Embedded Computing with ARM: Build or Buy?

Embedded Computing with ARM: Build or Buy?

Buy or develop yourself? That is a much less frequent question with x86 than ARM because x86 is supported by a number of form factor standards. ARM Cortex A-9 Freescale i.MX 6 processors are now also available on the sub-credit card sized μQseven form factor. But do developers really need a form factor standard for ARM processors?

by Dan Demers, congatec


RTC12 TDTW congatec Header

The x86 PC has dominated embedded computing for decades. For a long time, the next processor generation was better because it was faster. But at some point, the increase in energy consumption meant that the latest desktop processors became unusable in the embedded market. As a result, the industry switched to using the mobile processors that were powering notebooks. We are all familiar with the subsequent development towards tablets and smartphones.

The processors needed for these new consumer applications differ from traditional x86 PCs in several ways. Firstly, and because of the application, they consume significantly less energy. Secondly and because of that, they do not provide as many generic expansion options. Instead, they offer a more focused feature set with, for example, camera interfaces that are supported directly by the processor.

Focused Feature Set

Today, both processor architectures offer such focusing. With the increasing growth of the Internet of Things (IoT) and the associated expectation to be able to control the TV via smartphone, OEMs want to equip their devices, machines and plants with similarly intelligent IoT connectivity so that they can be controlled from anywhere in the world if necessary. Unlike consumer applications, which tend to have extremely large production quantities, batch sizes here are often much smaller.

The Freescale i.MX 6 processor, which distinguishes itself with a 10 year product longevity, a particularly compact 3.5W low-power design and excellent multimedia and computing performance, is integrated for instance in compact industrial controllers or POS systems and supermarket scales. These applications have much smaller quantities than smartphones or tablets. The same goes for digital signage systems in vehicles or parking and other ticket vending machines that benefit from the extended temperature range of -40°C to + 85°C.

Smaller batches

The lesser benefit is drawn from the economies of scale, the more important is the use of solutions that reduce development costs. For this reason, there have always been – even and especially in the ARM segment – a variety of pre- designed components and evaluation modules enabling developers to thoroughly test the latest processor technologies and integrate them into their custom designs. However, a true form factor standard, of which there are several for the x86 architecture, has never been formed.

That didn’t really matter because until now the next ARM processor generation by one or more manufacturers did not usually provide a truly comparable generic feature set with ‘simply’ improved performance. Each new target application was just too heterogeneous, which meant the I/Os also were too heterogeneous. So the reasons why no form factor emerged for the ARM segment lay in the processor architecture itself, the manufacturers’ different licensing requirements and the often very heterogeneous applications. This has now completely changed, thanks to the advent of ARM technology in smartphones, tablets and other comparable embedded applications that have features similar to x86, such as HDMI, USB or PCIe. It is exactly this uniform set of interfaces plus the similarity of applications that are the key drivers behind the formation of new standards for the tablet and smartphone class of ARM and x86 processors.

Standards Provide Economies of Scale

The Qseven (70 x 70 mm) Computer-on-Module specification by the Standardization Group for Embedded Technologies (SGET) has picked up on this trend and ARM-based modules have been available for some time. One example is the Freescale i.MX 6 based conga-QMX6 that can support as many as three displays via 2x LVDS and 1x HDMI 1.4. This was made possible by splitting LVDS into two independent display channels, each with 24 bits (Figure 1).

RTC12 TDTW congatec Fig1

Figure 1: The Freescale i.MX 6 based congatec Qseven module conga-QMX6 can support up to three displays via HDMI and 2x 24-bit LVDS.

For some designs, however, the 70 x 70 mm footprint turned out to be too large. This led to the introduction of modules in the μQseven format which is also standardized as part of the SGET specification for Qseven. Measuring 40 x 70 mm, μQseven is smaller than a credit card and suitable even for small handheld devices such as industrial-grade smartphones. This leaves virtually no embedded design that cannot be implemented with one of these standard modules (Figure 2).

RTC12 TDTW congatec Fig2

Figure 2: congatec’s first sub-credit card sized µQseven module (40 x 70 mm) is equipped with ARM Cortex A-9 based Freescale i.MX 6 processors and a standardized pinout.

The new conga-UMX6 µQseven modules from congatec are equipped with ARM Cortex A9 based Freescale i.MX 6 SoCs, which come with a long-term availability of at least 10 years, and range from 1 GHz single- to dual-core performance and up to 1 Gbyte of robust soldered memory. With OpenGL ES 1.1/2.0/3.0 and OpenVG 1.1, the integrated high performance graphics support appealing 2D and 3D applications with up to WUXGA resolutions (1920 x 1200). Thanks to hardware-accelerated video processing, the modules decode 1080p videos at 60Hz in real time and encode up to two 720p videos. Two independent displays get connection via 2x LVDS or alternatively via 1x LVDS and 1x HDMI 1.4. For application and data storage, the modules have one LVDS and one HDMI 1.4. For application and data storage, the modules have one SATA interface and an optional 32GB of SSD.

To connect application-specific I/Os, the new congatec µQseven modules provide 1x PCI Express 2.0, 5x USB 2.0, 1x Gbit Ethernet and 1x CAN Bus to the carrier board. I2S bus support further ensures jitter-free, high-quality audio transmission. The integrated board management controller offers – amongst other things – watchdog timer and power loss control, as well as support of monitoring, management and maintenance features for distributed IoT installations. Board Support Packages (BSPs) are available for Android and all common Linux distributions as well as Windows Embedded Compact 7. All BSPs are fully released and available for download from congatec’s GIT server.

