Spin-diode effect in vortex based nano-oscillators for hardware artificial intelligence applications

A hardware implementation of neural networks is considered through the study of artificial neurons and synapses. Firstly spin-transfer nano-oscillators have been investigated as physical neurons. Particular focus is made on vortex-based oscillators exited by an out-of-plane spin-polarized current. An analytical model describing the vortex dynamics is proposed based on Thiele equation. Nonlinearities related to the vortex core displacement are taken into account. Micromagnetic simulations are performed with mumax3 to allow comparison with theoretical results. Two oscillation regimes are identified, both falling in radiofrequency range. Below a critical current, relaxation is obtained and the device frequency evolves proportionally to the current density. Above this threshold value gyrotropic precessions are observed, leading to a nonzero steady orbit radius and nonlinear frequency evolution. Moreover a significant impact of the Ampère-Oested field on the dynamics is shown. Using these simulations data-driven corrections are made to the model. These fittings allow to derive with great accuracy the vortex properties both in transient and steady-state regime. Finally spin-diode effect in magnetic tunnel junctions is investigated. The semi-analytical model predictions are compared to experimental results obtained by partners. An unprecedented agreement is demonstrated. Secondly a memristor device mimicking synaptic behaviors has been explored. The latter is based on an interconnected silver nanowire network. Following a technique developed in past decades at UCLouvain a track-etched polycarbonate membrane is filled with Ag wires by electrodeposition. After this step the sample is fragmented by plasma etching-induced stresses. This allows to create gaps between Ag wire segments. Then voltage sweeps up to 100 V are performed. Characteristic memristive hysteresis is present in the I-V curves. Remanence is also observed. Sample resistance, in the Gohm range, decreases with successive sweeps. The creation of conductive Ag filament in the gaps separating segments is suggested. Those would result from field-induced ion migration along the nanopore walls. Spontaneous breaking of the bridging is remarked with time. Evidences of electron direct tunneling and field emission are shown in the first voltage scan. After tens of sweeps a dramatic change of properties is measured. The sample presents a resistance more than three orders of magnitude below its original value. Very peculiar behavior is shown in this regime, as the sample conductance is minimum for the largest applied bias.