The ADA4940 is being used to buffer the output of an AD9707 TxDAC. This is a circuit we've successfully used before with the ADA4932 differential amplifier. The wider output swing of the 4940 is very helpful in this application.
I have six prototypes, one has the INL and DNL we would expect but the others have poor INL. The output common mode offset of the 4940 changes with output signal level; with the offset typically at 0mV for much of the range but reaching 5mV around 0V. This offset change distorts the single ended output signal.
The feedback resistors are quite high and I initially suspected the device could be oscillating. However looking at the part with a spectrum analyser revealed a clean output. Lowering the resistors does fix the issue but I'd like to understand why to avoid a marginal design; the circuit shouldn't be near any limit of the amp or DAC (DAC load is nominally 400mV).
The attached images show the circuit diagram and the amplifier Vocm change. The test system was designed for the 9706 12bit device so there are 4088 steps instead of 16k. This shouldn't affect the result.