I'm designing a small PCB (max 5x5 cm) to bias and readout from an Avalanche Photodiode. Since the APD response varies greatly with the temperature, I want to use a cheap 12706 TEC together with a MAX1979 driver to achieve some kind of temperature control (0.1 is enough). However, I'm having trouble finding the best thermal pathway for the whole board - I was planning to use only one side for ICs and components, using the opposite side to place the TEC. However,

1 - Would I need to use via near the APD to conduct heat? Being a noise sensitive application, wouldn't cooled bias rails reduce signal noise?

2 - Can a thermisthor placed near the APD be enough for efficient temperature measurement? Or should I use a direct connection e.g. to the APD ground (I wouldn't want to disturbt the APD circuit)?

2 - With the number of R/Cs the layout requires, I could greatly reduce size PCB dimensions if I used a double-sided approach. Would there be any inconveninece in having the underside cooled (some rails would be cooler than others)?


1 Answer 1


Apparently you intend to put your TEC controller on the 5 cm x 5 cm board as your APD, and mount the board on the TEC. This strikes me as a very bad idea. Since the TEC controller works as a switcher, you are placing your very sensitive APD in close proximity to a multi-amp switching regulator. I suppose this is possible, but given what seems to be your level of expertise I would not predict good results. That said,

1) Using vias as thermal paths to reduce thermal resistance from the TEC to the APD is a really good idea. The same should be done with the thermal path to your thermistor. You do realize, I hope, that all your components, including the APD, will need to be SMTs. The amount of noise reduction you will get by using a single-stage TEC is unlikely to improve your noise performance significantly.

2) Placing your thermistor on the APD ground trace ought to work just fine for the sort of precision you want. You do realize that the thermal connection to the thermistor can be entirely independent of the electrical connection, right?

2) (the second 2) I really don't see how much you hope to gain by going double-sided on components. You'll be using a 5 cm x 5 cm board on a 4 cm x 4 cm TEC, which only leaves a 5 mm wide perimeter area. In direct answer to your question, you should get no particular harm or benefit, as long as you make sure the components are rated for the lower temperature they will experience. If you put a thermal insulator around the top of your pc board, with just a hole for the APD to peek through, you might get 50 C of cooling, which would bring everything down to about -10 or -20 C.

Another problem I see with your approach is exactly how you expect to interface the pc board to the TEC. Unless you epoxy it in place, you will need to provide pressure areas to allow clamping the board down, and you will need to do this to ensure good thermal contact. This will reduce the area available for components, which you say is already limited.

  • \$\begingroup\$ I understand that thermal connection is independent of electrical connection, and I was planning on keeping TEC control and APD circuits separate but for the shared thermal connection. 1 side with all the components, the other with a full ground plane and vias near the APD. Problem with keeping the TEC controller far way is that the thermistor bridge would still need to be placed near the APD - what issues could I run into due to the TEC controller proximity? \$\endgroup\$
    – joaocandre
    Jun 4, 2014 at 13:06
  • \$\begingroup\$ My fear of placing the thermistor on the APD ground would be the noise (thermal, electric from TEC controller) that it could add to the APD current. \$\endgroup\$
    – joaocandre
    Jun 4, 2014 at 13:22
  • \$\begingroup\$ "what issues could I run into due to the TEC controller proximity? " - As far as the temperature loop goes, not much. If you put a capacitor across the thermistor inputs at the controller, you will filter out the worst of high frequency noise - and what is left will basically be ignored by the low response time of the heating of the TED itself. That is, if a full-power step causes the temperature to drop by 1 degree / second, a 1%, 1 KHz square wave noise contribution will cause temperature to vary by 5 microdegrees. \$\endgroup\$ Jun 4, 2014 at 17:32
  • \$\begingroup\$ This answer made me realise that I in truth I do not need to include the TEC driver in the board, which can bring some unwanted interference and layout difficulties. Since I already have the APD bias regulator on the board, I'll just place the thermistor near the APD and control TEC voltage externally. \$\endgroup\$
    – joaocandre
    Jun 5, 2014 at 10:39

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.