The multicore impact on embedded design principles
June 01, 2012
3D process technology and scalable Thermal Design Power (TDP) are breaking the processing and power constraints of traditional embedded systems design...
Embedded systems design has more constraints than general-purpose computer design because the requirements of embedded systems platforms are fundamentally different. For cost, reliability, thermal management, support, and life-cycle reasons, everything in the design of an embedded system must be kept uncomplicated. Because of these constraints, embedded systems rarely have contingency reserves, and performance upgrades are usually limited to processing module exchanges. By and large, embedded systems require processor and design simplicity; the extended performance of multicore processors is rarely an option. It is available, but is generally reserved for higher-end systems.
This is about to change. Emerging multicore processor platforms are rewriting the rules of embedded systems design by addressing and eliminating many of the traditional embedded computing design constraints. The following discussion describes how new process and power-conservation technologies are enabling more flexible and powerful multicore processors to be used in embedded platforms without affecting or violating the fundamental principles of embedded systems design.
Increasing performance and efficiency
Recent technological developments are providing numerous benefits to embedded computing. First and foremost, processor design is no longer focused solely on dramatic increases in clock speed. It is also considering increased efficiency, lower power consumption, and more powerful integrated graphics performance – all music to the ears of embedded systems designers and hardware vendors.
Semiconductor process technology is now down to 22 nm, which increases efficiency due to shorter distances traveled on a molecular level, but is also approaching the physical limits of what’s possible with planar transistor designs. As a result, the market is now seeing a switch to 3D process technology, which provides a much larger surface area for electrons to travel, addresses leakage issues, and allows for fast switching.
All of this contributes to lower power consumption at the same level of performance (or more performance at the same level of consumption). In essence, the move from 2D to 3D transistors will allow Moore’s Law, which states that the number of transistors that can be placed in an integrated circuit doubles every 18 months, to remain valid for years to come. With modern process technology, users will see up to 15 percent more CPU performance at the same clock speed from the latest generations of multicore processors.
Equally important to embedded systems designers is the increasing performance from integrated graphics engines. In the past, embedded systems designers only had two choices: add discrete external graphics subsystems at extra cost and complexity or make do with the modest performance of processor-integrated graphics.
This is changing with the latest generations of multicore processors that include more and faster execution units for vastly improved 3D performance and transcoding speed. These multicore processors also support the latest version of Microsoft DirectX, as well as OpenGL, OpenCL, and other graphics standards. Consequently, embedded systems can now process much higher data loads and provide quicker, richer, and more complex visuals on multiple independent displays. And despite changes in process technology, chip manufacturers are increasingly providing cross-compatibility with earlier generations on both the socket and pin levels, allowing chips and/or chipsets to be upgraded without further design costs.
Enabling technologies enhance designs
In addition to offering advanced process technology and integrated subsystems with increased performance, today’s multicore processors come equipped with several enabling technologies that open doors to improved embedded systems designs.
The first is the emerging availability of scalable Thermal Design Power (TDP), which measures the maximum amount of power that the design’s cooling system must dissipate. In the past, TDP was static, and any given design had to thermally handle maximum heat output. Intel recently introduced scalable TDP to mobile processors so that if additional cooling is available, TDP can be increased; if less is available, it can be throttled. This gives embedded systems designers considerable design flexibility as advanced features and performance reserves can be scaled to low-, medium-, and high-power packages.
The second major feature is the trend toward improved integrated graphics in modern multicore processors. This has historically been a weak point in chips with integrated graphics, and often made discrete graphics necessary for specialized embedded computing applications. The latest designs offer higher numbers of more powerful execution units, as well as numerous architectural performance improvements and graphics-specific cache.
On the chipset side, more multicore processors are incorporating native USB 3.0 and PCI Express 3.0 support. This means that the throughput bottlenecks that often plagued low-power systems will become a thing of the past.
Basic tenets of embedded design
To see how these advancements will affect embedded computing and the move toward more powerful multicore processor implementations, let’s examine traditional embedded systems design principles and how modern multicore platforms affect them.
Unlike the world of general-purpose computers where more performance is always better, embedded systems have typically been designed to perform narrowly defined tasks that do not change over the system’s lifetime. On the plus side, this allowed designers to precisely match component performance to process requirements, with minimal hardware that did not exceed thermal constraints. On the negative side, choosing the simplest processor that could get the job done left no reserves and often required the addition of external functionality. Emerging 22 nm multicore processors help meet thermal constraints, provide power reserves, and ensure an increasing degree of advanced functionality is right in the CPU and chipset.
Due to their focus on well-defined tasks, embedded systems have been inherently simpler than general-purpose computing systems. On the plus side, this allowed embedded systems designers to maximize simplicity and stay away from needless hardware and software complexity. On the negative side, with embedded systems requirements becoming more complex, those simple designs became complex as well when external subsystems had to be added. A prime example is the addition of discrete external graphics subsystems that provide the graphics performance basic integrated designs simply cannot generate. With the latest multicore processors, industrial-strength graphics are now part of the package.
Managing heat looms large in most embedded computing design projects. Thermal stress is one of the primary causes of system failure and is of special importance in embedded systems that must perform within extreme temperature ranges. Furthermore, there are significant reliability implications when deciding between active and passive thermal cooling systems. Here again, complexity is the enemy and anything mechanical (such as fan-based cooling) can and will fail. New 22 nm process technology allows far greater performance and functionality while remaining within most thermal design constraints.
Unlike general-purpose computing systems, many embedded systems are not easily accessible for inspection and maintenance. At the same time, most embedded systems have far more stringent uptime requirements than general-purpose systems; 24/7/365 uptime is usually a must. Once more, minimizing complexity by eliminating as many possible points of failure is a primary design goal. Here, modern multicore technologies such as configurable TDP and component integration redefine the rules by reducing external points of failure.
Extended life cycle
While rapid obsolescence is accepted in consumer computing products, embedded systems have much longer life cycles, usually 3-5 years and often more. That’s mostly due to the inherent longevity of underlying solution logistics; the way an ATM or gaming system works, for example, might not change for several years. System load, however, often does change due to added software functionality or features requiring CPU module upgrades. Emerging multicore modules can have much larger performance reserves, reducing upgrade frequency.
An example of an embedded board utilizing the latest multicore processor technology is Advantech’s SOM-5892 (Figure 1), a COM Express basic CPU module that can be configured with standard, low-voltage, and ultra-low-voltage mobile Core i3, i5, and i7 processors with integrated Intel HD graphics and the new Mobile Intel QM77 Express chipset. The SOM-5892 supports a variety of I/O, including seven PCI Express x1, a PCI Express x16, two 300 MBps and two 600 MBps SATA channels, eight USB 2.0, four USB 3.0, 8-bit GPIO, HD audio, and a watchdog timer, as well as up to 16 GB of DDR3 or DDR3L RAM in two SODIMM slots.
Upholding design principles
A new class of powerful, feature-rich, yet ultra-efficient multicore processors is bringing vastly greater performance and functionality without violating the aforementioned fundamental embedded systems design principles. The implications for embedded systems designers are enormous, as evidenced by embedded vendors and their customers, who now strongly trend toward multicore systems.