Evaluation of Electrical Characteristics of Flash Type 2Χ2 Arrays Neuromorphic Transistors with Mesh Structural Floating Gate

With the rapid growth of the AI and big data industries, the importance of data processing, calculation, learning, and reasoning capabilities is emerging. Traditional von Neumann systems are limited by the bottleneck between computing unit and data storage, as well as high power consumption. As a solution to this, research on neuromorphic devices that mimic human brain neural networks is actively underway. In this study, we developed a flash-type 2Χ2 array neuromorphic transistor that expresses various synaptic weights with a single electrical measurement by applying a mesh-structured floating gate[1] [2] [3]. We also performed electrical characterization to confirm the excellent performance of repeatability, long-term plasticity, and reproducibility, which are the evaluation factors of synaptic devices. This shows the possibility of utilization in real industry.<br/><br/>In previous studies, we developed the structure of a single transistor and evaluated its electrical characteristics. The structure of a single transistor has a gate stack structure of Pt/Cr/HfO₂/Pt/Cr/HfO₂/SiO₂/Si using SOI wafer as a substrate. In particular, the Pt/Cr floating gate formed on the upper part of HfO₂/SiO₂ has a mesh structure. This is an island structure, similar to a quantum dots floating gate, that can operate successfully even if some of the mesh structure is damaged. Compared with a single floating gate, the mesh structure floating gate has better electron storage effect, higher surface density, and lower leakage current[4]. This was confirmed by electrical characterization.<br/><br/>Based on the information confirmed in previous studies, a 2Χ2 array neuromorphic transistor incorporating array technology was developed. The 2Χ2 array consists of parallel connections of single transistors. The electrical characteristics are evaluated by adjusting the applied voltage of the source, drain, and control gate regions of transistors arranged in a 2Χ2 array. The electrical characterization shows that the synaptic weights of arrayed transistors are more diverse than single transistor. It also shows good results in terms of repeatability, long-term plasticity, and reproducibility.<br/><br/>Through this study, it is possible to confirm the excellent performance of the 2Χ2array neuromorphic transistor expanded in previous studies. This means the implementation of a neural network and makes it possible to consider the possibility of expansion to a heavy arrays in the future. It also shows practicality in AI and big data industries.<br/><br/>Reference<br/>[1] S Kim et al., <i>Scientific Reports</i> volume 9, Article number: 11705 (2019)<br/>[2] H Choi et al., Electronics 2020, 9(1), 57<br/>[3] Y Park et al., IEEE Transactions on Electron Devices, vol. 66, no. 1, pp. 420-427, Jan. (2019)<br/>[4] H Chen, Y Zhou, S Han., Nano select, volume 2, Issue 7, Pages 1245-1265, July 2021