Micromagnetic simulations for spin-based neuromorphic computing

Abstract

Although modern A.I has enabled many new innovations such as voice assistant and facial recognitions, current A.I computers are still a long way from achieving human-like flexibility in problem solving and the ability to learn from unstructured stimuli with energy efficiency comparable to a human brain. Traditional von Neumann computers are highly inefficient at solving unstructured problems and as such, neuromorphic computers which mimic the biological neural network of a human brain have been researched extensively in a bid to create more powerful artificial neuromorphic computing systems that can compete with the human brain in unstructured problem-solving efficiency. Spin-based neuromorphic computing is a promising candidate that has the potential to create high-performance and low powered neuromorphic network. In our model, we make use of domain wall switching to model the Leaky-Integrate-and-Fire neurons. To further explore domain wall based neuromorphic computing, we ran simulations of our spin-based design to observe its Leaky-Integrate functionalities and analysed it suitability for use in neuromorphic computing.