Shifting sands: Trends in embedded systems design
Armed with over 15 years of data from the Embedded Systems Design subscriber study, ESD's publisher traces trends and preferences of embedded systems developers and industry vendors.
How sad a milestone it is for you depends upon your relationship with this particular print magazine and the rapidly changing world of print publications in general. Although this may be a farewell to print, I assure you that the energy and talent we put into serving you in print will increase online, now and for many years to come.
To be crystal clear, the magazine will not be printed and mailed from now on; nor will it be available as a digital edition. The website and community on Embedded.com, however, continues to thrive (with 3x the readership the magazine ever had). The Embedded Systems Conference (ESC), which was launched in 1989 by the same Miller Freeman publishing team that spawned the magazine, also continues to grow and evolve with new outposts in India and Brazil that have bright futures.
Why close the print publication, then, if other areas are thriving? The simple answer lies in pure economics. ESD, always a free magazine for qualified readers (known as controlled circulation in the media business), was supported by print advertising, and the number of advertisers who run print ads has fallen to a level that can't sustain the business model. Luckily for you, the reader, the same high-quality content that was the foundation of the magazine will continue online.
The second reason for this history is to share what ESP/ESD readers and ESC attendees have had to say over the last 15 years through our the annual subscriber survey. The survey, started by ESP magazine in the mid-1990s, became an annual rite of passage for the industry as we asked the magazine readers and conference attendees what hardware and software they were using and considering, and what their challenges were. Over the last five years, I've had the pleasure (and pain) of shepherding the survey through its many stages and then trying to make sense of it for readers, vendors, and industry watchers. The survey has its fans and detractors, and many people in the industry will see the data in a different light, so I encourage you to comment on this article on Embedded.com.
In 1997 (the earliest study we can find), we mailed a floppy disk (really, remember those?) and a crisp $2 bill to 1,000 readers of the magazine. In a direct reflection of how much the readers were engaged with the magazine, we got a 43.4% response rate--truly incredible and unlikely to be ever repeated in the history of market research. We're going to look then at the study data from 1997 to 2012 and stir some emotion and opinion about the embedded systems design market. Broadly, the trends we should look at are software, hardware, tools, and challenges.
In 1997, 80% of respondents were using C. Fifteen years later that number is still 65%, so C is still the dominant language used for programming embedded systems.
But what happened to C++? Back in the day, 35% were using C++ and now its 20%, so C++ didn't take hold. A Java "bubble" peaked in 2004 with 20% thinking they may use it next, but Java is now down in the noise again at 2%. The promise of Java in embedded systems was huge and logical at that point in time but never delivered on the "write once, run anywhere" promise. One of the great mysteries of the embedded market is why design teams don't migrate all code development to new tools like MATLAB or LabView and free themselves of the drudgery of C? Our guess is the legacy code base is large and needs to be supported and that it would take a very brave engineering team to abandon all that code and risk a failed project. My take is that one day many design teams will have to (and should) bite that bullet, painful and scary as it may seem.
Here is a topic that should get everyone fired up! Back in 1997, Wind River was the top used and considered RTOS (20 and 24% respectively) followed closely by pSOS. Wind River's VxWorks remained king of the hill until 2006 (peaked at 26%) but has seen a gradual decline since then and is now only used by 11% of our readers. pSOS was acquired by Wind River and despite valiant attempts by some companies to keep licensed versions alive, it faded away. There was a time that the industry thought Microsoft would make a big play for the embedded systems market but their interest was spasmodic peaking in 2006 (with WinCE), and they show few signs of interest in the market today. So what about open source you say? In 1997, we didn't even ask the "Linux" question but by this year 56% of you are using Linux in some shape or form and it seems to have stabilized at that level, the big growth years were from 2004 onward when we started to see double digit usage of some form of Linux.
One OS has maintained a roughly 20% hold on the market and that's "inhouse"--confusing I know, but it is whatever you define it as. All that homespun code and intellectual property is locked into your devices so you're understandably reluctant to throw it out. In 2012, the hot OS is Android; it reminds me of the Java wave discussed earlier and makes me wonder if Android will also be a passing fad.
Many people believe that the real embedded industry began with the broad availability (and affordability) of 8-bit microcontrollers primarily the 6800 from Motorola in the mid 1970s and the Intel 8051 from 1980 on. By the time we did our first survey in the 1990s, half of our readers had used or were planning to use the 8- or 16-bit versions of the Motorola 68HCxx or Intel 80xx controllers. There was a robust set of tool and emulator suppliers during those years but most have now merged, disappeared, or been acquired by a silicon vendor (more on this later).
