Spin-Glass Consisting of Nanomagnets Being Developed to Replicate Neural Network, a Strategy for Mathematical Thinking of Material Structure

A new study developing a rare form of matter also known as spin glass could ignite a new paradigm in artificial intelligence by enabling algorithms to be printed directly as physical hardware.

As indicated in a Phys.org report, the unique properties of spin glass allow a form of artificial intelligence that can recognize objects from partial images much similar to the brain does not exhibit promises for low-power computing, among other interesting capabilities.

According to a post-doctoral researcher in theoretical physics Michael Saccone from Los Alamos National Laboratory, the new study's lead author, their work accomplished the first experimental realization of an artificial spin-glass that consists of nanomagnets arranged to copy the neural network.

He added, their research lays the groundwork they need "to use these physical systems practically." Essentially, spin glasses are a strategy to think about material structure mathematically.

Spin-Glass

Describing their study published in the Nature Physics journal, Saccone said being free for the first time, tweaking the interaction within the said systems with the use of electron-beam lithography makes it possible to represent a variety of computing problems in spin-glass networks.

At the engineered materials' intersection and its computation, spin-glass systems are a kind of disordered systems of nanomagnets occurring from random interactions and competition between two types of magnetic order in the said material.

They show "frustration," which means, they do not settle into uniformly ordered configuration when their temperatures drop, and they own distinct thermodynamic and dynamic characteristics that can be harnessed for computing applications.

Saccone explained that theoretical models that describe spin glasses are broadly used in other complicated systems, like those that describe brain functions, error-correcting codes, or stock-market dynamics. Such wide interest in spin glasses offers strong motivation to produce an artificial spin glass.

Associative Memory

The researchers integrated theoretical and experimental work to fabricate and observe the artificial spin glass "as a proof-of-principle Hopfield neural network," which mathematically models associative memory to guide the artificial systems' disorder.

Hopfield and spin glass networks have developed symbiotically, one field feeding off the other. Meanwhile, associative memory, be it a Hopfield network or other neural network, forms associated with an object's two or more memory patterns.

If just a single memory is stimulated, for example, by receiving a partial illustration of a face as input, the network then, can recall the complete face. Different from the more traditional algorithms, associative memory does not necessitate a perfectly the same scenario to identify a memory.

The memory of these networks corresponds to the ground states of a spin system and is less disrupted by noise than other neural networks.

Essentially, this study by Saccone and the team validated that the material was a spin glass, evidence that will enable them to describe the properties of the system and the manner it's processing information, a Los Alamos National Laboratory specified.

AI developed in spin glass, explained Saccone, "would be messier" than traditional algorithms, although it is more flexible for some artificial intelligence applications.

Related information about nanomagnets is shown on CHMnanoed's YouTube video below:

Check out more news and information on Nanotechnology in Science Times.