London, UK – Researchers from UCL and Imperial College London have made significant strides in revolutionizing artificial intelligence (AI) by harnessing the power of chiral magnets. This breakthrough has the potential to dramatically reduce energy consumption in machine learning tasks and pave the way for more sustainable and adaptable computing technologies.
The study, published in the prestigious journal Nature Materials, focuses on a form of brain-inspired computing known as physical reservoir computing. By utilizing the inherent physical properties of chiral magnets, the researchers were able to adapt the material to suit different machine learning tasks by applying an external magnetic field and changing temperature.
Traditionally, computing systems consume large amounts of electricity due to the need for separate units for data storage and processing. This constant shuffling of information between the two units not only wastes energy but also generates heat. Machine learning, in particular, requires massive datasets for processing, resulting in substantial carbon dioxide emissions.
Physical reservoir computing offers an alternative approach by eliminating the need for distinct memory and processing units. This not only results in more efficient data processing but also presents a sustainable solution to conventional computing. Furthermore, this technology can be seamlessly integrated into existing circuitry, providing additional energy-efficient capabilities.
The research team employed a vector network analyzer to examine the energy absorption of chiral magnets at varying magnetic field strengths and temperatures. They discovered that different magnetic phases of chiral magnets excelled at different types of computing tasks. The skyrmion phase, characterized by magnetized particles swirling in a vortex-like pattern, exhibited a powerful memory capacity suitable for forecasting tasks. On the other hand, the conical phase exhibited little memory but possessed non-linearity that was ideal for transformation tasks and classification, such as identifying whether an image features a cat or a dog.
Dr. Oscar Lee from UCL, the lead author of the study, remarked, 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. The next step is to identify materials and device architectures that are commercially viable and scalable.
Dr. Jack Gartside from Imperial College London, a co-author of the study, highlighted the significance of the research, saying, Our collaborators at UCL recently identified a promising set of materials for powering unconventional computing. Working with the lead author Dr. Oscar Lee, our team at Imperial College London designed a neuromorphic computing architecture to leverage the complex material properties, matching the demands of diverse and challenging tasks. The results were remarkable, showcasing how reconfiguring physical phases can directly enhance neuromorphic computing performance.
The development of energy-efficient brain-like computing holds immense promise for various industries and applications, from powering smart devices to advancing robotics and autonomous systems. The integration of chiral magnets into computing technologies could usher in a new era of sustainable and adaptable AI, paving the way for a more energy-efficient future.
The study was supported by various organizations, including the Leverhulme Trust, Engineering and Physical Sciences Research Council (EPSRC), Royal Academy of Engineering, and the German Research Foundation (DFG).
As researchers continue their exploration of chiral magnets and their potential applications, the future looks bright for energy-efficient and brain-inspired computing.
