Cooling Integrated Circuits with Peltier Cells

1. Intro
Hi all,
I have a question about using peltier cells for both cooling and reading temperature in the same moment.
I need to cool down a couple of Integrated Circuits so I’ll put put a 20mmx20mm peltier cell over them but I cannot read the cold side of the cell because it’s sticked to the IC; I hope physics land me a hand here, providing I put a thermocouple to the hot side…

2. Relevant equations
S = - (Δv/Δt)
where:
S is the Seebeck coefficient (which I’ll need for later)
Δv is the voltage I’ll read with a multimeter
Δt is thermodynamic temperature which I’ll read with a thermocouple. If I understood correctly it should be the difference between the hot and cold side expressed in kelvin’s degrees

I’ll take all these measurements before placing the peltier over the IC to avoid experimental errors

3. The attempt at a solution
Now that we found S we place the cell on the IC with some thermal compund and we get:
Δt = - (Δv / S)
hence
Δt = - ((Vin - Vcc) / S)
because I expect Δv across the peltier cell should be Vcc + the voltage generated by the seebeck effect because the cell is in series.

then I convert °K to °C

Now that I have the temperature difference between the hot and cold side I can measure with a thermocouple the hot side, subtract the difference I just found and get the temperature of the cold side sticked to the IC.

Are my assumptions correct? will it work?

I know peltier are dangerous with electronics because of the condense but I recently used a dht11 which is a sensor which can read temperature and humidity.
I’ll use it to know how much voltage I can provide to the cell before generating condense using a table like that (I’ll drive it with a transistor) to keep it above the dew point (maybe 1 °C with a 5% tollerance).

I also know peltier are cooling better if the hot side is helped in dissipation, my thermocouple will find place between the peltier cell and a fan

Thank you all!

I suppose it would be a good assumption that in the steady state the delta-T across the peltier is dependent only on voltage. In this case, steady-state could be approximated by taking an average over 30 seconds or so.

Try some experiments with thermometers on both sides and see if this assumption is close enough for your calculations.

Why can you not put a temperature sensor in the unspecified-couple-of-integrated-circuits? If you have a spare P-N junction anywhere in the ICs, you can measure the temperature with that. That's how an LM355 sensor works - it's a transistor with known temperature dependency and you simply don't connect to the third leg of the transistor.

I'd seriously try to find a way of mounting the temperature to the cool side of the Peltier (or to a heatsink that's common to the ICs and the cool-side, etc.).

which I'll read with a thermocouple.

Of course a thermocouple will work, but an LM34 or LM35 is cheap and convenient. And in "real-world" applications, I'll bet an LM34/35 is just as accurate.

You can also get in-chip temperature from some AVR's

IMO that would be additional data.

I have a cheap 40mm Peltier. It's just a bit wider than a 328P DIP. There's room for 3 in sockets and digital as well as analog temperature sensors under that wafer. The Dallas Semiconductor 1-wire sensors take computations out of it and send you serial text data, and they're small and cheap for what they do.

One thing though. If you overcool the chip, water may condense where you'd wish it wouldn't. Take care.

im not sure you can rely on the hot side temp to give you a idea of the cold side. I use peltiers as heaters in incubators and at 100% you can rougly set the hot side from the cold side temperature. Anything under 100% pwm you end up with more heat which shows the inefficiency in the peltier is not linear. People have used peltiers for chip coolers with some limited success so it can be done.

If it’s any help, they come with Wattage ratings.

An under $5 TEC-12706 is rated to cool 50~72W heat source. Isn’t that more about what you want to know?

I'll give you some additional info: The piece of HW is the banana pi BPi-r1 which is a banana pi + switch. I'd like to use it as NAS/router/mediacenter/minipc/devboard.

The chips I'd like to cool are AXP209 (the PMU) and the BCM53125 (switch) as they gets hot when a sata hard disk + usb + wifi and a 3Amps PSU are used (expecially the AXP209) resulting in shortened HW life.

The A20 chip has a temperature monitor but it's most likely based on the frequency/work load instead of the real temperature.

@GoForSmoke: about overcooling... my last paragraph is all about preventing condense; do you think that method to prevent that wouldn't work? and TEC-12706 means 127 P-N couples @6A... I don't think I need all that power! o.o

LM34/35 are quite big... I could drill a hole in the dissipator I might decide to mount so I'd like a smaller solution. For maximum efficency I'd like to stick the cell directly to the IC's

I was asking you before even trying just to avoid wasting time on something already tried and proved not working; From what I studied and from what you say it could work so I'll buy some peltier cells and make some test with the LM35... There MUST be some steady correlation between the two sides!

There MUST be some steady correlation between the two sides!

I believe you are right…

Under steady state/equilibrium conditions the Peltier is dissipating a known amount of power (Wattage = Voltage x Current). That electrical power is 100% converted to a fixed net-total amount of heat energy. That heat will be dissipated (and hopefully some of it is conducted-away from the Peltier’s surface via a heatsink).

I don’t know how to calculate what the temperature rise will be (and there are usually lots of variables that are not accurately known) but it should be a constant and you might be able to determine it experimentally.

If the heat generated by the ICs you’re trying to cool isn’t constant, or the ambient temperature isn’t constant, I don’t think the hot-side will give you enough information.

Under dynamic conditions (when temperatures are changing) things get even more complicated because you have to take heat capacity into account (probably another unknown variable), and the instantaneous temperature of the hot side probably won’t tell you anything useful about the temperature of the cold side.

