Systems-on-chip: what are the limits?

Abstract

A quantitative comparison is made between the computational requirements of typical systems-on-chip and the computational capabilities of silicon. This is illustrated by the evolution of TV and other video appliances on the one hand and the progress of silicon technology on the other. As a basic benchmark figure the concept of the intrinsic computational efficiency (ICE) of silicon is introduced, and this is compared with the computational efficiency of commercial microprocessors and digital signal processors. It is shown that processors designed by application-specific architectural synthesis can approximate the ICE limit and that they exceed the computational efficiency of general-purpose devices by several orders of magnitude. To close the gap between flexibility and efficiency, the silicon system platform concept is introduced. Finally, it is shown how Moore's law of exponential growth together with Claasen's law of logarithmic usefulness make the perceived progression in systems a linear function of time.