Comment: I note that the test circuit in the ADG620 spec sheet shows that the specified charge injection is measured with the switch D (pin 1) driven by the test source and the load side is one of the S pins. I also note that we are operating the switch is the worst part of the charge injection-versus-signal-voltage graph (snapshot attached). Nonetheless, the amount of charge injection that would account for the change we see would have to be on the order of 1000x the range seen on the graph.
Questions: Since charge injection to the D pin is not specified, do you know what ballpark we should expect? Do you have any other switch that would drop in and do better. [Note the logic control is referenced to Gnd and Vdd (+3.3 V), while the Vss is at -4.5 V.] If the switch does turn out to be the problem and we cannot find a drop-in replacement, do you have a non-pin-for-pin that can handle the same control and bias voltages with low charge injection?
“Swapping pins 1 & 2 on the switch made no difference, it looks the same as the original way with about a 3 mV bump in the data when the 10k resistor is switched back in. So it looks the same whether the signal path is going from pin 2 to 1 or pin 1 to 2.”
I calculate that voltage change on 1 kohm load to translate to a 0.81 mV change on C203, which represents a charge change of nearly 18 nC. This is around 100 – 200 times the charge change expected per the data sheet, so it is not likely to be coming from the switch.
Do you have any past experience with mystery charge injection?