Abstract
Modern Artificial Intelligence (AI) relies on artificial neural networks, which attempt to emulate the functionalities of the human brain through a set of highly interconnected nodes that play the role of artificial neurons, and may revolutionize the way we interact with technology. Currently, the most robust artificial neural networks are constructed using appropriate software models on CMOS hardware. However, how calculations are carried out on computers differs significantly from how the brain processes information. The prominent modern alternative are the wave-based physical systems. They have been recently demonstrated to operate as recurrent neural networks, where interference patterns in the propagating waves can realize an all-to-all interconnection between points of the host medium, exploiting the rich nonlinear dynamics that mimics the action of artificial neurons by scattering and recombining input waves in order to extract the carried information. Especially spin-waves (magnons) in magnetic films are promising candidates for practical applications due to their low power usage, strong nonlinearity arising from magnetization dynamics, and established scalability as well as integrability of magnetic nanostructures. Spin waves are readily employed for performing logic operations and recent advances have been made towards magnonic artificial intelligence, where different types of nanoengineered magnon scattering reservoirs have been explored. However, realizing the full potential of these ideas requires precise manipulation of spin waves in nanostructures, which is still a challenge and needs to be promptly advanced for the benefit of functional magnonic devices.
In this project, we put forward magnonics in rapidly emerging 2D magnetic materials as a viable platform for neuromorphic and reservoir-computing applications. The magnetic properties of these atomically-thin, crystalline materials are extremely prone to electro-mechanical tuning, such as by lattice straining, gating, defect engineering, and/or layer stacking and heterostructuring. Furthermore, the recent observations of high-frequency (THz) spin-wave modes in monolayer CrI3 and room-temperature 2D ferromagnetism in several other materials put all the ingredients in place for the use of 2D magnetic materials as a technological platform for spin-wave-based neuromorphic computing. That said, theoretical and simulation insights are critically lacking in this field, which we aim to timely rectify in the present proposal. We will devise strategies to broadly and actively tune magnonic excitations and their propagation in selected 2D materials by nanoengineered structural and electronic stimuli, and engage to map out the viable realizations of neuromorphic computing in such materials, for which we will provide detailed theoretical recipes and in silico demonstrations. Considering that crystalline 2D materials offer a closest possible connection between the simulation environment and the practically measured quantities, our discoveries are bound to inspire experimental replication and further advances of magnonic technology.
Researcher(s)
Research team(s)
Project type(s)