Abstract: Modern digital systems consist of a complex mix of computational
resources, e.g. microprocessors, memory elements and reconfigurable logic.
System partitioning – the division of application tasks onto the system
resources – plays an important role for the optimization of the latency, area,
power and other performance metrics. This paper presents a novel approach for
this problem based on the Ant Colony Optimization, in which a collection of
agents cooperate using distributed and local heuristic information to
effectively explore the search space. The proposed model can be flexibly
extended to fit different design requirements. Experiments show that our
algorithm provides robust results that are qualitatively close to the optimal
with minor computational cost. Compared with the popularly used simulated
annealing approach, the proposed algorithm gives better solutions with
substantial reduction on execution time for large problem instances. Moreover,
a hybrid approach that combines our algorithm and SA achieves even better
results with great runtime reduction.
Keywords: System partitioning, ant colony optimization, hardware/software codesign, evolutionary computing, CAD
