In Depth Study Of The Parameters Influencing The Dynamics Of Resistive Switching In HfO2-Based Memristors
Please login to view abstract download link
Resistive switching devices based on HfO₂ are promising for non-volatile memories and neuromorphic computing. Beyond simply switching between two resistance states, precise control of conductance over time is essential for encoding synaptic weights in neuromorphic circuits (Fig.1). This requires a detailed understanding of the dynamic evolution of the memristor’s internal variables. Here, we present an in-depth analysis of the temporal dynamics of the set and reset transitions in TiN/Ti/HfO₂/W metal-insulator-metal (MIM) structures. By recording current transients under different conditions, we establish key differences between the two switching processes. The set transition follows a monotonic increase in current with an exponential decay in growth rate, whereas the reset transition exhibits a sigmoidal response (Fig. 2). We define the Time-to-Reset (τ_r) as the moment when the current variation reaches its maximum (Fig. 3). Our experimental results demonstrate that τ_r is highly sensitive to the applied voltage, the initial resistance state (RLRS), and the temperature. Increasing the applied voltage exponentially reduces τ_r, while a higher RLRS also leads to faster reset times. We model these dependencies using a plane (Fig 4), allowing precise predictions of τ_r under different initial conditions. Additionally, we investigate the impact of Joule heating by applying constant power signals rather than voltage signals, confirming that power directly controls reset times. From these measurements, the thermal resistance of the conductive filament, which influences switching speed, can be estimated (Fig. 5). Finally, temperature-dependent measurements reveal that lower temperatures accelerate the reset process (Fig. 6), likely due to the reduced thermal energy of oxygen vacancies. These findings provide valuable insights into the physics of resistive switching, offering a compact model for optimizing device performance in high-speed memory and neuromorphic applications.