Ten years ago, I wrote about how licensable processor cores were beginning to play a more important role in the industry. Among other trends, I observed that large chip companies were beginning to adopt licensable cores for application-specific chips such as cellphone baseband SoCs, rather than using proprietary cores that they developed in-house. This trend has certainly strengthened over the past ten years.
It’s one thing for a company making application-specific SoCs—which often incorporate numerous processor cores—to choose a licensable core rather than an in-house design. After all, in such chips a processor core is typically one of many utility functions, rather than the central, defining element of the chip. It’s quite another thing for a company making processor chips to adopt a licensable core rather than an in-house design; processor companies have traditionally differentiated their products in large part based on the superiority of their processor instruction set architectures and microarchitectures.
It’s therefore been striking to see a number of processor companies switching from in-house to licensable cores in the last few years. Recently, this has been particularly noticeable in the microcontroller market, where Microchip, Texas Instruments, Freescale, NXP, and STMicro have adopted licensable cores for their flagship microcontroller product lines.
Is this is a smart move? Or are chip companies simply looking for short-cuts to save money, and in the process giving up the ability to differentiate their products? Certainly some of the motivation for adopting a licensable core does come from chip companies’ relentless drive to cut costs. It costs real money to develop a processor core—typically tens of millions of dollars. And this isn’t a one-time expense; to remain competitive; a chip company must upgrade its processor core architecture frequently, requiring a fresh round of investment each time.
But I believe that most chip companies are thinking about the bigger picture when they make the decision to adopt a shared, licensable core rather than an in-house design. As processors are becoming more powerful—which of course is happening with $1 microcontrollers as well as $500 multi-core DSPs—the applications that run on them are becoming larger and more complex. Yet, for the most part, equipment manufacturers are not employing more engineers than they did in the past. So, to field these larger, more complex applications, system designers must work at a higher level of abstraction. In many cases, this means assembling application software largely from off-the-shelf components, rather than building it from the ground up.
A natural consequence of this trend is that the availability of off-the-shelf software components—operating system ports, device drivers, optimized multimedia codecs, etc.—is becoming an increasingly critical factor in processor selection for many system designers. But it’s exceedingly difficult for a single chip vendor to field all of the necessary software, even with the help of third-party developers. By using a shared processor architecture—especially one that is widely used—chip vendors effectively pool their resources. For example, ARM recently released a DSP software function library for the Cortex-M4 digital signal controller core. This library, of course, can be used by customers of any chip company building Cortex-M4-based cores. In the past, each of those chip companies would have had to develop its own DSP function library, though they would have gained little competitive differentiation by doing so.
Shifting engineering investments away from reinventing-the-wheel tasks like developing a competitive-but-unexceptional processor core and basic software development infrastructure, chip vendors can invest in more innovative—and arguably more valuable—areas. For example, NXP’s recent introduction of dual-core microcontrollers, where one core focuses on real-time I/O, freeing the other core for more computationally-demanding tasks, illustrates the potential for chip-level innovation, even in rather low-cost chips.
The trend towards shared architectures is by no means limited to low-cost chips, however. At the recent Consumer Electronics Show, there was much talk of high-performance, low-power processor chips based on licensable cores targeting applications such as tablet computers, set-top boxes, and servers. Companies like Qualcomm and Texas Instruments have already shown that they can field compelling processors based on licensed architectures for applications like smart phones and media players—and sell tens of millions of these chips. Clearly, other processor-centric chip companies like Freescale and NVIDIA now intend to do the same in various high-end processor markets, from automotive entertainment systems to tablet PCs to network servers.
By bypassing the traditional path of developing (and maintaining and refining) their own processor cores, these companies can innovate faster in other aspects of their products. It’s an increasingly sensible strategy for many chip suppliers.