Study and Design of a Leaky Integrate-and-Fire Neuron based on Domain Wall motion

One of the most promising approaches for the next generation technologies is neuromorphic computing, which takes inspiration from the brain working principles to design systems with faster computing speed and higher energy efficiency than conventional technology. This is why, in the past years, the possibility of creating artificial neurons and artificial synapses has attracted great interest, leading to a great increase in the research regarding this kind of devices. Among the proposals that can be found in the literature there are spintronic neurons which are particularly interesting since they can take advantages of both electrical and magnetic properties of electrons, allowing for a potential increase of energy efficiency. This thesis will concern the study and the design of a Leaky Integrate-and-Fire (LIF) neuron based on domain wall motion along a racetrack. The neuron’s inputs are current spikes in the heavy metal layer that move the domain wall in the ferromagnetic layer through Spin Orbit Torque (SOT). The SOT mechanism has been chosen over the Spin Transfer Torque since it is more efficient. The neuron fires an output spike when the domain wall reaches a Magnetic Tunnel Junction (MTJ) placed on the other side of the track. By applying a potential to a gate placed before the MTJ, one can create an adjustable threshold exploiting the Voltage Control Magnetic Anisotropy (VCMA) effect. Indeed, the VCMA effect allows to locally modify the anisotropy of the material creating an energy barrier that does not allow the further motion of the domain wall unless the current is above a certain threshold. In a real neuron if no input spike arrives for a certain time the membrane potential relaxes to a resting value. This leaking property of the neuron is usually implemented by the means of an external circuit. In this thesis two main solutions that have been proposed in literature which allow for an intrinsic leaking of the domain wall-based neuron are analysed. In particular, a racetrack with a graded magnetic anisotropy and a non-rectangular track are studied for this specific case. All the simulations are performed with the micromagnetic simulator Mumax3 and the resulting data are analysed in Matlab.