A Step Closer to Energy-Efficient Brain-Inspired Computing
A groundbreaking study led by researchers from UCL and Imperial College London has brought us closer to achieving energy-efficient brain-inspired computing. By harnessing the unique properties of chiral magnets, the team has demonstrated the adaptability of these materials for different machine-learning tasks.
Unlocking the Potential of Physical Reservoir Computing
Physical reservoir computing, a brain-inspired computing approach, has faced limitations due to its lack of reconfigurability. However, the researchers discovered that by applying an external magnetic field and adjusting the temperature, the physical properties of chiral magnets can be modified to suit various computing tasks.
Dr Oscar Lee, the lead author of the study, expressed excitement about the findings, stating, “This work brings us a step closer to realizing the full potential of physical reservoirs to create computers that not only require significantly less energy but also adapt their computational properties to perform optimally across various tasks, just like our brains.”
Towards Sustainable Computing
Traditional computing methods consume substantial amounts of electricity, primarily due to the separation of data storage and processing units. This constant shuffling of information between units wastes energy and generates heat. Machine learning, in particular, requires vast datasets, resulting in significant carbon dioxide emissions.
Physical reservoir computing, as a sustainable alternative, eliminates the need for distinct memory and processing units, enabling more efficient data processing. Additionally, it can be seamlessly integrated into existing circuitry, providing energy-efficient capabilities.
Unleashing the Power of Chiral Magnets
The study involved researchers from Japan and Germany who utilized a vector network analyzer to measure the energy absorption of chiral magnets at different magnetic field strengths and temperatures ranging from -269°C to room temperature.
The team discovered that different magnetic phases of chiral magnets excelled at different computing tasks. The skyrmion phase, characterized by swirling magnetized particles in a vortex-like pattern, exhibited a remarkable memory capacity suitable for forecasting tasks. On the other hand, the conical phase, with its non-linearity, proved ideal for transformation tasks and classification, such as distinguishing between cats and dogs.
Collaborative Success
Co-author Dr. Jack Gartside from Imperial College London highlighted the collaboration between UCL and Imperial College London, stating, “Our collaborators at UCL… recently identified a promising set of materials for powering unconventional computing. Working together, we designed a neuromorphic computing architecture that leverages the complex material properties to match the demands of a diverse set of challenging tasks.”
The study also involved researchers from the University of Tokyo and Technische Universität München and received support from various organizations, including the Leverhulme Trust, EPSRC, Royal Academy of Engineering, and the DFG.
Future Prospects
With this significant breakthrough, the next step is to identify commercially viable and scalable materials and device architectures. By harnessing the power of chiral magnets and their adaptability, energy-efficient brain-inspired computing could become a reality, revolutionizing the field of computing and paving the way for a more sustainable future.
Photo: Freepik.com
