PRIN 2022 CALL. DECREE NO. 104 OF 2 FEBRUARY 2022
TITOLO DEL PROGETTO: Metrology for spintronics: A machine learning approach for the reliable determination of the Dzyaloshinskii-Moriya interaction (MetroSpin)
CUP CODE: D53D23002360006
Budget: € 77.095
U.R. Coordinator: Vito Puliafito
Other Research Units or any Sub-Units:
National Metrology Research Institute of Turin (PI’s U.R.) and University of Perugia
Brief project description
Despite the strong scientific and industrial growth of spintronics, as a low-energy alternative to CMOS technology and compatible with it, the commercial diffusion of spintronic technologies is limited by the absence of reliable metrological standards for measuring the key parameters of devices.
The project proposed a study focused on the Dzyaloshinskii-Moriya interaction (DMI), a fundamental parameter for the stability of chiral spin structures used in low-power memories and logic devices. Following previous European studies that defined good experimental practices, a statistical approach capable of explaining the remaining discrepancies in measurements was still lacking.
The objective was therefore to analyse the reproducibility and repeatability of measurements, correlating data dispersion with defects and inhomogeneities in the samples, and to develop machine learning methods to extract DMI from simulated and experimental magnetic patterns. Artificial intelligence is being assessed as a possible metrological tool to improve measurement reliability, with potential impacts not only in spintronics but across all nanotechnology sectors facing similar issues.
Purpose
The project aimed to improve the reliability of spintronics measurements through a statistical approach based on machine learning, with particular attention to the measurement of the Dzyaloshinskii-Moriya interaction (DMI), a key parameter for innovative spintronic devices. The activities focused on reducing uncertainty and discrepancies among different measurement techniques, analyzing the physical origin of data dispersion, and developing automated analysis procedures based on artificial intelligence. The project also assessed the possible extension of these approaches to other areas of spintronics and nanotechnologies.
Expected results
The expected results of the project included a better understanding of the physical and statistical causes underlying the dispersion of DMI measurements, the identification of strategies to reduce experimental uncertainty and improve the reproducibility of measurement techniques, as well as the development of automated analysis tools based on machine learning. The project also aimed to demonstrate the potential of artificial intelligence as a metrological support for spintronics and, more generally, for applications in nanotechnologies characterized by similar measurement reliability issues.
Achieved results
The project successfully achieved three main objectives.
Accurate determination of the interfacial Dzyaloshinskii-Moriya interaction (iDMI). Guidelines were developed for the two main iDMI measurement techniques: Brillouin light spectroscopy (BLS) and the asymmetric expansion of magnetic bubbles measured via the magneto-optical Kerr effect (MOKE). The guidelines define the key technical aspects required to obtain accurate measurements, include reproducibility studies, analyze the dependence on intrinsic and extrinsic parameters, and compare different iDMI evaluation models. This provides a reliable tool for assessing the quality, performance, and energy consumption of spintronic devices.
Resolution of discrepancies between measurement techniques. Following the developed guidelines, it was possible to significantly reduce the measurement uncertainty for both methods. The large discrepancies in iDMI values reported in the literature were successfully attributed to the presence of chiral damping, which affects the bubble expansion technique differently from BLS. Thanks to a model developed within the project, it was possible to obtain consistent iDMI values within experimental uncertainty limits for both techniques. This result significantly improves the understanding of the physics of domain wall motion in the presence of iDMI.
Feasibility of statistical approaches and integration of artificial intelligence. The potential of using machine learning for rapid, high-throughput iDMI analysis, including statistical aspects, was demonstrated through a study on synthetic data simulating the bubble expansion technique. The model proved robust even in the presence of simulated defects and blurred images. Once suitable databases become available, this work may lead to the development of a fast and accessible AI-based tool for the classification of spintronic devices.
