Science

"Inexact" computer chip makes mistakes, but is 15x more efficient

"Inexact" computer chip makes mistakes, but is 15x more efficient
A prototype “inexact” computer chip that is around 15 times more efficient than current microchips (Avinash Lingamneni/Rice University/CSEM)
A prototype “inexact” computer chip that is around 15 times more efficient than current microchips (Avinash Lingamneni/Rice University/CSEM)
View 2 Images
A prototype “inexact” computer chip that is around 15 times more efficient than current microchips (Avinash Lingamneni/Rice University/CSEM)
1/2
A prototype “inexact” computer chip that is around 15 times more efficient than current microchips (Avinash Lingamneni/Rice University/CSEM)
Frames produced with video-processing software on traditional hardware (left), inexact processing hardware with a relative error of 0.54 percent (middle) and with a relative error of 7.58 percent (right) (Image: Rice University/CSEM/NTU)
2/2
Frames produced with video-processing software on traditional hardware (left), inexact processing hardware with a relative error of 0.54 percent (middle) and with a relative error of 7.58 percent (right) (Image: Rice University/CSEM/NTU)

Last year, a team of U.S. researchers applied the pruning shears to computer chips to trim away rarely used portions of digital circuits. The result was chips that made the occasional mistake, but were twice as fast, used half as much energy, and were half the size of the original. Now, building on the same “less is more” idea, the researchers have built an “inexact” prototype silicon chip they claim is at least 15 times more efficient than current technology in terms of speed, energy consumption and size.

In the traditionally exacting world of computing, it might seem counter intuitive to set out to develop a chip that is allowed to make a few errors. But by managing the probability of errors and restricting which calculations are allowed to produce errors, the research team led by Krishna Palem has been able to slash energy demands while also boosting performance.

In addition to removing certain processing components, the team also employed another innovation in the prototype chip to further cut energy demands called “confined voltage scaling,” which trades some performance gains by taking advantage of improvements in processing speed.

“In the latest tests, we showed that pruning could cut energy demands 3.5 times with chips that deviated from the correct value by an average of 0.25 percent,” said Avinash Lingamneni, a Rice graduate student and co-author of the study. “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 percent were up to 15 times more efficient.”

While you probably wouldn’t want to find any inexact chips in the cockpit of an airplane or a missile guidance system, there are plenty of applications where a certain margin of error is acceptable.

“Particular types of applications can tolerate quite a bit of error. For example, the human eye has a built-in mechanism for error correction,” says project co-investigator Christian Enz. “We used inexact adders to process images and found that relative errors up to 0.54 percent were almost indiscernible, and relative errors as high as 7.5 percent still produced discernible images.”

Frames produced with video-processing software on traditional hardware (left), inexact processing hardware with a relative error of 0.54 percent (middle) and with a relative error of 7.58 percent (right) (Image: Rice University/CSEM/NTU)
Frames produced with video-processing software on traditional hardware (left), inexact processing hardware with a relative error of 0.54 percent (middle) and with a relative error of 7.58 percent (right) (Image: Rice University/CSEM/NTU)

Palem says devices such as hearing aids, cameras and other electronic gadgets that use special-purpose “embedded” microchips are likely to be the first applications for the pruned processors.

The inexact design is also integral to the I-slate educational tablet being developed by the Rice-NTU Institute for Sustainable and Applied Infodynamics (ISAID). Intended for Indian classrooms where there is no power, the low-cost tablet is being designed to run on solar power from small panels like those found on solar-powered calculators by using pruned chips that cut power requirements in half.

Earlier this year, Indian officials in the Mahabubnagar District announced plans to put 50,000 I-slates into middle and hig school classrooms over the next three years. Palem expects the first I-slates, along with the first prototype hearing aids, to contain pruned chips will appear by 2013.

