Towards memristive crossbar-based neuromorphic HW accelerators for signal processing

Research progress in neuromorphic hardware, capable of biological perception and cognitive information processing, is leading the way towards a revolution in computing technology. Current research efforts have focused mainly on resistive switching devices, the electronic analog of synapses in artificial neural networks (ANNs), and the crossbar nanoarchitecture, for its huge connectivity and maximum integration density. In this context, this work presents the design and simulation of a memristive crossbar-based ANN for text recognition tasks, implementing a novel computing algorithm. In such case study, important issues during the application mapping process are identified and properly addressed at device and circuit level. The computing capabilities of the proposed system are highlighted through SPICE-level circuit simulations, which show excellent agreement with theoretical simulation results.