Neuromorphic optoelectronic computing system based on semiconducting metal oxide nanocrystallites

In pursuit of a computational system that matches the efficiency and speed of the human brain, overcoming the bottleneck of the von Neumann architecture is imperative. To this end, scientists are implementing neuromorphic computing, which emulates the biological principles of the human brain. We developed a hardware-in-the-loop neuromorphic computational block based on physical optoelectronic synaptic devices composed of ZnO, In₂O₃, WO₃, and SnO₂ nanocrystallite semiconductors. It serves to provide conductivity as synaptic plasticity. The dual optoelectronic approach of artificial synapses, in which optical signals modulate conductivity and electrical signals read it, enables the system architecture to avoid the programming (write) sneak-path currents that challenge electrically programmed memristive crossbar devices. In this proof-of-concept implementation, synaptic conductance modulation is realized physically in the optoelectronic devices, whereas leaky-integrate-and-fire neuron dynamics, thresholding, lateral inhibition, and learning-rule scheduling are implemented externally in software and microcontroller logic. The resulting hardware-in-the-loop platform is used to demonstrate device-assisted, STDP-inspired learning in a simple spike-based image-classification task. The study utilizes a fully connected, single-layer neural network with 16 input and 2 output neurons, which activate proportionally to the input data, thus classifying images.