One, I'm surprised that they include a USB PHY on board. Given that the machine has no externally accessible USB interface, that seems like a waste of power, BOM cost, and board area. The only thing I could imagine it being used for is through the mystical 5-pin programming header, for initial device programming; even still, I would imagine they'd prefer to cost-reduce and use JTAG (or another glueless interface) to do initial programming.
I could imagine that they'd use it for bringup, but even still, it seems like they'd want to have a non-bringup SKU for mass production that didn't have the USB PHY populated, and would just drop down the part if needed...
The other surprise was the MIPI DSI-DSI bridge. Having a DSI-DSI bridge that supports panel self-refresh allows them to keep the display alive even without having the OMAP (and its DRAM) turned on. The bridge seems to have 28Mbit of SRAM on it as a frame buffer, though, and that can't be cheap: I suppose the learning is that on OMAP, even keeping the DRAM powered on for a minimal refresh is more expensive in terms of power than a big external bridge.
Additionally, I'm vaguely curious about the external DSP; I wonder what they're doing with it? It's probably some operation to save power while the AP is turned off, but I wonder just how much power it really can save.
Neat device. It'll be interesting to hear the explanations of how people designed these first gen devices as they show up over the next decades.