The pursuit of energy-efficient computing inspired by the human brain increasingly focuses on harnessing the movement of magnetic domain walls, tiny boundaries within magnetic materials, as a means of processing information. Jeffrey Brock, Aleksandr Kurenkov, and Aleš Hrabec, at the Laboratory for Mesoscopic Systems at ETH Zurich, alongside Laura Heyderman and colleagues, now demonstrate a significant advance in controlling this movement within a specific class of materials called ferrimagnets. Their research reveals how carefully engineered magnetic interactions can drive domain walls forward spontaneously, offering a simpler and more tunable platform for building brain-inspired computing devices. By controlling the speed of this self-driven motion, the team achieves behaviours crucial for artificial neural networks, including the ability to accumulate and reset information, paving the way for scalable and energy-efficient neuromorphic architectures.

Significant interest exists in spontaneous domain wall motion as a physical mechanism to enable energy-efficient, next-generation brain-inspired computing architectures. Realising all behaviours required for neuromorphic computing within standard material systems remains a significant challenge, as these functionalities often rely on competing interactions. This research demonstrates how spontaneous domain wall motion, driven by locally engineered magnetic interactions in transition metal-rare earth ferrimagnets, can achieve numerous neuromorphic computing functionalities in devices with minimal complexity. Through experiments and simulations, the team shows how tuning the size and composition of these materials influences device performance and functionality.

GdCo Alloys for Neuromorphic Computing

This research focuses on developing a novel neuromorphic computing architecture using compensated ferrimagnetic materials, specifically GdCo alloys, to create artificial neurons and synapses. GdCo alloys exhibit magnetic compensation, where the magnetic moments of the constituent elements balance, resulting in a near-zero net magnetization. This is crucial for creating devices with low energy consumption and enhanced sensitivity. Researchers manipulate magnetic domain walls within these films to represent and process information; the position and movement of these walls act as the basis for artificial synapses and neurons.

By operating near the compensation point, the devices minimize energy dissipation, making them ideal for low-power neuromorphic computing. The research incorporates mechanisms for lateral inhibition and winner-take-all functionality, mimicking biological neural networks. This is achieved through careful design of the magnetic landscape and interactions between neighboring devices. Researchers also demonstrate voltage control of the magnetic order and domain wall motion, enabling efficient programming and operation of the devices. Key concepts underpinning this work include neuromorphic computing, which aims to build brain-inspired computing systems, and ferrimagnetism, a type of magnetism where opposing magnetic moments result in a net magnetic moment.

The researchers utilize techniques like thin film deposition and helium ion irradiation to create and modify the GdCo films. They visualize magnetic domains and track domain wall motion using the magneto-optical Kerr effect, and model the magnetic behavior of the devices using micromagnetic simulations. This research represents a significant step towards building energy-efficient and brain-inspired computing systems. Future research will focus on device optimization, network integration, algorithm implementation, and enhancing voltage control, ultimately paving the way for a new generation of computing systems.

Magnetic Domain Walls Emulate Biological Neurons

Researchers have demonstrated a new approach to building artificial neurons using magnetic domain walls within specifically engineered materials, offering a promising pathway towards more energy-efficient computing. These artificial neurons mimic the behavior of biological neurons by integrating input signals, leaking potential over time, and resetting to an initial state after firing, all crucial functions for neuromorphic computing. The team successfully created devices where the position of a magnetic domain wall represents the neuron’s potential, and its movement emulates the integration of incoming signals. The breakthrough lies in the use of transition metal-rare earth ferrimagnetic alloys, materials exhibiting exceptionally fast and efficient domain wall motion.

By precisely controlling the material composition, researchers engineered spontaneous domain wall movement without the need for external stimuli, a significant advancement over previous designs. This local control allows for fine-tuning of both the “leak” rate, how quickly the neuron’s potential decays, and the “reset” behavior after the neuron fires, essential characteristics for realistic neuron emulation. The researchers demonstrated that by integrating this spontaneous domain wall motion with spin-orbit torque, they could build structures exhibiting both leaky integration and passive reset, key functionalities for artificial neural networks. The speed and efficiency of domain wall movement in these materials, combined with the straightforward patterning process, provides a scalable platform for building complex neuromorphic architectures. The ability to control the leak and reset behaviors with such precision represents a substantial step forward in creating artificial neurons that more closely mimic their biological counterparts. By leveraging the non-volatility of magnetic domain walls, the team avoids the energy inefficiency associated with separating memory and processing in traditional computer architectures.

Domain Walls Emulate Biological Neuron Functionality

This research demonstrates a pathway to create more efficient and versatile brain-inspired computing architectures by harnessing spontaneous domain wall motion within specifically engineered magnetic materials. Researchers successfully controlled the speed of this motion through careful tuning of material composition, feature size, and chiral interactions in transition metal-rare earth ferrimagnets. This control, when combined with spin-orbit torque, enabled key functionalities for artificial neurons, including leaky integration and passive resetting of the domain wall position, essential behaviors for mimicking biological neural networks. The findings establish a scalable and readily implementable platform for domain wall-based computing, utilizing materials that can be deposited with high throughput and patterned using a straightforward laser-induced process. Future research directions include exploring the potential of magnetoionic gating to reversibly tune magnetic properties and further refine control over lateral exchange coupling, ultimately bringing increasingly sophisticated neuromorphic architectures within reach.

👉 More information
🗞 Engineering and exploiting self-driven domain wall motion in ferrimagnets for neuromorphic computing applications
🧠 ArXiv: https://arxiv.org/abs/2508.14252