Dynamic memristor for bioinspired sensory memory

Abstract

Heading toward a future society reliant on large-scale data processing, existing computing architectures face limitations due to the disparity in processing speeds between memory and the central processing unit (CPU), commonly referred to as the von Neumann bottleneck. Consequently, researchers have explored alternative computing architectures, with neuromorphic computing emerging as a promising, power-efficient solution. However, to further enhance data processing efficiency and expand memory capacity, additional studies are imperative. Sensory memory, a type of memory that briefly retains sensory information following the cessation of the original stimulus, plays a crucial role. Serving as a temporary buffer, it stores sensory impressions for a fraction of a millisecond to facilitate subsequent processing. Due to its temporary nature, sensory memory exhibits significant capacity while also enabling rapid synaptic plasticity. To leverage this efficiency, the TiN/WOx/FTO memristor demonstrated an average HRS/LRS memory window of 14.51 during a 100-cycle endurance test, without significant resistance fluctuations. Throughout consistent resistive switching, a memory window greater than 10 was consistently maintained with minimal fluctuations in resistance states, as indicated by the coefficient of variation values, thereby demonstrating the uniformity of the TiN/WOx/FTO memristor under continuous external bias application. Furthermore, cell-to-cell uniformity was confirmed by measuring 12 random cells over 30 cycles. Moreover, we employed our TiN/WOx/FTO device, known for its rapid facilitation, to investigate various synaptic behaviors. Our study examined the impact of spike arrival on synaptic weight changes under different pulse conditions, including amplitude, width, and interval. Additionally, we explored high-density memory adaptation through a 14 x 14 synapse image. Finally, we employed a four-state reservoir computing pattern recognition system to assess the computing capabilities and multifunctional behaviors of our memristor. In conclusion, our findings suggest that the TiN/WOx/FTO memristor holds promise as a volatile memory device capable of adapting sensory memory for use in neuromorphic computing applications