Seminar über die Physik der kondensierten Materie (SFB/TRR173 Spin+X und SFB/TR288 Kolloquium, TopDyn-Seminar)

June 10, 2025 at 1 p.m. in Noether Room (03-423)

Univ-Prof. Dr. Jure Demsar
Univ.-Prof. Dr. Hans-Joachim Elmers
Univ.-Prof. Dr. Mathias Kläui
Univ.-Prof. Dr. Thomas Palberg

Nanoscale Magnets for Non-volatile Memory, Hardware AI and Quantum Control
Jayasimha Atulasimha (Virginia Commonwealth University)


In a world where a single company like Google consumed more energy than a country with population of about 20 million [1] in 2019 and this is growing exponentially, it is essential to find energy efficient approaches to make our computing needs sustainable. One potential solution is the use of nanoscale magnetic computing devices. Towards this end, energy efficient approaches based on electrical field control of nanoscale magnetism are pursued in our group: (i) strain mediated switching of the magnetization of nanomagnets [2]; (ii) creation and annihilation of magnetic skyrmions using direct voltage control of magnetic anisotropy (VCMA) [3]; and more recently magnetoionic control [4]. Such nanoscale magnetic devices have application to non-volatile memory [3], hardware AI [4, 5, 6] and quantum control of spins [7,8,9].

We will discuss skyrmion mediated voltage control of nanoscale magnetization that has potential for extremely energy efficient non-volatile memory [3] and are robust to switching errors in the presence of thermal noise, material and device inhomogeneities, while scaling to lateral dimensions of 20 nm and below [3]. Furthermore, energy efficient AI hardware can be realized with nanomagnetic devices. Multi-state nanoscale domain wall racetracks can be used as highly quantized synapses in deep neural networks [5] and convolutional neural networks, with overall improvement in area, energy, and latency by 13.8, 9.6, and 3.5 times respectively [5] compared to purely CMOS implementations. Additionally, interacting nanomagnets can be used for analog [6] and digital reservoir computing [6] and long-term prediction of temporal data [6]. We will specifically discuss experimental implementation of reservoir computing with magnetoionic devices [4] that do not need conversion of signals to GHz unlike when Spin Torque Nano Oscillators (STNOs) are used.

In quantum computing, in addition to energy efficiency, one significant problem is implementing qubits in a scalable manner at temperatures of a few Kelvin. We argue that ensemble spin qubits may offer such a possibility [7]. Furthermore, by driving the magnetization of nanomagnets electrically, highly confined microwaves can be generated at the Larmor precession frequency of proximally located spins [8]. This can implement single-qubit quantum gates with fidelities approaching state-of-the-art in a scalable manner. Further confinement of microwaves using convergent-divergent skyrmion devices can implement even more localized and low footprint quantum control of spins [8]. New experimental and simulation results in these directions will be discussed [9].
References
[1] FORBES Editor’s pick, Oct 21, 2020,04:26pm EDT
[2] Nano Letters, 16, 1069, 2016; Nano Lett., 16, 5681, 2016; Appl. Phys. Lett. 121, 252401, 2022; https://arxiv.org/abs/2501.00980
[3] Nature Electronics 3, 539, 2020; Scientific Reports,11, 20914, 2021; Scientific Reports, 14, 17199, 2024
[4] https://arxiv.org/abs/2412.06964
[5] Nanotech. 31 145201, 2020, IEEE Access, 10, 84946, 2022; IEEE Trans. on Neural Networks and Learning Sys, 36, 4996, 2024.
[6] Appl. Phys. Lett. 121, 102402, 2022; Comm. Phys. 6, 215, 2023; Neuromorph. Comput. Eng. 2 044011; IEEE Access, 11,124725, 2023
[7] https://arxiv.org/abs/2503.12071
[8] Communication Physics 5, 284, 2022; Physical Phys. Rev. Applied 22, 06407, 2024.
[9] https://arxiv.org/abs/2407.14018