The research team, made up of experts from Rice University in Houston, Singapore’s Nanyang Technological University (NTU), Switzerland’s Center for Electronics and Microtechnology (CSEM) and the University of California, Berkeley, unveiled their prototype pruned chips at the ACM International Conference on Computing Frontiers in Calgliari, Italy, this week, where they picked up best-paper honors.

Source: Rice University

14 comments
14 comments
Tiltrotortech
If they link three or more together to use only the most common outputs, they will be able to compensate for most of the 'mistakes' while still being more efficient than a single chip. (If their efficiency numbers are right).
Imran Sheikh
so finally a Turbo for Processor :) doesen't this means that an inExact processing(15x fast) can be achieved in a regular Processor by patching it to Bios level and blocking less used section of it... the benefit it is "its reversible"
Dave B13
Seemed to be a horrible idea at first, then I ran into the use for photos above. Photos seemed like a good idea untill I studied the pictures. I'm very surprised I can tell the 0.54 percent error from the no-error picture. The error of 7.58 percent photo appearance difference is not an acceptable degradation. From this I can imagine using it for video. I think it would be great for robot vision. I can't think of any other application where I'd want a computer part dumbed down for "efficiency".
PrometheusGoneWild.com
Now if they just can make a regular chip that can decide to prune itself when requested. This would allow the programmer to decide when to dumb things down for specific processes..... For example for recording monitoring a security camera the chip could dumb down itself until it sees movement and then kick into high efficiency/low speed/high energy use.
Iván Imhof
Well, I'm not impressed. The third image is terrible, the middle one has less details, less sharp and more faded in colors at the first sight.
The whole IT technology is already loaded with bugs, you can't find any device or software able to do its function properly and flawlessly. What is more frustrating that usually nobody knows exactly what is the reason of the bugs, why it doesn't work like it should, and there are just trial-and-error ways to fix it.
Why we want to have more errors for some more speed? We already accepted that there is no guarantee for any software (read the EULA), and I have no doubt that companies will raise they "error tolerance" level to apply such chips, so we can experience more bugs, but at least they will come faster. :) Great prespective!
cachurro
I think you guys are not getting the picture. Today, our computers have many 'processors'. So adding a couple of inexact cores to a CPU could enable it to increase efficiency on demand by shutting down some of the exact ones. And it would keep the OS and any required program running in an exact core. So don't panic. Another task for these inexact computing is solving non polynomial problems that need heuristic approaches. Like finding the best route for mail delivery, air traffic, inferring phylogeny, etc. These approaches usually consist of many educated trials an errors and keeping the best solution so far, so one mistake will not be significant, and you can always re-check the best solutions. Oh, and these problems use a huge amount of computing power, so this is not a minor breakthrough!
Joel Detrow
Dave, those are frames rendered from video, so yeah, that's the idea. Obviously photos wouldn't necessarily be allowed to make so many mistakes. One thing I wonder about is the resolution of the video - I'm pretty sure the 7.58% error would be greatly muted in higher-resolution video such as 4 or 8K.
I can imagine graphics processing exploding once again with the use of this kind of pruning - enthusiasts who really care about image quality can set it so every pixel is exactly as it should be, but folks who don't care if some pixels are a few hair-shades off could play at substantially higher resolution & framerates. I bet games of the future will have options to set the % error in addition to all the other graphics options we get now. Very cool!
Dan Stillings
Somehow, I am reminded of Rosie on the Jetsons... :-)
Gregg Eshelman
To err is Human. To really screw things up requires a computer.
And here we have the proof, processing chips designed to make errors.
VoiceofReason
@Dave B13 I can think of many applications. Like Dennis said, it can have several chips inside and only use the high powered chip when it needs to. There could also be settings on the device. Some people would gladly accept a 7.58 loss if it meant not having to charge the device once a month instead of every several days.
Reread the article. You fixated on the most pruned chips. You mean to tell us that you won't accept a device that only varies 1/4 of a percent for 3.5 times less power usage?
Load More