How to develop a self-aware system from modularized design components is a major challenge on this theme. The authors introduce a hierarchical agent-based system architecture, which enables self-organization and self-adaptation upon parallel embedded systems. Self-organization is achieved by dynamic clusterization, which groups self-aware components into a cluster and continuously updates the organization to account for application changes and internal errors. Self-adaptation is performed by hierarchical agents, based on the run-time organization, to monitor corresponding levels of components and reconfigure the system to improve 

energy efficiency and dependability. The system architecture achieves functional scalability via partitioning of agent intelligence, and physical scalability via level-specific software/hardware co-design of agents.