It's a very interesting time in embedded processors.
For decades, embedded processors have continued to deliver more performance and more features, at ever-lower prices and power consumption levels. Today, embedded systems designers are leveraging these processors to create an incredibly diverse range of innovative products. Some of these products, like the Nest thermostat and the G-Box MX2 set-top box, target high-volume markets.
Given all the new embedded systems being designed (and the large market potential for some of them), it may come as a surprise that a number of chip suppliers are retreating from the embedded processor market.
Perhaps the most dramatic example of this is Texas Instruments. Over the past 10 years, TI leveraged its strengths in digital signal and mobile application processors to create a very broad portfolio of embedded processors, ranging from tiny microcontrollers to high-end multi-core chips. Some of these chips incorporated DSPs; others (like the AM3703 used in the Nest thermostat) did not.
But last month, TI announced 1,100 job cuts "in its embedded processing division and in Japan, where it said investments do not offer 'sustainable growth and returns'," according to the Dallas Morning News. TI is making these cuts despite 9 percent growth in embedded processing revenue in 2013 vs. 2012. It appears that TI sees more attractive opportunities in analog chips, which now account for 75% of its primary semiconductor revenue. Linley Gwennap, a respected, long-time observer of the processor industry, believes that TI will ultimately drop out of the embedded processor market.
Freescale, another long-time embedded processor stalwart with a very broad product portfolio, has also been narrowing its focus over the past several years, zeroing in on specific markets and applications where it believes it has the best odds of winning big.
As some suppliers retreat from the embedded processor market, can mobile application processors take up the slack? Perhaps. As I wrote in 2012:
Driven by the enormous volumes and intense competition in the smartphone and tablet markets, mobile application processors have evolved into marvels of engineering. In a very modest price, power and space budget, these chips deliver a remarkable amount of performance and functionality. And much of that performance and functionality is just as relevant in typical embedded applications as it is in smartphones. In addition, application processors targeting smartphones and tablets benefit from huge investments in software development infrastructure, including things like operating systems, device drivers, interface protocol stacks, and multimedia codecs and frameworks. And application processors are continuing to improve rapidly. As a result, it's becoming increasingly tempting to use application processors in embedded applications.
But as I also wrote in that earlier column, there's a catch:
But there are some serious obstacles to using application processors in embedded applications. Let's start with the big one: You can't buy them. Most application processor vendors will only sell their chips to ultra-high-volume smartphone and tablet makers. As strange as that may sound, it's true. And there's a very good reason for it: these processors are mind-bogglingly complex, and they've been designed for what really amounts to one application (OK, two, if you consider smartphones and tablets to be distinct applications). Incorporating these super-complex, highly specialized chips into other types of applications is very tricky.
So, while a few mobile application processor vendors, such as Qualcomm, may be dipping their toes in the embedded processor market, it will likely be some time before they're ready to engage with large numbers of customers with diverse applications.
What about PC processor suppliers? x86 CPUs have long been co-opted for a variety of embedded applications, especially for high-end systems like medical imaging devices where there's plenty of room for what often amounts to a ruggedized high-end PC. But that approach is clearly not going to cut it for thermostats and $99 set-top boxes.
Lately, however, AMD and Intel have shown new interest in embedded applications, with Intel introducing the Quark processor family and the Galileo and Edison embedded development platforms targeting cost-, power- and size-constrained applications. The rich application software development infrastructure associated with x86 processors is a significant advantage for AMD and Intel, but it remains to be seen whether these vendors have the appetite for long-term engagement with the embedded market, which is much more fragmented than their traditional PC and server markets.
FPGA powerhouses Xilinx and Altera are also interesting players to watch here. Both companies have introduced SoC devices combining multi-core ARM CPUs, an array of peripherals, and programmable logic. The programmable logic can be used both to create specialized coprocessors and to create additional standards-based or proprietary I/O interfaces. Although FPGA vendors have offered soft and hard processors in the past, the new SoCs are their first chips that are processors first and FPGAs second. As such, they have the potential to enable these vendors to compete head to head with traditional embedded processor suppliers.
Vendors exiting and entering markets is nothing new. But for embedded systems designers, the current heightened flux among suppliers creates both risks and opportunities. Primary among the risks is selecting a processor that turns out to have no roadmap for future improvements, forcing a costly change to a new processor for your next design. Primary among the opportunities is the chance to gain an advantage by adopting a new type of processor at the right time--not too early, when the technology is too immature, but early enough to leapfrog competitors.