For last few years, evolution of SoC and increasing density of integration has proposed to target the constraints of performance, consumption and design cost/time by inter-connecting a growing number of IP blocks. This complexity added to the increasing heterogeneity of these architectures leads us today to imagine new architectural and computation models more autonomous and easier to design.
The SATURN project (Self-Adaptive Technologies for Upgraded Reconfigurable Neural computing) takes place in this context of autonomous embedded systems and proposes to revise the actual fundamentals of embedded computing architectures. We propose in this project an original approach compared to the classical centralised approaches based on new computation paradigms. The SATURN project associates researchers from digital embedded systems and artificial intelligence in order to define this architecture of a new type for smart embedded platforms. This architecture will be based on a totally distributed control (decentralized), thus more flexible and more robust to unpredicted changes in the system, and its environment, dynamics. Moreover, the system will be doted of self-organisation capabilities allowing the dynamic allocation of computation, memory and communication resources. In order to reach these goals, the project will initiate new research activities and studies on bio-inspired mechanisms applied to digital integrated circuits.
The aim of our current work is to design an intelligent embedded robot controller that will be able to self-organize its elements in order to adapt its architecture to the robot behavior. Inspired from the feature integration theory , the robot will use three saliency maps. These three maps provide the robot with a sensorimotor cognitive capability in order to react and to adapt its behavior to the environment. The controller will be able to adapt the size of the different maps according to the states of its actuators and to the saliency of the information from the external environment. As a result of the Kohonen auto-organizing map model , the three saliency maps are constantly competing for the resources of the controller.
 A. Treisman, “A feature-integration theory of attention,” Cognitive Psychology, vol. 12, no. 1, pp. 97–136, Jan. 1980. [Online]. Available: http://dx.doi.org/10.1016/0010-0285(80)90005-5
 T. Kohonen, Self-Organization and Associative Memory. Springer-Verlag, 1989.