Fast Serial Interconnects— Will They Bypass Embedded or Bring it Along?


  • Page 1 of 1
    Bookmark and Share

It has long been a truism that a great deal of embedded technology has derived from things originally developed for the PC. A big factor in this has, of course, been cost. Once an interface technology, a connector, or an interconnect is adapted into the PC world with its huge volumes, costs plummet making it attractive for other uses and, where technically appropriate, these naturally include embedded. This has certainly been the case with a number of serial interconnect technologies such as USB and PCI Express.

Both of these have been enthusiastically embraced by the embedded computing industry and are today almost universal due to their costs, low power requirements, small size, speed and utility for moving data between modules and peripherals. Both are backed by industry consortia—the PCI SIG and the USB Implementers Forum—which support the technologies for large numbers of members across PC, enterprise, embedded and more in both groups. For these members the world is not standing still, and both of these along with other serial interconnects are continuing to strive for higher speeds. Ethernet is getting to 100 Gbit/s as well.

USB 3.0 now hits 5 Gbit/s and PCIe 3.0 is hitting a nominal rate of 8 Gbit/s and that is on one lane (x1). For high-speed graphics and other demanding applications, it can go to x16. Now Intel is actively promoting its Thunderbolt interconnect, which clocks in at 10 Gbit/s, and the PCI SIG has let it be known that it is working on a PCIe 4.0 spec that is targeted to get to 16 Gbit/s. Technically, PCI SIG talks in transfers (GT/s) because there are coding and other issues that can add some overhead to the actual transmission of data.

But you get the point. PCI SIG President Al Yanes notes that the higher speeds are addressing an ever narrower area on the pyramid of application performance requirements. The question is, does that observation apply to the short term only, or to the long term? Application requirements have a way of expanding to utilize all the available performance—it’s just that we do not yet know what those requirements will be. The tip of that pyramid may not always be so narrow.

Our question, of course, is how these developments apply to embedded systems. My own take on that is that the ever-increasing speeds will not mean much to what we consider “typical” embedded systems. But that does not necessarily apply to non-typical embedded systems, those we may not have thought of yet, or have failed to recognize as falling into our esoteric realm.

The faster speeds will, of course, provide better assurance of performance approaching real-time if not strictly deterministic levels. That would, however, apply to the sort of commands and data used by traditional control systems and these—at least those we are currently familiar with—are rarely overwhelming in their bandwidth demands. The big consumer of bandwidth is well known to be high resolution video and graphics. Video and graphics are currently not that big a part of “traditional” embedded systems—at least not in ABS braking systems or machine tool controllers. Nonetheless, developers are starting to become aware of some of the possibilities.

As embedded systems with multiple nodes connected via high-speed networks deliver more data, even in amounts that can be handled by current interconnects, there is increasing motivation to present that data in graphical form that can be dealt with interactively by the user. That will run up the bandwidth requirements big time if it happens. At this point, the higher speed interconnects will service consumer requirements for PCs and tablets. But we are seeing such devices as the iPhone and the iPad starting to serve as user interfaces for applications in embedded and medical arenas. The embedded systems that deliver data to such interfaces will—I predict—eventually be required to serve it up over the higher speed serial interconnects and also over wireless networks, which will have to keep up with those bandwidth requirements.

So maybe those roaringly fast interconnects will bypass our “traditional” embedded systems for a while, but just wait. A new generation of users weaned on the interactive video/graphical devices in their daily lives will soon begin to demand that they be able to interact with their work environments and with industrial systems the same way. And that use of graphical and video data will not only be implemented to satisfy those demands, it will also make the industrial systems they support more powerful as well.