Memristive devices for neuromorphic computing applications

Abstract

The performance of digital computers has begun to saturate due to material, size, and power limitations. In addition to solving these dilemmas, new computing paradigms are being investigated. This thesis explores neuromorphic computing as a possible new computational paradigm, specifically an all hardware approach based on biological neural processing. This thesis introduces neuromorphic computing, neurobiology, memristive devices for neuromorphic computing, and the memristive material lithium niobite (LiNbO2). It then discusses the synthesis of lithium niobite by room temperature sputtering as well as some basic physical, optical, chemical, and electrical properties and explores more complex properties of lithium niobite including the effects of high energy radiation on lithium niobite devices and the effects of light interacting with the mobile ions in lithium niobite. The thesis discusses three devices useful for mimicking sub-structures within a biological neuron. These devices are two-terminal lithium niobite memristors, lithium niobite based batteries, and a new memdiode based on Nb2O5. The thesis concludes by discussing device models for lithium niobite memristors and their application in several neuromorphic circuits to add biologically realistic behavior without increasing the complexity of the circuit.