Back to the program.
Data transfer optimization through algorithm of memory-architecture prediction
Fabio Tonti
University of Bologna
Faculty of Electronic Engineering
fabio_tonti@hotmail.com

Abstract

Power-saving design methodology brings many advantages thanks to the less strict requirements (batteries, cooling equipment, etc.) and the added value of letting the system be "mobile". Actually, real-life and real-time widespread media application are, power and area speaking, memory-driven: the largest consumption and surface occupied is due to the memory architecture. Inside the ATOMIUM project i've developed at IMEC a fast prediction of costs related to the introduction of certain constraints on the data allocation between memories, in order to let a more accurate decision on which data must be accessed at the same time (timing contraints on the execution of the given application) and which not. The proposed/predicted allocation, given in few seconds, is many times more optimised than the one given by the Memory Allocation and Assignment step of ATOMIUM (which it should predict) thanks to a very elementary and plain functioning.

Keywords

ATOMIUM, IMEC, prediction algorithm, to save power and area