'10,000 times faster than a human scientist' — New AI tool designed ultra-efficient heat-to-electricity generators at lightning speed, a breakthrough that could slash the cost of energy harvesters and help enable cheaper, high-performance home heat pumps

3 hours ago 7
TEGNet can design thermoelectric generators (Image credit: National Institute for Materials Science (NIMS)/Future)

  • AI tool speeds thermoelectric generator design while matching leading prototype performance
  • TEGNet cuts simulation time from thousands of seconds to fractions of one
  • Cheaper waste-heat harvesters could follow, although manufacturing still has to prove itself

Researchers in Japan have built an AI tool that can design thermoelectric generators far faster than standard simulation methods, pointing to cheaper ways of turning waste heat into electricity.

TEGNet was developed by Takao Mori and colleagues at Japan’s National Institute for Materials Science (NIMS) and the University of Tsukuba.

In the paper published in Nature, it predicted generator performance with more than 99% accuracy while using only 0.01% of the computing time needed by commercial finite-element solvers.

Article continues below

Acting as a fast emulator

Thermoelectric generators turn heat differences directly into electricity, with no turbines or moving parts.

They already power spacecraft, remote sensors and some isolated infrastructure, but cost and middling performance have kept them out of wider use in factories, refineries, vehicles and electronics.

Designing thermoelectric generators is slow because researchers have to balance materials, geometry, temperature conditions, electrical resistance and heat flow.

A conventional solver has to solve coupled physics equations again and again, which can take days or weeks for broad searches.

Sign up to the TechRadar Pro newsletter to get all the top news, opinion, features and guidance your business needs to succeed!

TEGNet learns from those simulations, then acts as a fast emulator. The paper says a typical material simulation took about 2,237 seconds in COMSOL, while TEGNet produced the same type of result in about 0.25 seconds.

The researchers used the AI to improve two types of generators, one built from stacked layers of different materials and another made from paired semiconductor materials that work together to produce electricity.

Lab-built prototypes reached conversion efficiencies of 9.3% and 8.7%, respectively, putting them among strong reported results for that temperature range.

That still doesn’t make thermoelectrics a cure-all. Heat-to-electricity conversion is capped by basic thermodynamics, and these devices need enough temperature difference to be useful.

The interesting part is cost. Mori told IEEE Spectrum that estimated costs suggest an industrially competitive power-generation cost could be possible “for the first time in thermoelectric history.”

TEGNet also identified designs that could use simpler fabrication and, in some cases, avoid bismuth telluride, a common but expensive thermoelectric material.

That could help waste-heat harvesters and high-performance home heat pumps become cheaper, although real-world manufacturing still has to prove the numbers.


Google logo on a black background next to text reading 'Click to follow TechRadar'

Follow TechRadar on Google News and add us as a preferred source to get our expert news, reviews, and opinion in your feeds.


Wayne Williams is a freelancer writing news for TechRadar Pro. He has been writing about computers, technology, and the web for 30 years. In that time he wrote for most of the UK’s PC magazines, and launched, edited and published a number of them too.

Read Entire Article