Design Space Exploration of Clustered Sparsely Connected MPSoC Platforms.

Area of Science:

• Computer Engineering
• Embedded Systems Design
• Algorithm Development

Background:

• Heterogeneous multiprocessor platforms are crucial for high-performance, low-power systems like IoT and mobile robotics.
• Limited interconnectivity in clustered platforms hinders feasible application mapping.

Purpose of the Study:

• To develop novel algorithms for design space exploration of sparsely connected heterogeneous multiprocessor platforms.
• To enhance the success rate of finding feasible application-to-platform mappings.

Main Methods:

• Modification of the Non-dominated Sorting Genetic Algorithm II (NSGA-II) and a prior SDSE mapping algorithm.
• Adjustments to infeasible solution penalization, chromosome decoding, and mapping strategies.
• Development of a synthetic benchmark for evaluating algorithms on clustered platforms.

Main Results:

• Four proposed algorithms achieved 100% success in finding feasible solutions for dedicated cluster platforms.
• For tile-like platforms, these algorithms showed an average success rate of 60%, with one reaching 84%.

Conclusions:

• The developed algorithms significantly improve the feasibility of application mapping on sparsely interconnected heterogeneous systems.
• The proposed synthetic benchmark provides a valuable resource for future research in this domain.