I didn't know about the Chumby at the time; I only read about it after the fact. But with the benefit of hindsight, I can think of one detail in the execution that probably didn't help the product's longevity. In 2009, at around the time that the iPhone 3GS and newer Android phones were using ARMv7-based processors and 256 MB of RAM, the Chumby One used an ARMv5-based chip (the i.MX233) that could only handle 64 MB of RAM. [1] Not exactly a forward-looking move. But I don't know if Bunnie could have done it differently. Was there any ARMv7 or even ARMv6-based chip available to startup companies at that time with the i.MX233's high analog integration and low cost?
The other big problem with Chumby was its use of an embedded Flash player to display content. Made it difficult for open-source developers to create widgets -- and Adobe's effective abandonment of that runtime in 2010 didn't help, either.
Not really. That little chip is /still/ about the fastest chip you're gonna get for any sort of hardware hacking project for which you want a display, and the ability to prototype using one of the super cheap pcb makers.
Really? Can you explain more about why the i.MX233 has those unique benefits? Might it be better to use something like the CHIP (https://getchip.com/), based on an Allwinner R8, instead? Yes, the CHIP is itself a PCB, but it's so small and has enough pin headers that I guess you could treat it as a system-on-module and solder it to a carrier board.
Starting a new project with an ARMv5-based processor today seems really stupid to me, even more so than in 2009. But maybe I'm just hyper-sensitive about obsolete hardware because a community that I care about has been burned in that area before. I'm talking about PDAs designed specifically for blind people. I explained more about that in a mailing list posting here:
The big advantage comes from its packaging and everything built into the chip. It comes in a 144 pin lqfp package, which means that routing pins can be done by a couple people with real minimal amounts of training, and it can be laid out onto a two-layer board, which tends to be a lot cheaper than a four layer board which a bga package pretty much requires. It's also got a dac and adc built in which has the circuitry to directly drive headphones, so you don't need amplification circuitry to do that. Similarly, it's got integrated power regulation, so you don't need those chips either. Finally, all the info on it is available in a datasheet you don't need to jump through hoops to read [1]. It's not as fast as a chip, but because of everything it has, it's much more flexible for when you are designing something.
The Allwinner R8 is available in eLQFP176 and also has a built in audio DAC, ADC and headphone amp. Of course, it requires external memory and all the routing headaches that go along with that, but so does the i.MX233. Power management is generally done by a companion chip from X-Powers, the AXP209.
If the CHIP from Next Thing Co. is any indication of the capabilities of the R8, the built-in audio DAC and headphone amplifier are somewhat lacking in low-end frequency response (i.e. bass) compared to what one can get in a PC, smartphone, or any other consumer-level audio player. Indeed, NTC told me on an email support thread that for their upcoming product, the Dashbot, they're using a dedicated DAC to get better audio.
I'm tempted to buy one of Olimex's i.MX233 boards to find out if the audio capabilities of that chip are better. But as I said, starting any new project today with an ARMv5-based processor just seems wrong. Technology moves on, even in applications that don't require a screen. I doubt that an ARMv5-based processor could run a modern speech recognition engine, even if only to listen to a wake word for something like the Alexa Voice Service or the Dashbot.
[1]: https://www.bunniestudios.com/blog/?p=611