Sounds odd. If the voltage you're measuring is consistantly 900mV regardless of the pull-up, that could mean that the MCU's IO is indeed driven high as a totem-pole output (and not as open drain), and the SDA line of the ATSHA204A is forcing it low. Have you tried looking at the SDA pin of the ATSHA204A while it's powered BUT SDA disconnected from the MCU? Is it then still low? You could issue a single shot acquisition triggered on Vcc when you're powering it on to see what's happening.
I did a measurement with SDA disconnected and I get 720mV. I cannot do the single shot acquisition at the moment easily, unfortunately.
The wakeup routine basically means that you have to pull the input low for ~60us and then high for 2.5ms.
The single wire "wake" token has 8 consecutive low bits, including the start bit. At the required 230.4 kbaud rate that only provides a 34.7us low pulse, so what am I missing?
Fair point, but the datasheet mentions this:
"The Wake token is special in that it requires an extra long low pulse of tWLO on the SDA pin (see Table AC Parameters – All I/O Interfaces), which cannot be confused with the shorter low pulses that occurduring a Data token (Zero, One, ZeroOut, or OneOut). "
According to the mentioned table the minimum time is 60us. I have tried different durations up to 100us at this point. The datasheet seems to suggest that the pulse can be longer than 60us.
One very far fetched thing I am thinking what could still be the case is that initially the ATSHA204A is working but loses sync at some point somehow. In that case I would expect to see full voltage swings though, which my logic analyzer should be able to pick up.