Implementation of BCM Learning Rule Based on Room Temperature Derived α-IGZO Synaptic Transistors
Min Zhu, Gang He, Can Fu, Qingqing Hu, Zhengquan Chen, Wenhao Wang, Shanshan Jiang
Abstract
In recent years, the implementation of neuromorphic computation on devices has attracted much attention. Bienenstock–Cooper–Munro (BCM) learning rule is considered as the most important synaptic model in biology and is more in line with the working principle of neuromorphic computational systems; therefore, it is of great significance to simulate BCM behavior on solid-state neuromorphic devices. In this work, we present an electric-double-layer (EDL) transistor based on α-IGZO with sodium alginate (SA) electrolyte film as the gate dielectric. The integrated transistor devices at room temperature have demonstrated good performance, including low subthreshold swing (SS) of 150 mV/decade, switching ratio greater than <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$10^{{6}}$ </tex-math></inline-formula> , and threshold voltage about 1.6 V. Some basic neuromorphic synaptic functions were also simulated, including excitatory postsynaptic current (EPSC), spike duration-dependent EPSC behavior, paired pulse facilitation (PPF), spike number-dependent EPSC behavior and the high-pass filtering. More importantly, the BCM learning rule containing historical frequency-dependent plasticity and adjustable frequency threshold is implemented on this device. This study contributes to the effective reduction of energy loss and protection of neural circuits from stimulus toxicity, which further provides an effective method to improve the efficiency of neural networks in artificial intelligence systems.