Peltier Thermoelectric

The Peltier effect, three sentences in

In 1834, the French watchmaker-turned-physicist Jean Charles Athanase Peltier passed a current through a junction of bismuth and antimony and noticed something the Seebeck effect (1821) did not predict: the junction either heated up or cooled down, depending on the direction of the current. The mechanism is simple in modern language. Charge carriers — electrons in n-type material, holes in p-type — carry not only charge but a quantum of entropy. When a current is forced from p-type into n-type across a junction, both carrier species drift away from the junction together, dragging their entropy with them; the junction therefore loses heat. At the opposite junction, where the carriers arrive, they dump that entropy and the junction warms.

The rate is quantitatively clean:

Q_peltier = Π · I
Π        = α · T          (Kelvin–Onsager relation, 1854 / 1931)

where Π is the Peltier coefficient of the material pair (units: volts), α the Seebeck coefficient (V/K), and T the absolute temperature in kelvin. The same α governs the reverse phenomenon: a temperature difference across the same junction drives an open-circuit voltage V = α·ΔT, which is the Seebeck effect. Peltier and Seebeck are not two effects — they are the same thermoelectric tensor, run in two directions, related by Onsager reciprocity.

The figure of merit ZT

You cannot build a useful module out of any material with a Seebeck coefficient. Three competing properties have to cooperate, and Abram Ioffe captured them in 1949 in the dimensionless figure of merit

ZT = α²σT / κ

Each ingredient pulls a different way:

• α (Seebeck coefficient, V/K). Wants to be large — each carrier should move a lot of entropy per unit charge. Squared, it dominates the numerator.
• σ (electrical conductivity). Wants to be large — ohmic losses in the legs heat the module from inside and waste pumping work.
• κ (thermal conductivity). Wants to be small — once you build a ΔT across the module, ordinary heat conduction tries to short-circuit it. κ = κ_electronic + κ_lattice; the lattice (phonon) piece is what nanostructuring attacks.
• T (absolute temperature). Appears because the available Carnot factor scales with T.

The cruelty is that α, σ, and κ are physically linked. Doping a semiconductor heavier raises σ but suppresses α (Pisarenko relation) and bumps κ_electronic via the Wiedemann-Franz law. Designing a good thermoelectric is therefore a Pareto-frontier exercise. For decades Bi₂Te₃ sat near ZT = 1 at 300 K — a number that limits commercial COP to roughly half of Carnot — and only since ~2000 have nanostructured samples (PbSeTe/PbTe quantum-dot superlattices, Bi₂Te₃/Sb₂Te₃ thin-film stacks, half-Heusler alloys, filled skutterudites) crossed ZT = 2 in the laboratory, with isolated reports above 2.5.

Coefficient of performance — how Peltier compares

For a refrigeration cycle the relevant efficiency is the coefficient of performance:

COP_cooling = Q_cold / W_input

For an ideal Carnot fridge, COP_Carnot = T_cold / (T_hot − T_cold). For a thermoelectric module, Ioffe's analysis gives the maximum COP at the optimal current as

COP_TE,max = (T_cold / ΔT) · [ √(1 + Z·T̄) − T_hot/T_cold ] / [ √(1 + Z·T̄) + 1 ]

where T̄ is the mean temperature. The bracketed factor is what ZT buys you — it equals one only when Z → ∞ (Carnot). For ZT = 1 you recover roughly 17% of Carnot; ZT = 2 raises it to ~30%; ZT = 4 to ~50%. The table compares typical performance:

