Engineering memristor : control over the fourth fundamental element for memory application and beyond
Date of Issue2017-01-04
School of Electrical and Electronic Engineering
In 1971 Prof. Leon Chua predicted that on top of the three known fundamental passive elements, the capacitor, the resistor, and the inductor that were discovered in 1745, 1827, and 1831 respectively, there should be a fourth fundamental two-terminals non-volatile device, which he called memristor after the name of memory-resistor. However, it was only a theory for nearly thirty years until the first paper published by HP Labs by the Stanley Williams group in Nature on May 1st, 2008. This announced the first materialization of the memristor, the fourth fundamental element device. Since then, memristor work has gained a tremendous increase in interest due to its unique idiosyncrasy and its special analog properties. This is especially shown in the recent neuromorphic computing since both neuron and memristor operate in ions. In many ways, a memristor is very similar to a synapse (a biological connection between two neurons), which is able to modulate and tune the efficiency of signal transmission between neurons with a high level of plasticity. Memristor based neuron chips can be astonishingly better than conventional CMOS digital approaches in pattern recognition in high speed, low power and parallel processing. The eventual goal is that computer chips can think like human brains too. Since the announcement of HP’s groundbreaking news, numerous papers aimed to analyze the special attributes of the memristor have been published. Moreover, various publications have been submitted for memristor fabrication, modeling, and applications. However, until today many fundamental questions, including an accurate understanding of the switching mechanism, availability for high quality selector devices, and implementation the fundamental element to the novel circuit design etc., are still in question. This work will try to study the fundamental memristor devices including material impact, device characterization, physical modeling, and selector device choice, then apply its unique property to novel applications using various process methods. The ultimate goal is to understand the memristor device deeply and gain control over this fourth fundamental device for memory applications and beyond. In the detailed work depicted here, a comprehensive material dependence of the memristor devices has been conducted and optimized through various physical analysis and electrical testing. Subsequently, a physical compact model for TaOx systems was developed and calibrated with electrical tests. This was the first two-state-variable device model with high accuracy for practical circuit design and simulation applications. A series of selector work aimed to minimize the sneak path issue was also investigated for the large scale crossbar application. Finally, the selected applications from this study will be introduced from the research work on hamming distance computation, vector production, and neuromorphic computing.