---
title: "What does a Peltier module make when it’s run as a generator?"
topic: "Peltier generator"
answer: "Simulated on a wrist, with 15 °C between body and room, the module gets only 0.58 °C of it and makes 32 mV with nothing connected, or 67 µW into the best load. Clamped between blocks at its 80 °C rating, it makes 0.75 W, from 37.4 W of heat going through it."
url: "https://nearfieldnotes.com/notes/peltier-generator/"
site: "Near Field Notes"
author: "Michael Walker"
note: 40
family: "Power"
published: "2026-11-15"
revised: "2026-11-15"
revision: 1
video: "https://nearfieldnotes.com/videos/peltier-generator/"
---

Note 40 · [Power](https://nearfieldnotes.com/families/power/) · Rev 1 · [Watch the video · 2:39](https://nearfieldnotes.com/videos/peltier-generator/)

Peltier generator

# What does a Peltier module make when it’s run as a generator?

**Short answer:** Simulated on a wrist, with 15 °C between body and room, the module gets only 0.58 °C of it and makes 32 mV with nothing connected, or 67 µW into the best load. Clamped between blocks at its 80 °C rating, it makes 0.75 W, from 37.4 W of heat going through it.

Run backward, a Peltier cooler is a voltage source: 49 mV for each degree between its blocks, behind about 2 Ω.

What limits it is how much of the temperature difference it gets. On a wrist, the body and the air take nearly all of it.

**Note 40’s opening figure: On a wrist.** The 15 °C from body to room, and the module’s share

![Note 40’s opening figure: On a wrist](https://nearfieldnotes.com/notes/peltier-generator/fig-opening.png)

*What the figure shows:* A Peltier module strapped to a wrist under a small finned radiator, and the 15 degrees C between the body's core at 37 and a 22 degree room drawn as one bar, simulated. The body takes 8.3 degrees of it and the radiator's air side 6.0, leaving the module 0.58 degrees. With nothing connected it makes 32 millivolts. Into the best load it gets 0.45 degrees and gives 16 millivolts and 67 microwatts.

Simulated. A TEC1-12706 under a finned radiator, body core at 37 °C, room at 22 °C. The last step connects the load that takes the most power.

## A Peltier module on a wrist

*TEC1-12706, 40 × 40 mm*

A Peltier module is a grid of small legs between two ceramic faces. Run a current through it and it pumps heat from one face to the other. That is how it works as a cooler.

Hold its two faces at different temperatures instead, and it makes a voltage. That is a [thermoelectric generator](https://nearfieldnotes.com/glossary/#g-thermoelectric-generator).

This one is the TEC1-12706, the common 40 mm cooler, with 127 pairs of legs.

**Fig. 40a: The module, opened.**

![Fig. 40a: The module, opened](https://nearfieldnotes.com/notes/peltier-generator/fig-40a.png)

*What the figure shows:* Figure 40a: the TEC1-12706 module rendered with its top ceramic face lifted off. Underneath, a grid of small upright legs stands on copper tabs on the lower ceramic face, 254 legs in all, 127 pairs, with a red and a black lead at the front corners.

Rendered with the top ceramic face lifted off. The legs alternate between n-type and p-type, joined in series by copper tabs.

Strap it to the outside of a wrist, under a small finned radiator. The body’s core is at 37 °C and the room at 22 °C, so 15 °C sits between them.

Heat leaves the body through the skin, the module and the radiator, one after another, and out into the room.

**Fig. 40b: The wrist, cut away.**

![Fig. 40b: The wrist, cut away](https://nearfieldnotes.com/notes/peltier-generator/fig-40b.png)

*What the figure shows:* Figure 40b: the wrist rendered with a slice cut away. From the bottom: the forearm's cross-section, with its two bones, muscle and a layer of fat and skin; the module's two thin ceramic faces with the legs between them, lying on the skin; an aluminum radiator with upright fins on top; and the strap holding it all. The body's core is at 37 degrees C and the room at 22.

Rendered with a slice cut away. The module lies on the skin, with the radiator on top.

Heat through a stack like that behaves like current through [resistors in series](https://nearfieldnotes.com/glossary/#g-thermal-resistance), with temperature in place of voltage.

Under the module’s 16 cm², the body is 27.5 K/W, from Leonov’s measurements on the outer wrist. The radiator’s air side is 20 K/W. The module itself is only 1.9 K/W.

Electrically, the module is a voltage source behind its own resistance: 54 mV for each degree across its legs, behind 2.1 Ω.

**Fig. 40c: The same wrist, as two circuits.**

![Fig. 40c: The same wrist, as two circuits](https://nearfieldnotes.com/notes/peltier-generator/fig-40c.png)

*What the figure shows:* Figure 40c: the wrist as two circuits. For heat, three resistances in series between the body's core at 37 degrees C and the room at 22: the body, 27.5 kelvin per watt under the module's 16 square centimeters; the module, 1.9; and the radiator's air side, 20. About 305 milliwatts of heat flows through them. For electricity, the module is a voltage source of 54 millivolts for each degree across its legs, behind its own resistance of 2.1 ohms, driving a load.

The heat path’s values are the wrist’s. The module’s resistance is 2.1 Ω at skin temperature and rises as it warms.