Cooling technology	Working principle	COP (room temp, small ΔT)	Notes
Domestic refrigerator	Vapor compression (R-600a)	3 – 4	Compressor + working fluid; loudest, largest
Window AC, mini-split	Vapor compression (R-410A)	3 – 5	Same physics, optimised for steady cooling load
Stirling-cycle cryocooler	Gas-cycle regenerative	~1 – 2	Used in IR detector cooling, no working fluid leak
Bi₂Te₃ Peltier (ZT≈1)	Thermoelectric, solid state	0.5 – 1.5	No moving parts, silent, reversible
Multistage cascade Peltier	Thermoelectric, 4-6 stages	0.05 – 0.2	Reaches ΔT ≈ 130 K; mostly for IR detectors
Magnetic / electrocaloric (lab)	Caloric effect on solid	2 – 4 (projected)	Still pre-commercial in 2026

The honest takeaway: thermoelectric cooling is not the right choice for raw efficiency. It wins when one or more of the following matters more than COP — compactness, silence, vacuum survival, sub-second response, the absence of a working fluid, or the ability to reverse direction by flipping the current. CPU coolers, PCR thermocyclers, scientific CCDs, satellite focal planes, and laser-diode mounts all hit at least one of those.

Inside a commercial module

A standard 40 × 40 mm Bi₂Te₃ Peltier module is a flat sandwich:

• Top & bottom ceramic plates. Usually alumina (Al₂O₃) or aluminium nitride (AlN). They electrically insulate the legs from the heat sink and load while passing heat through. AlN, with thermal conductivity ~170 W/m·K against alumina's ~25, is preferred for high-power modules.
• Solder-tinned copper pads. Patterned on the inner faces of the ceramics. Each pad bridges the top (or bottom) of one p-leg to one n-leg, putting them electrically in series.
• Semiconductor legs. 100 to 300 alternating p- and n-type Bi₂Te₃ pellets, typically 1 mm × 1 mm cross-section and 1-2 mm tall. The p-type is Bi₂Te₃ doped with Sb; the n-type is Bi₂Te₃ doped with Se.
• Electrical series, thermal parallel. Current snakes through every leg in series, so the total Peltier pumping is N·Π·I for N couples. But each couple's heat pump runs between the same hot and cold faces, so the heat fluxes add — the module is thermally a single fat thermocouple of effective area equal to the cumulative leg area.
• Two lead wires. Brought out one side of the module, almost always red (positive — cooling on the opposite face).

A typical 127-couple module (TEC1-12706, the most popular hobbyist module on Earth) is rated 12 V, 6 A, pumps ~60 W of heat at ΔT = 0, and produces a no-load ΔT of about 65 K. At maximum ΔT it pumps essentially zero — there is a Q_max vs ΔT_max trade-off built into every module's data sheet.

Multi-stage cascades for big ΔT

The single-stage ΔT ceiling of about 70 K is set by ZT, not by anything you can buy your way out of. To reach lower temperatures, modules are cascaded: stage 1 pumps heat from the cold object to its own hot face; stage 2, sitting on top of stage 1, pumps that hot face to a colder hot-side; and so on. Stage 2 must be smaller (it carries less heat than stage 1), stage 3 smaller still — the cascade has a tapered, stepped pyramid look. Six-stage commercial cascades reach ΔT ≈ 130 K, enough to drop room temperature 25 °C surroundings to roughly -100 °C, which is the sweet spot for cooling astronomical CCDs and infrared detectors without cryogens.

Run it backwards — the Seebeck side and RTGs

The same junction that pumps heat under applied current generates a voltage when heat is dropped across it. This is Seebeck (V = α·ΔT for one couple), and it is the engineering basis of the radioisotope thermoelectric generator (RTG), the workhorse of deep-space power for fifty years.

An RTG is a heat source plus a thermoelectric stack plus a radiator. The standard heat source is a plutonium-238 dioxide ceramic pellet, sized for the mission — Voyager's RTGs carried ~4.5 kg of PuO₂ each, providing ~2 kW of thermal power. Pu-238 alpha-decays with a half-life of 87.7 years, so after fifty years 67% of the original heat is still being delivered. The hot face of the thermoelectric stack rides on the source; the cold face is bolted to a finned radiator pointed at deep space, where blackbody emission at ~300 K is enough to maintain the gradient. Voyager's RTGs used Si-Ge alloy (ZT ≈ 1 at 1000 K hot side), producing 470 W of electrical power at launch in 1977. After 48 years of decay, fission-product buildup, and thermocouple aging, they still produce ~225 W — enough to keep the Plasma Wave Subsystem and a few science instruments alive into the 2030s.

