Proposal of Stateful Relilability Counter in Small-World Cellular Neural Networks

Cellular neural networks (CNN) is a neural network model linked to only neighborhoods. CNN is suited for image processing such as noise reduction and edge detection. Small world cellular neural networks (SWCNN) is a CNN extended by adding a small world link, which is global short-cut. SWCNN has better performance than CNN. One of weak points of SWCNN is fault tolerance. We proposed multiple SWCNN layers in order to improve fault tolerance of SWCNN. However, it is not sufficient because only stop failure is considered. In this paper, we propose stateful reliability counter for triple modular redundancy (stateful RC-TMR) method in order to improve tolerance.