## A cooler run backward

*54 mV per °C across the legs*

Warm one face and the charge carriers in the legs move faster at that end. More of them drift toward the cold end than drift back, and they crowd there.

In the n-type legs the carriers are electrons, so the cold end goes negative. In the p-type legs they are holes, so it goes positive. That is the [Seebeck effect](https://nearfieldnotes.com/glossary/#g-seebeck-effect), the cooler’s Peltier effect run the other way.

Each pair gives 0.43 mV for each degree across it. The 127 pairs are in series, so the module gives 54 mV per degree.

**Fig. 40d: One pair of legs.**

![Fig. 40d: One pair of legs](https://nearfieldnotes.com/notes/peltier-generator/fig-40d.png)

*What the figure shows:* Figure 40d: one pair of the module's legs between its cold face above and its hot face below. The n-type leg on the left and the p-type leg on the right are joined by a copper tab at the hot end. Heat comes in at the hot face. In the n-type leg electrons, and in the p-type leg holes, drift toward the cold end and crowd there, so the n leg's cold end is negative and the p leg's positive: 0.43 millivolts for each degree between the faces. 127 pairs in series give 54 millivolts per degree.

A drawing, not to scale. The carriers are drawn far larger, and far fewer, than they are.

On a bench the module sits clamped between two blocks. The joint on each face, grease and ceramic, takes a little of the difference, so between the blocks it reads 49 mV per degree. Harz University measured 49 to 51.

Its resistance is 2.05 Ω at room temperature and 2.33 Ω with 55 °C across it. The legs’ resistance rises as they warm.

The model behind this page is a cooler model of this module, fitted to its datasheet and run backward. One more number is fitted to a bench: 0.10 K/W for the joint on each face. Its voltages land within 2% of Harz’s four measurements.

**Fig. 40e: Voltage with nothing connected.**

![Fig. 40e: Voltage with nothing connected](https://nearfieldnotes.com/notes/peltier-generator/fig-40e.png)

*What the figure shows:* Figure 40e: the module's voltage with nothing connected, against the temperature difference between the two blocks it is clamped between, from 0 to 100 degrees C with the cold block at 25. The model's line rises almost straight, 49 millivolts per degree, to 2.75 V at 55 degrees and 4.94 V at 100. Harz University's four measured points sit on it: 1.99 V at 40, 0.97 V at 20, 0.51 V at 10, 0.25 V at 5 degrees.

Line: the model, with the cold block at 25 °C. Dots: Harz University’s measurements, with the cold side at 30 °C.

**The numbers**

Harz’s cold side was at 30 °C; the model is run at their temperatures. The line on the chart has its cold block at 25 °C.

| Between the blocks | Harz, measured | Model | Model without the joints | Harz, resistance |
|---|---|---|---|---|
| 40 °C | 1.994 V | 2.005 V | 2.239 V | 2.66 Ω |
| 20 °C | 0.975 V | 0.997 V | 1.109 V | 2.52 Ω |
| 10 °C | 0.512 V | 0.496 V | 0.551 V | 2.50 Ω |
| 5 °C | 0.248 V | 0.247 V | 0.275 V | 2.30 Ω |

## At its 80 °C rating

*0.75 W from 37.4 W of heat*

Makers rate a standard cooler like this one for 80 °C. The solder inside melts at 138 °C.

So put its hot face against a block held at 80 °C, and its cold face against one held at 25 °C. That is 55 °C across, the most it is rated for with its cold side near room temperature.

**Fig. 40f: The module on a bench.**

![Fig. 40f: The module on a bench](https://nearfieldnotes.com/notes/peltier-generator/fig-40f.png)

*What the figure shows:* Figure 40f: the module on a bench, rendered from the side. A copper block with a heater in it presses on the module's hot face, held in a frame of two aluminum plates and four threaded rods; the module's other face sits on a finned heatsink with a fan. Thermocouple leads come out of the copper block and the module's red and black leads hang below.

Rendered. In the model the hot block is held at 80 °C and the heatsink at 25 °C.

With nothing connected the module reads 2.75 V. Drawing current pulls the voltage down almost in a straight line, to nothing at 1.09 A shorted.

Power peaks at 0.75 W, at 1.38 V and 0.55 A, half the [open-circuit voltage](https://nearfieldnotes.com/glossary/#g-ocv). The load is then 2.52 Ω, 1.08 times the module’s own resistance.

**Fig. 40g: From open circuit to a short, 55 °C across.**

![Fig. 40g: From open circuit to a short, 55 °C across](https://nearfieldnotes.com/notes/peltier-generator/fig-40g.png)

*What the figure shows:* Figure 40g: the module between a hot block at 80 degrees C and a cold one at 25, simulated, as the current drawn from it goes from nothing to a short. Above, the voltage at its leads falls almost in a straight line from 2.75 V with nothing connected to zero at 1.09 A shorted. Below, the power into the load rises to 0.75 W at 0.55 A and 1.38 V, half the open-circuit voltage, and falls back to zero.

Simulated, blocks at 80 and 25 °C. Each point is a load left on long enough for the temperatures to settle.

