Errors Make Some Circuits Better
Circuit design is usually no place for mistakes. But new research shows that introducing a controlled quantity of errors into a simple circuit can double its speed while also halving its energy consumption and size.
The researchers behind the work are using the design method to create hearing aids that they hope will have much longer battery lives. The methods could also improve the efficiency of other specialized circuits used in displays and cameras.
Researchers led by Krishna Palem, a professor of computing at Rice University, have designed an algorithm that modifies a circuit’s design to make it more efficient, given a set rate of errors that can be tolerated. Researchers from Palem’s lab presented the work last week at the DATE11 conference in Grenoble, France.
Allowing for a predetermined rate of errors can lead to major efficiency gains without a noticeable drop in performance. As long as the errors are introduced in a controlled way, and the most important parts of an operation are protected from error, small errors are tolerable in many applications—for example, in audio and graphical signal processing. A single such computational mistake might result in a tiny, momentary distortion in an image or sound that most people would not be able to detect.
Lowering the voltage a circuit uses in order to decrease power consumption will introduce errors. When the voltage is lower, some parts of a circuit run slower than the rest, leading to mistakes. Computer scientists have made chips that vary the voltage of different parts of the circuit on the fly. But these designs are complex and increase the size of a chip.
“You can think of a circuit like a network of roads,” says Palem. As information flows through a circuit, some paths have heavy traffic, some hardly any. The Rice group’s algorithm analyzes a circuit to identify paths that can be “pruned,” while only introducing tolerable errors. “We ran audio files through the circuit, and looked for zones of high, medium, and low activity over a series of diagnostic trials,” explains Palem.
The Rice group then collaborated with researchers at the Switzerland Center for Electronics and Microtechnology to fabricate and test the pruned circuits. They found that the new circuit runs twice as fast on half the energy, and had an 8 percent error magnitude. “You get much more back than you give away,” says Palem. This error rate is in the ballpark of what’s tolerable for perceptual tasks such as vision and hearing.
Previous work on allowing errors in circuit design has not been so systematic, says Subhasish Mitra, assistant professor of electrical engineering and computer science at Stanford University. He notes, though, that so far the Rice group has proven the design method with a very simple circuit. Mitra expects that doing this type of design with more complex systems will prove a challenge.
For example, researchers would love to extend battery life in laptops or cell phones using this type of approach. But these devices have complex microprocessors, made up of many circuits integrated into many cores. “When you’re building an overall system, you have to make sure you’re adding value and that the system is robust to withstand the errors,” Mitra says.
Palem hopes to prove the pruned circuit concept in a simple system first: the digital signal processing blocks of a hearing aid. His group is working with neuroscientists at Nanyang Technological University in Singapore who are modeling human hearing in test subjects. “We don’t know yet which information the ear cares about,” says Palem. In about six months, the neuroscience studies will be done and Palem’s group will feed information about the human ear’s error tolerance into the circuit design process. “We hope to have a design by the end of the year,” he says.
Geoffrey Hinton tells us why he’s now scared of the tech he helped build
“I have suddenly switched my views on whether these things are going to be more intelligent than us.”
ChatGPT is going to change education, not destroy it
The narrative around cheating students doesn’t tell the whole story. Meet the teachers who think generative AI could actually make learning better.
Meet the people who use Notion to plan their whole lives
The workplace tool’s appeal extends far beyond organizing work projects. Many users find it’s just as useful for managing their free time.
Learning to code isn’t enough
Historically, learn-to-code efforts have provided opportunities for the few, but new efforts are aiming to be inclusive.
Get the latest updates from
MIT Technology Review
Discover special offers, top stories, upcoming events, and more.