Controlling Thermoelectric coolers

[wesg]

I'm looking to use a thermoelectric cooler (TEC/Peltier cooler) for a CPU cooling test project, and want to drive it with an Arduino so I can create a feedback loop in software (hope to have voltage and current sensors streaming data to Python). Does anyone have experience driving these devices?

Most operate between 12-16V and specific units can use as much as 26A. What are my options to drive it? This mosfet from Sparkfun can handle 30A so I'm assuming that's good, but would a dedicated motor driver be better? At higher currents, I think cooling would begin to be a problem.

[franklin97355]

A properly selected MOSFET should have a fairly low on resistance and thus would not heat excessivly.

[john444]

Hi Wesg,

Here is a little rule-of-thumb advise:
TECs make better heaters than coolers.
In order to get cooling, plan on lots of heatsink on the hot side, with a good fan.
It would be easy to PWM drive the TEC using a FET. I wouldn't consider a motor driver.
For the FET - most often low on-resistance is more important than the voltage and current specs.
If the FET does not dissipate excessive heat when 'on' then you are probably operating within the current spec.
Also, you can parallel FETs to get the net 'on' dissipation down. Just keep the external resistances similar.
Do the arithmetic and keep the power to 1-W or so or you will need a heatsink on the FET.
Choose a FET with a low turn-on threshold voltage like the IRFB8721 that Adafruit sells (ID 355).
It is a good general purpose unit that the Arduino can directly drive with it's 5-V output.
Most FETs require about 10-V to turn on so they would require an additional transistor.
For good measure, Lady Ada has the FET data sheet and a tutorial on using it.
The tutorial is about driving LEDs but the same general concepts apply.
Good Luck

[philba]

Selecting a mosfet can be a bit tricky. Estimating power dissipation (and thus heat) involves a fairly complex iterative process due to heating effects on resistive Rds and switching Rds. Most mosfet datasheets have a maximum safe operating area graph that you can use to make an educated guess. In the mosfet you linked to, it looks like at 16V the best continuous current it could handle is about 6A which would put the mosfet at something above 150C. Depending on duty cycle, you can go higher, though. Plus, a heat sink will make a lot difference.

Also, don't just take the current rating with out looking at the data sheet. The one you linked to claims 30A but the datasheet tells you that is only for a continuous current at a Vdss of around 2.5V. (safe operating area chart again)

Bottom line - there's no easy answer but don't rely on the current rating alone.

[wesg]

Thanks everyone, some good info here.

John, I'm going to be using the heat side of the TEC specifically, as I'm building a system to measure the cooling performance of other devices, so controlling the heat source is crucial. The TECs I'm hoping to use have a pretty high current rating, so I'm pretty sure I'll have to push more than 1W.

After the original post, I found this shield which looks far more convenient for my project. I think it uses the same MOSFETs as before.

I might add some heat sinks like the one in the adj. power supply kit and that should take care of the heat.

[lyndon]

If you're only interested in getting heat from the TEC, then why not just use a resistor? Or did I misunderstand what you're doing?

[thefatmoop]

It's not that complicated to select a mosfet... Just make sure you're driving it correctly.
Use a mosfet to do pwm on the peltier. Then find an xbox 360 power supply, the older the better. The power supply i found on ebay is 12v, 16.5 amps.
http://reibot.org/2011/09/06/a-beginner ... he-mosfet/

Then again if your computer's PSU can handle it, run the peltier off of your 12v rail.

[john444]

The TECs I'm hoping to use have a pretty high current rating, so I'm pretty sure I'll have to push more than 1W.

We are going to presume the TEC can handle what ever it is rated for.
It is is the dissipation of the FET you don't want to become excessive.
When the FET's gate is 'low' there is very high resistance Source to Drain.
Therefor the FET has essentially no dissipation even though it has full voltage across it.
When the FET's gate is 'high' there is a low resistance Source to Drain, so there is fairly low dissipation in the FET.

For example, if your TEC will draw 26-A at 12-V and you use the Sparkfun FET with 0.05-ohm on-resistance
26-A * 0.05-ohm = 1.3-V (across the FET), 1.3-V * 26-A = 33.8W of heat to be dissipated by the FET.
The remainder of the 12-V or 10.7-V will be across the TEC.
If you use the Adafruit FET with 0.009-ohm on-resistance then
26-A * 0.009 = 0.234-V, 0.234-V * 26-A = 6.084-W. Much easier to handle. With 11.766-V across the TEC.
These calculations are ignoring the wiring, connection resistances and the non-linear characteristics of the TEC.

There are still some things that can go wrong. But, you can avoid them by keeping the gate leads short and driving the gates quickly.

The Sparkfun shield would be handy with the terminal blocks on it. I would not hesitate to use it. But be prepared to beef-up the circuit traces and I would substitute the Adafruit FETs. Connect all 4 FETs in parallel and use a 1x2" piece of aluminum across them. They should then easily handle the 1½-W they each need to dissipate.

[philba]

but lets be clear, even 6W dissipated by a TO220 package will leave a smoldering mess on your board. The adafruit MOSFET datasheet gives a thermal resistance junct to ambient of 62 C/W. That means that each watt dissipated will raise the temperature by 62 C. 6*62 = 372 C above ambient. Be sure to get it on video as it ought to be pretty dramatic. The SFE MOSFET would blow even more spectacularly. So, yes, selecting a mosfet for this application IS tricky. Parallel several of them, use a big heatsink, thermal grease and maybe even forced air. Good luck.

