Performance Limits on Biochemical Computation

The Problem: Recent work addresses the possibility of designing biochemical systems that compute, using concen-trations of distinct chemical species as logic signals. Some researchers claim the performance of such biochemical computers will eventually rival or exceed electronic computers for some tasks. We critically evaluate this claim from a theoretical perspective, by deriving bounds on the performance of biochemical computers and comparing these bounds to the performance of current electronic computers. Motivation: Current models of chemical computation differ from the electronic model: signals are not carried on wires, but are co-distributed in solution and diffuse freely. In fact signal molecule diffusion is an integral part of chemical computer action. In electronic computation, the sequesterment of signals to wires allows distinction of one signal from another. In chemical computation, since signals are spatially mixed, they must be distinguished by some other mechanism. Typ-ically a different chemical species is chosen for each signal, and the differences in chemical activity distinguish the signals. The physics of this chemical distinction mechanism, combined with some reasonable assumptions about the classes of molecules being used as signals, some information theoretic tools, and simple diffusion kinetics, can give us funda-mental bounds on the (complexity, speed, error) envelope within which a chemical computer must operate