I'm really excited for the iMX8 though. It looks like it has much better specs, and if they can remain as open source and hobby friendly as the iMX6 then I'm all for it, even if it costs slightly more.
Plus, if you ever ship at scale, it's pretty easy to get your hands on a large quantity of them!
And there's no shame with using a processor/RAM/eMMC combo on a SOM, either. Doing the high speed DDR layout on 6 or 8 layers of PCB is not a task for novices. Saving the low-speed interface work for a larger 2-layer carrier PCB makes the project go really smoothly (just watch the EMI off those right angle card-edge connectors!)
There are plenty of companies selling the IMX in all kinds of configurations for really decent volume prices. TechNexion and Boundary have been my favorites so far.
I've designed it in a thermal imaging product, where it's 20 bit wide DVP support came in very handy.
NXP's stance is probably what made others shy away from the chips.
Minor hassle, however.
TI's stuff is open. TI's manuals are awesome. It already exists. There are user iterations that are different from the TI reference. The board is a much cheaper 6 layer board.
What niche does this serve that a Beaglebone would not?
Anyone know why they are designed this way? Very wiggly traces.
Differential pairs carry polar opposite signals so that any noise can be canceled out by comparing the two. While parallel bus traces need to be routed to 1/20-1/40th of a wavelength, the differential pairs need to be routed to within 20mils of each other with impedance controlled traces (which depends on distance to each other and other traces as well as material variability, hence the tight routing).
The GnuBee is an open source nas that is actually cheaper than everything else on the market. I suspect because its a fairly simple device and the proprietary versions are super marked up as well as the creator being located inside Shenzhen