In a recent nanotechnology advancement, scientists found that a new kind of artificial intelligence could be developed through magnets. These magnets, which are specialized to reach nanoscale labels, were modified to form a neural network that acts similar to the human brain.
Nanotechnology Meets Artificial Intelligence
Specialists from Imperial College London innovated the novel nanotechnology. According to the study's authors, the model could be an efficient option that uses only a small amount of energy to create modern AI-powered systems.
Based on the estimates, the energy usage of international artificial intelligence systems doubles every 3.5 months. For example, when an AI solves just a simple Rubik cube, the simulation could take most of the energy stored in two nuclear power stations in just one hour. The paper shows comprehensive evidence that could cut this energy expenditure while keeping the quality of performance of AI models when built with nanomagnets.
According to the experts, the nanomagnets could work best for tasks related to 'time-series prediction.' These tasks focus on medical solutions, especially for those with diabetes that require proper predictions and regulations on their insulin levels.
Most technologies based on artificial intelligence are equipped with neural networks that replicate how the natural brain of humans works. When neural networks activate, materials serving as 'neurons' connect and relay information with each other for processing.
The neural networks consist of numerous arithmetic solutions that curate the possible outcomes of a certain process. In the past, the concept of neural networks was attempted to be developed in magnetic instruments, but the lack of resources and limiting capabilities of older technologies did not allow scientists to proceed.
Nanomagnets as Neural Networks Now Possible
The challenging process of putting information in and out of a magnetic structure seemed impossible, which eventually led to the idea being buried for decades. Today, there are many AI simulations possible through silicon-based models. This paved the way to test the possibility of the once-forgotten magnetic neural network.
In the new study, experts were able to actualize the idea. Now, the model was developed to carry and store the data throughout the magnetic networks. Moreover, this system could keep the information quality and run the process in parallel to the modern neural networks while consuming less energy by up to 100,000 times.
Nanomagnets could come in various states, and this function depends on where their directions are pointed to. Experts used a specialized magnetic field to change the state of the nanomagnets and their directions, allowing the input field to process properly and result in a dedicated outcome.
The Imperial College London's Department of Physics customized this process to count each of the changing states throughout the magnets as the field passes through while giving the right 'answer.'
In a report by PhysOrg, author Jack Gartside explained that the new technique is a step to exceeding current AI solutions based on computer software known to run with energy-intensive simulations. Kilian Stenning, a co-author of the paper, said that the magnets could indeed give the needed data, and this approach serves as evidence that the laws of physics could become the computers themselves.
The study was published in the journal Nature Nanotechnology, titled "Reconfigurable training and reservoir computing in an artificial spin-vortex ice via spin-wave fingerprinting."
Check out more news and information on Nanotechnology in Science Times.
