An important goal for computer science is to find practical, scalable models of computation that are as efficient as is permitted by the laws of physics. Given a constant upper bound on entropy density, physics implies fundamental constraints on the efficiency of any computation that produces entropy. As a result, it appears that the most efficient possible computer architectures must use reversible primitive operations which produce arbitrarily little entropy. In this paper we show that a 3-D mesh of reversible processing elements, constructible using existing technology, scales better than any physically possible computer based on irreversible primitives.
展开▼