TPA6120 isn't too bad to solder with hot air - I've made an amp with one using on home etched double-sided PCB (extended the pad underneath to the sides of the part, ran a bunch of manual-bits-of-wire vias to the ground plan underneath).
The main downsides to this chip are it's unnecessary speed (be careful with the feedback - no parallel capacitance and use the datasheet suggested feedback resistor!), and the 10 ohm output impedance requirement.
For my amp I wrapped it within the feedback loop of an audio-specced fet input op-amp (some TI one, nothing particularly amazing, but it was in the parts box...) so DC characteristics were determined by the input op-amp, not the TPA part. As I was running the TPA chip in gain-of-2 mode I was also careful to set the external op-amp's gain above 2 (4 or 5 IIRC) to make sure it was running at at least unity gain for stability. Worked perfectly, I got good audio performance, very stiff current drive capability and low output impedance. There are however a couple of caveats to this method though:
- Any op-amp/buffer inside a feedback loop should be faster than the outer part, as to not disturb it's stability (no issue in this case, the TPA6120 screams along)
- Make sure everything it running at a stable gain
- If the output is shorted then you've got a grunty op-amp dumping the full supply rail into it's output resistors, which will proceed to go up in a puff of magic smoke. To fix this I put an ohm or so in series with the headphones - makes the output impedance worse but at least a short will only apply 10x the normal output voltage over the 10R resistor rather than the full supply rail.
Other notes:
- I used a low noise switcher to make a negative 12V supply from the 12V input, and followed this with linear regulation to ~10V. Can't hear the difference compared to a linear supply (actually a little quieter, see below)
- Be very careful to separate audio traces from other power traces. Star grounding is your friend. You can hear stuff that is invisible on a scope - my initial prototype with a linear power supply hummed because a tiny bit of the 50Hz input power was flowing through an audio ground due to a layout mistake.
- I also incorporated a simple headphone protection circuit - the output is switched through a 5V coil relay with an 15V zener in series. If the total supply voltage drops below ~15V the headphones are open-circuited, hopefully avoiding any DC if one of the supplies fails.
I'd probably do things a bit differently if i built on again, probably using some of the LME parts mentioned earlier in the thread, or at least work out a way to deal with the TPA6120 output impedance requirement more elegantly. Then again, I got everything designed and built in a weekend from parts on hand - the only things I bought were a plastic case and the potentiometer from the local Jaycar store.