I've done a number of hardware projects for other companies in the past ranging from million unit production runs to "artisan" product development (less than 1K - 5K units)
Some things I've picked up over time you might want to keep in mind:
- First, the bad news is that your current prototype is raspberry-pi based. Unfortunately this platform is the least-production friendly of the ARM experimenter-board options right now. Numerous teams have encountered that Broadcom can only handle inquires for this SoC in volumes of at least 100K+ units (and even that is a tiny run for them) The SoC also uses a complex pop stacking design - limiting your production partner options significantly. Remember, it's not just about getting your hands on the chips, but working with a silicon partner that is geared up to work with small teams making inquires when your custom board doesn't work - or even getting access to data sheets.
- If you're still early enough in the software and feature design phase, look at something like the Beaglebone (similar to Raspberry pi, but based on a similar TI based chip) or dev boards from Freescale or Atmel. These platforms tend to be more "small volume production friendly" - with the ability to source 1, 500, 1,000, or 10,000 units as needed from distributors like Digi-key or Mouser, etc.
- For a connected device running linux, the required processor, sdram, and ethernet phy options out there are going to push you towards fine-pitch bga components. At first glance, these look seductively inexpensive from distributors (e.g.: maybe $2 for your SDRAM chips in volume) - on the back of a napkin it seems like you should be able to sell your product for $50…
What you'll find is that these components are rarely "small-run" friendly. They'll push you into very complex high frequency PCB design and testing, limited assembly partners, plus minimum part buys. (and you'll rapidly eat up your cost savings here)
The reason for the technical details above are to lay the groundwork for the next part of the discussion….
It's a lot easier to get into the small-volume hardware / devices business these days, but productizing something complex at small volumes (10K units) is expensive. You'll need to make sure your business model can support this.
The raspberry pi is developed in close partnership with Broadcom and they've produced close to a million units. The whole project has a lot of people wondering how the math works out and if it's actually subsidized by broadcom right now.
In an earlier project, I had estimated that for another team to take the Raspberry pi schematics and build an *identical board*, the per unit-cost at 1K units would at least be double. (~ $50 - $60) TI sometimes provides this information about their dev-boards. For example, TI's guidance on the Beagleboard (similar specs as the pi) was that a team could expect the part+production cost for a 1K unit run of a customized adaptation at around $100 per unit. Remember that's still minus the shell, power adapter, and packaging
So - if your idea supports selling the hardware component for ~$200, a $100 per-unit production cost probably can make sense.
The other approach to think of is building your custom hardware board around a module. Module vendors like Gumstix, LogicPD, Phytec, etc… produce the core complex parts of a design in large quantities (Soc, SDRAM, Flash, Ethernet PHY, wifi+wifi testing) then you only need to work with someone to design a simple baseboard with the ports you need and in the shape that you need. There was even some discussion about coming out with a version of the pi in module form - so you could source it without ports and design a baseboard around just what you need
Ok, I'm surprised the text entry field on this website hasn't cut me off by now…. I can go on and on… l'd be happy to chat about your project sometime - just let me know