CNNs with bistable-like non-volatile memristors: a novel mem-computing paradigm for the IoT era

This work presents a novel mem-computing paradigm for the Internet-of-Things era. The paradigm revolves around the peculiar feature of memristors to process information and store data in the same physical location in order to enhance the performance of Cellular Nonlinear Networks, especially in view of a future hardware implementation and its integration with state-of-the-art high-resolution visual sensor arrays. A bistable-like non-volatile memristor is used in each dynamic processor of the cellular network in place for the linear resistor of a standard realization. The resulting cell is then endowed with a couple of possible Dynamic Route Maps, one for each of the two memristor states. This peculiar property endows the network with new functionalities as compared to a standard nonlinear dynamic array.