Tuning the Relaxation Time of Diffusive Memristors for Neuromorphic Applications

A diffusive memristor is a volatile resistance switch with a metal/oxide/metal structure. It goes to a low resistance state when an electrical stimulus is applied and automatically relaxes back to its original high resistance state if the stimulus is removed. The switching behavior is attributed to the formation and rupture of a localized conducting channel within the oxide layer. Because of their structural similarity with biological ion channels, diffusive memristors have been successfully used to emulate typical synaptic and neuronal behavior such as spike-timing-dependent plasticity, leaky integrate and fire, etc. However, a remaining issue of the diffusive memristor is the non-uniform and non-controllable relaxation time (the time it takes to go from low to high resistance states), which limits the wide-range adoption of such devices in large arrays for real-world problems. In this work, we designed and fabricated diffusive memristors with uniform and tunable relaxation times. We adopted a double oxide switch layer that led to an over ten-fold improvement in the cycle-to-cycle uniformity. By connecting the device to different resistors, we tuned the relaxation time up to three orders of magnitude. This controllable and uniform relaxation process was utilized to generate the time surfaces in the hierarchy of time surface (HOTS) algorithm for pattern recognition.