Second-Order Conditioning Emulated in an Artificial Synaptic Network

Abstract

Associative learning plays a significant role in shaping human behaviors. Complex cognitions such as image recognition and pattern recognition in real-time, abstract thinking, among others, are easily accomplished with high energy efficiencies, outperforming the superior machines existing today. Thus, brain-inspired synaptic devices are proposed to enhance the computation speed and efficiency, which are lacking in the conventional von Neumann architecture and hence a promising approach for neuromorphic artificial intelligence. Here, an artificial synaptic network (ASN) is explored for emulating higher-order learning without any CMOS supporting circuits. A self-assembled Ag dewetted island network resembling the bioneural network is utilized to fabricate the synaptic device (Ag-ASN). Under an electric field, Ag migration results in the formation of filaments, leading to the emulation of synaptic behavior such as short-term plasticity (STP) and long-term plasticity (LTP). By tuning the input signal, a rehearsal- and compliance-based STP to LTP transition was realized. Along with the diverse nanogap, the formation of various filaments in response to the different electrical stimuli leads to the mimicking of Pavlov’s dog experiment and a louder bell concept. Excitingly, the complex second-order conditioning was emulated for the first time using the synaptic device without any supporting circuits.