The semiconductor industry has been on a head-spinning merger binge lately. NXP is acquiring Freescale. Avago is acquiring Broadcom. Intel is acquiring Altera.
Much has been written about the motivations for these mergers, and about the implications for investors in the merged firms. But so far, little has been said publicly about the consequences for customers. For designers of embedded systems, the most obvious result of these mergers is fewer suppliers to choose from. In itself, fewer suppliers is not necessarily a bad thing, especially if it means that the remaining suppliers are healthy. But for most system designers, this industry consolidation also means fewer processors to choose from.
Before the current wave of acquisitions, as I wrote last year, several of the largest embedded processor suppliers were already pruning their product portfolios and narrowing the number of markets they target. I believe that industry consolidation will accelerate this trend, for several reasons.
First, larger companies necessarily tend to focus on larger markets. A behemoth like Intel, for example, with its roughly $55 billion in annual revenue, can't afford to chase after $1 billion markets. Second, executives of the merged companies will be under pressure from investors to realize cost savings – which will probably mean developing fewer products and servicing fewer markets.
The increasing complexity of typical embedded processors also plays a role here. Twenty years ago, a typical embedded system incorporated a general-purpose processor working together with numerous additional chips: DSPs, display controllers, graphics processors, etc. In contrast, these days most of the functionality of a typical embedded system is delivered by a single embedded processor system-on-chip. For example, take a look at the Nest thermostat guts. Besides the Texas Instruments processor SoC, there's not much there: a power management chip, DRAM, flash memory, and some wireless networking chips. In addition to the Cortex-A8 CPU, the TI SoC includes a graphics accelerator, display processor, camera interface, memory controller, and dozens of peripheral interfaces. (And that's a five-year-old chip.)
As processor chips become more complex, they tend to become more application-specific, incorporating features like encryption engines, video compression hardware, and packet processing accelerators. In addition, the application software development infrastructure supplied with them also tends to become more complex and application-specific. This includes things like operating systems, development boards, device drivers, and optimized software component libraries.
Increasing complexity and application-specificity requires chip suppliers to make larger "bets" on new products, which means that fewer new chips will get developed. With fewer suppliers developing fewer products for fewer markets, most embedded system designers will be faced with fewer processor options. As a result, it will become increasingly important for system developers to adapt processor chips for applications other than those they were designed for. Clever software implementing "out of the box" approaches will be necessary in many cases.
Due to this growing need to bend chips for varied applications, chip and tool features that enable flexibility will become very valuable. On the hardware side, this can include things like graphics processing units that can be harnessed as parallel processing engines, or on-chip FPGA blocks that can be used to create custom co-processors. On the software side, robust tools, industry-standard APIs like OpenCL, and open source frameworks and example applications will become even more critical than they are today.
Among the surviving embedded processor suppliers, those that enable flexibility for their customers will gain a significant advantage in the coming years. And among embedded system designers, the skills needed to "fit a square peg into a round hole" will be key.