**Fig. 40h: Power against temperature difference.**

![Fig. 40h: Power against temperature difference](https://nearfieldnotes.com/notes/peltier-generator/fig-40h.png)

*What the figure shows:* Figure 40h: the most power the module gives, against the temperature difference between its blocks, from 0 to 100 degrees C with the cold block at 25, simulated. It grows as the square of the difference: 0.11 W at 20, 0.75 W at 55, where the hot block reaches the module's 80 degree rating, and 2.2 W at 100. The University of Portland's six measured points, from 0.11 W at 20.6 to 1.17 W at 68.1 degrees, and Harz University's four sit on or near the line.

Line: the model, with the cold block at 25 °C. Marks: two papers’ bench tests of the same part. The model gives 0.81 to 1.03 times Portland’s power and 1.00 to 1.11 times Harz’s.

**The numbers**

The model is run at each paper’s own temperatures and load. Harz’s power is at a load equal to the resistance they measured.

| Test | Measured | Model | Model over measured |
|---|---|---|---|
| Portland, 20.6 °C | 0.11 W | 0.11 W | 1.03 |
| Portland, 26.6 °C | 0.23 W | 0.19 W | 0.81 |
| Portland, 35.3 °C | 0.36 W | 0.32 W | 0.90 |
| Portland, 44.9 °C | 0.55 W | 0.52 W | 0.94 |
| Portland, 55.9 °C | 0.81 W | 0.78 W | 0.97 |
| Portland, 68.1 °C | 1.17 W | 1.13 W | 0.97 |
| Harz, 40 °C | 0.373 W | 0.403 W | 1.08 |
| Harz, 20 °C | 0.094 W | 0.104 W | 1.11 |
| Harz, 10 °C | 0.026 W | 0.026 W | 1.01 |
| Harz, 5 °C | 0.007 W | 0.007 W | 1.00 |

Stock module at its rating

### 0.75 W with 55 °C across

**Field marks.** Hot block at 80 °C, cold side at 25 °C: 2.75 V with nothing connected, and 0.75 W into a 2.52 Ω load.

**Look-alike.** A generator module. It makes power the same way, but it is built for a hot side of 200 °C and more, with solder to match.

**Range.** Power grows as the square of the difference: 0.11 W at 20 °C and 2.2 W at 100 °C. But 100 °C across means a 125 °C hot block, 45 °C past the rating and 13 °C short of the solder’s melting point.

**Habitat.** Warm surfaces under 80 °C. On an 80 °C surface with a fan heatsink it gets 39 °C of the 55 °C and makes 0.51 W. On a plain heatsink, 0.17 W.

## Where the heat goes

*2.0% comes out as electricity*

At the best load, 37.4 W of heat goes in at the hot face. Only 0.75 W comes out as electricity, 2.0%.

Most of the heat simply leaks through the legs, 27.0 W of it. The current carries another 10.3 W across.

These are the cooler’s two costs from [Note 39](https://nearfieldnotes.com/notes/peltier-pwm/) again. The legs conduct heat whether or not they make power, and their resistance turns some of the current back into heat.

**Fig. 40i: Where 37 W of heat goes.**

![Fig. 40i: Where 37 W of heat goes](https://nearfieldnotes.com/notes/peltier-generator/fig-40i.png)

*What the figure shows:* Figure 40i: where the heat goes with 55 degrees C between the blocks and the best load connected, simulated, as two bars to one scale. In at the hot face: 37.4 W, of which 27.0 W leaks straight through the legs and 10.3 W is carried by the current. Out: 36.6 W through the cold face and 0.75 W as electricity, 2.0 percent.

Simulated, best load, blocks at 80 and 25 °C. The two bars are to one scale.

**The numbers**

Best load, blocks at 80 and 25 °C, the legs at 76.1 and 28.8 °C. *S* is the module’s Seebeck coefficient, *K* its legs’ conductance.

|  | Heat |
|---|---|
| Leaks through the legs, *K* Δ*T* | 27.0 W |
| Carried by the current, *S* *I* *T*_h | 10.7 W |
| Half the legs’ own heat, back to the hot face | −0.35 W |
| **In at the hot face** | **37.4 W** |
| Out the cold face | 36.6 W |
| Out as electricity | 0.75 W |

No heat engine can beat the Carnot limit: the temperature difference over the hot side’s absolute temperature. Between 80 and 25 °C that is 15.6%.

The module gets about an eighth of it. The textbook formula for a generator’s best efficiency, with the module’s figure of merit, ZT, of 0.77, gives 2.03%, against the model’s 2.01%.

Hotter only helps so far. At 100 °C across it would reach 3.0%.

**Fig. 40j: Efficiency, against the limit.**

![Fig. 40j: Efficiency, against the limit](https://nearfieldnotes.com/notes/peltier-generator/fig-40j.png)

*What the figure shows:* Figure 40j: the share of the heat through the module that comes out as electricity at the best load, against the temperature difference between the blocks with the cold one at 25 degrees C, simulated, beside the limit for any heat engine working between the same two temperatures. At 55 degrees the module reaches 2.0 percent and the limit is 15.6. At 100 degrees, 3.0 against 25.1.

Simulated, cold block at 25 °C. The limit is for any engine working between the same two block temperatures.

