The international team published the first proof that nanomagnet networks can be used for AI-like processing in a paper that was published today, May 5, 2022 in the journal Nature Nanotechnology. Researchers demonstrated that nanomagnets could be used to perform ‘time-series predictions’ tasks such as regulating insulin levels and predicting diabetes.
Artificial intelligence using neural network aims to mimic the brain’s workings, in which neurons communicate with each other to retain and process information. Many of the math used to power neural network was first invented by physicists to explain how magnets interact. However, it was difficult at that time to use magnets directly because researchers didn’t know how data was entered and received.
Instead, software running on silicon-based computers was used for simulating magnet interactions and, in turn, the brain. The team has been able to use magnets to store and process data, cutting out the middleman in the simulation. This could potentially lead to huge energy savings.
Nanomagnetic states
According to their direction, nanomagnets may be in different’states’. The state of a magnet changes when it is applied to a network nanomagnets. This depends on its properties and the states of nearby magnets.
A team of Imperial Department of Physics researchers was then able to devise a method to count the number of magnetic particles in each state after the field passes through. This gave the ‘answer.
Jack Gartside, co-first author of this study, said that “We’ve been trying for a while to solve the problem of how magnetic computing can input data, ask questions, and give an answer.” It’s now possible to get rid of the computational software that performs the energy-intensive simulation.
Kilian Stenning, co-first author, said: “How magnets interact gives you all the information that we need; laws of Physics themselves become the computer.”
Dr. Will Branford, team leader, stated that it was a long-term goal of realizing computer hardware inspired from the software algorithms by Sherrington and Kirkpatrick. Although it was impossible to control the spins of atoms in conventional magnets, we were able scale up the spins into nanopatterned arrangements and have achieved the required control and readout.
Reduce energy costs
AI can be used in many contexts, including voice recognition and self-driving vehicles. However, even simple tasks like solving a Rubik’s cube can require a lot of energy to train AI. To train AI to solve Rubik’s Cube took about the same energy as two hour-long runs of nuclear power stations.
This is because of inefficient electron transport during processing and storage. Nanomagnets don’t rely upon the physical transport of particles such as electrons but instead process and transmit information in the form a’magnon wave, where each magnet has an effect on the state of the neighboring magnets.
This allows for much lower energy consumption and allows information to be stored and processed together as opposed to being performed in separate steps like conventional computers. This innovation could make nanomagnetic computer up to 100,000x more efficient than traditional computing.
Artificial Intelligence at the Edge
Next, the team will teach the system by using real-world data such as ECG signals. The goal is to eventually make it into an actual computing device. Magnetic systems could eventually be integrated into traditional computers to increase energy efficiency and improve performance for high-performance processing tasks.
They are also energy efficient, so they could be powered by renewable energy and used for ‘AI at edge’, which is processing data right where it is being collected (e.g., weather stations in Antarctica) rather than sending it back into large data centers.
They could also be used in wearable devices to process biometric information on the body. This includes predicting and controlling insulin levels for diabetics or detecting abnormal heartbeats.