Mission	Launch	RTG type	Initial electrical power
Pioneer 10 / 11	1972 / 1973	4× SNAP-19	~155 W total
Voyager 1 / 2	1977	3× MHW-RTG	~470 W each
Galileo	1989	2× GPHS-RTG	~570 W
Cassini	1997	3× GPHS-RTG	~885 W
New Horizons	2006	1× GPHS-RTG	~245 W
Curiosity (Mars)	2011	1× MMRTG	~125 W
Perseverance (Mars)	2020	1× MMRTG	~110 W
Dragonfly (Titan)	2028 planned	1× MMRTG	~110 W

RTGs are chosen not for efficiency — they convert about 6-7% of thermal to electrical — but because they work for decades with zero moving parts, in radiation environments that would destroy photovoltaics, beyond the Sun's reach where panels can no longer collect, and with a power profile insensitive to dust, eclipse, or temperature swings of hundreds of kelvin. The same property that makes Peltier coolers irreplaceable in CCDs makes Seebeck generators irreplaceable in interplanetary probes.

Where Peltier and Seebeck show up

• CPU and laser-diode coolers. Compact 40 × 40 mm modules clamp under a water block or fin stack to drop a CPU below ambient for overclocking demos, or stabilise the wavelength of a DFB laser to ± 0.01 °C for telecom and metrology.
• PCR thermocyclers. The flagship application in molecular biology. Peltier modules ramp a sample block between ~50 °C, ~72 °C, and ~95 °C in seconds, denaturing and annealing DNA over 30-40 cycles. The reversible direction of pumping is essential — vapor-compression cannot ramp this fast.
• Scientific CCD & infrared detector cooling. Astronomical CCDs (e.g. on the Atacama Cosmology Telescope, Gemini imagers), CMOS sensors in cooled microscope cameras, and HgCdTe IR arrays sit on multi-stage Peltier cascades to reach -50 °C to -100 °C, suppressing dark current.
• Portable refrigerators and beverage coolers. Plug-in camping fridges, in-car coolers, "mini fridges" for cosmetics. The COP is mediocre but the form factor is irresistible at small volumes.
• Dehumidifiers, photonic biosensors. Anywhere a small cold spot is needed in a sealed instrument.
• Waste-heat harvesting. Bismuth-telluride strips on exhaust pipes, automotive turbochargers, and industrial chimneys recover a few percent of the heat as electricity. Most commercial efforts have stalled at marginal economics, but the niche persists for self-powered wireless sensors.
• RTGs in space. Voyager, Cassini, New Horizons, Curiosity, Perseverance, and (planned) Dragonfly all run on Pu-238 + thermoelectric stacks. The cardiac pacemakers of the 1970s used the same physics with Pu-238 batteries.

Designer's checklist and common pitfalls

• Heat-sink the hot side first. Every watt pumped from cold must leave through hot, plus every watt of resistive Joule heating. A 60 W cooler typically dissipates 120-180 W on its hot face. An undersized hot-side sink will not make the cold side cold — it will make both sides hot.
• Mind the optimum current. COP is maximised at a current well below the rated maximum. The data-sheet "I_max" gives maximum ΔT at zero load, not maximum efficiency. Below ~70% of I_max the module is more efficient but pumps less heat.
• Insulate the cold side. Any thermal leak from the warm environment to the cold face is heat the module has to pump in addition to the actual load. A bare cold face condenses water from room air, which then leaks heat back via the puddle — closed-cell foam or vacuum gaps are mandatory below ambient.
• Use thermal paste, not air gap. Each ceramic interface needs ≤ 0.1 mm of high-conductivity paste; an air gap of even 0.05 mm wipes out the module's pumping budget at modest ΔT.
• Beware reverse polarity. Reversing current reverses pumping; you can boil what you were trying to cool. If the application calls for reversal (PCR, polarity-switched temperature controllers), the heat sink must handle both faces being "hot" at different times.
• Cascade efficiently or not at all. Each stage adds a multiplicative COP loss. A 4-stage cascade with stage-COP 0.7 each gives system COP ~0.24. For ΔT under ~70 K, do not cascade.
• Calibrate around ZT(T). ZT is not constant. Bi₂Te₃ peaks near 300 K and falls sharply above 100 °C; PbTe and SiGe replace it for hot-side temperatures above ~500 K. Mixing materials by stage is common in RTGs.

