Adaptive Spiking Neural Network Neuromorphic Hardware for Interfacing Between Emerging Neuron and Synaptic Devices

This paper proposes an SNN-based system with 2T1R architecture capable of interfacing with emerging neuromorphic devices that emulate crucial components such as neurons and synapses in neural network. This system proposes a spike regenerator that enables the effective use of neuron device spike signals and a Spike-Timing-Dependent Plasticity (STDP) generator to update synaptic device utilizing update pulse train for the STDP. This compact system allows simple adjustments of regenerated waveforms and STDP pulse widths to meet various device specifications. The measurement results using hBN-based pre-synaptic neuron device and Cu:Te-based CBRAM synaptic device demonstrate the feasibility of learning through the system. This system supports on-chip learning network operations with various devices as well as hBN and Cu:Te devices. In the future research, this system can be expanded for large-scale array implementations.