**The numbers**

The limit is the Carnot efficiency, the temperature difference over the hot block’s absolute temperature.

| Between the blocks | This module | Limit | Share of the limit |
|---|---|---|---|
| 10 °C | 0.42% | 3.2% | 13% |
| 20 °C | 0.82% | 6.3% | 13% |
| 40 °C | 1.54% | 11.8% | 13% |
| 55 °C | 2.01% | 15.6% | 13% |
| 80 °C | 2.64% | 21.2% | 12% |
| 100 °C | 3.00% | 25.1% | 12% |

## Drawing current brakes it

*28 W, 37 W and 46 W*

Drawing current makes the module carry more heat. With both blocks held 55 °C apart, 28 W goes through it open, 37 W at the best load and 46 W shorted.

The current carries heat from the hot face to the cold one, as it did in the cooler, and here that is the way the heat already flows. A motor turned as a generator does the same thing: draw current and its shaft gets harder to turn.

Where nothing holds the hot side, that extra heat flow cools it. With a fixed 40 W into the hot face and a fan heatsink on the other, the temperature across the module falls from 63 °C open to 51 °C at the best load and 39 °C shorted.

**Fig. 40k: The heat and the temperature, as the current grows.**

![Fig. 40k: The heat and the temperature, as the current grows](https://nearfieldnotes.com/notes/peltier-generator/fig-40k.png)

*What the figure shows:* Figure 40k: what drawing current does to the heat, simulated, from nothing connected to a short. Above, with both blocks held 55 degrees C apart, the heat through the module climbs from 28.1 W open to 37.4 W at the best load and 45.9 W shorted. Below, with a fixed 40 W into the hot face and the fan heatsink on the other, the temperature across the module falls from 63.0 degrees open to 50.8 at the best load and 38.8 shorted.

Simulated. The fixed 40 W puts the hot face at up to 104 °C, past the module’s rating, to show the shape. Montecucco and colleagues found the same sag on a generator module fed a fixed 156 W: 195 °C across open, 155 °C at the best load and 124 °C shorted.

So where the heat is fixed, the best load moves up. Held blocks want 1.08 times the module’s resistance. The fixed 40 W wants 1.59, and the wrist 1.80.

A load equal to the module’s resistance, the textbook [matched load](https://nearfieldnotes.com/glossary/#g-matched-load), then gives up 5% and 8% of the power. In all three cases the best point sits at half the voltage the module settles to with nothing connected.

Harvester chips work from the open-circuit voltage too. TI’s BQ25504 stops drawing for 256 ms every 16 s to sample it, then holds the module at a set share of it, typically 50% for a thermoelectric harvester. A quarter of a second is too short for the temperatures to settle, so where the heat is fixed it draws a little more current than the best. Montecucco and colleagues found that costs 2.9% of the power.

**Fig. 40l: Power against load.**

![Fig. 40l: Power against load](https://nearfieldnotes.com/notes/peltier-generator/fig-40l.png)

*What the figure shows:* Figure 40l: the power the module gives as a share of its best, against the load's resistance over the module's own, on a log scale, for three ways of holding the heat, simulated. Both blocks held: best at 1.08 times the module's resistance; a load equal to it gives 99.8 percent. Fixed 40 W: best at 1.59 times the module's resistance; a load equal to it gives 94.8 percent. Wrist: best at 1.80 times the module's resistance; a load equal to it gives 91.9 percent.

Simulated. The dots mark each case’s best load.

**The numbers**

Both blocks held 55 °C apart; a fixed 40 W into the hot face with the fan heatsink on the other; the wrist with the finned radiator.

| Heat held by | Best load over its resistance | Best voltage over settled open-circuit | Lost at a load equal to it |
|---|---|---|---|
| Both blocks held | 1.08 | 0.500 | 0.2% |
| Fixed 40 W | 1.59 | 0.495 | 5.2% |
| Wrist | 1.80 | 0.500 | 8.1% |

## Where the 15 °C goes

*3% to the module at the best load*

Back on the wrist, the three resistances share the 15 °C in proportion to their size. With nothing connected, the body takes 8.3 °C, the air side 6.0 °C and the module 0.58 °C, which makes 32 mV.

The module is 1.9 K/W against 47.7 K/W for everything else in its path, so it gets only a sliver: about 2.1 mV for each degree between body and room. Connect the best load and its share falls to 0.45 °C, because the current carries heat across it.

A better radiator moves some of the drop onto the module. Bare, it gets 0.24 °C at the best load and makes 19 µW. With the finned radiator, 0.45 °C and 67 µW. Walking, with air moving past the fins, 0.57 °C and 107 µW.

With the finned radiator, 19 mW/cm² of heat flows out through the skin under it, the average Leonov measured on wrists with a radiator.

**Fig. 40m: Where 15 °C goes on a wrist.**

![Fig. 40m: Where 15 °C goes on a wrist](https://nearfieldnotes.com/notes/peltier-generator/fig-40m.png)

*What the figure shows:* Figure 40m: how the 15 degrees C from body core to room splits on the wrist with the best load connected, simulated, for three air sides, each a bar from the core on the left to the room on the right. Bare module, air side 62 K/W: body 4.5 degrees, module 0.24, air side 10.2; 19 microwatts. Finned radiator, air side 20 K/W: body 8.4 degrees, module 0.45, air side 6.1; 67 microwatts. Walking, air side 10 K/W: body 10.5 degrees, module 0.57, air side 3.8; 107 microwatts.

Simulated, with the best load connected. The body takes 8.4 °C and the air side 6.1 °C with the finned radiator, a little more than with nothing connected.

