Assembly
I built the first prototype this morning. A few notes:
- Solder paste is annoying to apply by hand; it's best to think of it as tiny, sticky solder balls, rather than a viscous liquid. The best that I've come up with is to use a tiny screwdriver to chop off a little blob of paste, then try to drop it onto the board. It doesn't really smear, nor does it stick to the board or flux very well.
- It doesn't really matter how much paste you use, so long as there's a bit of a blob there. It'll spread out when you heat it.
- My SMT soldering tip is no good. It just doesn't get hot enough. This could be because it's not exactly the right tip for the soldering iron, but it seems to fit pretty well over the heater element. I had much better results with the full-sized tip.
- The soldering procedure goes something like: apply solder blobs, use tweezers to place part, use tweezers to hold down part, heat the pad/pin to cause the solder to reflow, then heat the other side.
- If you don't hold the part down with the tweezers you can sometimes get it to float perfectly into place over the pad, but just as often it'll float away. If you hold it down, every part will look bad, but you'll maintain your sanity.
- QFN parts are hard to solder by hand. You need to rely on the part floating into position, because the pins are too small to line up by hand. I also couldn't see the pads with solder paste all over the place. Pleasingly, the solder stays off the blank PCB, so the lack over solder mask isn't too much of a handicap. In hindsight, I should've gotten SOIC parts. They're slightly larger, but far easier to hand solder, and make it easier to lay out the PCB.
- If your soldering iron sucks (like mine did until I realised it was the tiny tip), it helps a lot to solder the side with less thermal mass (smaller tracks) first. This holds the part in position while you try to heat up the other side. Much of my design is high-current tracks, so the thermal mass of the tracks is significant (or at least, it was with the crap tip).
- I've had the board sitting in front of me on my desk for the last few hours, and I'm still marvelling at how tiny it is. The soldering was surprisingly easy - with the exception of the QFN part - so if you're umming and ahhing about it, I'd recommend just jumping in.
The board needs to be reflowed; the QFN IC is misaligned and only soldered on two sides. I also haven't tried to solder the underside pad. I bought a cheap hotplate and gave the infamous skillet reflow method a go. No problems there. It seemed to be a bit slow to start - there was plenty of flux boiling off, but no obvious solder reflowing. I got a bit impatient and started tapping some larger parts. The solder was already molten, and probably had been for some time. It took a while to cool off, too. The IC has probably had a lot more than its recommended ten seconds of soldering heat, but oh well. It appears to be pretty well aligned now, although there's still not quite enough solder on two of the pads. I also have no idea of what's going on underneath, so there'll be a lot of probing with a multimeter before I test it out.
My poor little prototype sitting on the hotplate: Wow, I need to clean the kitchen 
Burn baby, burn! 
All soldered and ready for testing
Testing for this board may be a bit tricky. I normally run things off a current-limited power supply or stick a resistor in series with the supply - it guards against damage in case of a short circuit or mistake somewhere. In this case, the realistic current draw is very high, and the circuit likely won't do anything useful if I constrain the input current. I think my best bet is going to be to run off a high supply voltage, and slowly ramp up the current limit. If I notice excess heating or no output supply, something's wrong.
Of course, once I have the production process fairly reliable, I'll be happy enough to just connect them directly to a power source and toss the faulty boards into a 'to fix' pile. The materials cost of the board will be quite low; labour will be the dominating cost.
Update: I checked with a microscope, and the entire IC is soldered perfectly with no bridges. I can't see under the thing, but I haven't been able to locate any shorts with a multimeter.