The effect of bending on the BCM learning rule-related performance of Au/WO<i> _x </i>: Al(9 at.%)/Ti memristive synapse

Abstract Binary oxide-based memristors, characterized by their simple ‘sandwich’ structure and compatibility with flexible manufacturing processes, have emerged as promising candidates for emulating neural synapses in wearable neuromorphic chips. For these applications, mechanical durability under bending deformation is paramount. Consequently, investigating the bending-dependent performance of memristive synapses is of substantial significance. The Bienenstock-Cooper-Munro (BCM) learning rule, recognized as the most accurate synaptic plasticity model to date, serves as the foundation for this investigation. This study systematically examines the bending effects on BCM learning rule-related performance in Au/WO x : Al(9 at.%)/Ti memristive synapses. Experimental results reveal that the devices maintain frequency-dependent plasticity and demonstrate sliding frequency threshold characteristics across various bending conditions (flat, with bending radius of 12, 10, and 7.5 mm). Notably, the frequency threshold range exhibits a progressive expansion with decreasing bending radius. This observed phenomenon can be attributed to the reduction in diffusion activation energy of oxygen vacancies within the WO x : Al resistive switching layer under appropriate deformation, which subsequently accelerates oxygen vacancy diffusion and induces a shift in threshold frequency toward higher values. These findings offer significant insights and provide substantial support for advancing memristive synapse development in wearable neuromorphic computing chips.