Implementation of reservoir computing using volatile WOx-based memristor

In this study, we investigate a Ni/WOx/ITO-glass memristor device to verify short-term memory characteristics for reservoir computing systems. We verify the chemical and material compositions of each layer using transmission electron microscopy (TEM) image and X-ray photoelectron spectroscopy (XPS). The device has a characteristic that the current decreases with time, but shows a reverse current decay phenomenon. In addition, potentiation and depression data are obtained through modulated pulses and measurement methods. Based on this result, meaningful pattern recognition accuracy is obtained. Also, it is proved that the gradual conductance modulation can be controlled through pulse amplitude and time interval between the pulses. Finally, reservoir computing is realized based on short-term characteristics of the device. All 16 states of 4 bits have been implemented, and it is proved that the changed state can be classified using a simple learning algorithm after reading it with pulses. We also propose to make the system to consume low power.

Automated summary

In this study, a Ni/WOx/ITO-glass memristor device was used to investigate its short-term memory characteristics for reservoir computing systems. The chemical and material compositions of each layer were verified using transmission electron microscopy (TEM) image and X-ray photoelectron spectroscopy (XPS). The device exhibited a characteristic where the current decreased with time but showed a reverse current decay phenomenon. Potentiation and depression data were obtained through modulated pulses and measurement methods, leading to meaningful pattern recognition accuracy. The study also demonstrated that gradual conductance modulation could be controlled through pulse amplitude and time interval between the pulses. Reservoir computing was successfully realized based on the short-term characteristics of the device, with all 16 states of 4 bits being implemented and classified using a simple learning algorithm after reading them with pulses. The study also proposed a method to reduce power consumption in the system.