Synaptic Properties and Short‐Term Memory Dynamics of TiO<sub>2</sub>/WO<sub>x</sub> Heterojunction Memristor for Reservoir Computing
Hyojin So, Jungwoo Lee, Chandreswar Mahata, Sangwan Kim, Sungjun Kim
Abstract
Abstract A hard breakdown phenomenon occurs in the TiN/WO X /Pt device owing to the metallic nature of the WO X layer deposited by pulsed direct current (DC) sputtering. In particular, analog resistive switching (RS) is achieved as the defect states of the naturally occurring TiON layer (oxygen vacancies region) between TiN and TiO 2 fluctuate based on the polarity of the bias. Interestingly, the TiN/TiO 2 /WO X /Pt device displays gradual, bipolar, SET and RESET operations during DC voltage sweep cycling without requiring an electroforming process. Excellent linearity in potentiation and depression is demonstrated via identical pulse trains based on the analog RS behavior. Additionally, the neuromorphic system simulation achieved a pattern‐recognition accuracy of over 95% when conductance is employed as the weight in the neural network. Furthermore, essential synaptic functions, such as spike‐rate‐dependent plasticity (SRDP), spike‐number‐dependent plasticity (SNDP), the transition from short‐term plasticity to long‐term plasticity, “learning‐experience” behaviors, and paired‐pulse facilitation (PPF), are demonstrated to emulate biological synapses for neuromorphic computing applications. Lastly, a reservoir computing system (RC) is implemented using the short‐term memory effect of the TiN/TiO 2 /WO X /Pt device. Specifically, it is deployed to differentiate all 16 (4‐bit) states using various pulse trains, and a simple algorithm is suggested to implement a low‐power consumption system.