MIT designs computing component that uses waste heat ‘as a form of information’
MIT scientists have published a proof of concept for new analog computer components that could allow electronic devices to process data using the heat they generate.
In a study published on January 29 in the journal Applied physical examinationResearchers have designed microscopic silicon structures that precisely control how heat spreads across a chip’s surface.
The approach represents a form of analog computingin which continuous physical values – in this case, temperature and heat flow – are used to process information instead of binary 1s and 0s.
This technique could be used to detect heat sources and measure temperature changes in electronic devices without increasing power consumption. It would also eliminate the need for multiple temperature sensors that take up space on a chip, the researchers said.
Provided the design can be scaled, the team hopes it can one day be integrated into microelectronic systems to perform high-powered computing tasks, such as artificial intelligence (AI), more energy efficient.
“Most of the time when you’re doing calculations in an electronic device, heat is waste. You often want to eliminate as much heat as possible. But here we took the opposite approach by using heat as a form of information itself and showing that computing with heat is possible,” lead author of the study, Caio Silvastudent in physics at MIT, said in a statement.
The work builds on 2022 MIT research on the design of nanostructured materials capable of controlling heat flow.
Hot chip
As heat flows through the silicon from hotter to cooler regions, the internal geometry of the structures determines how much heat reaches each exit point.
The thermal power at these points can be measured and converted into a standard electrical signal using conventional on-chip sensors. The resulting signal can then be processed by other parts of a system, the scientists explained.
In simulations, the structures performed simple matrix-vector multiplication with over 99% accuracy, the team said in the study.
Matrix multiplication is the basis of many machine learning and signal processing tasks, although the team noted that scaling this approach to large language models (LLMs) would require millions of linked silicon structures to work together.
The team next wants to explore applications in thermal management, heat source detection, and monitoring temperature gradients in microelectronics, where the new structures could prevent chips from being damaged without requiring additional power.
Co-author of the study Giuseppe Romanoresearcher at MIT’s Institute for Soldier Nanotechnologies, added in the release: “We could directly detect such heat sources with these structures, and we can simply plug them in without the need for digital components.”
