Nonplanar Atomic Layer Deposition (ALD)-Niobium Oxide (NbOx) Neurons for Oscillatory Neural Network Applications

Abstract

This article presents a nonplanar niobium oxide (NbOx) neuron device fabricated using an atomic layer deposition (ALD) method for use in oscillatory neural networks (ONNs). Potentially, such nonplanar geometry allows for high-density arrays. The desired threshold switching (TS) characteristics are achieved through an interfacial method using a thin titanium (Ti) layer. X-ray photoelectron spectroscopy (XPS) analysis confirms that the oxyphilic Ti layer reduces Nb2O5 to the desired NbO2 stoichiometry, which is crucial for the device's functionality. The device demonstrates S-type negative differential resistance (NDR) under current-controlled operation, and its self-oscillation capabilities are verified within an electrical oscillator circuit. The oscillation frequency is shown to increase linearly with the applied voltage (Vd). The nonplanar structure and the use of a Ti layer to tune the material's properties allow for the realization of a compact, low-power neuromorphic device without the need for additional thermal annealing. Finally, the study demonstrates the practical application of these NbOx neurons in an ONN architecture for pattern recognition. The system successfully recognizes binary representations of digits. The network's functionality relies on phase synchronization between neurons, where a binary pattern is encoded by applying a delayed VDD to each neuron pixel.