COM Express mini, Qseven, SMARC meet at the crossroads of an evolving processor landscape
COM Express Type 10, Qseven, and SMARC are all vying for position as ARM-based processors take hold in SFFs.
While ARM Systems-on-Chip (SoCs) are bringing a legacy of mobility to “mini” Computer-On-Module (COM) standards, power reductions in x86 technology have made Intel and AMD-based options increasingly suitable for the shrinking PCBs needed in ultra-portable applications. This PC/104 and Small Form Factors interview examines how trends in processor technology are affecting the bantamweights of the COM class with a look at the COM Express Type 10, Qseven, and Smart Mobility ARChitecture (SMARC) specifications. Edited excerpts follow.
How have advances in silicon technology affected the “mini” Computer-On-Module (COM) class?
LONDON, Kontron: When COMs first appeared in the market, the performance attributes of a module were restricted to the thermal power dissipation. The [original COM Express] specification allows for a high level of dissipation of 100 W+, but to support this level the module would have to have a significant and unreasonably sized thermal heat sink to properly cool. Subsequently, designers tend to focus on limiting their designs to a more modest 45 W – 50 W Thermal Design Power (TDP). For this reason, most embedded module suppliers opt to use the mobile-processor SKUs that tend to have significantly more available thermal options compared with their desktop brethren.
Now that silicon lithography geometries have advanced to allow more gates, hence processing power, but at lower TDP, the overall performance capability of a given module has improved significantly without materially adding to the TDP overhead. The net result is that designers are getting more performance without having to accommodate a higher TDP in their application.
DEMERS, congatec: With the introduction of the Intel Atom processors, especially the second generation and now the third, you’re seeing modules that are in the $200 range that are really fast and have a lot of memory. It’s a different world. This is what’s also helping modules to make sense in the higher volume designs, versus people designing their own board because we’re only talking a couple hundred bucks; we’re not talking $500 – $1,000. So that helps. That whole, “I can’t get an Intel platform for less than $500,” well, now you can, and it’s not a dog.
If you go back to 2008 when Qseven was first designed and the specification first gelled, everybody agreed on a 12 W ceiling for total power draw. So that kind of weeds out applications and draws a line in the sand for designers. Most applications that we’re seeing people are trying to drive down [power consumption] as far as they can; people love the 5-W world, especially on the x86 side of things. On the ARM side of things, everybody would laugh at me for saying that because they’re like, “no, we want 2 W, 3 W.” But in the x86 world for all the guys that have done Intel and AMD for decades, if you can get 5 W, 6 W, 7 W, that’s a whole new solution for them. You’re talking fanless, mating the silicon directly to a cooling device that’s part of the chassis. Doing designs those types of ways is something that the ARM guys have always done but the x86 guys now get to do.
When you get into [COM Express] Type 10 or Qseven, there’s still a range and there’s still some scalability there (Figure 1). You could have a single-core Atom processor up to a quad-core Atom processor – but it’s not always that the customer wants that quad-core because a lot of times in these smaller systems they might actually mean smaller resale prices at the Original Equipment Manufacturer (OEM) level, so they’re a little more cost-conscious and they are more power-conscious. So we start seeing the mid-level SKUs being pretty popular: Things like dual-core versus quad-core, things like 2 GB of memory versus 4 GB of memory.
How has the rise of ARM-based processors impacted a space that has been traditionally x86-centric?
LONDON, Kontron: The impact of ARM SoCs in this class of COMs for embedded platforms is not cannibalizing the other COMs so much as it is augmenting the options that designers can draw from. As SMARC continues to be adopted by other embedded manufacturers, more and more ARM-based solutions are emerging. The intent of SMARC was not to displace x86 solutions. In fact, Kontron recently announced an x86-based SMARC module to not only meet market needs but also to highlight the versatility of the SMARC specification (Figure 2).
The developers are reaping the benefit of a broad and growing product offering from an expanding ecosystem of embedded COMs manufacturers who are committed to providing standards-based platforms and system building block solutions.
