École Nationale Supérieure de
Techniques Avancées

Abstract

The continuous increase of capacity integration on one side, the exponential increase of embedded application complexity on the other side led to system on chip (SoC) then to multiprocessor system on chip (MPSoC). The fundamental problem associated with the large scale system on chip is the one of design methodologies and the resulting design productivity crisis preventing efficient exploitation of these circuits. This design productivity crisis is the result of ad-hoc and manual design while the problem should be established as a multiobjective optimization problem which should be solved by automated optimization techniques. In this PhD thesis, we present a design methodology for multiprocessor systems on chip implementation on programmable circuits, whose originality is based on three main points: (1) multiobjective evolutionary algorithm based exploration of the design space in order to undertake an intelligent solution search, (2) the use of programmable logic circuits for fast performance evaluation through emulation which is far superior to simulation and finally (3) the use of MPSoC synthesis from a high level parallel programming language (Occam) and a monitoring on chip framework. Case studies on circuit’s implementations prove the efficiency of such a design methodology to solve the problem of the design productivity crisis.