"Inexact" chips save power by fudging the maths
Scientists show off chips that are 15 times more efficient, but get sums wrong
Computer scientists have unveiled a computer chip that turns traditional thinking about mathematical accuracy on its head by fudging calculations.
The Rice University researchers say their “inexact” chip could be useful because it uses dramatically less power than conventional accurate processors.
The scientists claim the prototypes unveiled are 15 times more efficient because they allow occasional errors and could be used in some applications without having a negative effect.
The concept works by allowing processing components — such as hardware for adding and multiplying numbers — to make a few mistakes, which means they are not working as hard so use less power and get through tasks more quickly.
The researchers say the technology is able to "manage the probability of errors and limit which calculations produce errors", in a move that "can simultaneously cut energy demands and dramatically boost performance".
Particular types of applications can tolerate quite a bit of error. For example, the human eye has a built-in mechanism for error correction
One example of the idea in action can be seen with "pruning, or trimming away some of the rarely used portions of digital circuits on a microchip".
Obviously the chips wouldn't be suitable for applications requiring high accuracy, such as financial trading or computer modelling, but the researchers say other tasks can be more flexible.
“Particular types of applications can tolerate quite a bit of error. For example, the human eye has a built-in mechanism for error correction,” said researcher Christian Enz.
“We used inexact adders to process images and found that relative errors up to 0.54% were almost indiscernible, and relative errors as high as 7.5% still produced discernible images.”
According to the team, initial uses for the pruning technology would be in application-specific processors, such as embedded chips used in hearing aids and cameras.
“In the latest tests, we showed that pruning could cut energy demands by 3.5 times in chips that deviated from the correct value by an average of 0.25%,” the researchers said.
“When we factored in size and speed gains, these chips were 7.5 times more efficient than regular chips. Chips that got wrong answers with a larger deviation of about 8% were up to 15 times more efficient.”