To pad or not to pad (q. about using pad vs. straight surface)

Hello Vitaliy,

Sorry, but I do not find that picture below too clear. Actually rather confusing.

The old rule is that there be the least amount of parasitic inductance from the ground plane to one side of a decoupling cap, and from the supply rail to the other side. If with "pad" you mean the SMT solder lands then the connections must be as short as possible while maintaining thermal relief rules.

Well, we all do :-)

But if you already have it running you should be able to gain some extra margin by short bypasses, clever placement and shielding.

Vitaliy, With what you have explained I would offer the following.

Since you are putting a decoupling/filtering cap on the signal/power, of what worry is the parasitic capacitance? The parasitic capacitance is undoubtedly much smaller than the capacitor you are using, particularly on the power rails. More capacitance is better in that case, that is also why closely spaced gnd/vcc planes work very well for the higher frequency decoupling, not quite the same as your case but just presented the concept for your consideration.

As for the assembly of such capacitor placement, as long as you can solder then anything is fair game. The only concern would come in in the form of reliability. Soldering the end cap of an SMT cap on top of a trace would not be as reliable as using a pad where the solder can form a full meniscus off the end of the cap and thus supply a highly reliable joint. This assumes that your trace is not as wide as the pad would have been length wise, otherwise there is no point in this whole discussion if the pad was smaller than the trace to begin with because adding the pad would have made absolutely no difference to the circuit whatsoever.

Connections to large blocks of copper are generally always better for general circuit performance. Ever wonder why all of those sample/evaluation boards from chip manufacturers are typically done removing only isolation channels between non-connected components/signals? Some of those circuits are difficult to get operating to the full spec'd performance if you use more typical traces to connect the circuitry.

The largest problem for oscillations in a circuit like you are describing is typically dealt with through good part placement and thus good routing. Keeping inputs away from outputs, keeping feedback loops very short and direct, etc.. Since your currents are so low, you might consider guard tracks between the input signal(s) and other nearby copper. After making you board you should make sure it is well dried (bake it in an oven for several hours at up to 100C), then conformal coat it so that contaminants can't interfere with your small currents. Although these days uAmps aren't that small, they could still give you problems doing this yourself using (I assume) hobbyist techniques/materials/processes.