LAUSR

Abstract This paper presents the design and implementation of a low power and ultrafast spike-timing dependent plasticity (STDP) of the spiking neural network (SNN) in a crossbar structure based on the ferroelectric tunnel memristor (FTM) where the nonlinear switching behaviors are mathematically described by a physics-based compact model. The FTM is used as 1R structure without a cell selector to implement the memristive synaptic weight multiplications. The implementation details for the fully asynchronous communication/learning of the FTM as an artificial synapse are demonstrated. The FTM dynamics and the designed leaky-integrate-and-fire neurons are interconnected allowing a scalable STDP learning. FTM changes its resistance in crossbar structure depending on the spike timing. The achieved results of the proposed analog implementation prove that the application of spikes as neural actions manipulates the STDP, which follows a multiplicative type of learning rule. A comparison of STDP results based on FTM device used in 1R and 1T1R (i.e. with a select transistor) structures in a crossbar network is demonstrated. This system presents a good scaling behavior, a low temporal processing delay along with an ultrafast plasticity at a low energy processing cost (∼0.125 pJ). This gives new opportunities for building compact, ultrafast and power-efficient neuromorphic computing systems.