An energy-complexity model for VLSI computations
An energy complexity model for CSP programs to be implemented in CMOS VLSI is developed. This model predicts with some accuracy the energy dissipation of the "standard" asynchronous VLSI implementation of a CSP program, associated to a given trace of that program. This energy complexity is...
| Summary: | An energy complexity model for CSP programs to be implemented in CMOS VLSI is developed. This model predicts with some accuracy the energy dissipation of the "standard" asynchronous VLSI implementation of a CSP program, associated to a given trace of that program. This energy complexity is used in the analysis of CSP programs, in order to optimize this high level representation of asynchronous circuits for energy efficiency. A lower bound to the energy complexity of a CSP program is derived, based on the information theoretical entropy per symbol of the input/output behavior of the CSP program. This lower bound abstracts the specification of the circuit (that is, its input/output behavior), from the implementation of the specification (that is, the text of the program), and therefore applies to any program that meets the specification. A number of techniques are presented to write programs of low energy complexity, and are applied to several examples.
To link the high level representation of circuits to the CMOS representation, several circuits are analyzed to provide standard translations for basic CSP operators into CMOS. In particular, a method for pipelining bus transfers using the sense-amplifier of the bus as a register is proposed.
|
|---|