Novel highly nonlinear memristive circuit elements for neural networks

Highly nonlinear dynamics arises from novel serially and antiserially connected memristive switches. Antiserially connected memristors offer a device with built-in spiking neuron behavior. Serially connected memristors offer new functionality of a cascade of sigmoid staircase curves to represent multi-level spike-timing dependent plasticity (STDP). In a programming operation, a serial arrangement of two switches displays multiple time-delays between threshold transitions showing three distinct current levels and an arrangement of three switches four current levels spanning 6 orders of magnitude. Anti-serially connected resistive switches, aka resistive floating electrode device (RFED), can generate well-controlled spikes as a result of the history of the dynamic input. Both types of composite switches can be packed into a single intersection of a memristive crossbar architecture of 4F². The switches have been manufactured as a multi-stack of Cu, TaO_x, Pt materials with 32 nm of oxygen-deficient TaO_x in a crossbar architecture.