A critical analysis of the quantum point contact model of conduction in Ta2O5-based resistive switching memories

• Autor(a) principal: Andrade, Jaime Manuel Maia
• Data de Publicação: 2018
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/10773/26119
• topic: Resistive switching; Ta2O5; Thin oxide films; Memristive systems; Quantum point contact; Conduction mechanisms

Redox-based resistive random access memories (ReRAMs) are promising candidates to implement a new class of memories, called storage class memories (SCMs). These are meant to achieve small latency times, at an affordable price, fitting in between Flash memories and dynamic RAMs (DRAMs). ReRAMs are also being applied in the neural network field of research, given their ability to emulate synaptic plasticity.Thus, there is a growing interest in studying this class of devices, known as memristive, or resistive switching, devices. This work focuses on the conduction mechanisms proposed to model the electrical current in RS devices. One in particular, called quantum point contact (QPC), was studied in depth. With this intent, a Pt/Ta/Ta2O5/Pt memrisive device was studied, and current-voltage (−) curves for both resistance states obtained. This was repeated for various values of applied current compliance. A mathematical method for applying the QPC model was then developed, involving the Gauss-Newton and Levenberg-Marquardt algorithms. The latter was used with a heuristic approach to the regularization weight, to facilitate its application en masse. The convergence rates, influence of starting parameters and goodness of fit were all measured and accounted for in developing this approach. Two approximations of this model were considered. In the first, only the first subband in the conducting channel contributes to the conduction. In the second, the barrier height is fixed, in addition to the first approximation. The original model was found to be hard to apply: the starting parameters had a large influence on the fitting results, and the algorithm was not robust. The first approximation was able to provide good fits to the data, and to do so better than the other conduction mechanisms considered. However, its physical basis was criticized, and certain considerations in interpreting the results must be taken. The second approximation was argued against. It was able to provide adequate fits to the experimental data, but the parameters’ evolution failed to match the model’s predictions.