On synapse intelligence emulated in a self-formed artificial synaptic network†
Abstract
A synaptic device forms an essential component of hardware-based neuromorphic artificial intelligence. These brain-inspired intelligent devices are more efficient in terms of speed, energy and space. Most present-day synaptic devices attempt to emulate only basic synaptic functions and involve intricate fabrication processes. Often, they are accompanied by CMOS circuits to accomplish complex learning. Here we report emulation of several of the human-like behavior patterns in a synaptic device fabricated via self-forming. Such a process is endowed with several advantages such as simplicity, scalability and cost-effectiveness. Moreover, the fabricated device comprises a hierarchical structure resembling the biological neural network with synapses and neurotransmitters. This biomimicking device can emulate various behaviors from basic synaptic actions such as STP, LTP, potentiation and depression to complex learning activities such as associative learning, interest-based learning, supervision and impression of supervision. Besides, an important phenomenon in biological systems, synaptic fatigue and recovery, is also demonstrated. An interactive Pavlov's dog prototype has been fabricated, demonstrating the potential utility of the device in brain-inspired neuromorphic applications.