Elucidation and Optimization of Resistive Random Access Memory Switching Behavior for Advanced Computing Application

RRAM has recently emerged as a strong candidate for non-volatile memory (NVM). Beyond memory applications, RRAM holds promise for use in performing logic functions, mimicking neuromorphic activities, enabling multi-level switching, and as one of the key elements of hardware based encryption or signal processing systems. It has been shown previously that RRAM resistance levels can be changed by adjusting compliance current or voltage level. This characteristic makes RRAM suitable for use in setting the synaptic weight in neuromorphic computing circuits. RRAM is also considered as a key element in hardware encryption systems, to produce unique and reproducible signals. However, a key challenge to implement RRAM in these applications is significant cycle to cycle performance variability. We sought to develop RRAM that can be tuned to different resistance levels gradually, with high reliability, and low variability. To achieve this goal, we focused on elucidating the conduction mechanisms underlying the resistive switching behavior for these devices. Electrical conduction mechanisms were determined by curve fitting I-V data using different current conduction equations. Temperature studies were also performed to corroborate these data. It was found that Schottky barrier height and width modulation was one of the key parameters that could be tuned to achieve different resistance levels, and for switching resistance states, primarily via oxygen vacancy movement. Oxygen exchange layers with different electronegativity were placed between top electrode and the oxide layer of TaOx devices to determine the effect of oxygen vacancy concentrations and gradients in these devices. It was found that devices with OELs with lower electronegativity tend to yield greater separation in the OFF vs. ON state resistance levels. As an extension of this work, TaOx based RRAM with Hf as the OEL was fabricated and could be tuned to different resistance level using pulse width and height modulation, yielding excellent uniformity and reliability. These findings improve our understanding of conduction within TaO x-based RRAM devices, providing a physical basis for switching in these devices. The value of this work lies in the demonstration of devices with excellent performance and demonstrated devices constitute a significant step toward real-world applications.