The looming obsolescence of Moore's Law is forcing scientists to look for other ways to improve computer performance and innovation.
The effects of this trend will likely be felt by mobile computing devices later than by other types of computing, predicts University of Michigan professor Thomas Wenisch. He says two more generations of mobile devices could be produced before being affected, but a more immediate impact would be on billion-dollar data centers that underlie many useful mobile applications.
Lawrence Berkeley National Laboratory's Horst Simon, co-recipient of the 1988 and 2009 ACM Gordon Bell Prizes, says the approaching limit to transistor density will rekindle interest in rethinking the fundamental computer architecture among supercomputer and data center designers, but this will require a redrafting of many types of software and an overhaul of programmer habits. Wenisch says Intel and other firms will need to take inventive approaches to ramping up computing power, and the possibilities include working harder to improve chip design and specializing chips to accelerate specific algorithms.
One likely inevitable trend is strong demand for silicon tuned for calculations that are vital to deep learning. Microsoft and Intel are focusing on running code on reconfigurable field-programmable gate array chips to boost efficiency.
Massachusetts Institute of Technology professor Neil Thompson raises the question of whether computing advances similar to those achieved via Moore's Law will be possible once it has ended.
From Technology Review
View Full Article
Abstracts Copyright © 2016 Information Inc., Bethesda, Maryland, USA
No entries found