Tungsten Oxide Memristive Devices for Neuromorphic Applications.

Metal-oxide Resistive Memory;Memristor;Synaptic Emulation;Neuromorphic;Electrical Engineering;Engineering;Electrical Engineering

Chang, TingPeterson, Becky Lorenz ;

University of Michigan

关键词: Metal-oxide Resistive Memory;    Memristor;    Synaptic Emulation;    Neuromorphic;    Electrical Engineering;    Engineering;    Electrical Engineering;   

Others  :  https://deepblue.lib.umich.edu/bitstream/handle/2027.42/95956/satich_1.pdf?sequence=1&isAllowed=y

瑞士|英语

来源: The Illinois Digital Environment for Access to Learning and Scholarship

【 摘 要 】

Inspired by the superb efficiency of biological systems, solid-state neural network systems have attracted much attention for their potential to learn and function in complex environments. Memristors, with several unique properties, are exceptional candidates for emulating artificial synapses and thus for building artificial neural networks. This thesis work explores the material properties, device characteristics, synaptic plasticity implementations, and CMOS integration of tungsten oxide (WOX) nanoscale memristors, advancing this technology for neuromorphic applications.Device fabrication, electrical studies, and material analyses of WOX memristors are presented. Such devices offer simple fabrication, low-power, high-density, scalability, connectivity, and CMOS-compatibility, making them ideal candidates for neuromorphic applications. Bipolar analog resistive switching (RS) is observed as a result of oxygen vacancy (VOX) migration within the WOX film upon an applied electric field. Material characterizations suggest that the concentration and distribution of VOX are directly related to the stoichiometry and morphology of the film. The switching characteristics are formulated into a set of memristive equations and further implemented in commercial simulation softwares, enabling simulations at the circuit level.Based on the understanding of the switching dynamics of WOX memristors, important synaptic functions are demonstrated in these electronic devices. Bipolar analog RS implies that the memristor conductance (synaptic weight) can be potentiated (strengthened) and depressed (weakened) continuously by applying biases of opposite polarities. The inherent diffusion of VOX results in limited retention in WOX memristors. However, the retention can be enhanced by repetitively stimulating the memristor, known as the memory enhancement effect which mimics the transition of short-term memory to long-term memory in biological systems. Rate-dependent plasticity is studied by varying the interval between stimulations, and the results are analogous to post-tetanic potentiation and paired-pulse facilitation in neurobiology. Sliding threshold effect is also observed in WOX memristors, indicating that conductance modulation is a dynamic process depending on the entire stimulation history, as stated in the BCM rule for biological systems. Heterosynaptic plasticity is also implemented through the design of a three-terminal memristor geometry. Finally, successful integration of WOX memristors and with CMOS circuitry is carried out, facilitating the creation and characterization of hybrid memristor/CMOS neural network systems.