A neural network for prediction of high intensity resonance modes in magnetic multilayers

Abstract

The use of magnetic materials as building blocks for frequency applications makes it possible to fabricate micrometer and nanometer high frequency devices. Moreover, devices with multiple high intensity modes for multiband devices can be designed by using magnetic multilayers. However, as the number of layers increases the multilayer becomes more complex, making it very difficult to find optimal configurations due to a big number of possible configurations. Fortunately, over the past decade a surge in the applicability and accessibility of machine learning algorithms and neural networks has been observed, which allow to analyse big quantities of data in search of complex patterns not always evident to humans. In this work, a theoretical model is used to generate approximately 10 × 106 data points, which in turn are used to train a neural network to calculate the number of high intensity resonance modes of three ferromagnetically coupled magnetic layers with an accuracy of over 99.8%. The neural network is then used to identify a configuration of the multilayer which provides the maximum number of high-intensity modes, and comparisons with the theoretical model are presented. Finally, the correlations between parameter were calculated over 600 million of data points, and clear guidelines for obtention of two high intensity resonance modes were identified. These results provide a simple way to find a configuration of the trilayer that have a high number of high intensity modes, thus greatly simplifying the design process of magnetic multi-band frequency devices.