Reconfigurable, Non‐Volatile Switching in WO ₃ Film for Resistive Memory and Multistate Programming Toward Energy‐Efficient Neuromorphic Computing Applications

ABSTRACT Non‐volatile memristive device compatible for futuristic memory, data storage and in‐memory computing with good exceptional energy efficiency will be an integral part of neuromorphic architecture. Tungsten oxide () is a versatile metal oxide displaying memristive characteristics where resistance states can be controlled through oxygen vacancy concentration holds great potential for such low energy neuromorphic devices. Here, we report a WO 3 based resistive switching memory device showing exceptional stability in switching with respect to number of dc‐switching cycles (12 10 cycles) and retention for more than 5000 sec at +0.2 V. Based on voltage stress measurements the device offers low voltage switching operation (+0.72 SET V, –0.12 RESET V), a large ON/OFF ratio (10 3 ), low energy consumption (2.1 per event, and dynamic range of 7. Additionally, primary synaptic characteristics such as paired pulse felicitation (PPF) and paired pulse depression (PPD) are shown which indicates that the based devices are suitable to neuromorphic applications. Interestingly, a transition between short term memory (STM) to long term memory (LTM) is seen as a function of stimulation duration. The learning and forgetting curves show very good linearity with image recognition capability using the MNIST data set. Recognition accuracy of 88% is achieved with respect to ideal device. This work demonstrates the effective use of WO 3 based memristive device for low energy consuming neuromorphic computing applications.