2

I have a camera sensor (bare die) that must go on a very tiny PCB that also contains electronics for data serialization that dissipate enough power to make the PCB pretty hot (40-60 deg. C). I want to keep the camera sensor cool (at least 25 deg C or colder) to reduce thermal noise. The PCB size and weight must be minimized so simply making the PCB larger, or adding a large passive heatsink are not options. I was wondering if a small Peltier device could be thermally coupled the the PCB, the die placed on the other side, and then chip to board wireboding used to connect the sensor to the PCB, just as described in this article https://sci-hub.se/10.1109/edaps.2009.5403990:

enter image description here

Aside from this paper, I can find very little examples of people doing this in practice and it makes me worried that I'm missing another option or better technique for getting local cooling on a bare die in small lightweight formfactor. I would appreciate any insight or alternative designs people have encountered.

jonnew
  • 81
  • 6
  • 1
    Do you actually need to cool the sensor? For newer sensors operating at >5-10 fps thermal noise is essentially zero even tens of degrees above ambient. Cooling helps for very long integration time (~1 second or longer) and for obsolete sensors. Have you done the calculations to see what your SNR will be with and without cooling? – user1850479 Sep 11 '21 at 14:52
  • 2
    Adding a TEC is going to vastly **increase** the amount of heat you ultimately need to get rid of, by a factor of at least 3 or 4 (TECs are horribly inefficient). You need to consider this problem in the context of the entire camera package. – Dave Tweed Sep 11 '21 at 15:11
  • 1
    What type of sensor is it? IR detectors like microbolometers typically do better with temperature control, but modern CCD and CMOS detectors don’t usually need this. TECs are commonly used for FPA temperature control, but they require a heatsink on the hot side to conduct the heat away from the detector, otherwise it will just get hotter as @Dave said. – Ryan Sep 11 '21 at 15:17
  • @user1850479 Yes, I do. This is a custom chip. Its used for biomedical imaging applications and thermal noise is and issue even with CDS and other techniques. We have very short photon integration times (1 msec). – jonnew Sep 11 '21 at 15:23
  • @DaveTweed, yes, I'm aware of this but I don't particularly care that the camera is extremely hot as long as my pixels and ADCs are cold. – jonnew Sep 11 '21 at 15:23
  • @Ryan Its CMOS device. Its a prototype chip that performs on-board compressive sensing. The issue is that its sitting in a miniature microscope that has to be carried on the head of a mouse. The entire system must be about 1-2 grams. – jonnew Sep 11 '21 at 15:25
  • 1
    Sorry to belabor this point, but are you sure that is thermal noise and not read noise? CDS will help with read noise, but not thermal noise. Being thermal noise limited at 1ms integration time would be extremely poor by 1980s standards, and seems unreasonable in 2020. – user1850479 Sep 11 '21 at 15:26
  • 1
    I'm just picturing all the condensation concentrated on the surface of your sensor... –  Sep 11 '21 at 15:30
  • @user1850479 You might be correct and in fact I'm not the camera designer. However empirically, with our chip, we need to keep the temperature down. We have the chip and it performs its function properly at room temp but not particularly well when its hot. As far as I understand it, there are reasons for this outside the normal ones you might be used to like the fact that we forgot to put a guard ring around the pixel array and we have thermally-dependent effect that is causing blooming effect in the pixel array. We are working with the fab to resolve this but for now its what we have. – jonnew Sep 11 '21 at 15:31
  • @user_1818839 Yes, I had this thought as well. That might be a deal breaker. – jonnew Sep 11 '21 at 15:32
  • 2
    Ok so not conventional thermal noise but more of a constraint on chip operating temperature. Since you are so mass limited, I would try to thermally decouple the other onboard electronics as much as possible (e.g. thermal reliefs in the ground plane between them to reduce thermal conductivity as much as possible without actually disconnecting them). Then I would put a small heatsink made of lightweight metal (e.g. magnesium) directly on the ground connections of the sensor. I have seen this approach used for CMOS sensors, can it can result in impressively low temp even without TEC. – user1850479 Sep 11 '21 at 15:59
  • @user1850479 Thank you very much for this suggestion. I should have mentioned the chip imperfections in the question. Indeed I will try my best to get thermal isolation. Can I ask why you suggest a heatsink rather than active cooling? I understand that the overall power dissipation will increase but if I use a really tiny TEC (e.g. https://www.tec-microsystems.com/products/thermoelectric-coolers/1md03-series-thermoelectric-coolers.html) then I can get the cooling exactly where I need it or am I totally missing something? – jonnew Sep 11 '21 at 16:03
  • 2
    Mounting sensors on TECs is a widely used technique, but I think it will be hard here since your PCB is already very hot. The additional power from the TEC is going to require even more heatsink mass, which it doesn't sound like you can afford. – user1850479 Sep 11 '21 at 16:08
  • Yes, I think I understand: I guess it is balance between if the deltaT I'm able to get out of the TEC is any better with that of thermal isolation and passive heatshink and your argument is that its going to be tough. Thank your for your input. – jonnew Sep 11 '21 at 16:14
  • @jonnew I would also suggest doing non-uniformity correction if you’re no already. It may help with the blooming effect you’re seeing. – Ryan Sep 11 '21 at 20:09
  • 1
    As others have said, the TEC will most likely increase the mass of the assembly due to the increased heatsink requirements, not decrease it. Is there any way you can blow air on the assembly? Decoupling the sensor from the heat source should be your first step, use a short ribbon cable. – Drew Sep 11 '21 at 23:07

1 Answers1

0

One method is to create a PCB island dedicated for the sensor. The island is able to significantly reduce the heat from other ICs.

Tony Tao
  • 1