Realization of cellular neural networks from a neuron component library

The cellular neural network (CNN) architecture utilizes some features of fully connected analog neural networks, along with the nearest neighbor interactions found in cellular automata. In our research, an already built neuronal sigmoid activation function was used to realize the basic component of a CNN, a "cell". Our eventual goal is to facilitate rapid prototyping, a design technique that enables the use of the basic building block of CNN, a "cell" available in the standard library, for the hardware realization of CNNs. Our experience shows that the application-specific needs of the basic analog cell must be taken into account from the beginning of the design process in order to avoid extensive redesign. This is important to remember as analog designers attempt to utilize techniques such as hierarchical design and hardware description languages (HDLs).