**The numbers**

Temperatures at the best load. Body 27.5 K/W under the module’s 16 cm².

| Air side | Its resistance | Body | Both joints | Module | Air side | Open circuit | Best load | Heat, mW/cm² |
|---|---|---|---|---|---|---|---|---|
| Bare module | 62 K/W | 4.53 °C | 0.03 °C | 0.24 °C | 10.20 °C | 17.1 mV | 19 µW | 10.3 |
| Finned radiator | 20 K/W | 8.39 °C | 0.06 °C | 0.45 °C | 6.10 °C | 31.7 mV | 67 µW | 19.1 |
| Walking | 10 K/W | 10.52 °C | 0.08 °C | 0.57 °C | 3.83 °C | 39.8 mV | 107 µW | 23.9 |

The most power comes when the module’s thermal resistance is about the size of everything else in its path, so that it takes about half the drop. Snyder’s chapter on energy harvesting gives the same rule.

A stock module is far too small a resistance for a wrist. Longer legs make it larger.

**Fig. 40n: Longer legs, in three places.**

![Fig. 40n: Longer legs, in three places](https://nearfieldnotes.com/notes/peltier-generator/fig-40n.png)

*What the figure shows:* Figure 40n: the most power the module would give with its legs made longer or shorter, as a multiple of the stock module's, against leg length from a tenth to 300 times the stock legs on a log scale, simulated, in three places. Wrist: best at 34 times as long, 8.71 times the stock module's power. 80 °C surface, fan heatsink: best at 0.42 times as long, 1.21 times the stock module's power. 80 °C surface, plain heatsink: best at 1.5 times as long, 1.04 times the stock module's power.

Simulated: the wrist with the finned radiator, and an 80 °C surface with a fan heatsink or a plain one. The same 127 pairs, each leg made longer or shorter, its resistance and thermal resistance scaled to match.

**The numbers**

Longer legs: the same 127 pairs, each leg’s resistance and thermal resistance scaled with its length. The surfaces are held at 80 °C in a 25 °C room.

| Where | Everything else in the path | Best legs | Module then | Stock module | Best legs give | Stock’s share of best |
|---|---|---|---|---|---|---|
| Wrist | 47.71 K/W | 34× | 64.21 K/W | 67 µW | 584 µW | 11% |
| 80 °C surface, fan heatsink | 0.51 K/W | 0.42× | 0.71 K/W | 0.51 W | 0.62 W | 83% |
| 80 °C surface, plain heatsink | 1.71 K/W | 1.5× | 2.49 K/W | 0.17 W | 0.18 W | 96% |

Module with longer legs

### 8.7 times the power on a wrist

**Field marks.** With each leg 33.5 times as long, the module rises to 64 K/W, a little more than the rest of the path, and takes 7.5 °C of the 15 °C. It gives 584 µW at 235 mV.

**Look-alike.** A module built for small temperature differences, like the one in Citizen’s watch: many thin legs, packed tight. That raises its thermal resistance and its voltage per degree together.

**Range.** Longer legs pay only where the rest of the path is a large resistance. On an 80 °C surface the stock legs already give 83% of the best with a fan heatsink and 96% with a plain one.

**Habitat.** Body heat. Citizen’s 1999 watch packed 2,268 pairs of legs into about a square centimeter, for 0.8 V per degree. On a wrist it saw about 1.2 °C and made 12 µW on average. A quartz movement needs about 1 µW.

## Stretch the legs yourself

*0.3 to 300 times as long*

Set the legs’ length, on the wrist with the finned radiator. Each length is its own run.

Short legs get almost none of the drop. Very long ones take most of it, but their own resistance eats the power. The best is 33.5 times the stock length.

**Note 40’s explorer: Legs on a wrist.**

![Note 40’s explorer: Legs on a wrist](https://nearfieldnotes.com/notes/peltier-generator/fig-explore.png)

*What the figure shows:* The module on the wrist with its legs made longer, at the chosen length: its power as a multiple of the stock module's along the line of all lengths, and the 15 degrees C from body core to room split between the body, the module and the air side.

Each length is its own run at the best load. Above, the power against leg length; below, where the 15 °C goes.

**The numbers**

On the wrist with the finned radiator. Temperatures and power at the best load.

| Legs | Across the legs | Module | Open circuit | Best load | Best load’s resistance |
|---|---|---|---|---|---|
| 0.3× as long | 0.14 °C | 0.6 K/W | 10 mV | 21 µW | 1.1 Ω |
| 0.5× as long | 0.23 °C | 1.0 K/W | 16 mV | 35 µW | 1.9 Ω |
| Stock legs | 0.45 °C | 1.9 K/W | 32 mV | 67 µW | 3.7 Ω |
| 2× as long | 0.87 °C | 3.8 K/W | 61 mV | 127 µW | 7.4 Ω |
| 3× as long | 1.3 °C | 5.8 K/W | 88 mV | 179 µW | 10.9 Ω |
| 5× as long | 2.0 °C | 9.6 K/W | 137 mV | 267 µW | 17.6 Ω |
| 10× as long | 3.5 °C | 19.2 K/W | 235 mV | 416 µW | 33.1 Ω |
| 20× as long | 5.6 °C | 38.4 K/W | 365 mV | 548 µW | 60.8 Ω |
| 33.5× as long | 7.5 °C | 64.2 K/W | 470 mV | 584 µW | 94.5 Ω |
| 50× as long | 8.9 °C | 95.8 K/W | 547 mV | 562 µW | 133.0 Ω |
| 100× as long | 11 °C | 191.4 K/W | 656 mV | 442 µW | 243.3 Ω |
| 200× as long | 13 °C | 382.7 K/W | 729 mV | 291 µW | 456.3 Ω |
| 300× as long | 13 °C | 574.0 K/W | 757 mV | 215 µW | 666.9 Ω |

