Dynamic memristors: from device to applications

Abstract:

Memristors have been extensively studied as a promising candidate for the next generation non-volatile memory technology. More recently, memristors are also becoming extremely popular in neuromorphic applications because of its striking resemblance to biological synapses. The memristor was firstly proposed conceptually as the fourth basic electric circuit element whose resistance is dependent on the history of electrical stimulation. Physical implementations of memristors are normally based on a solid state nanoscale sandwiched metal-insulator-metal (MIM) structure, and the resistance change is achieved by controlling the filament growth inside the insulating/switching layer. Devices based on such resistive switching mechanism are also termed resistive random access memory (RRAM) devices. They offer advantages of simple structure, high density, low power, good endurance, etc.

In this dissertation, two kinds of memristor devices based on the MIM structure will be discussed first, using Ag2S and WOx as the switching material, respectively. The WOx device allows incremental modulation of the device conductance, and enables efficient hardware emulation of important biological synaptic learning functions. Neural networks based on the memristor crossbar array has been used to successfully perform image reconstruction tasks based on the sparse coding algorithm. Additionally, interesting short term decay dynamics can be observed in both Ag2S and WOx based devices. Different from the requirements of non-volatile memory which aims for long term memory storage, the volatile nature of these devices can be used to encode neural spiking information in the temporal domain. These devices are termed "dynamic memristors" and can be used in novel computing schemes such as reservoir computing for temporal information processing including speech recognition.