Michel estimates that 3D packaging could, in principle, reduce computer volume by a factor of 1,000, and power consumption by a factor of 100, compared to current 2D architectures. But the introduction of brain-like, 'bionic' packaging structures, he says, could cut power needs by another factor of 30 or so, and volumes by another factor of 1,000. The heat output would also drop: 1-petaflop computers, which are now large enough to occupy a small warehouse, could be shrunk to a volume of 10 liters.
If computer engineers aspire to the awesome heights of zetaflop computing (1021 flops), a brain-like structure will be necessary: with today's architectures, such a device would be larger than Mount Everest and consume more power than the current total global demand. Only with a method such as bionic packaging does zetaflop computing seem remotely feasible. Michel and his colleagues believe that such innovations should enable computers to reach the efficiency — if not necessarily the capability — of the human brain by around 2060. That is something to think about.