Working Towards a Revolutionary Computer Professor Erik Folven is working on the SpinENGINE project, full project name: Harnessing the Emergent Properties of Nanomagnet Ensembles for Massively Parallel Data Analysis. Specialized computer chips may in the future be based on magnetism and have an architecture inspired by the brain. In all kinds of scientific fields, the ability to harvest huge amounts of raw data is exponentially improving. It has put demands on the processing and analysis of data that are often beyond the reasonable capabilities for conventional computing methods. Whilst neural networks have been a game-changer for accomplishments in, for example, Artificial Intelligence, the energy required for training advanced models is high, indeed too high to be sustainable. The energy cost of AI, for example, is said to be doubling, in terms of global progress, every 3.5 months. Thus, new kinds of energy-efficient hardware need to be explored. A magnetic solution Neuromorphic hardware, the kind that the SpinENGINE project is attempting, works on the premise of parallel data processing of very large amounts of data at once, instead of in sequence, one step at a time. This could be the key to making headway in the next generation of low energy computers. The computational demands for data analysis required with many tasks today
would benefit from this envisaged new kind of computer hardware. The interdisciplinary team of researchers working on the SpinENGINE project is reimagining how a computer can be built, with nanomagnets, to be energy-efficient and much faster than conventional computers. Nanomagnets can be in different states depending on their magnetization direction. By applying a magnetic field to a network of nanomagnets, they will change state depending on the properties of the input field and the states of the other magnets nearby.Nanomagnets are non-linear in their response to most stimuli such as magnetic fields, electric currents, temperature, and laser pulses, and could be more than 100,000 times more energy-efficient than conventional computers. They are also capable of storing information without a power source, scalable in their architecture and relatively easy to fabricate. Professor Erik Folven and a multidisciplinary team of scientists across Europe working on the SpinENGINE project, are experimenting with nanomagnet ensembles
to understand their viability and their farreaching potential. The project aims to demonstrate emergent behaviour from these nanomagnet ensembles in a non-linear way based on ‘reservoir computing’ – which is inspired by the workings of biological systems like neural networks in brains. Reservoir computing has its roots in machine learning, where it has outperformed state of the art methods for a range of computational tasks. It relies on neural networks and hinges on the concept of a ‘reservoir’ of ‘neurons’ that does not require training and can process large amounts of data at the same time. It becomes a computational substrate that transforms input data into high dimensional representations – specifically making it excellent at capturing or recognising patterns. Simple linear processing techniques can then be trained to extract an output. Reservoir computing is particularly suitable for machine learning of time varying phenomena and is being considered for sensing and prediction applications which are useful from a smart sensing perspective.
The SpinENGINE team.
www.euresearcher.com
73