In recent years, the energy consumption related to information technology has increased exponentially, partly due to the advent of artificial intelligence. Novel hardware approaches are needed to tackle this issue. Spintronics is emerging as one most promising alternatives to the conventional microelectronics. Within the framework of spintronics, nanoscale magnetic textures known as magnetic skyrmions have attracted a great interest due to their potential applications in information storage, logic and also unconventional computing technologies. However, several limitations have been encountered regarding their electrical manipulation and detection in ferromagnets, which need to be overcome before moving toward applications. In this thesis, we address the electrical manipulation and detection of magnetic skyrmions in magnetic tracks and magnetic tunnel junctions for memory and neuromorphic computing applications. In the first part of this work, we study the current-induced dynamics of magnetic skyrmions in synthetic antiferromagnets, which are composed of two thin ferromagnetic layers antiferromagnetically coupled. We demonstrate that skyrmions in such materials can move about one order of magnitude faster than in ferromagnets, reaching speeds up to 900 m/s. This effect is explained by the compensation of the so-called gyrotropic force exerted on the skyrmions as a result of the antiferromagnetic coupling.
In the second part of the thesis, we study the electrical manipulation of skyrmions in magnetic tunnel junctions, which represents another major challenge for applications. In particular, we demonstrate the electrical control of the nucleation and annihilation process of magnetic skyrmions in magnetic tunnel junctions. To this end, we performed operando magnetic imaging by combining the electrical detection with x-ray microscopy. We also investigated the dynamics of the nucleation and annihilation process using time-resolved magnetic microscopy experiments. These results open a promising path toward a multi-state memory based on magnetic skyrmions. In the third and final part of this thesis, we exploit spin textures in magnetic tunnel junctions to achieve complex temporal pattern recognition using the physical reservoir computing paradigm. For this purpose, we use the spin texture in a magnetic tunnel junction and its dynamical response to electrical stimuli as the computational and memory elements of a neural network. We first show that the voltage induced dynamics of the spin texture exhibit the non-linearity and short-term memory properties required from a physical system to be exploited as a physical reservoir computing. We then demonstrate that such a system can perform temporal signal classification, such as distinguishing in between sine and square waves, with high accuracy, and achieve good performances on other benchmark tasks for reservoir computing systems. These results open a pathway toward nanoscale, low power artificial intelligence hardware based on the manipulation of the spin textures.

Plus d'information :https://www.spintec.fr/phd-defense-electrical-manipulation-of-topological-spin-textures-for-memory-and-neuromorphic-application/
​

Pour suivre la soutenance ​​​en visioconférence : https://univ-grenoble-alpes-fr.zoom.us/j/91833631116?pwd=DJBtFSKUDnzgNjVKN1gFGLWqW5z0mh.1

​ ​ ​​