Creating complexity: toward atomic spin-based neural hardware

ended

The growing trend in global electricity consumption has created a new challenge for materials-based science: to find computational paradigms toward ICT that are not only smaller and faster, but also energy-efficient. A new source of inspiration is the human brain, which consumes a mere 20 W of energy, while a supercomputer consumes about 10 MW. The emerging field of brain-inspired hardware aims at utilizing physical phenomena in high-quality materials toward pattern recognition and energy- efficient ICT. The goal of this project is to adapt the principles of magnetism toward brain-inspired hardware, utilizing individual and coupled atomic spins. The ultimate aim of SPINAPSE is to probe the feasibility and create proof-of-concept systems, which demonstrate computational principles such as pattern recognition. I define three objectives, which address understanding magnetism in the three most prominent neural models: (1) Hopfield model, (2) Perceptron, (3) Reservoir computing. The strategy is to utilize the so-called spin workbench, based on low-temperature scanning tunneling microscopy, as a platform to create tailored spin arrays with atomic-scale control. This method combines single atom magnetic imaging and atom-scale fabrication, enabling the control of the magnetic interactions and dynamics between ensembles of atoms, atom by atom. We will construct bottom-up magnetic nanostructures to implement all-spin and atomic-scale based neural hardware. We will deliver a new state of the art in magnetic imaging, including (a) developing the spin workbench with a newly built 30 mK magnetic STM facility, defining a new state of the art in magnetic imaging worldwide, and (b) time-resolved imaging to probe the magnetization dynamics of stochastic spin arrays at milliKelvin temperatures. The outcome of SPINAPSE will deliver a new state of the art, new fundamental understandings, and create proof-of-concept technologies for atomic-scale brain-inspired hardware.