Affiliation:
1. Department of AI Semiconductor Engineering Korea University Sejong 30019 Republic of Korea
2. Division of Electronics and Electrical Engineering Dongguk University Seoul 04620 Republic of Korea
3. Department of Electrical and Computer Engineering and Inter‐university Semiconductor Research Center (ISRC) Seoul National University Seoul 08826 Republic of Korea
Abstract
AbstractUtilizing memristors to increase the density of crossbar arrays requires reducing dependency on transistors. This paper presents an approach where the current limiting function is integrated within the memristor by inducing an AlOx/TaOx layer, thereby limiting overshoot current during filament formation. The reaction between TaOx and Al can be accelerated through annealing, which optimizes the on/off ratio and reduces device‐to‐device variation. Additionally, AlN is inserted to inhibit the movement of oxygen ions to the bottom electrode, improving conductive filament reoxidation. Furthermore, biological synaptic properties are examined using electrical pulse schemes, revealing multibit characteristics of >5‐bit. After the structure optimization, 24 × 24 crossbar arrays are fabricated, allowing 100% of cells to achieve self‐compliance filament formation without hard breakdown. Moreover, the crossbar array demonstrates an on/off ratio of over 4 × 102. Additionally, a multibit‐encoded neuromorphic system is proposed based on the device's multibit capability. The number of synapses can be significantly reduced by grouping input data into a single memristor device. When comparing classification accuracies, 97.14% and 95.54% are observed without and with encoding. The improvements in device structure and encoding method presented in this study enable highly integrated crossbar arrays and efficient neuromorphic systems.
Funder
National Research Foundation of Korea
Ministry of Science and ICT, South Korea