Thermoelectrics and the Waste Heat Economy

TL;DR

The US economy rejects about 67% of the primary energy it consumes as waste heat.¹ Industrial flue gas, datacenter exhaust, vehicle tailpipes, gas turbines — all of it. A thermoelectric generator that converts even 5–8% of a 400°C waste stream into electricity flips the economics of dozens of industrial processes. The materials to do it exist in the lab. None of them are in volume production.

The figure of merit ZT has been the gating number for forty years. Bi₂Te₃ holds the commercial ceiling at ZT ≈ 1 near room temperature.² SnSe single crystals reached ZT ≈ 2.6 at 923 K in 2014.³ Half-Heuslers and skutterudites hit ZT ≈ 1.4–1.8 in the mid-temperature range.⁴ The lab record materials are real. The supply chain that turns them into modules is not.

This is a manufacturing problem masquerading as a materials problem.

§1 — Where the heat is

Three streams matter at scale:

1. Industrial process heat (200–600°C): Cement kilns, steel reheat furnaces, glass tanks, chemical reformers. The DOE estimates 1.4 quad/yr of recoverable industrial waste heat in the US alone.⁵
2. Power generation flue gas (300–500°C): Combined-cycle gas plants reject 35–45% of fuel energy as exhaust heat after the steam bottoming cycle. Simple-cycle peakers reject more.
3. Datacenter and HVAC exhaust (40–80°C): Low-grade, but the volume is enormous. A 100 MW datacenter rejects 70–80 MW continuously. Most of it goes through cooling towers or chillers and gets thrown to the atmosphere.

The first two streams are the addressable market for ZT > 1.5 mid-temperature thermoelectrics. The third needs a different material class — narrow-gap semiconductors or organics tuned for ΔT < 50 K — and the economics are tighter.

§2 — Why ZT has stalled

ZT = S²σT/κ. To raise it you increase the Seebeck coefficient S, increase electrical conductivity σ, and decrease thermal conductivity κ — three knobs that fight each other. Heavily doping a semiconductor raises σ but flattens S. Adding nanostructure scattering centers cuts κ but also cuts σ. Every “breakthrough” of the last twenty years has been a clever way to break one of these correlations:

• Nanostructured Bi₂Te₃ superlattices (2001, MIT/RTI): ZT ≈ 2.4 in a thin film by phonon scattering at interfaces.⁶ Never made it past coupon scale.
• PbTe with Tl resonant levels (2008, Heremans): ZT ≈ 1.5 by reshaping the density of states near the Fermi level.⁷
• SnSe single crystals (2014, Zhao): ZT ≈ 2.6 from anomalously low intrinsic κ.³ Hard to grow large single crystals; polycrystal SnSe drops back to ZT ≈ 1.

The pattern: the lab measurement passes peer review, the material can’t be scaled, the result becomes a citation rather than a product.

§3 — The substrate problem

A thermoelectric module is a sandwich. P-type and N-type legs are soldered between two ceramic plates with metalized contacts. The ceramic has to:

1. Hold mechanical tolerance across thermal cycling from cold side to hot side
2. Match the CTE of the leg material closely enough not to crack the solder joints over 10⁵ cycles
3. Conduct heat into the legs efficiently on the hot face and out of them on the cold face
4. Cost less than the legs themselves at volume

Alumina is the default. It’s cheap, it’s available, it doesn’t match anything’s CTE particularly well. AlN is better thermally and matches Si/SiGe legs but costs 5–10× more. The substrate alone is often 30–40% of the module BOM at production scale.⁸

This is the crack that lets a materials company in. If you can deliver a substrate that pairs with a specific leg chemistry and survives the thermal cycling profile of an industrial application, the module manufacturer doesn’t have to redesign their leg material — they just buy your substrate. The leg work has been done. The integration work hasn’t.

§4 — Where the next ten years go

Three things have to happen for thermoelectrics to move from niche (Peltier coolers, RTGs, automotive exhaust harvesters) to grid-relevant:

1. Mid-temperature half-Heuslers in volume. ZrNiSn, HfCoSb, TiNiSn — the chemistry works, the lab data is solid, the production lines don’t exist outside Japan and Germany. Whoever builds the first US half-Heusler line owns the industrial waste-heat market for a decade.
2. A standard substrate for the 300–500°C window. Something between alumina and AlN, ideally with tunable CTE for half-Heusler vs. skutterudite vs. SiGe. This is a materials problem with a clean spec sheet.
3. Automotive as the volume driver. BMW shipped a TEG in a 5-series prototype in 2014.⁹ Nothing made it to production at the time because the cost-per-watt was wrong. With heavy-duty trucks under tighter EPA regulation post-2027, the math changes.

The companies that win are the ones that pair a leg chemistry with a substrate and a thermal-cycling validation package. Selling raw material doesn’t move the needle. Selling a qualified module pair does.

§5 — Why we care

Armadillo Labs is a materials company. Thermoelectrics is one of three pillars where the lab-to-line gap is the binding constraint, not the underlying physics. Our screening tools were built for superconductors, but the same machinery — phonon-mediated transport, electron-phonon coupling, lattice thermal conductivity — runs the thermoelectric problem in reverse. A material that’s bad for a superconductor (high κ, weak coupling, dispersive phonons) is often a candidate for a thermoelectric, and vice versa.

The roadmap is the same: identify the material, validate it computationally, partner with a fab, qualify a module. The customers are different. The supply chain bottlenecks are different. The physics is the same physics.

---

Footnotes

1. Lawrence Livermore National Laboratory, “Estimated U.S. Energy Consumption” Sankey diagrams, annual series. https://flowcharts.llnl.gov/ ↩

2. Goldsmid, H. J. Introduction to Thermoelectricity (Springer, 2nd ed., 2016), Ch. 5. ↩

3. Zhao, L.-D. et al. “Ultralow thermal conductivity and high thermoelectric figure of merit in SnSe crystals.” Nature 508, 373–377 (2014). DOI: 10.1038/nature13184 ↩ ↩²

4. Rogl, G. & Rogl, P. “Skutterudites, a most promising group of thermoelectric materials.” Current Opinion in Green and Sustainable Chemistry 4, 50–57 (2017). ↩

5. U.S. DOE, “Waste Heat Recovery: Technology and Opportunities in U.S. Industry” (BCS Inc., March 2008), Table 2.1. ↩

6. Venkatasubramanian, R. et al. “Thin-film thermoelectric devices with high room-temperature figures of merit.” Nature 413, 597–602 (2001). ↩

7. Heremans, J. P. et al. “Enhancement of thermoelectric efficiency in PbTe by distortion of the electronic density of states.” Science 321, 554–557 (2008). ↩

8. Composite of public BOM disclosures from Marlow Industries, Ferrotec, and Komatsu thermoelectric module datasheets, 2022–2024. ↩

9. BMW Group / DLR collaboration on the Thermoelectric Generator (TEG) for the BMW 5-series, presented at the U.S. DOE Thermoelectrics Applications Workshop, 2014. ↩