## Starting from 20 mV

*LTC3108*

A stock module on a wrist makes about 30 mV. An ordinary harvesting boost converter can’t start from that. TI’s BQ25504, BQ25505 and BQ25570 need about 600 mV for a [cold start](https://nearfieldnotes.com/glossary/#g-cold-start).

The LTC3108 starts from 20 mV typical, 50 mV at most. Its switch and a 1:100 step-up transformer form an oscillator, with 330 pF from the secondary back to the switch’s gate.

The switch conducts with no voltage on its gate, so current flows in the primary before any supply exists. Chipworks’ teardown found it to be a depletion-mode transistor. The datasheet says only that the chip starts from 20 mV.

**Fig. 40o: The LTC3108 on the module.**

![Fig. 40o: The LTC3108 on the module](https://nearfieldnotes.com/notes/peltier-generator/fig-40o.png)

*What the figure shows:* Figure 40o: the LTC3108 on the module, after the datasheet's application circuit. The module, a source behind about 2 ohms, feeds the primary of a 1 to 100 step-up transformer, whose other end goes to the chip's SW pin, its internal switch to ground. One end of the secondary is grounded; the other feeds the chip's C1 pin, its charge pump, through 1 nanofarad, and its C2 pin, the switch's gate, through 330 picofarads. The chip's VOUT gives 3.3 V.

After the datasheet’s application circuit, which also puts 220 µF across the module, 1 µF on VAUX and 470 µF on VOUT.

On this module a typical part starts with 0.55 °C between the blocks, 27 mV with nothing connected. A worst-case part needs 1.6 °C. At 2 °C it delivers 0.39 mW of the 1.1 mW the module could give.

On the wrist it only just starts. A typical part sees 20 mV at its pin and delivers 12.6 µW. A worst-case part wouldn’t start.

The 1:100 transformer suits the lowest voltages. At 10 °C it delivers 2.6 mW of 27 mW, and the datasheet points to a lower ratio for higher input voltages.

**Fig. 40p: What the LTC3108 delivers.**

![Fig. 40p: What the LTC3108 delivers](https://nearfieldnotes.com/notes/peltier-generator/fig-40p.png)

*What the figure shows:* Figure 40p: the power the module could give into its best load, and the power the LTC3108 delivers from it with a 1 to 100 transformer, against the temperature difference between the blocks from 0.1 to 10 degrees C, both on log scales, from the datasheet's typical curves. The chip starts at 0.55 degrees with a typical part and 1.6 with a worst-case one. At 2 degrees it delivers 0.39 milliwatts of the 1.08 available; at 10 degrees, 2.6 of 26.9.

From the datasheet’s typical curves for the 1:100 transformer, read by eye, with the module between the bench’s blocks.

**The numbers**

From the datasheet’s typical curves for the 1:100 transformer, read by eye.

| Between the blocks | Open circuit | At the chip | Chip’s input resistance | Available | Out of the LTC3108 |
|---|---|---|---|---|---|
| 0.5 °C | 24 mV | 18 mV | 6.5 Ω | 68 µW | Doesn’t start |
| 1.0 °C | 48 mV | 34 mV | 5.0 Ω | 272 µW | 85 µW |
| 2.0 °C | 95 mV | 60 mV | 3.6 Ω | 1.1 mW | 394 µW |
| 5.0 °C | 238 mV | 132 mV | 2.6 Ω | 6.8 mW | 1.6 mW |
| 10.0 °C | 476 mV | 264 mV | 2.6 Ω | 27 mW | 2.6 mW |

## Where it wins

*A few percent*

A few percent is as good as a thermoelectric generator gets. Marlow’s TG12-4, a module built as a generator, reaches 4.97% with its hot side at 230 °C.

So they win where nothing else can run. More than 35,000 burn gas along pipelines and at wellheads, turning 2.3 to 3.1% of the fuel’s heat into electricity. A review by Da Lio and Lazzaretto gives them 20 years of life.

Voyager 1 left in 1977 with three radioisotope generators making 470 W between them, at 6.5%. By 2023 NASA put it at 225 W.

**Fig. 40q: How good a thermoelectric generator gets.**

![Fig. 40q: How good a thermoelectric generator gets](https://nearfieldnotes.com/notes/peltier-generator/fig-40q.png)

*What the figure shows:* Figure 40q: how much of the heat comes out as electricity in four thermoelectric generators. This module with its hot block at its 80 degree C rating: 2.0 percent, simulated. Gas-fired generators on pipelines: 2.3 to 3.1 percent. A Marlow TG12-4 generator module with its hot side at 230 degrees: 4.97 percent. Voyager's radioisotope generators at launch: 6.5 percent.

