Instead of building USB Miners why won't you build something like this.
Put the most chips you can on one board, since its lower power it probably won't start a fire like the Prisma. I can see it running at 5-6Th/s and using only 1000Watts.
Please RTFT before asking questions that have already been answered all over the place.
FROM POST #1:
The primary goal is to build a simple board which would be USB-connected to a controller, and capable of adjusting both core voltages and clock speeds using cgminer flags. We're looking at a single board capable of 300GH at 150W, downclockable to around 150GH 50W. At mid- or low-range settings it could be run off a brick with a quiet 120mm fan and heatsink and be a decen Jalapeno-formfactor home desk miner. The board will be designed specifically so that four of them would mount to an S1 chassis. Couple that with a 4-port USB hub and a Pi or something as the controller, and you have an "S1 Upgrade Kit" which will aim for 1.3TH at 600W clockable down to 600GH about 160W.
...
A secondary goal would be to design another board which could mount to a Prisma chassis. Modifying our design to a similar power density as a Prisma (1KW board-level) we could see a 2.4GH miner downclockable to about 1.1TH at 300W.
Regarding large-form chips, Spondoolies is the only company that's actually had good luck with 'em. Everyone else (well, I guess except KFC sorta) has pretty much crashed and burned. ASICMiner, Bitmain, Avalon all did quite well with small distributed chips instead of single giant chips. I'm hoping most outfits keep doing things that way, as it reduces complexity, increases modularity and flexibility and decreases requirements for cooling and power density. So basically, from where I'm sitting it looks pretty stupid to keep using that chip type. I like to think our design is the best design and our design is approximately impossible with giant chips.