Both - specifically, that the ground end of the decoupling capacitors needs to have a short path to the ground pad on the RP2040. Current flows round a loop, and the loop here is from the + end of the cap, through the RP2040 power pin, through he RP2040 logic, out of the ground pin on the RP2040, and back to the - end of the cap. You need to assume that with a modern high-speed device like the RP2040 all the power is drawn from the decoupling capacitors (in high-frequency pulses) and the rest of your power wiring is just to keep the capacitors 'topped up'.Do you mean ground connections as in the ground connections that on the end of the 3V3 decoupling capacitors? Or are you referring to the ground pad in the center of the chip. I’m fairly competent with board design, I just so happened to try something new here.
Power is typically the most critical aspect of an RP2040 design; the next two being the crystal (but that's typically easier to get right, all on the top side, unless you have mechanical constraints stopping you putting it right next to the chip), and the flash (again straightforward, but it needs to take priority over other things, and think about placement of the resistor for the boot button, which acts as a dividing line between high-speed and low-speed areas).
In general on a 2-layer design, I aim to get nearly all the wiring done on the top side, with just _short_ links on the bottom, such that the bottom can become mostly groundplane and there aren't any long-distance connections slicing the groundplane in half.
30 years ago I'd have flooded the spare space on the top side with power, but with modern devices using less static power and all the dynamic power having to come from capacitors that doesn't make sense any more, so I fill the top side with ground as well and use extra vias to use the top-side ground to 'stitch together' any significant gaps in the groundplane caused by tracks on the bottom side.
Statistics: Posted by arg001 — Wed Jan 31, 2024 9:55 am