Hi there,

I'm using a H-Bridge to drive a thermoelectric cooler (TEC or Peltier) module. I'm not here to discuss control circuitry/drive voltages/slew rates/switching losses, which I know all are important.

I've accounted for all of these and optimised my circuit to make sure these are all as efficient and effective as possible. The last remaining problem is \$ R_{ds(on)} \$ losses in the P- and N-Channel MOSFETs in the H-Bridge. A H-Bridge is used in order to heat/cool different sides of the Peltier plate, depending on the direction of the current (just an FYI for those who didn't know).


simulate this circuit – Schematic created using CircuitLab

(Note: This circuit is simplified. There are various extra components such as flyback diodes, series current shunt resistor, etc. in the actual circuit)

As you can see, this is a fairly high-power system. With a 24V rail over a 30V peltier, rated for 12A (I've shorted it through a power supply and it never exceeded 10A, however). This means each FET will (excluding switching losses) dissipate power somewhere in the order of \$P = I^2\times R = 100 \times R_{ds(on)} \$ Watts - nothing to laugh at!

I've never calculated a component's heating before, and so I wanted to make sure I am doing it right here. I know this has been done on other threads countless times, but I just wanted some guidance and to make sure I haven't missed anything from following those as guides. For simplicity sake, I will just do the P-FET.

Accurate \$ R_{ds(on)} \$

First step is to find an actual \$ R_{ds(on)} \$ that I'll be most likely operating at. I'm driving the P-FET with MOSFET Drivers, all the way to the rails. So, \$ V_{gs} = -24V \$ (yes, I know the FET is only rated for max 25V, and I know ideally I would use an isolated gate driver with an N-FET or some sort of voltage divider to lessen the drive voltage etc., but for now this is the case for a simplistic prototype). From the datasheet graphs:

enter image description here

Since they only graph to -15V, I'm over-specing and basing my calculations off that. This gives \$ R_{ds(on)} = \frac{1}{10 \div 0.2} = 20m\Omega \$, which you can see from Figure 2 will be even better as the FET gets hotter and hotter, with a constant drive voltage (up to 30A, which I'm going nowhere near). So this is really absolute worst-case.


With this, the power loss in the FET will be:

\$ P = I^2 \times R = 10^2 \times 0.02 = 2W \$


From this, we can use the thermal resistance given in the datasheet to calculate heat. Now, this is a fairly complicated figure, and obviously depends heavily on PCB design. That is why, fortunately, this datasheet specs a thermal resistance, junction to ambient, for a 1 inch squared pad of 2oz copper - exactly what I have done on my PCB design for the FET.

enter image description here

Again over-specing, I will use a compromise between these two figures of 50 degrees per Watt. So, this gives:

\$ T_J - T_A = \Theta_{JA} \times P = 50 \times 2 = 100 \$ degrees Celsius

\$ \therefore \$

\$ T_J = 100 + 25 = 125 \$ degrees Celsius


There are a lot of assumptions here, which is (I guess) expected without any experimentation beforehand. But, I'd like to think I've over-spec'd what I can, and assumed worst-case here. Since the device can operate up to 175 degrees Celsius, should this design be okay? On top of that, this also assumes this thing is being driven constantly, with absolutely no cooling - which isn't the case. This will be run sporadically, with different duty cycles, and with 3 small fans cooling it.

Thanks for your time.

  • 1
    \$\begingroup\$ Since what you have calculated is your worst case Junction temperature, you are on the right track. If you are using surface mount FETs then having 1 in^2 of heat dissipation may be ok, but you could use TO220 packages with a small profile heatsink. That would potentially get you down to 10-12 degC J-A or better. \$\endgroup\$ Jul 2, 2018 at 4:44
  • \$\begingroup\$ Okay awesome, glad I'm at least on the right track. Due to spec constraints I've done with D2-PAK surface mount ones, but in future I could reconsider classic TO-220 with mini-heatsinks on each. Thanks. \$\endgroup\$
    – DSWG
    Jul 2, 2018 at 23:17

1 Answer 1


TI has done a good job on this app note http://www.ti.com.cn/general/cn/docs/lit/getliterature.tsp?baseLiteratureNumber=snva419&fileType=pdf

The app note contains rules of thumb and forced convection calculations.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.