Advances in Engineering: Reinforcement Guided Genetic Algorithm for MPSoC Scheduling

Strategic Insights -

Researchers from Southern Technical University, in collaboration with Mahmood A. Al-Shareeda and Ahmad Taher Azar, have developed a novel hybrid application mapping framework that integrates Genetic Algorithm (GA) with Reinforcement Learning (RL) to optimize task allocation in 2D Network-on-Chip based Multiprocessor System-on-Chip (NoC-based MPSoC) architectures. This innovative approach, called Reinforcement Guided Genetic Algorithm (GA+RL), demonstrates significant improvements in runtime efficiency and communication costs.

Key Takeaways:

  • The GA+RL framework integrates Genetic Algorithm with Reinforcement Learning to optimize task allocation in NoC-based MPSoC architectures.
  • Experimental results demonstrate that GA+RL outperforms the baseline GA, achieving a minimum communication cost of 116.354 in the TGFF-G5 benchmark (80 cores), representing over 52% improvement.
  • GA+RL shows lower standard deviations and earlier convergence generations compared to the original GA.
  • The framework incorporates Explainable AI (XAI) techniques using SHapley Additive exPlanations (SHAP) to analyze feature contributions and improve interpretability.
  • The GA+RL model demonstrates greater transparency and consistency, aiding design-time decisions.
  • This research has direct implications for industrial platforms, including Kalray MPPA-256, Intel SCC, Adapteva Epiphany, and Tilera TILE-Gx.
  • The framework supports real-time scheduling in agriculture IoT systems, enabling energy-efficient deployment of smart technologies to enhance food production and sustainability.

Statistics:

  • 52% improvement in communication cost achieved by GA+RL compared to the original GA in the TGFF-G5 benchmark (80 cores).
  • 116.354 minimum communication cost achieved by GA+RL in the TGFF-G5 benchmark (80 cores).
  • 244.645 communication cost achieved by the original GA in the TGFF-G5 benchmark (80 cores).
  • 80 cores in the TGFF-G5 benchmark used for evaluation.
  • 12x11 NoC mesh size used for evaluation.
  • 3x3 NoC mesh size used for evaluation.

Sources:

  • Reinforcement Guided Genetic Algorithm for Application Mapping in Network-on-Chip Architectures: Toward Transparent and Efficient MPSoC Scheduling. IEEE Access, 2025, 13():177520-177536. (IEEE Access - http://ieeexplore.ieee.org/servlet/opac?punumber=6287639)
  • Southern Technical University, Basra, Iraq, Asia, Engineering, Cybersecurity, Genetic Algorithms.