Abstract
The recent trend of development of the Internet of Things (IoT) technology calls for the implementation of Artificial Intelligence at the Edge. This imposes the devising of suitable computing engines overcoming the limitations of common von Neumann architectures. In this respect, oscillatory neural networks (ONNs) are appealing ultra-low-power neuromorphic architectures promising to perform sophisticated tasks, such as pattern recognition, in edge devices. Key elements of ONNs are the oscillators mimicking the neurons. The use of beyond-CMOS devices based on vanadium dioxide (VO2) to realize these oscillators is an innovative step for the ONN technology, compared to traditional realizations through CMOS-based oscillators. The core mechanism operating in VO2 devices is the insultor-to-metal transition (IMT) of VO2, which undergoes a phase transition triggered by the temperature induced by self-heating from a high-resistive monoclinic (M1) crystal structure to a low-resistive tetragonal rutile-like (R) one. The use of commercial TCAD software tools to simulate VO2 devices is essential prerequisite to boost the development of ONN technology. This requires in turn dedicated TCAD modeling approaches to simulate thermal-induced resistive switching effects in VO2 devices. In the present contribution, results will be shown over 3D electrothermal simulations of a relaxation oscillator based on a VO2 device and the impact of material properties and geometry over its self-oscillatory behavior will be investigated.