My hypothesis is that the freezer motor has to be right at the freezing point and the tray has to have few nucleation points such that when a small spek of water freezes the phase change disperses enough heat to prevent them from immediately following suit, the small flake of ice then rises to the surface. As it continues to cool, further freezing is more likely to occur in the existing crystal. Through a combination of ice's buoyancy and the surface tension of liquid water, the crystal gets pushed upward compared to liquid water.

I wonder if the fridge motor's vibration plays any role on where the fingers form (due to resonance patterns).