Transition boards become the future of I/O stacks

Last quarter, we discussed the role of transition carrier boards in saving stackable I/O from declining relevance. Stackable I/O board vendors themselves use Computer-On-Module (COM) form factors such as ETX to provide legacy PCI and ISA buses to the broad ecosystem of proven PC/104 I/O cards. Now there is a second wave of transition carriers poised to propel I/O into the future using legacy-free COM Express CPUs. This trend invigorates the debate over how to differentiate and optimize small form factors (SFFs) in the marketplace.

The recent advent of side-by-side expansion I/O sites on transition cards may well represent the largest breakthrough in SFF embedded computing in 20 years. A fresh outlook puts the I/O first, rather than the CPU first.

Old-school philosophy

To understand the way forward, let’s first examine the 20-year-old thinking present in the latest stackable bus architectures. First is the notion that all chipset interfaces need to be brought up the stack. If there are eight PCI Express lanes, eight USB ports, two SATA ports, and one PCI Express x16 interface, nearly all must be made available to stacking I/O expansion cards. Next is the idea that two of the four sides of the board should be filled up (consumed) with stacking connectors, even though space-consuming wide parallel buses are now being replaced by high-speed serial interconnects.

A large number of I/O ports then necessitates a tall stack. In the corner case of a hot CPU card, a hot PCIe x16 graphics card, and a hot power supply card in the stack, the board-to-board Z-axis spacing needs to remain the same or even grow in order to get all the heat out of the stack. By starting with a CPU mindset, the size, weight, power, and cost (SWaP-C) of the solution starts going up, up, and away.

The new world order

At last, a breakthrough comes by considering the I/O first, instead of the CPU first. Last quarter’s column showed how choosing to preserve existing PC/104 I/O cards leads to an elegant solution with a transition board that mechanically adapts the ETX module’s PCI and ISA buses to the stackable formats.

What about brand new applications that don’t use legacy I/O? The same way of thinking applies, starting with the I/O itself. An abundance of tiny PCI Express Mini Cards (mini-PCIe cards) exist with laptop PC-type I/O onboard, and lately this standard is recruiting embedded I/O like A/D in the same form factor. Working backward toward the CPU, the next step is to create a baseboard or carrier board with multiple mini-PCIe sites (sockets) spaced horizontally, side-by-side instead of stacked.

So that brings us to the final puzzle piece: Low-speed I/O. New off-the-shelf I/O cards, as well as a framework for OEMs to develop their own custom I/O cards, are clearly needed. The question remains: How does one hook slow I/O into a modern high-speed system? Forcing these onto PCI Express and USB leads to device-driver concerns and bridging complexity, which hampers determinism and increases power consumption. Instead, slow I/O is very easy to interface to buses like I2C, SPI, and even classic UART serial ports, all of which suddenly find favor in this Internet of Things (IoT) era. Creating device drivers is much easier as well.

The SBC35-CC405-3815 board from WinSystems (see Figure 1) embodies this way of thinking. Leveraging high-volume, cost-effective COM Express CPU modules plugged in below with a convenient heat spreader plate at the bottom, this transition board has plenty of room to route the high-speed I/O to side-by-side I/O sites rather than forcing a tall stack. This board employs mini-PCIe and m-SATA sockets and creates a simple low-speed expansion interface, called IO60, for a low-speed mezzanine I/O card. The IO60 interface is intended for right-size bandwidth implementations for a broad set of typical embedded I/O.

21
Figure 1: WinSystems’ SBC35-CC405-3815 greatly reduces stack height.
(Click graphic to zoom)

The design goal and mindset have everything to do with optimizing and differentiating SFF embedded systems. Very different implementations derive from whether the CPU or the I/O is prioritized first. Transition boards leverage mainstream, cost-effective COM Express CPUs and spoke out a fair amount of the I/O directly to connectors to keep the system height (stack size) reasonable.

Small Form Factor Special Interest Group 408-480-7900 info@sff-sig.org