From the makers’ sheets, JPL’s reference book and, for this module, the simulation. Their hot sides differ: this module stops at 80 °C.

## Will a stock module do?

*Three questions*

It comes down to how much of the temperature difference the module gets, how hot its hot side runs, and how many millivolts it makes.

Key to a Peltier module as a generator

### What sets its power, and what reads it?

1. 1a Large heatsinks or blocks on both faces Go to 2
2. 1b Skin, still air or a small radiator in the heat’s path Go to 3
3. 2a The hot side stays under 80 °C Stock module, loaded at 1.1 to 1.6 times its resistance
4. 2b The hot side runs hotter A generator module rated for it
5. 3a Tens of millivolts from the module LTC3108 with a 1:100 transformer, or legs made for it
6. 3b More than about 0.6 V A harvesting boost converter held at half the open-circuit voltage

A cooler’s solder melts at 138 °C, and generator modules are built for 200 °C and more. The bench in Fig. 40f ends at 2a, and the wrist in Fig. 40b at 3a.

#### Conditions

- **Simulated:** Every voltage, power, heat flow and temperature on this page comes from a model of the module, checked against two published bench tests of the same part. None was measured on a bench for this page.
- **Module:** TEC1-12706, 127 pairs. A standard cooler model, fitted to its datasheet, with Ferrotec’s curves for how the legs’ properties change above room temperature. One fitted number: 0.10 K/W for the joint on each face, grease and ceramic. “Across the legs” leaves those joints out; they take 0.03 °C each on the wrist.
- **Check:** Harz University, 2024: voltage within 2% of four points from 5 to 40 °C, and power 1.00 to 1.11 times theirs. University of Portland, 2012: power 0.81 to 1.03 times theirs, and 2.41% efficiency against their 2.22% at 68.1 °C, which counts the heater’s whole power. Without the joints the model reads 12% high on voltage.
- **Bench:** Blocks held at 80 and 25 °C unless a figure says otherwise. The fixed-heat case puts 40 W into the hot face, with a 0.30 K/W fan heatsink on the cold face in a 25 °C room. The hot surfaces are held at 80 °C, with the same fan heatsink or a 1.5 K/W plain one.
- **Wrist:** Body core 37 °C, room 22 °C. The body under the module’s 16 cm²: 27.5 K/W, from Leonov’s 440 cm²·K/W for the outer wrist. The air side: 62 K/W bare, 20 K/W with the finned radiator, 10 K/W walking.
- **LTC3108:** Its input resistance and efficiency against input voltage, read by eye from the datasheet’s typical curves for the 1:100 transformer. It starts at 20 mV on its pin, typical, or 50 mV worst case.
- **Render:** Blender. The wrist and the bench are drawn for this page.

Model results, checked against published bench tests. Nothing on this page was measured on a bench for it.

#### Limits

- The joint resistance is fitted to Harz’s bench. A different clamp, grease or block would give a different number, and every bench figure moves with it.
- Ferrotec’s curves come from a 71-pair module, scaled to this one at room temperature. The datasheet gives 2.0 Ω ±10% for the TEC1-12706, and Harz measured 2.3 to 2.7 Ω on theirs.
- The fixed 40 W case runs the hot face past the 80 °C rating, up to 104 °C with nothing connected. It is there for the shape, not as a way to run the part.
- The wrist uses one body resistance, Leonov’s average, and still air values for each radiator. Real skin changes with blood flow, the room and the person.
- Longer legs keep the same material, count and faces. A real harvesting module also changes the legs’ area, how tightly they pack, and its faces.
- The LTC3108 curves are read by eye, and its start points are the datasheet’s typical and worst-case values only.
- The pipeline, Marlow and Voyager efficiencies are their makers’ and operators’ published numbers, at hot sides far above this module’s.

#### Files

- sim/teg.py The module as a generator, the check against the two papers, the bench, the hot surfaces, the fixed heat, the wrist, longer legs and the LTC3108
- sim/site.py The heat taken apart, every interface on the wrist, power against load and the leg lengths of the explorer
- render/wrist.py The wrist and its cutaway in Blender
- render/bench.py The bench in Blender