What is changing — nanostructures and beyond

After fifty stagnant years near ZT ≈ 1, the field broke out around 2000 with two ideas. First, quantum confinement: a thin enough film changes the density of states of carriers in ways that boost α without crashing σ; Harman et al. reported PbSeTe/PbTe quantum-dot superlattices with ZT ≈ 1.6 at 300 K (Science, 2002). Second — and more practically — phonon engineering: by introducing nanoscale grain boundaries, point defects, and rattling cage atoms (filled skutterudites, clathrates, half-Heuslers), the lattice thermal conductivity κ_lattice can be suppressed below the amorphous limit while leaving electronic transport intact. Modern half-Heusler alloys reach ZT ≈ 1.5 at 700-900 K; SnSe single crystals have reported ZT > 2.5 at 800 K (Zhao et al., Nature 2014, with subsequent controversy and replication efforts). Polycrystalline bulk samples now routinely report ZT ≈ 1.5 to 2 in laboratories.

The 2026 commercial picture remains conservative: most modules you can buy off the shelf are still Bi₂Te₃ with ZT ≈ 1. The disconnect between laboratory ZT > 2 and shipping product ZT ≈ 1 reflects the difficulty of scaling nanostructured materials reproducibly and the very large embedded base of Bi₂Te₃ tooling. The likely first crossover application is high-temperature waste-heat recovery for trucks and industrial chimneys, where 5-10% conversion efficiency is finally beginning to clear the cost threshold.

Worked example: cooling a small Peltier-driven beverage box

Suppose we want a 10-litre insulated box to sit at 5 °C while ambient is 25 °C — a ΔT of 20 K. The wall conduction load through 5 cm of expanded polystyrene (k ≈ 0.035 W/m·K) over ~0.3 m² of surface is

Q_load = k · A · ΔT / L
       = 0.035 · 0.3 · 20 / 0.05
       ≈ 4.2 W

A single TEC1-12706 module pumps about 30 W at ΔT = 20 K (off its Q vs ΔT curve), so it has a comfortable factor-7 margin. At its sweet-spot operating current of ~3 A (50% of I_max) and 12 V, electrical input is 36 W. COP ≈ 30 / 36 ≈ 0.8 at this operating point — about a quarter of a domestic fridge but adequate for a 10 L box drawing under 40 W. Hot-side dissipation is the sum of pumping plus input: about 66 W. An 80 × 80 mm aluminium heat sink with a 50 mm fan handles this at ΔT_sink ≈ 8 K above ambient, putting the hot face at ~33 °C and the cold face at the desired ~5 °C with margin. This is roughly the design that ships in every camping cooler under $80 today.

Where to read more

• Goldsmid, H. J. Introduction to Thermoelectricity (2nd ed., Springer, 2016) — the standard graduate text.
• Rowe, D. M. (ed.) Thermoelectrics Handbook: Macro to Nano (CRC Press, 2006) — encyclopedic.
• Snyder & Toberer, "Complex thermoelectric materials," Nature Materials 7, 105 (2008) — the canonical modern review.
• NASA RPS office, rps.nasa.gov — current and historical RTG hardware documentation.