HELENIUS, Hectronic: At the same time that ARM is moving up in performance, x86 is going in the other direction. When people understand that we are in this spot I think that most companies will begin focusing more on understanding the two ecosystems and the effort that is needed to make an ARM system versus an x86-based system.
The majority of our customers are either using Qseven [exclusively] for x86 or Qseven [exclusively] for ARM; however, there are customers that are moving in both directions. There is a large and growing number of requests for high-end ARM processors on Qseven or SMARC, and a lot of first-generation ARM-based Qseven modules have been designed according to Qseven Revision 1.2, though I’m expecting that most new modules should be designed according to Revision 2.0.
Hectronic has designed our Revision 2.0 modules so that they in most cases will work without any problem on a Revision 1.2 carrier (Figure 3). However, there are always compatibility challenges between different revisions and different manufacturers, and supporting ARM and x86 will add additional challenges.
Given the small size and wide range of processor support, what are the pinout and connector implications in the “mini” COM module space?
LONDON, Kontron: The SMARC pinout is optimized to support specific I/O functionality that is more commonly identified with ARM technology. ARM has typically been employed to support more of the ”non-PC-like” interfaces such as MIPI, CSI, and parallel TFT display buses found in many of today’s mobile and handheld devices. The MXM3 card edge connector used on SMARC modules has 314 pins available for I/O and power connections, compared with other form factors in its class that are limited to 230 pins (Table 1).
The extra pin count on SMARC allows for greater functionality, as well as future expandability functions as the specification continues to evolve and expand. This additional pin count gives designers freedom to add more functionality into their application in less total surface when comparing SMARC with Qseven. Similarly, the MXM3 connector has more pins than the 220 on the COM Express Type 1 and Type 10 A-B connector, but has fewer compared to the COM Express Type 2 and Type 6 pinout configurations.
DEMERS, congatec: If you look at the A-B connector on COM Express, which is the only connector that’s on Type 10, and you look at the edge connector on Qseven or SMARC, the signals are very, very similar. Now, obviously it’s a different pinout – one’s a board-to-board connector, one’s an edge connector, so they’re not all going to be the same traces.
One of the nice things about Qseven and COM Express is there are multiple heights of these connectors, so you have an ability to scale in overall height (Figure 4). One of the ideas behind COM Express and Qseven from day one was, “let’s give the user multiple heights on standoffs and connectors so they can judge whether or not they want to have components on the top side of their carrier board that would reside right under the module. So it gives some more flexibility to the designer. There’s even a Qseven height connector that, if you were to use it, you would have to have a cutout in your carrier board because it’s so short that the module components would actually hit your PCB even if you didn’t have components on your carrier board. That’s how low it is. So at the end of the day, if you look at that sandwich, that’s almost a monolithic board at that point.
How is a two-board COM approach able to meet the cost-conscious designs of the mini/mobile space?
LONDON, Kontron: While it is true that designers can allocate cost targets to certain functions of the system, one has to consider the entire functional solution as a whole. In this case, the combination of a module and carrier board serves as a functional unit versus discrete functions. Comparing a two-board solution to a single-board solution is a valid comparison, as there may indeed be a premium for the two boards versus a single board. However, one has to offset that premium with the advantages gained by having a flexible, reusable carrier board that will scale with the application rather than requiring a dedicated PCB for each product SKU within a product line.
DEMERS, congatec: I kind of equate it to the idea of a car manufacturer like Audi. They have an A3, an A4, an A5, and an A6. So there are four product families and as consumers we all know that one is better than the other because you’re moving up the food chain. But if you look at those products, they put the same engine in all four of them. They’re reusing technology, which drives cost down. So, if you’re an OEM designing multiple products with the same carrier board, that is cheaper than designing four products with a different motherboard in each one.
That module and carrier design takes, in many cases, six to nine months less time to design than a single monolithic board. So that’s significant design time – design time is money. Getting to market quicker is money. You’re saving a lot getting to market quicker and reducing the amount of effort to get there. So there are certainly cost savings involved, which, depending on your system, can be in the six figures, easy.