Periodic Reporting for period 1 - NIMFEIA (Nonlinear Magnons for Reservoir Computing in Reciprocal Space)
Berichtszeitraum: 2022-10-01 bis 2024-03-31
NIMFEIA aims to provide a hardware solution for brain-inspired computing using magnetic materials on the nanoscale combined with advanced spintronics technologies. It is based on the disruptive idea of reciprocal space computing, utilizing nonlinear interactions of quantized magnetic excitations. In the presence of nontrivial spin structures, such as magnetic vortices, these nonlinear interactions can be efficiently harnessed for reservoir computing tasks like pattern recognition and time series prediction with minimal pre-processing of input data. By making use of quantum interactions in reciprocal space, computing is done in individual spatially confined devices, such as a single magnetic disc, without the need to transport information data in real space.
We will demonstrate the ground-breaking nature of our proposal by meeting four core objectives.
1. Demonstrate the core principles of reservoir computing using GHz-regime spin waves by quantifying and designing nonlinear interactions in reciprocal space (TRL 3);
2. Develop an experimental and scalable proof-of-concept device using industrially compatible processes (TRL 4);
3. Demonstrate the utility of the magnon reservoir on a selected pervasive real-world use case, namely gesture and feature recognition from radar data (TRL 4-5);
4. Scale the magnon reservoir to the THz regime by the use of synthetic and pure antiferromagnetic materials (TRL 1-2).
While objectives 1-3 will focus on the validation of the novel technology in laboratory and also industrially relevant environment, objective 4 will provide the groundwork for pushing this technology towards THz frequency operation and 6G compatibility.
NIMFEIA will lay the foundations for a new paradigm in nanomagnetic and spintronic technologies that go beyond traditional applications in binary storage and Boolean logic, radiofrequency signal processing, as well as field sensing. It will address major current technological challenges by proposing an energy-efficient computing scheme for edge computing.
We will demonstrate the ground-breaking nature of our proposal by meeting four core objectives.
1. Demonstrate the core principles of reservoir computing using GHz-regime spin waves by quantifying and designing nonlinear interactions in reciprocal space (TRL 3);
2. Develop an experimental and scalable proof-of-concept device using industrially compatible processes (TRL 4);
3. Demonstrate the utility of the magnon reservoir on a selected pervasive real-world use case, namely gesture and feature recognition from radar data (TRL 4-5);
4. Scale the magnon reservoir to the THz regime by the use of synthetic and pure antiferromagnetic materials (TRL 1-2).
While objectives 1-3 will focus on the validation of the novel technology in laboratory and also industrially relevant environment, objective 4 will provide the groundwork for pushing this technology towards THz frequency operation and 6G compatibility.
NIMFEIA will lay the foundations for a new paradigm in nanomagnetic and spintronic technologies that go beyond traditional applications in binary storage and Boolean logic, radiofrequency signal processing, as well as field sensing. It will address major current technological challenges by proposing an energy-efficient computing scheme for edge computing.
During the first reporting period, the NIMFEIA consortium made significant progress towards achieving the defined objectives:
Objective 1: The material parameters for NiFe, the standard metallic ferromagnet used for NIMFEIA's vortex-based reservoir, were optimized and a manifold of different samples was fabricated exploring different geometries for realizing magnon- and skyrmion-based reservoirs. The proof-of-principle for reservoir computing using a vortex-based magnon reservoir and a skyrmion reservoir was demonstrated experimentally and in micromagnetic simulations. The antenna design for addressing the vortex-based magnon reservoir was optimized to allow its scalability and targeting the vortex core gyration for more versatile dynamics in different frequency ranges.
Objective 2: The material stack for magnetic tunnel junctions (MTJs) was customized and optimized for NIMFEIA's goals by depositing blanked multilayer films on 300-mm wafer technology and by determining characteristic material parameters such as saturation magnetization, magneto resistance and coercive field. Using this customized multilayer stack, fully structured MTJ arrays (30 bits) were patterned which allow contacting one single MTJ in laboratory environment. To allow for a minimal distance between the MTJ and the magnon reservoir, which will be patterned on top of the MTJ, the contacting Cu electrode was thinned down, a process that was optimized for the NIMFEIA project.
Objective 3: Radar data which represents four different gestures and consists of 4,800 gestures in total was distinguished using a single-skyrmion reservoir. For each gesture, the Fourier transform of the radar data was converted into a time-dependent voltage and fed into the reservoir, uniquely manipulating the skyrmion. Reading-out the skyrmion's position, the reservoir demonstrated classification quality at the level of software-based approaches but offers the advantages of ultra-low power consumption.
Objective 4: It was realized that to describe nonlinear magnonics in antiferromagnets to the THz regime, a new description of spin dynamics is required. Instead of dynamics of the net magnetization and Neel vector, commonly used to describe magnons with long wavelengths and in the GHz regime, dynamics at short wavelengths should be formulated on the basis of spin correlations. It was demonstrated that only a description of spin correlations can adequately describe the selection rules for the excitation of magnons at the edge of the Brillouin zone.
Objective 1: The material parameters for NiFe, the standard metallic ferromagnet used for NIMFEIA's vortex-based reservoir, were optimized and a manifold of different samples was fabricated exploring different geometries for realizing magnon- and skyrmion-based reservoirs. The proof-of-principle for reservoir computing using a vortex-based magnon reservoir and a skyrmion reservoir was demonstrated experimentally and in micromagnetic simulations. The antenna design for addressing the vortex-based magnon reservoir was optimized to allow its scalability and targeting the vortex core gyration for more versatile dynamics in different frequency ranges.
Objective 2: The material stack for magnetic tunnel junctions (MTJs) was customized and optimized for NIMFEIA's goals by depositing blanked multilayer films on 300-mm wafer technology and by determining characteristic material parameters such as saturation magnetization, magneto resistance and coercive field. Using this customized multilayer stack, fully structured MTJ arrays (30 bits) were patterned which allow contacting one single MTJ in laboratory environment. To allow for a minimal distance between the MTJ and the magnon reservoir, which will be patterned on top of the MTJ, the contacting Cu electrode was thinned down, a process that was optimized for the NIMFEIA project.
Objective 3: Radar data which represents four different gestures and consists of 4,800 gestures in total was distinguished using a single-skyrmion reservoir. For each gesture, the Fourier transform of the radar data was converted into a time-dependent voltage and fed into the reservoir, uniquely manipulating the skyrmion. Reading-out the skyrmion's position, the reservoir demonstrated classification quality at the level of software-based approaches but offers the advantages of ultra-low power consumption.
Objective 4: It was realized that to describe nonlinear magnonics in antiferromagnets to the THz regime, a new description of spin dynamics is required. Instead of dynamics of the net magnetization and Neel vector, commonly used to describe magnons with long wavelengths and in the GHz regime, dynamics at short wavelengths should be formulated on the basis of spin correlations. It was demonstrated that only a description of spin correlations can adequately describe the selection rules for the excitation of magnons at the edge of the Brillouin zone.
A this stage of the project it is too early to draw conclusions on potential impacts going beyond the state of the art.