Embedded Files
Mission:
Mission:
The project aims to apply machine learning (ML) to study chiral spin structures, a recent focus in spintronics. These structures are important for low-energy memory and logic devices, and their stability depends on the Dzyaloshinskii-Moriya interaction (DMI), an anisotropic exchange interaction between neighboring spins. While recent research has explored measurement techniques and protocols for DMI, discrepancies still exist. This project plans to:
The project aims to apply machine learning (ML) to study chiral spin structures, a recent focus in spintronics. These structures are important for low-energy memory and logic devices, and their stability depends on the Dzyaloshinskii-Moriya interaction (DMI), an anisotropic exchange interaction between neighboring spins. While recent research has explored measurement techniques and protocols for DMI, discrepancies still exist. This project plans to:
1) Investigate measurement statistics, reproducibility, and repeatability, and their relationship to sample inhomogeneities and defects.
1) Investigate measurement statistics, reproducibility, and repeatability, and their relationship to sample inhomogeneities and defects.
2) Use ML to evaluate DMI from magnetic domain patterns, training the system with simulated patterns (bubble domains) for comparison with experimental data. Parameter sensitivity analysis from ML will offer insights into data spread's physical origins.
2) Use ML to evaluate DMI from magnetic domain patterns, training the system with simulated patterns (bubble domains) for comparison with experimental data. Parameter sensitivity analysis from ML will offer insights into data spread's physical origins.
The goal is to improve measurement reliability using artificial intelligence and determine if AI can become a metrological tool or service for the industry. The project's findings will benefit the spintronics and metrology research community, as well as nanotechnology-related fields facing similar measurement challenges.
The goal is to improve measurement reliability using artificial intelligence and determine if AI can become a metrological tool or service for the industry. The project's findings will benefit the spintronics and metrology research community, as well as nanotechnology-related fields facing similar measurement challenges.
Concept of the study. Micromagnetic simulations are used to generate thousands of training images. There are five relevant material parameters: Ms, K , A, D, and σ (anisotropy dispersion). Here D and σ are varied in the simulations so that the system can learn domain patterns with different D and σ. After the supervised training, the system is feeded with an experimentally obtained image of magnetic domains to extract D and σ. From DOI: 10.1038/s41524-020-00485-2.
Need for the project
The overall need for the project is the growing commercialization and industrial use of spintronics and the with coming need of industry, society and governments for standardization and regulations. Spintronics is lately gaining attention in the metrology community, and there is recent interest at the IEC level, however measurement protocols, a good practice guide and a common nomenclature are currently missing. The complex measurement procedures and techniques proposed by the scientific community do not fulfill the need of industry for fast and efficient quality control of production. Therefore, EEI, especially of smaller size, is discouraged to invest in spintronics, not having available straightforward tools for characterization and classification of materials and devices. There is a need for:
1) Accurate DMI measurements. The DMI is currently the key-parameter for novel applications in the magnetic memory and logic sector, extremely promising for creating new computing concepts and low energy data storage. Without reliable and accurate measurements no device classification for their quality or energy consumption can be performed;
2) Understanding measurement discrepancies. Quite often in spintronics, nominal similar samples show large discrepancies in measuring key-parameters. This happens not only for the DMI, but also for the spin Hall angle or the spin Seebeck coefficient. At the moment, it is not completely understood if these discrepancies are due to differences in the measurement methods or inhomogeneities of the samples themselves;
3) Need for statistical approaches and AI. While the scientific community is trying to gain a deeper understanding of the mechanisms underlying novel spintronic effects, industry started pilot production of innovative memories and sensors.
The scientific approach, although indispensable, requires often complex and rarely available methods for characterizing materials and devices. This does not allow for a rapid and high throughput analysis considering statistical aspects. AI is a perfect tool for large data classification and a limited tool for physical interpretation. Joining the two approaches may help to create a new line of materials characterization.
