Communications of the ACM
Nanomagnets Can Choose a Wine, Could Quench AI's Thirst for Energy
Researchers leveraged the low-power, inherently binary operation of magnetic tunnel junctions to demonstrate neural network hardware inference based on passive arrays of MTJs.
Credit: Getty Images
Scientists at the U.S. National Institute of Standards and Technology, the University of Maryland, and Western Digital Technologies have developed artificial intelligence (AI) devices that could consume less energy and operate faster than other hardware. The researchers programmed a neural network from magnetic tunnel junctions (MTJs) and trained it to taste wines virtually.
The team used 148 of 178 wines produced from three grape varieties, with each wine possessing 13 properties; each property was assigned a value between 0 and 1 for the network to appraise when distinguishing between wines. The network then "tasted" a dataset that included 30 previously unknown wines, yielding a 95.3% success rate and committing just two errors on the untasted wines.
The results indicate an array of MTJ devices could potentially be scaled up and assembled into new AI systems.
From National Institute of Standards and Technology
View Full Article
Abstracts Copyright © 2022 SmithBucklin, Washington, DC, USA
No entries found