I'm prototyping a low-noise photodiode frontend using the ADA4817-1 with 2 MΩ transimpedance gain.
Despite considerable efforts to reduce feedback capacitance, I can't seem to get more than ~730 kHz bandwidth (-3 dB). Using the updated SPICE model from Using ADA4817 as TIA stage for fast pulses, I'm hoping for around 2 MHz instead.
The measured bandwidth corresponds to about 100 fF parasitic capacitance between output and inverting input. Parasitics are of course tricky to manage at this level, but the carefully laid out feedback path consists of two 0402 thin film resistors in series, with ground fills below them. The shunt capacitance should be significantly better than 50 fF this way.
This leads me to suspect that the op-amp/package itself contributes the largest part of the 0.1pF. Is there any data on the stray capacitance from output to inverting input (i.e. die, wirebonds, the package and the PCB footprint on top of a ground plane)?
It's a shame the ADA4817 is only available with the low-distortion pinout; it stands to reason that another package, e.g. SOT-23 (such as for the ADA4895-1), would perform much better in this regard.
(Interestingly, I'm observing the almost the same bandwidth with both the optimised PCB layout using the LFCSP variant and 0402 resistors, and the SOIC package with the feedback path consisting of 4 0603 resistors soldered between then pins in a bridge shape.)