Look closely at a computer chip, and you’ll see circuits. Zoom in further, and you’ll find atoms moving around in patterned ways that correspond to the 1’s and 0’s of binary code. These collections of atoms consume energy, perhaps provided by a battery, to perform tasks and execute algorithms — and they produce waste heat.
Yet experts still don’t understand how exactly energy flows in and out of these atoms during computation. To study this, they must figure out how to apply the laws of thermodynamics — the generic rules of heat, temperature, and energy that physicists first used to understand gases and engines — to computers. This August, SFI Professor David Wolpert and other SFI researchers have organized two workshops convening physicists, computer scientists, and biologists to discuss this question. At the workshops, they will discuss how to establish a mathematical language to describe the many microscopic processes that occur during computation.
One practical motivation behind the workshop is that, by pinpointing the physical processes that use or waste the most energy, you can engineer them to achieve higher efficiency. Energy-efficient computers grow increasingly relevant as local governments work to reduce their carbon footprints.
“Right now, five percent of energy consumption in the US comes from computation,” says Wolpert. Furthermore, industries will need strategies to reduce waste heat in future powerful computers known as exascale computers. “These computers would generate so much heat that they would melt,” Wolpert says.
The organizers hope that the multidisciplinary attendees can exchange ideas to inspire new research questions. They’ve invited biologists because many biological systems — if you think about it — are computers too. Cells, for example, receive inputs, execute algorithms, and even know how to repair themselves.
“I’m interested in how well evolution has come up with solutions for energy efficiency for computation in single cells,” says Chris Kempes, an SFI Omidyar Fellow and co-organizer. Kempes’ example involves the microscale of life, but the workshops will span many scales.
For example, Wolpert is interested in discussing a strategy the human brain uses, known as approximate computing: when your brain is sloppy but still gets the job done. If you’re crossing the street, your brain doesn’t need to register the color of every car passing by. “Your brain can be imprecise because there’s no big outcome if you screw up,” Wolpert says.
Approximate computing saves energy, and researchers want to learn how the brain does it and how they can implement similar strategies in computer algorithms. The goal of the workshops is to think about computation from a fresh angle by combining perspectives from multiple disciplines. Physicists know the rules that govern collections of atoms; computer scientists know how to design algorithms; biologists know how organisms function. “We hope there will be a bunch of new questions we haven’t even thought of,” Wolpert says.
Read about the workshops