Neuromorphic spiking information processing based on neuron-like excitable effect has achieved rapid development in recent years due to its advantages such as ultra-high operation speed, programming-free implementation and low power consumption. However, the current physical platforms lack building blocks like compilers, logic gates, and more importantly, data memory. These factors become the shackles to construct a full-physical layer neural network. In this paper, a neuromorphic regenerative memory scheme is proposed based on a time-delayed broadband nonlinear optoelectronic oscillator (OEO), which enables reshaping and regenerating on-off keying encoding sequences. Through biasing the dual-drive Mach-Zehnder electro-optic modulator in the OEO cavity near its minimum transmission point, the OEO can work in excitable regime, where localized states are maintained for robust nonlinear spiking response. Both simulation and experiment are carried out to demonstrate the proposed scheme, where the simulation results and the experimental results fit in with each other. The proposed OEO-based neuromorphic regenerative memory scheme exhibits long-term response ability for short-term excitation, which shows an enormous application potential for high-speed neuromorphic information buffering, optoelectronic interconnection and computing.
Read full abstract