Enlarging the frequency threshold range of Bienenstock–Cooper–Munro rules in WOx-based memristive synapses by Al doping

Abstract

Memristors are considered to be one of the optimal choices for simulating synapses and fabricating artificial neuromorphic chips. The Bienenstock–Cooper–Munro (BCM) rules represent one of the most accurate models of synaptic plasticity. Implementing the spike rate dependent plasticity and sliding frequency threshold feature in artificial synapses is crucial for realizing the BCM rules. Specifically, enlarging the tunable range of the sliding frequency threshold, which refers to the frequency threshold range of artificial synapses through the modification of the device itself, is conducive to advancing the development of high performance artificial synaptic devices. In this study, the spike rate dependent plasticity and sliding frequency threshold feature were realized in WO_x-based memristive synapses doped with Al with concentration ranging from 0 to 15 at%. Moreover, the frequency threshold range can be adjusted by varying the Al doping concentration. Specifically, when the Al doping concentration is approximately 9 at%, the device exhibits a broader frequency threshold range of 23 kHz to 103 kHz. The test results for the relaxation time constant and the density of oxygen vacancies suggest that Al:WO_x (9 at%) exhibits a high diffusion rate and a high density of oxygen vacancies. These characteristics are likely to result in an increased frequency threshold and may be the primary reason for the large tunable range of the frequency threshold. This study offers a method to regulate the frequency threshold range by modifying the device itself, and provides support for the development of neuromorphic computing systems.