Shane Rau, IDC
The embedded microprocessor market has attracted significant investment over the past couple of years. Despite the economic recession, demand for more programmability, higher performance, and lower power consumption in electronic systems outside of traditional computing areas like PCs and servers has meant that the embedded processor market hasn’t decelerated as much as the overall semiconductor market.
In 2009, embedded processor market revenues declined approximately –5%, compared to the overall semiconductor market revenue decline of –12%. In 2010, particularly in the second half, several trends will combine with general economic health to accelerate demand for embedded processors significantly.
DEFINING THE MICROPROCESSOR
An embedded microprocessor can be defined as a host or central logic processing semiconductor that furnishes intelligence to an electronic system serving applications outside of general-purpose computing. This includes processors for classic, deeply embedded machines such as test and measurement devices and medical devices, but also mainstream systems such as cell phones, point-of-sale devices, and set-top boxes.
What the vast majority of these processors have in common are designs tailoring them to a specific application. They tend to have a central processor core or cores surrounded by application-specific features, such as the Ethernet interfaces and security accelerators found in embedded processors for networking applications.
If customized, these processors are ASICs. If off-the-shelf, these processors are ASSPs. ASICs and ASSPs in embedded systems are typically complete, single-chip solutions, meaning systems-on-a-chip (SoCs). The few system types in the embedded systems space that use standalone processors, such as an Intel Core 2 Duo, tend to be the most PC-like of embedded systems, such as thin-clients. IDC estimates that nearly 79% of embedded processors shipped in 2009 were ASSPs, 17% were ASICs, and 4% were standalone processors.
DRIVEN BY THE WEB
A single overarching trend creating demand for embedded processors is the Internet. Today, nearly 1.4 billion users access the Internet, and by 2012 more than 2 billion will be online. What makes the Internet compelling is not only the increasing number of users accessing it but also the number and diversity of devices that each user has to access it and the content that they’re creating, storing, and moving with those devices. Notably, more than half of these users will access the Internet with mobile devices. Increasing percentages of those users will be from emerging markets like China, India, Russia, and Latin America.
Another overarching trend enabling this demand is increasing silicon capabilities, ably represented by Moore’s Law. Advances in process technology mean that a typical 45-nm processor with 820 million transistors today will be a 22-nm processor with 1.5 billion transistors in 2012.
These developments will enable multiple virtues, such as more application-specific features, more cores for more performance, lower power consumption, and lower cost. As electronic devices and applications grow in complexity and new usage models emerge that require more access to media and Internet connectivity, Moore’s Law will continue to feed them.
The demand that the Internet creates and the supply that Moore’s Law enables are converging into multiple design, manufacturing, technological, and business model trends that affect the entire embedded systems supply chain. Supporting an increasingly large and diverse base of users will require a tremendous amount of investment at every link in the supply chain.
More mobile devices means that system OEMs and semiconductor suppliers will bear more design and manufacturing costs to implement technology that can, at the same time, offer better performance, optimize power for mobility, integrate more features, and reduce cost. In turn, the effort to offer better performance is leading to the use of multiple processor cores per die.
Over the past five years, multicore has replaced frequency as the primary way to enable higher performance and to support multiple applications to run in parallel. Over the next decade, multicore design will change the performance measures and complexity of computing that we use today. Already, mobile phones use multiple heterogeneous cores, game consoles are using multiple embedded microprocessors and data engines, and next-generation Internet Protocol set-top boxes enabling video services will use more than two cores in each device.
THE VALUE OF IP
Simultaneous with the design and manufacturing pressure is the pressure to acquire the intellectual property (IP) necessary to implement in the widening variety of embedded processors vendors must support. OEMs and semiconductor suppliers continually must determine whether to create or buy their IP.
Not only will embedded processor vendors like Broadcom and Nvidia continue to turn to processor IP providers like MIPS and ARM, they may also look to their dedicated foundry partners, such as TSMC and UMC, who may additionally play the role of aggregators of IP. Given the wafer scale of foundries, a foundry could subsidize or give away IP to stimulate an increase in wafer production from foundry customers.
And the IP isn’t only in the area of features to be rendered in silicon. Embedded processor vendors, responsible for complete solutions that enable fast time-to-market for customers, are strongly trending towards the platform approach. This approach demands the arrangement of standards-based IP according to the needs of a given usage model.
As a result, the platform approach emphasizes IP acquisition and provision of complete, system-level solutions. This means responsibility not only for hardware and software, but also for user interfaces, reference designs, system integration services, and back-end technical support for customers.
A parallel trend implied by the SoC and IP acquisition trends is increasing competition and consolidation among competitors in the embedded processor space. We expect traditional embedded architectures, such as ARM, MIPS, and PowerPC and their customers, such as Cavium, Freescale, and Applied Micro, to challenge each other in their traditional camps.
For example, ARM maintains a clear leadership position in mobile devices. But as it focuses more on device segments beyond mobile, it will compete more with MIPS, ARC Cores, and Tensilica, requiring it to invest more in software and platforms to differentiate. Furthermore, Intel is a threat over the next couple of years with its x86-based Atom microprocessor, which is more reason for ARM, MIPS, and other companies to invest in broadening their customer bases, ecosystems, brands, technologies, and low-power leadership.
WHAT’S NEXT?
Embedded microprocessors have evolved from limited-function products to become the centerpiece of every major device platform. The direction of the market and opportunity is shaping up to be extremely favorable for embedded processor companies. In 2009, more than 3.9 billion embedded processor cores were shipped, which is approximately 3% higher than in 2008, despite the recession.
Over the next five years, the volume will double as more devices connect to the Internet and demand more performance to enable multiple applications to run on each. To take advantage of the opportunity, though, all embedded processor companies must have proven multicore architectures with performance scalability and low power. They also must invest more in their ecosystem, IP, design, and platforms to support the challenges and opportunities of their customers and partners in the embedded space.