Hewlett Packard Enterprise (HPE) is showing off a prototype computer designed to imitate the parallelism of the human brain using circuit boards and memory chips.
"We're mimicking that architecture of parallel computation using our memristor technology and a specially designed architecture," says Hewlett Packard Lab researcher Cat Graves.
In the manner of synapses, learning and retention on memristor circuits are shaped by current and data flow characteristics, which "has the potential to be incredibly more power efficient, save a lot of time, reduce computing complexity, and not be clogging up the bandwidth," Graves notes.
In the HPE prototype, computation occurs in cells where data is stored--similar to neurons--and then links are set up between cells, much like synapses.
Graves says calculations, or "vector matrix multiplications," can be highly parallel in such an arrangement. She notes these calculations form the core of computationally intensive algorithms and applications such as image filtering, speech recognition, and deep-learning systems.
The memristors are configured in a grid-like pattern connected to the Dot Product Engine, and researchers can switch grid setups on the testbed to determine which patterns are best suited for different kinds of algorithms.
The prototype computer performed 8,000 calculations in one clock cycle in one specific memristor configuration.
From IDG News Service
View Full Article
Abstracts Copyright © 2016 Information Inc., Bethesda, Maryland, USA
No entries found