Recent advancements in artificial intelligence systems have been propelled spectacularly by the progress in machine learning techniques, particularly deep neural networks and spiking neural networks. However, such software and CMOS-based approaches present challenges in terms of energy efficiency and scalability. To address these issues, there has been growing interest in the development of energy-efficient ML techniques centered around the restricted Boltzmann machine (RBM). The RBM capitalizes on the Contrastive Divergence, a local learning rule that reduces computational load and energy consumption. Additionally, the RBM can serve as a foundational unit for the deep belief net (DBN). In this study, a simple stochastic neuron device composed of the Ovonic threshold switch (OTS) connected in series with a resistor (Rload) is proposed. Demonstrating probabilistic switching that follows a sigmoid function, this behavior can be adjusted based on the width and interval of the input pulses. Through simulation studies, the device demonstrated successful application in the recognition and reconstruction of handwritten digits.

Stochastic artificial neuron based on Ovonic Threshold Switch (OTS) and its applications for Restricted Boltzmann Machine (RBM)