2013 Embedded Market Study
Summary: Embedded.com's parent company has released the results of a new survey, and it has some fascinating data.
Surveys are funny things. Gallop presents data explaining that Americans feel strongly about X, yet at election time the vote goes in the opposite direction. The same is true for data collected about the embedded industry; the numbers are usually extremely noisy and untrustworthy.
But the data from UBM's annual survey of this market has remarkable year-to-year stability. Trends are just that: fit a line to the data and the standard deviation from that line will be very small. This year 2098 people responded, and some of the results are quite fascinating.
For instance, 11% fewer engineers are working on communications and networking equipment than just two years ago. UBM attributes this to the rise of the cloud; many companies have eliminated their on-site data centers in favor of Amazon's and others' cloud services. No matter how you may feel about this technology, it is reshaping at least this corner of our industry.
Five years ago 75% of us reported that our projects had real-time needs. That has steadily dropped in each of the succeeding years to 68%. Is this a result of Linux stealing market share? Nope. It seems the use of that OS is slightly off, too. People aren't avoiding RTOSes because of cost; only 2% report the price to be a deterrent. 79% of respondents who don't use one eschew them because the project just doesn't need real-time multitasking.
While overall RTOS use is declining, there's also a shift to open source versions. Puzzling, that, since cost is not an issue. Are the proprietary vendors providing poor support?
Get any group of engineers together and they'll complain about the schedule more than anything else. Turns out "schedule" is our second-favorite swear word. The winning expletive goes to "debugging tools,", though we loathed these more in prior years. Maybe the vendors are listening.
For some time I've been prognosticating that the next big thing will be embedded security. So far my batting average is 0.000. But managers' worries over this has jumped by a factor of four in three years. It's still way down on the list of things that keep them awake at night (number one being "integrating new technology or tools," followed by "managing code size/complexity").
Multicore is still a marketing person's pipe dream, at least in the embedded world, with only 15% reporting the use of any kind of multicore chip. Over half of all projects use just a single processor.
In this "this is really weird" department, 32 bits is slightly down, while both 8 and 16 bit use is up. The variation is small, but over 5 years the annual change is markedly consistent. I wonder if this is an artifact of the increasing complexity of our projects, where in some cases the developer is so divorced from the powerful (32 bit) hardware they don't consider themselves embedded engineers - there could be some selection bias. Or all of the pundits could be wrong: though 32 bit CPU shipments are huge and growing fast, many of these are in mobile devices manufactured in enormous volumes. Perhaps the growth in the number of 32 bit projects is slower than growth in 8/16 bits.
The survey asked people which 32 bit chips they could consider for their next project. 17% are planning on TI's OMAP. But that part has been deprecated for general use.
One set of data lists which processors engineers are using. Another gives their votes for the best CPU ecosystem (IDE, etc.). The latter is consistently much lower than the former; it seems that engineers are pretty unhappy with the support for the particular CPU they are using, and find the grass greener pretty much everywhere else.
I was quite surprised that FPGA/programmable logic use has fallen from 45% in 2009 to 31% today. Indeed, in "FPGA programming for the masses" in April's issue of Communications of the ACM the authors complain that the tools are too difficult, and propose alternatives to woo software folks into the FPGA fold.
Subversion is the clear winner in version control systems, with twice as many users as the next (Git and CVS are a virtual tie). Other than Clearcase and Perforce, the use of other VCSes is roughly zero.
The survey results are available (for a fee). I'm told interested parties should contact email@example.com.
Published April 17, 2013