Memristor-based neural networks: Synaptic versus neuronalstochasticity

In neuromorphic circuits, stochasticity in the cortex can be mapped into the synaptic orneuronal components. The hardware emulation of these stochasticneural networksare currently being extensively studied using resistive memories or memristors. The ionicprocess involved in the underlying switching behavior of the memristive elements isconsidered as the main source of stochasticity of its operation. Building on its inherentvariability, the memristor is incorporated into abstract models of stochasticneurons andsynapses. Twoapproaches of stochasticneural networksare investigated. Aside from the size and area perspective, the impact on the systemperformance, in terms of accuracy, recognition rates, and learning, among these twoapproaches and where the memristor would fall into place are the main comparison points tobe considered.