There still will be success with boards, in 2 areas: the low end (say, <$300), and the high end (say, >$3000). But the middle is disappearing, slowly but surely.
The low end will have things like BeagleBone, and the Raspberry Pi, and development boards for all sorts of microcontrollers. These low end boards will proliferate because SoCs have lots of popular interfaces, and there are low-cost expansion modules to add that one feature a project needs.
The high end will still consist of things like big multiprocessor, multicore, and programmable logic system boards for communications and signal processing infrastructure. These high end boards will still find homes in areas like defense applications and broadcasting - niches that need powerful processing in relatively limited volumes, with very special purpose I/O interfaces.
But it's the middle that is a problem. It won't die immediately - the projects using these boards have long life cycles, five years or more, in some cases much more. It'll be more like the death scene of the Paul Reubens character Amilyn in Buffy the Vampire Slayer. Ehhhh, Ehhhhhh. EHHHHHHH. Ehhh.
There are at least four dynamics at work:
- The same features and performance that used to be on an X86 or Power Architecture mid range board are now in the ARM or MIPS SoC on a low end board.
- Just as PCs became the value-add platform for industrial, medical, point-of-sale and other apps a couple decades ago, the tablet is about to become the platform of choice for many applications, running an app and connecting to the cloud to collect data from sensors near and far.
- Low volume custom boards can be designed relatively quickly by grabbing a chipset or SoC and adding an FPGA and the IP of choice.
- High volume custom boards are really an exercise in designing an SoC, again with the IP of choice.
EHHHHHHH. Ehhhhh. Death by trickle, shrinking at 3 to 5% a year for a while. There might be hope for a really small module, like COM Express or its relatives, but the big bad backplane and stackable form factors will struggle to hold ground. Just wait until the defense budget cuts start coming (something people are loathe to admit, "but UAVs are growing" ... these budget shortfalls will come from somewhere, folks).
I chuckled when I read that the TI OMAP family was "dead", the proclamation from many bloggers after TI said recently they are refocusing away from mobile and toward embedded applications. Squeezed out, yes - from one end by the Apple/Samsung/ Qualcomm love triangle, and from the other end by the onslaught of the new Chinese/Taiwanese mobile SoC designs we've been talking about. But TI and vendors like them are headed straight for low-end boards with their SoC technology.
The volumes in these embedded applications are much more fragmented - tens of thousands of projects consuming hundreds to thousands of parts, instead a few projects consuming a few million. That's why cloud connectivity, expansion interfaces, configurability via FPGA, or outright SoC design where board design used to live, are vital.
That tablet comment is worth revisiting. Where one used to have to design some cool handheld terminal with hardware to attach a barcode scanner, or a scale, or a temperature sensor, those are now all connected via wireless. That means the "closed" tablet that might have connectors for USB or HDMI isn't really closed, it's open to the cloud for expansion. Most of that effort that went into large scale backplanes or stacking for system I/O expansion isn't needed any more. Expansion will be on a small scale, like the connectors on the edge of the BeagleBone above.
Ehhhh. The middle is a bad place to be. TI got out of their middle mobile SoC situation to get into the low end CPU board situation leveraging their technology, and it's a move that will pay off for them long term. It's no coincidence that Freescale is putting tons of energy into ARM microcontrollers, similar thinking. Intel would be wise to invest a lot more resources to push their mobile SoC into embedded; they are already making their Atom push cannibalizing sockets for some of their bigger CPUs.
What do you think? Are your teams doing less board design and more FPGA and SoC design? Does the cloud shift the embedded equation from a box with a lot of I/O to a tablet with an app and wireless connections to data? How long will the risk-adverse defense industry stay on their old technology before they move? Are defense budget cuts unthinkable? Are more powerful MCUs taking sockets that used to need a microprocessor? Or is the mid-range CPU board market just resting up before making a comeback?