The MULTICUBE Design Flow - Multi-objective Design Space Exploration of Multiprocessor SoC Architectures - page 7

there are more than one objective to be optimised (maximized or minimized), mean-
ing that the outcome of the optimisation process is not a single solution but a set
of solutions. This set of solutions, which is called the Pareto front, represents the
best trade-off between the different (and possibly contradictory) objectives. The set
of algorithms implemented includes state-of-the-art algorithms widely used in the
field of multi-objective optimisation (ranging from evolutionary and genetic algo-
rithms up to simulated annealing and particle swarm algorithms), enhanced versions
of algorithms that were specifically targeted in the project for the multi-core SoC
design, and new developed algorithms. The multi-objective optimisation algorithms
developed in the MULTICUBE project are described in more detail in Chap. 3 of
this topic.

1.2.2 The Power/Performance Estimation Framework

The Power/Performance Estimation Framework (see Fig. 1.1 ) consists of a method-
ology and related tools that have been set up to provide accurate estimates for
complexity, timing and power consumption at different abstraction levels and for
different use cases. A set of tools has been used for the system modeling and estima-
tion of several metrics such as energy consumption and execution time of the target
MPSoC platforms among which:

￿ Multicube SCoPE: an extension of the open-source high-level SCoPE perfor-

mance and power evaluation framework [ 7 ] developed by University of Cantabria
for performing HW/SW co-simulation. Multicube SCoPE enables the definition
of SystemC platform template models to evaluate performance and power con-
sumption. Multicube SCoPE efficiency comes from the fact that performance and
power estimations of the software side are performed at the application source
code level by using back-annotation. The back-annotated software components
are then linked to the hardware components by using standard SystemC interfaces.
This modeling style is called Timing Approximate . Software back-annotation
avoids instruction set simulation therefore decreasing of several orders of mag-
nitude the simulation time and maintaining a fairly good accuracy with respect
to cycle-accurate simulation. Multicube SCoPE also provides some hooks for
enabling C/C++ software code to invoke operating system primitives compliant
with POSIX and MicroC/OS. Multicube SCoPE is described in more detail in
Chap. 2 of this topic.

￿ A proprietary set of simulation tools developed by IMEC as SystemC-based

transaction-level multi-core simulator built on top of the CoWare virtual proto-
typing environment to support platform-based design approach. The TLM-based
prototype models an ADRES multi-core [ 6 ] and has been integrated with both
modeFRONTIER and Multicube Explorer tools. The platform is composed of
a variable number of processor nodes and memory nodes. All processor nodes