and you can see the slave device cannot overdrive the master, which master is actively driving log. 1 to the SDA line.
The fix is to alter the pinctrl settings and switch SDA + SCL lines to open-drain mode.
Original code:
i2c0_default: i2c0_default {
group1 {
psels = <NRF_PSEL(TWIM_SDA, 0, 21)>,
<NRF_PSEL(TWIM_SCL, 0, 20)>;
};
};
Fixed code:
i2c0_default: i2c0_default {
group1 {
psels = <NRF_PSEL(TWIM_SDA, 0, 21)>,
<NRF_PSEL(TWIM_SCL, 0, 20)>;
nordic,drive-mode = <NRF_DRIVE_S0D1>;
};
};
Our original configuration was inspired by other dtsi / pinctrl examples in the NCS v2.0.0 tree. Apparently, all of them do not use the open-drain drive, and despite the slave might "seem to work", from a hardware perspective this is quite a critical misconfiguration. Not only this problem increases the power budget for the I2C transaction, but it might also exceed the slave chip specifications.
When the chip has a slightly weaker drive strength than nRF52, the logic threshold might be on the edge and the system may work in an unpredictable manner.
Thank you. P.
