Man-made a rare type of substance known as spin glass, according to a new study, could usher in a new era of artificial intelligence by allowing algorithms to be printed as physical hardware. The special characteristics of spin glass enable a type of AI that, like the brain, can distinguish objects from incomplete images and has the potential for low-power computing, among other things.
Our work led to the first experimental realization of an artificial spin glass made of nanomagnets arranged to replicate a neural network. Our article lays the foundation we need to use these physical systems in a practical way..
Michael Saccone, Study Lead Author and Postdoctoral Researcher, Theoretical Physics, Los Alamos National Laboratory
The study was published in the journal Natural Physics.
Spin glasses are a mathematical approach to thinking about material structure. Saccone explained that the ability to modify the interaction within these systems using electron beam lithography makes it possible for the very first time to represent a range of computational problems in spin glass gratings.
Spin glass systems are a form of disordered system of nanomagnets that result from random connections and competition between two types of magnetic order in the material. They sit at the intersection of engineering materials and computation.
When their temperature decreases, they exhibit “frustration”, meaning that they do not settle into a uniformly ordered arrangement, and they have different thermodynamic and dynamic properties that can be used in computer applications.
Theoretical models describing spin glasses are widely used in other complex systems, such as those describing brain functions, error-correcting codes, or stock market dynamics. This great interest in spin glasses provides a strong motivation to generate an artificial spin glass..
Michael Saccone, Study Lead Author and Postdoctoral Researcher, Theoretical Physics, Los Alamos National Laboratory
The study team integrated theoretical and experimental investigation, to construct and analyze artificial spin glass as a proof-of-principle Hopfield neural network, which mathematically simulates associative memory helping to regulate the instability of artificial spin systems. .
The spin-glass and Hopfield networks have evolved in a symbiotic relationship, with one field feeding the other. Associative memory connects two or more memory patterns associated with an item, either in a Hopfield network or other types of neural networks.
The network can remember the entire face if only one memory is enabled, for example by obtaining a partial image of an input face. Associative memory, unlike more standard algorithms, does not require a completely identical circumstance to detect memory.
The memories of these networks are similar to the ground states of a spin system, and they are less affected by noise than other neural networks.
Research by Saccone and his team verified that the material was spin glass, providing evidence that will allow them to explain the attributes of the system and how it processes data. According to Saccone, spin-glass AI algorithms would be “messier” than standard algorithms, but more versatile for particular AI applications.
The research was funded by the laboratory-led research and development program at Los Alamos National Laboratory.
Journal reference:
Saccone, M. et al. (2022) Direct observation of a dynamic glass transition in a nanomagnetic artificial Hopfield grating. Natural Physics. doi.org/10.1038/s41567-022-01538-7.
Source: https://www.lanl.gov/