In the first survey (1997), which was dominated by Motorola and Intel, we see the emergence of Microchip, which had 20% market share and Zilog with 15%. Interestingly the use of 32-bit parts was in its infancy, but most designers had at least tried or were considering migrating to the 32-bit architecture (a few were looking at 64-bit and still are!). By the time we get to this year's study only 13% of readers were using 8 bit, 16% using 16 bit, and 63% are at 32 bit. We have been asked many times over the years how long 8- and 16-bit parts will endure and frankly it could be for many more years as those parts continue to evolve and prices fall.
What we're seeing is strong growth in the number of ARM-based processors (SOCs really), and the number of vendors who offer their own ARM variants is increasing all the time. Let's not forget that in our top 10 Microchip has made a good showing with the MIPS architecture and Intel is still in the running with ATOM, which it's pushing down the power curve to stay in the hunt.
There was a time in embedded when both large companies and startups saw opportunity in tools, debuggers, OSes, and emulators, the magazine was full of companies advertising their products, such as Nohau, Huntsville Micro, and even Hewlett-Packard. So where are all those vendors today? To understand any market, you need to follow the money and ultimately in embedded the money has remained with the hardware vendors. They have either purchased, licensed, or developed many of the tools needed to get that all-important design win. This trend has been hard on independent software vendors of all kinds and with Intel's acquisition of Wind River last year, it does make us wonder what the future looks like for those few brave software businesses who are caught between open source and the hardware giants battling for share. Despite this, your favorite tools are consistently compilers, oscilloscopes, and debuggers. We think this is because it's where you spend the bulk of your time, so you have to learn to use and to live with them.
FPGAs, Memory, and LCDs
Over the years it looked like the phenomenal success of Xilinx and Altera and the lure of programmability would attract designers, but this year's study points to a declining interest in FPGA's. The perennial criticism of FPGA's in embedded was always threefold; too hard to program, too expensive, and too power hungry. It's going to take some creativity and innovation for the FPGA vendors to overcome these prejudices. FPGAs have been incredibly successful in the communications and military markets where programmability is a huge advantage, but we're seeing new generations of ARM-based microcontrollers with clever programmable communications and analog blocks being offered at very low prices that will keep the FPGA vendors on their toes.
Over the years we've also asked readers about their use of memory and LCDs but if we've learned one thing, it's how the specs, not brand, drive these decisions. Few respondents have expressed any preference for a vendor. After all, engineers are driven by rational and logical decisionmaking, or so they tell us!
Over the last few years as it became clear that embedded systems are really a collection of many kinds of software, hardware, and tool chains, so we've asked the "ecosystem" question. What we mean by this is which vendors offer you the best environment for getting your design completed and its very telling that in the results (unaided by the way) that of the top 10 answers only 1 was a software vendor (actually a tool vendor, IAR). This finding seems contrary to the data we see which shows most of the effort (60 to 70%) goes into writing code so why would the hardware vendor's ecosystem be so top of mind? Is it the only place where all the interdependencies in design come together?
Finally, the consensus is that the largest challenge (and overwhelming enemy) of design teams face is the "schedule." For three years in a row, 58% of all projects were late or cancelled. That's a pretty poor average in any industry. Why is this happening? Recent studies point to a skill-level issue. But, I predict it may also be all that legacy code sinking the project timeline. This year, we asked about project management tools; after Microsoft Project, a full 40% of you were using Excel as a project tool, which shocked us but was roundly defended by some teams because its cheap and everyone has it!
This year's study (all 89 slides) is available for download along with our annual webinar, which distills all our findings into a quick 30-minute PowerPoint (online at www.eetimes.com/4369712).
Therefore, after 15 years of surveying this market, what can we say about the current state of the industry and where it's heading? The embedded systems market is very conservative and changes slowly, but we can see some broad trends:
- More systems are connected to the web with 65% of design now including some form of Wi-Fi connectivity (Bluetooth, cellular and ZigBee make up the next three).
- Embedded designers have even more reasons to move to 32-bit processors these days (and the price differential shrinks every day).
- Perhaps the biggest shift is the move to open-source code. The numbers show that more design teams (37%) think their next OS will be some flavor of open-source compared with 31% using a commercial OS. This trend is fueled by many silicon vendors who offer their latest hardware with a free OS/RTOS loaded up and ready to go. It's hard to beat free, and if designers become comfortable with the support issues and tool chains, its going to become commonplace for design teams to take the hardware/software bundle.
A much deeper question is "what happens to the for-profit embedded software business?" I don't have the answer but the remaining players (remember Wind River is now owned by Intel) are going to have to examine their business models intently over the next year or so and really figure out how to add value that designers will pay for.
I'm sure I've missed something here that may be particularly important to you, so, leave a comment on Embedded.com when this article is posted.
David Blaza is vice president of UBM Electronics.