This Tiny Chip Could Turn Computers' Waste Heat Into More Performance
A processor that does matrix multiplication calculations using heat rather than electricity could herald a major efficiency gain.
Engineers from the Massachusetts Institute of Technology (MIT) have come up with a device that could turn the waste heat thrown out by electronic computing devices into more compute performance — by doing their calculations thermally, rather than electronically.
"Most of the time, when you are performing computations in an electronic device, heat is the waste product," lead author Caio Silva explains of the project, which leads to a heavy environmental impact as data centers are forced to use vast amounts of electricity and water to cool their compute-dense racks. "You often want to get rid of as much heat as you can. But here, we’ve taken the opposite approach by using heat as a form of information itself and showing that computing with heat is possible."
The team's work builds on a software tool created for "inverse design," in which the functionality is defined and the software figures out how it could be achieved — allowing for the design of tiny silicon structures that can be tailored for specific heat conduction and distribution. The result: an analog processor capable of performing matrix multiplication operations using heat, rather than electricity, with a claimed accuracy over 99%.
"Finding the right topology for a given matrix is challenging," Silva admits of the issues facing the technology in scaling up from proof-of-concept to a device which could sit beside existing electronic processors to turn their waste heat into additional performance. "We beat this problem by developing an optimization algorithm that ensures the topology being developed is as close as possible to the desired matrix without having any weird parts."
Senior author Giuseppe Romano, however, suggests that the heat-processing technology could find a real-world use-case sooner through application to specific tasks like thermal management and temperature gradient detection. "Temperature gradients can cause thermal expansion and damage a circuit or even cause an entire device to fail," Romano explains. "If we have a localized heat source where we don’t want a heat source, it means we have a problem. We could directly detect such heat sources with these structures, and we can just plug them in without needing any digital components."
The team's work has been published under closed-access terms in the journal Physical Review Applied; an open-access preprint is available on Cornell's arXiv server.
Main article image courtesy of Jose-Luis Olivares/MIT.