Study on NbOx-based non-linear and self-rectifying analog charge trap memristor for a high density crossbar application

Abstract

We developed a Pt/TaO$_x$/NbO$_x$/AlO$_x$/Ti memristor that shows both the ~10$^5$ of high self-rectifying ratio and ~10$^3$ of non-linear characteristics at below 1uA of programming current, which are desired for the high density and low power memristive crossbar applications. The device showed a continuous and uniform analog switching behavior, which makes it suitable for multi-bit memory or neuromorphic applications. Its retention and endurance characteristics are exceeding 2ⅹ10$^5$ seconds at 150℃ and >10$^5$ cycles, respectively. We propose an in-depth understanding on the switching mechanism based on the experiments and physical analysis. In addition, we demonstrate the optimized array-level methodology to maximize the characteristic of the memristor in the neuromorphic computing.