Trimming out the deadwood in computer chips could make them twice as fast and half the size, without damaging real-life application performance, a team of computer scientists has said.
The technique could be used to build chips for specific applications, such as in cameras and media players by trimming away areas of the chip that are not used in those applications, the scientists said.
"This is the first time someone has taken an integrated circuit and said, 'Let's get rid of the part that we don't need'," said principal investigator Krishna Palem. "It's a physical trimming of the chip, not a conceptual thing.
"We can boost performance and cut energy use simultaneously if we prune the unnecessary portions of the digital application-specific integrated circuits that are typically used in hearing aids, cameras and other multimedia devices."
According to the team – made up of scientists from the Rice-NTU Institute for Sustainable and Applied Infodynamics and Switzerland's Center for Electronics and Microtechnology – the concept relies on a technique that is deceptively simple.
"We basically run the applications on the chips and see where it is busy, and where it's not being used by an application you can cut it out," Palem told PC Pro. "It identifies where there is nothing happening during a process."
The scientists said initial tests had indicated that the pruned circuits would be at least twice as fast and consume about half the energy and space of traditional circuits.
The final result is a chip that is an example of a concept called "inexact hardware" that makes more mistakes - with as many as 8% more errors - but is able to process application-specific information much more quickly.
“By cleverly managing the probability of errors and by limiting which calculations produce errors, we have have found they can simultaneously cut energy demands and boost performance,” Palem explained.
"To put the 8% into context, we know that many perceptive types of tasks found in vision or hearing applications can easily tolerate error magnitudes of up to 10%," Palem said.
The team said it had already produced a prototype chip and that the end results could be chips that would enable batteries to run “four to five times longer”.
This article originally appeared at pcpro.co.uk
