Specifically the 1284P, but several other MCUs would require a cross-connection between GND pins on opposite sides and VCC to AVCC. Like the 40 pin DIP: GND pins 11 to 31 would cross the connection of pins 10 to 30.
So I am wondering why they didn't choose pin locations to allow two parallel connections when AVCC or AREF need to go to VCC, without crossing the GND connection. In this scenario, pins 10 and 11 would be swapped out.
I always assumed it was so you'd have a prayer of saving the chip if you plugged it in backwards, instead of it shorting supply and ground though the chip.
I believe if one could look at the die of the processor itself, the location of the power and Gnd connections would show power & gnd coming in to various sections of the chip, perhaps to help with distribution. I know AVCC goes to the corner of the chip where the ADC circuit is for example.
In SMD parts, there are multiple VCC pads. In DIP parts, the die pads may be wired to the same leg. Or perhaps only one, as Atmel says the current per VCC pin is 200mA - DIP 1284 only has 1 VCC pin, while SMD has 4.
CrossRoads:
I believe if one could look at the die of the processor itself, the location of the power and Gnd connections would show power & gnd coming in to various sections of the chip, perhaps to help with distribution.
It will be for precisely that reason. You do not want all the current focussed on one area on the chip - having power and ground supplied to opposite sides of the chip obviously distributes the current more evenly and you do not want bonding wires crossing over merely to provide that.
Interesting, thank you all for your input.