Does Q = MC/sup 2/? (On the relationship between Quality in electronic design and the Model of Colloidal Computing)

Abstract

This paper introduces colloidal computing as an alternative to the classical view on computing systems in terms of design feasibility, application adaptability and better energy-performance trade-offs. In colloidal computing, simple per computational particles are dispersed into a communication medium which is inexpensive, (perhaps) unreliable, yet sufficiently fast. This type of clustering into computationally intensive kernels with loose inter-particle communication, but tight intra-particle communication is typical not only for the underlying hardware, but also for the actual application which runs on it. We believe that the colloidal model is appropriate to describe the next generation of embedded systems. For these systems, a significantly better design quality can be obtained via run-time trade-offs and application-driven adaptability, as opposed to classical systems where optimizations are sought in a rather static manner.