Artificial Optic-Neural Synaptic Device for Colored-Pattern Recognition

S. Seo, J-H Park
Sungkyunkwan University,

Keywords: 2D material, neuromorphic, synaptic device, neural networks


In real-time image signal processing, the spike-based neural processing in the human vision system outperforms the conventional frame-based approach, in terms of energy efficiency (1-100 fJ/synaptic operation) and parallel computing capability (~80 billions of neurons). Due to the systematic superiority of brain-inspired spiking neural process, many advanced researches on artificial neural networks (ANNs) for pattern recognition, especially employing various memristors as synapses, have been carried out. However, the priority has been given to prove the device potential for the emulation of synaptic dynamics, not to functionalize further synaptic devices for more complex learning by combining biometric sensors with synaptic devices. Here, we demonstrate optic-neural synaptic (ONS) device by implementing synaptic and optical sensing functions together on a single van der Waals (vdW) material, consequently mimicking the colored pattern recognition capability of human vision system through an optic-neural network (ONN). Our synaptic device shows weight update trajectory close to linear curve (Nonlinearity factor = 1.4/1.4 for weight increase/decrease) and large number of stable conduction states (>600 states and around 1% variation at each state) at low voltage pulse (0.3 V pulse), thereby enabling more accurate and energy-efficient classifications of colored patterns. This demonstration is an important one-step forward towards much larger and more complex artificial neural networks (ANNs) that comprises neural sensing and training functions.