Story perspectives

Revolutionary Thermodynamic Computing Harnesses Heat for Machine Learning

3/7/2026

Summary

Story summary

Researchers at Lawrence Berkeley National Laboratory developed a framework for thermodynamic computing that uses thermal noise as power for machine learning.
The approach lets computers operate at room temperature by exploiting random vibrations of electrons.
The team trained thermodynamic computers with digital simulations to perform complex calculations.
A genetic algorithm identified optimal parameters, enabling low-energy inference once hardware is built.

1 of 1

Story summary

Researchers at Lawrence Berkeley National Laboratory developed a framework for thermodynamic computing that uses thermal noise as power for machine learning.
The approach lets computers operate at room temperature by exploiting random vibrations of electrons.
The team trained thermodynamic computers with digital simulations to perform complex calculations.
A genetic algorithm identified optimal parameters, enabling low-energy inference once hardware is built.