Remote thermoelectric cooler/generator

You are probably familiar with thermocouples, thermopiles, and their more recent applications as thermoelectric coolers and generators (using the Peltier and Seebeck effects, respectively). (If not, check out links [1]–[5].) These devices take a difference in temperature between two nearby locations and turn it into a voltage difference between two terminals, or vice versa. They work by using junctions between dissimilar electrical conductors, which just happen to have this property. Thermocouples are generally used just for temperature measurement, because their output voltage is too low to use for power.

But you can put many thermocouples in series to make a thermopile, which has a useful output voltage, in which case you can use it as a thermoelectric generator for powering low-power devices. And, instead of applying heat to one thermal terminal and cold to the other, you can apply voltage to its electrical terminals, causing it to pump heat from one thermal terminal to another, resulting in a thermoelectric cooler, which is usually used for cooling things on one side, but can be reversed with a switch to heat them instead.

While thermocouples used for temperature measurement can be and commonly are arbitrarily long (tautology?) without any significant degradation in performance [6][7], thermopiles (even those used for temperature or heat flux measurement rather than power transduction) are not, AFAICT (from the lack of Google search results for queries like "long thermopile" and "remote thermoelectric (cooler|generator)"), but I see no reason why they couldn't be. For a metal thermopile, composed of several pieces of thermocouple wire connected appropriately, you could just build it out of longer pieces of thermocouple wire, so you can put the thermal terminals farther apart. (Both electrical terminals can easily, and by default would, still be at the same end.)

But modern TECs/TEGs are made of semiconductors rather than metal wire (excellent visual aids at [8]), which you generally can't easily make into wire. However, you can just connect wires in the middle of the semiconductors, between the hot and cold junctions, to extend the whole device. The junctions between the semiconductor and the wire will also exhibit thermoelectric effects, of course, but they should cancel out, because at one end there's a semiconductor-to-wire junction and at the other end there's a wire-to-semiconductor junction. This could be accomplished by an individual wirebond (or even electrically conductive glue) on each thermocouple's each terminal, but it might be cheaper to make the thermocouples on, or attach them to, a large semiconductor wafer, to which the wirebonds are made. You could also use this method to extend a metal thermopile, because thermocouple wire is more expensive than regular wire. Just make sure all of the thermocouple-to-mundane wire junctions at each end are isothermal, I guess.

But modern TECs/TEGs have many, many thermocouples. They're electrically in series, as I mentioned above, but they're thermally in parallel. This means that if you want to extend them with wires, you'll need as many wires as you have thermocouples, resulting in a thick cable running between the hot side and the cold side. This, besides being awkward to handle or install, could serve as a thermal path for unwanted thermal leakage from the hot side to the cold side, reducing the thermoelectric device's efficiency. But, while the thermocouples are in thermal parallel, I don't think their electrical connections between the hot and cold junctions need to be in thermal parallel as well (and, at the end of this writeup, I realize that they probably don't even need to be thermally separate in any case—you can cross-connect thermal or electrical true parallel strings at any point without changing the operation). But they do still need to be electrically in series… but we can simulate that! To reduce the number of wires, we can try multiplexing their electrical connections over fewer wires. This could be done with a TDMA scheme [9], where each thermocouple gets connected to one shared pair of wires for 1/n of the time (where n is the number of thermocouples), with capacitors for smoothing the voltage experienced by each thermocouple when the others are taking their turns. Depending on switching frequency, this could be implemented with (planar) distributed element filters at each end, integrated as a layer behind the thermocouples, or as a planar arrangement of discrete capacitors, possibly on a PCB.

That is unless I've misunderstood the working of a thermopile. In the previous paragraph, I assumed that no thermal connection between the hot and cold sides of a thermopile is necessary, and the thermal power is transferred between the two purely in the form of electricity, a transduction happening at either end of the electrical connection. Effectively, if you could see heat flow, but not electrical current, it would look as if heat is magically being destroyed at one end and created at the other, with no flow between them. That makes sense from the operation of temperature-measurement thermocouples: a temperature difference between the working junction at the distal end of the thermocouple wire and the "cold junction" inside the measurement device results in a voltage, with the longitudinal thermal conductivity of the thermocouple wire appearing unimportant. But maybe it just appears unimportant because the current (both electrical and thermal) is extremely low in a thermocouple used for temperature measurement, and a power thermopile does need significant thermal conductivity in that direction.

If a thermopile does in fact require a thermal connection between each thermocouple on one side and the two it's connected to on the other side, then the thermal path through the parallel wire bundle that I mentioned above as being unwanted is actually necessary instead, and that—and its susceptibility to thermal interference—could be the reason that such remote thermopiles don't exist. That would also make the multiplexed version inoperative. To make it work again, the electrical wires would need to be supplemented by long, thin thermal conductors to thermally connect the thermocouples. But these could also be multiplexed! Just use a very fast thermal conductor (such as diamond wire, graphene, or carbon nanotube, all expectable to be capable of achieving thermal conduction speeds of kilometers per second but all not yet in production) and some kind of rapid thermal switching device—hmm… I don't know what to use there… MEMS?—at each end. Or you could use a slightly slower but cheaper and still minimally inductive thermal conductor such as a loop heatpipe [10], and synchronously switch it at each end with compensation for the travel time of the thermal pulses. But wait… you probably don't actually need to multiplex the thermal circuit, because the thermopiles at each end should all be at the same temperature, so multiplexing would accomplish nothing. So, if a thermal path between the hot side thermocouples and the cold side thermocouples is actually necessary, then I guess a single plain old heatpipe should be able to do the job.

And heatpipes are common. So why don't these exist (yet)?

N/A [2019-04]