Grainy Cr x O y RRAM with Thermal Stability for Enabling Emerging Computing Applications

With the increasing demand for high-performance data processing, traditional silicon-based integrated circuits and emerging memory technologies are in urgent need as device miniaturization approached physical scaling limits. Resistive random-access memory (RRAM) is a promising candidate for neuromorphic computing and nonvolatile memory applications for its high scalability and low cost. With the rapid growth of AI workloads, ensuring device-level thermal robustness in non-volatile memory technologies has become increasingly critical. In this work, Cr x O y -based RRAM devices were fabricated by localized prepatterned metal sources by chemical vapor deposition at 650 °C, as redistributed localized oxide deposition. The grainy oxide RRAM devices exhibit reliable bipolar switching with high uniformity and an ON/OFF ratio exceeding 10 4, while operated under 25 to 70 °C with an increased memory window attributed to the grainy structures with a thermal treatment process, which stabilized the filamentary-based resistive switching under high temperature. Linear conductance modulation with long-term potentiation (LTP), long-term depression (LTD), and training accuracy on the MNIST data set has resulted in 92.4% of pattern recognition accuracy. This study provides a unique technique in fabricating the active switching layer as Cr x O y with improved operating temperature robustness, good memory window, and analog behaviors for future in-memory computing applications.