'Inexact chip' allows errors to increase efficiency - massively
 

After fifty years of attempting to create accurate calculations, chip makers should be embracing the odd mistake in order to create more powerful and efficient processors.

That is the view of a group of researchers at the Rice University in the USA, claiming that an “inexact” chip can actually get a boost by allowing for the odd error.

In fact, the team claims that speed and energy consumption of inexact chips mean fifteen times more efficiency than current chips.

They were able to achieve this massive boost by getting rid of some parts that would usually make up a circuit, as well as cutting the power use for hardware to make calculations.

This means errors. But, through clever management of the probability of errors and limiting which calculations produce them, the chip designers were able to create massive performance boosts, with only a marginal drop in accuracy.

Of course, many computers rely totally on complete accuracy, but the errors manifest themselves in ways that might not make much difference in certain applications.

For example, in processing an image the chips could provide a lower quality picture, with the grey matter lodged between our two ears processing into a workable image.

As ARM has shown over the past few years, efficiency is the name of the game, and the chip designs could well see plenty of applications in low powered devices.

At the very least, in a world where computers are fast approaching human levels of intelligence, it is nice to know that there is room for the odd idiot too.

http://news.techeye.net/chips/inexac...ency-massively