'Stateful' threshold switching for neuromorphic learning

Memristors have promising prospects in developing neuromorphic chips that parallel the brain-level power efficiency and brain-like computational functions. However, the limited available ON/OFF states and high switching voltage in conventional resistive switching (RS) constrain its practical and flexible implementations to emulate biological synaptic functions with low power consumption. We present 'stateful' threshold switching (TS) within the millivoltage range depending on the resistive states of RS, which originates from the charging/discharging parasitic elements of a memristive circuit. Fundamental neuromorphic learning can be facilely implemented based on a single memristor by utilizing four resistive states in 'stateful' TS. Besides the metaplasticity of synaptic learning-forgetting behaviors, multifunctional associative learning, involving acquisition, extinction, recovery, generalization and protective inhibition, was realized with nonpolar operation and power consumption of 5.71 pW. The featured 'stateful' TS with flexible tunability, enriched states, and ultralow operating voltage will provide new directions toward a massive storage unit and bio-inspired neuromorphic system.

Automated summary

Researchers propose a method of using memristors for neuromorphic computing, which can achieve energy efficiency and simulate the functions of biological neurons. By introducing threshold switching with different resistive states, different learning and forgetting behaviors can be achieved, with flexible tunability and ultra-low power consumption.