In the era of big data, the ever increasing energy demands of data storage and computation present a fundamental roadblock in achieving a net-zero carbon future. This is due to the proliferation of machine learning (ML) & artificial intelligence algorithms and the fundamental inefficiencies of their CMOS host. This has a sparked the exploration of novel computing systems which are better suited to the parallel, non-linear computations required for ML.
Nanomagnetic systems have emerged as promising candidates for next-generation computing systems due to their passive long-term memory and non-linear, wireless interactions inherent in the physics
itself. They exhibit collective spin-wave dynamics in the GHz-range, well-suited for technological
integration. Realising the full potential of nanomagnetic systems requires methods of controlling the magnetisation state of individual nanomagnets and using this state-control for computation.
In this thesis, novel methods of controlling the state of individual nanomagnets are developed. Low power (mW) all-optical switching of individual nanomagnets is achieved through linearly polarised laser illumination, removing power-consuming electromagnet requirements. A magnetic tip is used to write all available states in circular nanomagnetic elements (nanodisks). Increasing and controlling the range of accessible states in bar-shaped nanomagnets is achieved by engineering their size and shape. These methods of state control are used for two types of low-power computation. A nanodisk-based
reconfigurable magnonic crystal is proposed where information propagates along state-controlled pathways. Spin-wave bending, phase-shifting and interferometry is demonstrated. State engineering and spin-wave dynamic readout is used to present the first realisation of neuromorphic computation in large, strongly-interacting nanomagnetic arrays. The nanomagnetic array is capable of non-linear
signal transformation and prediction of chaotic time-series.
These results usher in a new paradigm of nanomagnetic state-control and computation, enabling a host of low-power next-generation computing schemes.