Sustainable Development Goals
Abstract/Objectives
This project plans to prepare metal halide resistive random access memory (RRAM) devices using a vacuum deposition process, which contain cesium bromide, cesium iodide, and yttrium trifluoride. These lead-free metal halide materials, when combined with silver electrodes, may form resistive switching memories dominated by either silver conducting filaments or halide vacancies. This study aims to explore the versatility of metal halide materials for RRAM applications. Additionally, the performance of these devices will be evaluated using different metrics such as the on/off ratio, retention time, and endurance cycling. Finally, the team will investigate the multi-level storage capability of these RRAM devices and apply the results to neural network computing.
Results/Contributions

The resistive switching behavior of cesium bromide and cesium iodide RRAM devices effectively suppresses current in the off state due to their inherent insulating properties. The addition of a thin layer of molybdenum trioxide improves the stability and electrical properties of the devices. The switching mechanism in these devices is dominated by silver conducting filaments, with a very low switching voltage (<0.18 V), extremely long retention time (>106 s), excellent endurance cycling (>105 cycles), and fast switching speed (<200 ns). The team also tested the multi-level storage capability of these devices using different compliance currents and applied them to MNIST handwritten recognition network testing, achieving a recognition rate of over 90%.

In contrast, the resistive switching mechanism in yttrium trifluoride RRAM devices is dominated by fluoride ion vacancies due to its amorphous structure, which prevents silver ions from effectively passing through. This device has two operating modes and can serve as both analog and digital memory. It also has a long retention time (>105 s) and stable endurance cycling. Notably, the retention time performance of the yttrium trifluoride RRAM devices is even better (>106 s) when used for multi-level storage, achieving a high on/off ratio of 7.2 and a very high recognition rate of 97% in MNIST handwritten recognition network testing.

Lead-free metal halide RRAM devices may have different operating mechanisms due to their crystal phase, but they all exhibit excellent performance, indicating their potential value in memory applications and the possibility of breaking through the von-Neumann bottleneck.

Keywords
Resistive random access memoryLead-free metal halide materialsMulti-level storageNeural network computing
Contact Information
林皓武教授
hwlin@mx.nthu.edu.tw