Spintronic oscillator networks for unconventional computing accelerators

Abstract

Networks of coupled oscillators are fundamental to various natural and technological processes. For instance, networks of biological neurons exhibiting oscillatory behavior are crucial for cognitive functions, memory, and perception. Brain waves, arising from the synchronized activity of neural populations, play a crucial role in decision-making and learning. Hence, oscillator networks have gained significant interest for their ability to perform complex computations through collective dynamics.

This thesis explores spintronic and magnonics-based oscillators, which offer advantages over electronic counterparts by utilizing magnetization dynamics instead of charge currents. These oscillators are smaller, potentially more energy-efficient, and can be engineered into networks for unconventional computing architectures. Leveraging their nonlinear behavior, we advance two key approaches for implementing spintronic oscillator networks.

First, we further develop spatially-resolved networks, where arrays of spin Hall nano-oscillators (SHNOs) have been shown to work as proof-of-concept spintronic Ising machines. Through experimental, theoretical, and computational studies, we demonstrate that voltage control over the local magnetic anisotropy can be used to fine-tune both individual SHNO behavior and coupling between SHNOs, enabling flexible and scalable architectures for spatially-resolved SHNO-based Ising machines. Second, we demonstrate a time-multiplexed oscillator network, where spin wave pulses propagate in a Yttrium Iron Garnet (YIG) delay line. This demonstration acts as a proof of concept for the potential that the implementation of magnonic systems holds. We further develop our approach, optimizing its performance through numerical modeling and expanding its computational complexity by adding a global biasing term to our experimental demonstration.

The results presented in this thesis highlight the potential of spintronic oscillators as key building blocks for next-generation unconventional computing technologies.