Frustrated magnetic systems as a basis for the neural computing
Promotor: Prof. Dr. Bartel Van Waeyenberge
Supervision : Mykola Dvornik
1 or 2 students
The aim of the proposed work is to explore the possibility of employing frustrated magnetic media to create magnetic neural networks – simulating the mechanism behind the brain’s operation. The clusters of magnetic nanoelements will be employed as neurons, while local and non-local magnetic interactions will act as synapses. Neural networks are one type of adaptive system, whose topological connections change depending on the propagating signal. This is usually achieved by establishing pathways between neurons, so that the structure of the network defines its functionality. The same could be achieved in frustrated magnetic systems by employing their natural ability support magnetic domains and their propagation, without changes to the energy state. Furthermore, it has been already demonstrated that in artificial spin ice it is possible to create analogues to Dirac’s strings. Such strings could also act as pathways, correlating states between different artificial magnetic monopoles localized in well separated parts of the system.
The aim is to explore the dynamics of such domains and strings and in particular their ability to form complex networks with neural properties.
The work will start with a comprehensive literature review on the topic of frustrated magnetic media research to get familiar with properties of the structures and roughly estimate parameters space for the micromagnetic simulations.
Micromagnetic simulations using in-house developed software MuMax2 running on in-house built 30TFlops GPU cluster. The ultimate goal is to demonstrate numerically a magnetic structure where logical operations (such as XOR, NOT, etc.) could be dynamically programmed in the magnetic state, rather then topology.
Optionally, experimental investigation of frustrated magnetic media by magnetic force microscopy (MFM) and resistivity measurements can be set-up.
Location: De Sterre S1, and/or home