Also, I should mention we're looking to use a temp sensor that sits a few thousandths shorter than the ASICs, and park it right in the center of them, so it should make fair contact with the heatsink and get a decent measure of that temperature instead of, say, the temperature of the opposite side of the PCB.
If you melt it you're doing something very wrong.
What we've talked about doing in the past, it'll depend on what Novak implements but we've talked about doing fan control independent of cgminer. This avoids problems like the S5 meltdown, where the chips keep running hot for some reason but cgminer is down and so is the fan. If the onboard controller, which reads the temp sensor directly and also controls the fan speed directly (and actually even knows how many watts the chip string is pulling), can be given a curve or profile for thermostatic control and run the fans based on that without cgminer intervention, burnup should be fairly idiot-proof. The controller would have a default profile and you could, probably through cgminer, push a custom profile if you wanted to do something a bit different. The only way this might screw up is if you ran without a cooler and the temp sensor was not accurately reading temperature - but at that point most of the chips' heat being dissipated will hit the PCB, and the temp sensor is sorta soldered to the PCB right smack in the middle so it'd probably still be close enough to true that it'd shut down before catching fire.
Yes, the pod ended up bigger than I was initially planning. If I wanted to make a double-sided board and put only the ASICs and their associated bits on top, then tuck all my power and controls underneath, I could have fit it all in a 10x10 pretty easily. I'd much rather build a single-sided board (by which I mean, I'm capable of building a single-sided board) and making room for a CPU cooler pretty much means I have to mark out a 9cm-diameter circle with zero tall parts and let's not forget the legs coming off it for screwing to the board. By the end of it, there's almost no room on the 10x10cm square for anything except the cooler and ASICs. I could fit most of the controls on there if I had to, but not the power circuit and none of the jacks and headers. So instead we make the board a bit bigger and now I have all sorts of room. Should be able to stick rubber feet on it.
Someone might look at the U3 and say "oh well those guys fit all kinds of crap on a tiny board". To which I respond yes, but they only had 4 ASICs and those were all on a single power rail and they had a custom-milled heatsink to fit around stuff. The total parts count of my thing is probably four to six times the parts count of a U3, and it's built with two power jacks and two USB jacks and can fit a dozen different heatsinks with two, three or four-wire fans. It's intentionally general which means I need more room to cover all the necessary bases.
Also, and here's a great part - all the controls we're putting on this board not only make it awesomely flexible and tweakable, but they'll translate almost directly to the S1-sized board I hope to be building next. So there's more good reason to make this thing slightly more awesome. I mean, as if "because that's the kind of feature set every miner should have anyway" wasn't good enough reason.