‘Nanomagnetic’ computing can provide low-energy AI, researchers show | imperial news

A grid of long ovals containing small arrows pointing in different directions and representing nanomagnets.

Researchers have shown that artificial intelligence is possible using tiny nanomagnets that interact like neurons in the brain.

The new method, developed by a team led by researchers at Imperial College London, could dramatically reduce the energy cost of artificial intelligence (AI), which is currently doubling globally every 3.5 months.

The way the magnets interact gives us all the information we need; the very laws of physics become the computer. Kilian Stenning

In an article published today in Nature Nanotechnology, the international team produced the first proof that networks of nanomagnets can be used to perform AI-like processing. The researchers showed that the nanomagnets can be used for “time series prediction” tasks, such as predicting and regulating insulin levels in diabetic patients.

Artificial intelligence using ‘neural networks’ aims to replicate the way parts of the brain work, where neurons communicate with each other to process and retain information. Much of the math used to power neural networks was originally invented by physicists to describe the way magnets interact, but at the time it was too difficult to use magnets directly as researchers didn’t know how to input data and get information.

Instead, software running on traditional silicon-based computers was used to simulate magnetic interactions, in turn simulating the brain. Now the team has been able to use the magnets themselves to process and store data, cutting out the software simulation middleman and potentially offering huge energy savings.

nanomagnetic states

Nanomagnets can come in various “states”, depending on their direction. Applying a magnetic field to a network of nanomagnets changes the state of the magnets based on the properties of the input field, but also the states of the surrounding magnets.

A graph of fuzzy and strong lines.
A map of the states of nanomagnets in an experiment

The team, led by researchers from the Imperial Department of Physics, were able to devise a technique to count the number of magnets in each state once the field has passed, giving the ‘answer’.

Study co-author Dr Jack Gartside said: “We have been trying to solve the problem of how to enter data, ask a question and get an answer from magnetic computing for a long time. Now that we’ve shown it can be done, it paves the way for us to get rid of the computer software that does the power-hungry simulation.”

Co-author Kilian Stenning added: “The way the magnets interact gives us all the information we need; the very laws of physics become the computer.”

Team captain doctor will branford he said: “It has been a long-term goal to make computing hardware inspired by Sherrington and Kirkpatrick’s software algorithms. It was not possible to use the spins of atoms in conventional magnets, but by scaling the spins in arrays with nanometric patterns we have been able to achieve the necessary control and readout.”

Cut the cost of energy

AI is now used in a variety of contexts, from voice recognition to self-driving cars. But training AI to perform even relatively simple tasks can take enormous amounts of energy. For example, training the AI ​​to solve a Rubik’s cube took the energy equivalent of two nuclear power plants running for an hour.

Much of the energy used to accomplish this in conventional silicon-chip computers is wasted in inefficient electron transport during processing and storage in memory. However, nanomagnets are not based on the physical transport of particles such as electrons, but instead process and transfer information in the form of a ‘magnonic’ wave, where each magnet affects the state of neighboring magnets.

This means that much less energy is wasted and that processing and storage of information can be done together, rather than being separate processes as in conventional computers. This innovation could make nanomagnetic computing up to 100,000 times more efficient than conventional computing.

AI on the edge

The team will then teach the system using real-world data, such as ECG signals, and hope to turn it into a real computing device. Eventually, magnetic systems could be integrated into conventional computers to improve power efficiency for processing-intensive tasks.

Their energy efficiency also means they are feasible to run on renewable energy and be used to do ‘AI at the edge’ – processing data where it is collected, like weather stations in Antarctica, rather than sending it back to big data centres. .

It also means they could be used in wearable devices to process biometric data in the body, such as predicting and regulating insulin levels for people with diabetes or detecting abnormal heartbeats.

Reconfigurable Training and Reservoir Computation in a Spin-Vortex Artificial Ice via Spin-Wave Fingerprinting‘ by Jack C. Gartside, Kilian D. Stenning, Alex Vanstone, Holly H. Holder, Daan M. Arroo, Troy Dion, Francesco Caravelli, Hidekazu Kurebayashi, and Will R. Branford is published at Nature Nanotechnology.

Leave a Reply

Your email address will not be published.