Feedback control modelling for learning reconfigurable embedded systems

We observe that upcoming strongly recongurable embedded systems will be soon available for complex multi-application systems. Besides, the QoS/Power/Real-time trade-os cannot be totally handled at design time due to the growing uncertainty conditions. This paper focuses on the auto-configuration decision modelling and proposes an original and generic method based on control theory including stability analysis. Our approach addresses the question of local vs global reconguration decision at hardware, software and RTOS levels. We tackle the question of uncertainty with a learning system approach and the tradeoff between accuracy and complexity with run-time light estimators based on signal processing theory. A simulator applied to 3D synthesis is presented as a first proof of concept.