Combined Micromagnetic Simulation and Machine Learning Approach to Analysis of Polycrystalline Bilayer With Exchange Bias

Exchange bias is a complex interface phenomenon often used in multiple applications ranging from magnetic field sensors to spintronics and neuromorphic computing for inducing a unidirectional anisotropy in a ferromagnetic layer exchange coupled to an antiferromagnetic layer. Despite the significant progress in understanding mechanisms behind the exchange bias, predicting and optimizing hysteresis properties of the pinned layer in real-life systems with complex crystalline and magnetic structure is still a challenge. In this work we use a combined machine learning and micromagnetic simulation approach for building a unified predictive model giving macroscopic hysteresis properties of the pinned layer for a given set of magnetic and structural parameters. The approximator can be considered as an unknown function, which can be used for finding local or global extrema for coercivity or exchange bias field. We believe that a similar approach can be applied to other computer models or high-quality experimental data for advanced analysis of functional dependencies of hysteresis properties as well as evaluation of computer model used for simulation. A machine learning model of a bilayer with exchange bias could be helpful for decreasing the computation time, optimizing layers materials and parameters, and minimizing the number of test samples.