University of Michigan computer scientists David Blaauw and Dennis Sylvester are researching micro mote computers, which are one-cubic-millimeter computing sensors that can collect data and conduct analysis on board, unlike other smart devices that must send data to the cloud for analysis.
Blaauw and Sylvester were able to embed flash memory into the micro mote, creating an energy-efficient device with 1 MB of storage.
One of the researchers' micro motes incorporates a deep-learning processor that operates a neural network on only 288 microwatts of power. Neural networks typically demand large memory banks and considerable processing power. The researchers were able to scale down the computer's size by redesigning the chip architecture to minimize data movement.
Blaauw and Sylvester say deep-learning processors could be integrated into motion-detection systems, face-recognition systems, and other Internet of Things devices.
From IEEE Spectrum
View Full Article
Abstracts Copyright © 2017 Information Inc., Bethesda, Maryland, USA