Fault Tolerant Small-World Cellular Neural Networks for Intermitted Faults

A Cellular Neural Network (CNN) is a neural network model linked only to neighborhoods and which is suitable for image processing, such as noise reduction and edge detection. A Small World Cellular Neural Network (SWCNN) is an extended CNN to which has been added a small world link, which is a global short-cut. The SWCNN has better performance than the CNN. One of the weaknesses of the SWCNN has low fault tolerance. If the the neuron is failed, the SWCNN shows lower fault tolerance than the CNN. Previously, we proposed TMR and Reliability Counter (RC) for fault tolerance the SWCNN. In this paper, we propose the Stateful Reliability Counter (Stateful RC) method to improve tolerance. The Stateful RC has a failure state of the last histrory. The Stateful RC for TMR has higher fault tolerant than TMR and RC in the low repaire rate.