D values measured in the international round robin comparison by MOKE (INRiM, University of Leeds) and BLS (NIST, KRISS and UniPG). The uncertainties vary from a few percent to about 40% depending on the sample quality. Large discrepancies between the two methods occur: BLS measures 7-8 times larger values. (from https://arxiv.org/abs/2201.04925)
Work Packages (WPs)
WP1: Sample preparation and measurements (UNIPG):
WP1: Sample preparation and measurements (UNIPG):
WP1 aims to assess the actual discrepancies of measurements of the DMI and to establish guidelines for reducing the measurement uncertainty. The WP builds on the recent results of the EMPIR project TOPS, which showed that the two most popular measurement methods (a) asymmetric bubble domain expansion by MOKE and b) spin wave dispersion non reciprocity by BLS) are potential candidates for measurement services for industry. One of the results of TOPS was to give guidelines for a reliable measurement employing these methods. While the measurement uncertainty of a single measurement is excellent for high quality samples, there are remaining open questions, such as the large discrepancy between MOKE and BLS measurements of identical samples or why one method works better for certain types of samples. WP1 will tackle these problems by a) preparing well characterised samples with different DMI values and properties, b) determining and comparing the actual measurement uncertainty of a single measurement performed by the two methods with the aim to propose practices for improving it and c) studying repeatability and reproducibility of the two techniques correlated to spatial inhomogeneities of the samples. WP1 will be complemented by WP2, which will interpret the results of WP1 on a theoretical basis. Furthermore, WP1 will build the physical basis for a successful ML approach applied in WP3, since the validation of the ML approach requires the availability of accurate and reliable measurements.
WP2: Theory and models (POLIBA):
WP2: Theory and models (POLIBA):
WP2 aims to correlate the measurement discrepancies and uncertainties observed in WP1 to physical parameters and to build a solid theoretical basis for the ML approach in WP3. Task 1 of WP2 is dedicated to a review of the different analytical models for the evaluation of the DMI from the measurement and proposes innovative approaches taking into account e.g. the magnetic history of the sample and/or defects and inhomogeneities. Task 2 is for micromagnetic simulations, which represent an indispensable tool for testing magnetic systems. Simulations will be performed to reproduce the asymmetric bubble expansion technique, and they will complement the statistical approach in WP1 Task3 by considering defects and inhomogeneities. A parameter sensitivity analysis will be carried out for the validation of the analytical model and the benchmark with experimental achievements, which represent Task 3. WP2 is essential for a successful ML approach applied in WP3, since ML requires a clear and unique connection between the training dataset and the material parameters provided by accurate analytical or numerical models.
WP3: Metrology and machine learning (INRiM):
WP3: Metrology and machine learning (INRiM):
WP3 aims to evaluate from WP1 and WP2 the actual measurement reliability and to employ an innovative approach to the measurement analysis of the DMI. This approach will help to overcome the difficulties due to the statistical spread of measurement data and has therefore the ability to revolutionise metrology in fields where these play an important role (nanotechnology, spintronics, etc.). In particular, four steps are needed: i) standardise the procedures and the data analysis to be able to determine the uncertainty budget (Task1), ii) apply ML methods to trial and test simulation and experimental data to determine the DMI (Task2); iii) evaluate various sources of noise and errors to be able to perform a Uncertainty Quantification and a parameter sensitivity analysis (Task3); iv) validate the overall ML approach, and set a few guidelines for estimation of spintronics parameters.
WP4: Creating impact and project management (INRiM)
WP4: Creating impact and project management (INRiM)
The project encompasses various tasks including knowledge transfer, training, data management, open access. Under WP4 Task1, initiatives such as creating a project website, sending newsletters, forming a stakeholder committee, publishing papers, participating in conferences, and developing webinars or online courses are planned. In WP4 Task2, activities involve preparing a good practice guide and assessing measurement reproducibility. WP4 Task3 focuses on research data management, open access publishing, and developing a data management plan. In WP4 Task4, project coordination and management entail regular interaction among leaders, monthly reporting, holding meetings, and conducting risk assessments. These tasks aim to ensure effective knowledge dissemination, stakeholder engagement, data integrity, and project progress.
Page updated
Report abuse