#### Sources

- [Mecke and Kußmann, Harz University, “Thermoelectric generator for energy production from renewable sources”, RE&PQJ 22, 2024: a TEC1-12706’s open-circuit voltage and resistance from 5 to 40 °C, measured](https://repqj.com/index.php/repqj/article/download/3932/3580/8503)
- [O’Halloran and Rodrigues, University of Portland, “Power and Efficiency Measurement in a Thermoelectric Generator”, ASEE 2012: a TEC1-12706’s power and efficiency from 20.6 to 68.1 °C, measured](https://peer.asee.org/power-and-efficiency-measurement-in-a-thermoelectric-generator)
- [TEC1-12706 datasheet, Conrad item 189115: 127 pairs, 40 × 40 mm, 2.0 Ω ±10%, the ratings the model is fitted to, and BiSn solder at 138 °C in its standard version](https://asset.conrad.com/media10/add/160267/c1/-/en/000189115DS02/datablad-189115-tru-components-tec1-12706-peltier-element-15-v-64-a-65-w-l-x-b-x-h-40-x-40-x-38-mm.pdf); [Thermonamic’s copy](https://dessy.ru/include/images/ware/pdf/RDKT0083.PDF) gives the same, with the resistance as 1.8 to 2.2 Ω
- [TE Technology, FAQ: standard modules rated for 80 °C](https://tetech.com/faqs/)
- [Ferrotec, Thermoelectric Technical Reference, “Mathematical Modeling of Modules”: curves for a module’s Seebeck coefficient, resistance and thermal conductance from −100 to 150 °C, taken from a 71-pair module](https://thermal.ferrotec.com/technology/thermoelectric-reference-guide/thermalref11/)
- [Leonov, “Human Machine and Thermoelectric Energy Scavenging for Wearable Devices”, ISRN Renewable Energy, 2011: 440 cm²·K/W under a device on the outer wrist, and 19 mW/cm² through a wrist device with a radiator](https://doi.org/10.5402/2011/785380)
- [Watanabe, Murakami and Yamada, Citizen, “Development of Eco-Drive Thermo”, Micromechatronics 44(1), 2000: 2,268 pairs, 0.8 V per degree, about 1.2 °C on a wrist, 12 µW on average, and a quartz watch’s 1 µW](https://www.jstage.jst.go.jp/article/micromechatronics/44/1/44_KJ00001887786/_pdf/-char/en)
- [Snyder, “Thermoelectric Energy Harvesting”, book chapter: matched thermal resistance, with half the temperature difference across the module](https://thermoelectrics.matsci.northwestern.edu/publications/pdf/TEEnergyHarvestingBookChapter.pdf)
- [Montecucco, Siviter and Knox, “Constant heat characterisation and geometrical optimisation of thermoelectric generators”, Applied Energy 149, 2015: 195, 155 and 124 °C across at a fixed 156 W, the best point near half the settled open-circuit voltage, and 2.9% less power held at half a quick sample of it](https://eprints.gla.ac.uk/107320/1/107320.pdf)
- [Analog Devices LTC3108 datasheet, rev. D: start from 20 mV typical and 50 mV maximum, the 1:100 transformer, the application circuit and the typical curves](https://www.analog.com/media/en/technical-documentation/data-sheets/LTC3108.pdf)
- [Allan, “Chipworks Analysis: Linear Technology’s LTC3108 DC-DC Converter For Energy Harvesting Apps”, Electronic Design, 2012: the depletion-mode switch](https://www.electronicdesign.com/markets/energy/article/21767854/chipworks-analysis-linear-technologys-ltc3108-dcdc-converter-for-energy-harvesting-apps)
- [TI BQ25504 datasheet: the 600 mV cold start, the open-circuit sample of 256 ms every 16 s, and the 50% setting for thermoelectric harvesters; the BQ25505 and BQ25570 sheets give the same cold start](https://www.ti.com/lit/ds/symlink/bq25504.pdf)
- [Marlow TG12-4 datasheet: 4.97% with its hot side at 230 °C and cold side at 50 °C](https://www.coherent.com/resources/datasheet/materials/single-stage-thermoelectric-generator-tg12-4-ds.pdf)
- [Global Power Technologies, overview, 2018: more than 35,000 generators deployed](https://www.lincenergysystems.com/images/GPT-Overview-2018-c.pdf)
- [Global Power Technologies, model 5030 generator sheet](https://www.lincenergysystems.com/images/Model-5030-Specification-Sheet-c.pdf), with the [5060](https://www.lincenergysystems.com/images/Model-5060-Specification-Sheet-c.pdf), [5120](https://www.lincenergysystems.com/images/Model-5120-Specification-Sheet-c.pdf) and [5220](https://www.lincenergysystems.com/images/Model-5220-Specification-Sheet-c.pdf) sheets: each model’s top rated power against its gas use gives 2.3 to 3.1% of the gas’s heat
- [Da Lio and Lazzaretto, “Remote Power Generation for Applications to Natural Gas Grid”, Energies 15, 2022: thermoelectric generators at 2 to 3%, lasting 20 years](https://psecommunity.org/wp-content/plugins/wpor/includes/file/2302/LAPSE-2023.11698-1v1.pdf)
- [Idaho National Laboratory, “Atomic Power in Space II”, 2015: Voyager 1’s three generators, 470.1 W at launch](https://www.osti.gov/servlets/purl/1893110)
- [Lee and Bairstow, JPL, “Radioisotope Power Systems Reference Book for Mission Designers and Planners”, JPL Publication 15-6, 2015: the MHW-RTG flown on Voyager, 6.5% efficient at the start of the mission](https://ntrs.nasa.gov/api/citations/20160001769/downloads/20160001769.pdf)
- [NASA, Voyager spacecraft: 225 W in 2023](https://science.nasa.gov/mission/voyager/spacecraft/)

What should the next note be? [Ask for one](https://nearfieldnotes.com/#ask-for-a-note)

---

Source: https://nearfieldnotes.com/notes/peltier-generator/

Cite as: Michael Walker, “Peltier generator”, Near Field Notes, Note 40, Rev 1, 15 Nov 2026. https://nearfieldnotes.com/notes/peltier-generator/

Figures: free to reuse under CC BY 4.0 (https://nearfieldnotes.com/license/). Each one's image is linked under its name. Credit Michael Walker, Near Field Notes, and link the figure: this address plus #fig- and its number, like https://nearfieldnotes.com/notes/peltier-generator/#fig-40a.