Given that the end result is an OEM-specific design, why is standardization so important? Why not use other predesigned, proprietary ODM modules and boards? As always, it is of course a question of cost and time savings as well as staying independent from individual manufacturers. The first μQseven module with Freescale i.MX 6 is sold not just by one manufacturer, but has been available from a second source right from the start. By now, there are almost a dozen providers of Qseven and μQseven modules with Freescale i.MX 6[i]. It is therefore safe to assume that pricing will always be competitive.

In addition, an ecosystem has evolved that ranges from carrier boards for custom designs and evaluation kits to design guides and carrier board design training. Then, there is also a developer community that can learn from each other and even re-use models for carrier board designs when there are no directly competing users. Such an approach is therefore significantly more open than the proprietary board or module solutions provided by individual manufacturers.

Another factor that shouldn’t be underestimated is that modules are available for many years allowing OEMs to replace a module with the same functionality if the existing processor design is discontinued. ETX modules are a good example to prove the benefits of long-term availability in the case of x86. They are still available with full ISA and PCI bus support, even though the PC standards turned their back on these buses more than 10 years ago. Customers can relatively safely assume that the ecosystem will include some providers who are prepared to service the long-tail of these technologies after the hype is over and who will support legacy applications for years, even decades to come. As a result, OEM customers are able to sell their original applications for several decades, thereby increasing their return on investment. And that’s a more common scenario for a range of industrial applications than you might think. Ultimately, this is an added value that no proprietary ARM design can offer. So there are many reasons to rely on embedded form factor standards for ARM such as Qseven and μQseven.

Standard Providers Offer Better Service

Once a user has decided in favor of standardization and the use of modules, the question remains which provider to buy from. Next to the offered module selection, the company’s market position, focus on board-level products and accompanying embedded design & manufacturing services are important factors to consider, so there will be no competition at the system level. In addition to these strategic decisions is also important that the day-to-day contact runs smoothly and efficiently during the Design-in phase. Companies that provide comprehensive documentation along with industry-standard driver implementations, while also offering personal integration support when this package is not sufficient for the customer, certainly have a distinct competitive advantage. They enable OEMs to integrate new processor technologies more quickly and efficiently into their own applications.

Qseven and μQseven specify modern serial I/O interfaces such as PCIe, Gigabit Ethernet, USB, SATA, SDIO and the CAN bus. In addition, they also support platform-specific I/Os, such as the LPC bus for x86. In line with the latest graphics interface trends, Qseven offers multi-display support for up to three independent screens via HDMI V1.4 and 24-bit LVDS dual channel, and up to 4k resolutions (3840 x 2160).

Qseven specifies a maximum power dissipation of 12W, even though current ARM platforms such as the Freescale i.MX6 processor never actually reach this limit. 12W is reaching the upper threshold for fanless designs. While the 12W limit restricts developers’ choices, it also adds greater security, because the specification of the thermal interface opens up the possibility of seamless interchangeability. With the appropriate standardization, modules as well as the cooling systems become swappable across different CPU technologies. This allows the development of new product lines in the field of heatsinks, which reduces costs. Anyone who has experienced or been responsible for a mechanical design change (e.g. heatsink) at a process-heavy company, is sure to appreciate this.

Future Evolution

In the midst of this, computer technology is still evolving in big steps. The processor structures (yes, surprisingly, Moore’s Law is still valid) continue to shrink, resulting in more computing power with less energy consumption. The interfaces are also changing. In recent years, parallel data buses have virtually disappeared to be replaced by fast differential serial interfaces. Qseven never provided any parallel interfaces, not even in the original version of the specification. As a consequence, the specification has so far needed few changes.

Minor modifications are currently underway to ensure that the Qseven standard remains future-oriented. For instance, the Qseven SGET team is working on replacing the LPC bus with ESPI. The LPC bus is ultimately a descendant of the ancient ISA bus, which will not be supported by x86 processors or chipsets for much longer. ESPI (often referred to as QSPI or QuadSPI) is the successor and – unlike LPC – it is also supported by a wide range of ARM-based processors. This will ensure even greater continuity between ARM and x86 architectures for Qseven in days to come.

The future Qseven specification also takes advantage of the strong trend towards ever lower power consumption. Fewer pins are required for the power supply, and the free pin space is already being reserved for future new interfaces. To round off Qseven 3.0, a further USB 3.0 port is added while the errata for USB OTG, the pinout and the position of the MIPI CSI camera interfaces, which are currently published as stand-alone documents, will be integrated into the main document.

Qseven 3.0 is therefore well prepared for future challenges. It provides four PCI Express x1 lanes, five USB 2.0 and three USB 3.0 (incl. 2.0) plus SPI, ESPI, I2C and GPIO/SDIO for generic and intrinsic extensions. Storage media can be connected via two SATA. DP/HDM, eDP/LVDS (2×24-bit) specify numerous video interfaces for several independent high-resolution displays, Ethernet for (IoT) connectivity, two MIPI-CSI for the connection of low-cost cameras and HDA/I2S for sound. So, why not try out an ARM-based Qseven module instead of relying on a proprietary solution?

congatec, San Diego, CA. (858) 457-2600. www.congatec.com