TWI clock pull up

[aras]

The data sheet recommends that both the TWI clock and data lines use pull ups and the GPIOs should be set to S0D1, essentially open collector drive. I can see the need for this on the data line, but if my I2C device does not do any clock stretching, is it safe to omit the pullup on the clock line and set the GPIO to S0S1?

I have tried this on my board and have seen no ill effects.

[aras]

I'm not so sure about that.

I set up my clock pin as follows:

  NRF_GPIO->PIN_CNF[twiInstance[twiHandle].clockPinNumber] =     \
    (GPIO_PIN_CNF_SENSE_Disabled << GPIO_PIN_CNF_SENSE_Pos) \
  | (GPIO_PIN_CNF_DRIVE_S0S1     << GPIO_PIN_CNF_DRIVE_Pos) \
  | (NRF_GPIO_PIN_NOPULL    	 << GPIO_PIN_CNF_PULL_Pos)  \
  | (GPIO_PIN_CNF_INPUT_Disconnect  << GPIO_PIN_CNF_INPUT_Pos) \
  | (GPIO_PIN_CNF_DIR_Output      << GPIO_PIN_CNF_DIR_Pos);

There is no external pullup. Attached is the scope trace of the clock pin.

IMG_20160902_102434.jpg

Of course, this only works if there is only one master and the slave device does not do any clock stretching.

[Ole Bauck]

You are sure the slave does not have internal pull-ups? Try measuring without the slave attached.

[aras]

I verified with the slave device removed. No pull-up. No other connection. Exactly the same result. Chip is N51822, CFACA0, 1506AK

[Ole Bauck]

Now I understand. You cannot set the drive to S0S1. If you do you will have a short when the slave pulls the clock down to ACK the master. I will update my original answer with more info.

[aras]

The slave pulls the DATA, not clock. Is this correct?

[Ole Bauck]

Sorry for the late answer, the slave should pull both lines.