Skip to content
Artificial intelligence. Clearly. RSS feed
AI NEWS AI

News. Evidence. Perspective.

ResearchExplainer

AI-designed generators turn heat into electricity with unequal legs

TEGNet helps design working generators by changing the proportions and combinations of their semiconductor legs.

A small generator with unequal semiconductor legs converted 8.7% of its incoming heat into electricity in laboratory measurements. Its hot side was held at 593 K (about 320°C) and its cold side at 293 K (about 20°C); the tests ran under vacuum. The proportions came from an AI design system called TEGNet.[1]

A thermoelectric generator uses a temperature difference to produce electricity without moving parts. Its semiconductor legs carry electrical current and conduct heat at the same time. Changing their geometry therefore changes two jobs together. A longer leg impedes heat flow; changing cross-sectional area changes the electrical and thermal behaviour. Choosing dimensions belongs alongside choosing the material.[2]

TEGNet learns from detailed simulations to predict voltage and the heat leaving a leg’s cold end. Separate models represent different materials. The researchers combine them using conservation of energy and the requirement that connected sections carry the same current, allowing rapid searches through complete device designs.[1]

Why combine materials? A generator spans a temperature range, so its components must operate together across that range. Jeffrey Snyder and Tristan Ursell formalised this problem in 2003 through a property called the compatibility factor. Their analysis showed that materials with dissimilar factors cannot simply be joined into an efficient segmented generator. A material’s individual thermoelectric merit does not settle how well it will cooperate with its neighbours.[3]

Earlier three-dimensional modelling made the geometry issue concrete. A 2016 study found the best efficiency in its tested combinations when the two types of semiconductor leg had different cross-sectional areas. The geometry producing the highest efficiency also differed from the one producing the most power per unit area. Maximising the fraction of heat recovered and maximising watts from a limited footprint are separate engineering objectives.[2]

The 2016 work also examined losses at material junctions. Those junctions add electrical and thermal resistance and can shift the temperatures experienced by each segment. Careful assembly and suitable dimensions help the constituent materials deliver their performance as a working generator.[2]

The fabricated two-pair generator used square legs 3.3 and 2.9 millimetres wide, both 6.6 millimetres long. A separate segmented device reached 9.3% efficiency under the same endpoint temperatures. Those laboratory results give the design search a physical outcome: a better combination of materials, dimensions and operating conditions.[1]

Follow the evidence

Sources & further reading

  1. Composable thermoelectric neural emulators Nature · 15 Apr 2026 · Accessed 20 Oct 2026
  2. Segmented thermoelectric generator modelling Scientific Reports · 7 Apr 2016 · Accessed 20 Oct 2026
  3. Thermoelectric efficiency and compatibility Physical Review Letters · 2 Oct 2003 · Accessed 20 Oct 2026