By the way, Rds increases when the temperature of a MOSFET goes up. The simple static calculations are ok for ballpark estimates but you really need to know what to expect at higher temperatures. The Adafruit datasheet shows Rds being about 40% higher at 100 C. Also the data sheet gives an Rds of .013 ohms average at a Vgs of 4.5V (not .009). Max is .016 ohms and it makes sense to design with that in mind. Otherwise, you run the risk of designing a blue smoke emitting device.
http://www.irf.com/product-info/datasheets/data/irlb8721pbf.pdf

Frankly, if you are just doing this for heat, I'd get a bunch of nichrome wire, run it on 120 VAC and phase control it with a TRIAC. A lot simpler approach.

[john444]

Philba,
It is probably good that you are clarifying.

even 6W dissipated by a TO220 package will leave a smoldering mess on your board.

You are exagurating slightly. At 6-W you may not see flames but it will definatelty be excessive for the TO-220. But I think I recommended to keep a FET's disipation below 1-W w/o a heatsink.

Maybe you missed the part where WESG asked for advise on using TECs and the SparkFun FET.
My intent is to assist WESG with his project and maybe other forum readers with similar interests. If I made a factual error then jump on it. It is my opinion that using a FET is not that hard. Which is why I offered some simplified guideance for choosing one FET over another.

Debating the issues around using / designing with FETs is not on topic.
Contact me off the forum. Maybe we can combine our ideas and work up something that will be useful for our fellow hobbiests.
John

[john444]

Hi Wesg,

I'm going to be using the heat side of the TEC specifically, as I'm building a system to measure the cooling performance of other devices, so controlling the heat source is crucial.

Like Lyndon I am wondering why you wish to use the TEC.
It seems like determining heat put into your cooling device would be easier with a simple resistor.
Maybe a pretty big resistor (12-V @ 26-A = 312-W, about ½-ohm) but still just a resistor.

Also, like Philba said:

Frankly, if you are just doing this for heat, I'd get a bunch of nichrome wire, run it on 120 VAC and phase control it with a TRIAC.

Maybe not that far (120-V @ 2.6-A = 312-W, about 46-ohm) but doubling the voltage to 24-V will ½ the current for the same heat.

Are there some physical constraints you need to meet?
Will you be sensing temperature and using it for feedback for temperature control?

[wesg]

Hi Wesg,

I'm going to be using the heat side of the TEC specifically, as I'm building a system to measure the cooling performance of other devices, so controlling the heat source is crucial.

Like Lyndon I am wondering why you wish to use the TEC.
It seems like determining heat put into your cooling device would be easier with a simple resistor.
Maybe a pretty big resistor (12-V @ 26-A = 312-W, about ½-ohm) but still just a resistor.

Also, like Philba said:

Frankly, if you are just doing this for heat, I'd get a bunch of nichrome wire, run it on 120 VAC and phase control it with a TRIAC.

Maybe not that far (120-V @ 2.6-A = 312-W, about 46-ohm) but doubling the voltage to 24-V will ½ the current for the same heat.

Are there some physical constraints you need to meet?
Will you be sensing temperature and using it for feedback for temperature control?

I'm attempting to use the TECs as stand-ins for CPUs and GPUs in order to benchmark various PC cooling equipment. The TECs will likely be installed into a static metallic structure and the cooling gear will be installed above that. I figured that by measuring current and voltage and using a control system I can calibrate the TECs evenly across each test. I like the TECs because they can produce significant heat and seem easy enough to control. Also being physically close to the dimensions of a CPU makes it convenient.

[john444]

Hi Wesg,

I figured that by measuring current and voltage and using a control system I can calibrate the TECs evenly across each test.

I see. Yes, you should be able to set a particular temperature and PWM to adjust the effective current.
With a resistance, the voltage and current are directly proportional and you can easily calculate the wattage.
Not so with a TEC. You have a bunch of strings of parallel diodes. If you PWM it shouldn't matter though.
It becomes the % of on time that determines the wattage. Temperature rise at a set % on-time will depend on other factors.

to benchmark various PC cooling equipment.

You can either set a fixed temp and record the % on-time or set a fixed on-time and record the temp rise.

However, using a TEC, the dissipation on the cold side needs to be consistent.
Frost and moisture may cause you problems in this area. You will have to check.

Also being physically close to the dimensions of a CPU makes it convenient.

Definitely more convenient.

Good Luck, John

[lyndon]

If I were doing this, I'd cut an aluminum blank to the dimensions of a CPU and secure a few high wattage resistors to one side with thermal paste. Much simpler (to say nothing of cheaper!) than a Peltier device IMO.

[philba]

Debating the issues around using / designing with FETs is not on topic.
Contact me off the forum. Maybe we can combine our ideas and work up something that will be useful for our fellow hobbiests.
John

I hadn't come back to this thread but think I need to say that the issues I raised are, indeed, on point. You said, basically, that FETs are easy. While for simple uses, they are. But when running them up to their limits, the design criteria becomes challenging. Especially in regards to thermal issues wrt specifications like Rds that are key design points. What might look ok with a simple analysis becomes a dead MOSFET in practice because iterative analysis was ignored. In general, heat is ignored by too many beginning designers who then scratch their heads when the design fails.

Hey, anyone can do anything they want. And, maybe a blown mosfet or two is a good learning exercise.