You don't have to feed Peltier's full power. They won't make full cooling but you don't have to do that. You do have to feed them more than they'd produce from the existing heat difference but that's not much compared to what it takes to cool.

You can either keep the wafers above the dew point or waterproof your circuits with sealant.

Suppose that you got an aluminum bar and mounted your components upside down on top? One wafer with its own heat sink underneath could cool that shared heat sink for the chips.

Peltiers are a cool idea but there's a reason why they aren't used in phones or cars or TVs or any other piece of electronics outside of a laboratory. They simply don't do the job of keeping chips cool.

If you need to keep regular microchips cool then you should put a good heatsink on them with good thermal paste and a good fan blowing air over the heatsink. In this context, "good" might be a tiny little stick-on heatsink. The energy put into the fan is ten or a hundred times more effective at cooling the chips than a Peltier.

it's like I must expect a big delay when IC changes temperature... something like: the IC gets hot -> the cold side gets warmer -> the hot side gets hotter will it be so slow?!

the heatsink is about 2mm height (base+blades). if I drill a hole in the middle of it and place a tmp35 would it fit? (PTH not SOP). looks ok from here -> http://cdn.sparkfun.com/datasheets/Sensors/Temp/TMP35_36_37.pdf

@MorganS: other guys said heatsink isn't enough... I don't even know if a fan could be enough in summer; next week we're expecting 38 °C here!!!

The power you give it moves the heat. Hook a wafer up to 12V 1A and see how fast the faces get hot and cold. That is how fast it should move heat at 1A, more pushes harder.

You have to have an effective heat sink for the wafer or it will self-destruct!

If the background is 38C, you will need active cooling to get any cooler.

MorganS:
Peltiers are a cool idea but there’s a reason why they aren’t used in phones or cars or TVs or any other piece of electronics outside of a laboratory. They simply don’t do the job of keeping chips cool.

If you need to keep regular microchips cool then you should put a good heatsink on them with good thermal paste and a good fan blowing air over the heatsink. In this context, “good” might be a tiny little stick-on heatsink. The energy put into the fan is ten or a hundred times more effective at cooling the chips than a Peltier.

They worked with 486’s and early Pentiums but IIRC the newer chips in the late 90’s had too small a hot spot which is why I got a couple coolers on deep discount back then. Before that, coolers were overclock enablers.

I’d rather see Seebeck solutions than Peltier. Heat from a junction in the sink could be removed to bigger junctions in the case itself or even outside possibly submerged in liquid. It’s not as fast but it powers itself.

Here is a good selection guide and technical reference for choosing Peltier modules as chip coolers: http://www.teamwavelength.com/downloads/notes/an-tc09.pdf

a peltier does not cool it simply moves heat from the cold side to the hot side. If you can not remove the heat from a heat sink with a fan then how are you going to remove the extra heat produced by the peltier. In a pc application where you have a massive cooler that can remove more heat than the chip produces then you can go for extra cooling as the peltier will bring the chip temperature down about 20 degrees cooler than the hot side sink. At 20% power its not to efficient so you will end up with lots of heat and very little cooling. any lower than 9% power you will start to get serious heat bleed back. If you turn it off then you have a ceramic insulator between the chip and the sink which im guess would be bad.

I would personally look at liquid cooling as that's a better way to move the heat to a area where a larger fan can be used.

Liquid cooling is used by overclock extremists .... because it works so well!

Here's the TEC1-12706 datasheet.

It's got some graphs and data the OP might find useful, hopefully more than I do.

gpop1: a peltier does not cool it simply moves heat from the cold side to the hot side. If you can not remove the heat from a heat sink with a fan then how are you going to remove the extra heat produced by the peltier. In a pc application where you have a massive cooler that can remove more heat than the chip produces then you can go for extra cooling as the peltier will bring the chip temperature down about 20 degrees cooler than the hot side sink. At 20% power its not to efficient so you will end up with lots of heat and very little cooling. any lower than 9% power you will start to get serious heat bleed back. If you turn it off then you have a ceramic insulator between the chip and the sink which im guess would be bad.

I would personally look at liquid cooling as that's a better way to move the heat to a area where a larger fan can be used.

we have to debate on the concept of heat: it's "simply" particle's energy. IC's particles are excited by the current passage, then this energy goes from the IC to the cold side of the peltier through induction (and convection in the heatsink groove if we consider the air as a fluid) because of the difference of particle's excitement state. Then Peltier cells does transfer that energy from one side to the other and if you can take it out with other cooling system (e.g. a fan, or even another peltier's cell!) you can even get below the abient temperature! so peltiers actually cools if that means transferring the IC's particle energy to the peltier's cold side. Anyway I told from the first post I'll also use a fan.

To prove what I say (peltier actually cool stuff) I invite you to see this. That's what you can achieve using a peltier to cool another one and a pinch of radioactive material.

anyway I found a lot of interesting cues here and in a physics forum on this very experiment. I'll soon try it and I hope to come back with good news! (my only concern is the current needed :D)

A fan blowing on a bare Peltier wafer is not enough for much cooling at all. The fan is to cool a bigger heat sink attached to the wafer’s hot side.

Peltier does move heat with power, making more heat total to remove or cooling is wasted, you might as well try to air condition the house by opening the refrigerator door.

BTW, the thing to look up is Seebeck Current. If you leave using the current for anything but heat transfer, it will do that. If you give a push in the opposite direction you have the Peltier principle.
These complex wafers are just a way to get the voltage into the useful range by adding series resistances and so dropping the need for enormous currents at millivolts or less.